Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Predicting Behavior With Implicit Measures: Disillusioning Findings, Reasonable Explanations, and Sophisticated Solutions

Predicting Behavior With Implicit Measures: Disillusioning Findings, Reasonable Explanations, and... REVIEW published: 08 November 2019 doi: 10.3389/fpsyg.2019.02483 Predicting Behavior With Implicit Measures: Disillusioning Findings, Reasonable Explanations, and Sophisticated Solutions 1 1 1 2 Franziska Meissner *, Laura Anne Grigutsch , Nicolas Koranyi , Florian Müller and Klaus Rothermund 1 2 General Psychology II, Institute of Psychology, Friedrich Schiller University Jena, Jena, Germany, Department for the Psychology of Human Movement and Sport, Institute for Sports Science, Friedrich Schiller University Jena, Jena, Germany Two decades ago, the introduction of the Implicit Association Test (IAT) sparked enthusiastic reactions. With implicit measures like the IAT, researchers hoped to finally be able to bridge the gap between self-reported attitudes on one hand and behavior on the other. Twenty years of research and several meta-analyses later, however, we have to conclude that neither the IAT nor its derivatives have fulfilled these expectations. Their predictive value Edited by: for behavioral criteria is weak and their incremental validity over and above self-report Zheng Jin, measures is negligible. In our review, we present an overview of explanations for these Zhengzhou Normal University, China unsatisfactory findings and delineate promising ways forward. Over the years, several Reviewed by: reasons for the IAT’s weak predictive validity have been proposed. They point to four Xiaoming Wang, potentially problematic features: First, the IAT is by no means a pure measure of individual Qufu Normal University, China differences in associations but suffers from extraneous influences like recoding. Hence, Colin Smith, the predictive validity of IAT-scores should not be confused with the predictive validity of University of Florida, associations. Second, with the IAT, we usually aim to measure evaluation (“liking”) instead United States of motivation (“wanting”). Yet, behavior might be determined much more often by the latter *Correspondence: Franziska Meissner than the former. Third, the IAT focuses on measuring associations instead of propositional franziska.meissner@uni-jena.de beliefs and thus taps into a construct that might be too unspecific to account for behavior. Finally, studies on predictive validity are often characterized by a mismatch between Specialty section: This article was submitted to predictor and criterion (e.g., while behavior is highly context-specific, the IAT usually takes Cognitive Science, into account neither the situation nor the domain). Recent research, however, also revealed a section of the journal Frontiers in Psychology advances addressing each of these problems, namely (1) procedural and analytical Received: 05 June 2019 advances to control for recoding in the IAT, (2) measurement procedures to assess implicit Accepted: 21 October 2019 wanting, (3) measurement procedures to assess implicit beliefs, and (4) approaches to Published: 08 November 2019 increase the fit between implicit measures and behavioral criteria (e.g., by incorporating Citation: contextual information). Implicit measures like the IAT hold an enormous potential. In order Meissner F, Grigutsch LA, Koranyi N, Müller F and Rothermund K (2019) to allow them to fulfill this potential, however, we have to refine our understanding of these Predicting Behavior With Implicit measures, and we should incorporate recent conceptual and methodological advancements. Measures: Disillusioning Findings, Reasonable Explanations, and This review provides specific recommendations on how to do so. Sophisticated Solutions. Front. Psychol. 10:2483. Keywords: implicit measures, predictive validity, IAT, attitude-behavior gap, multinomial processing tree models, doi: 10.3389/fpsyg.2019.02483 wanting vs. liking, propositions vs. associations, context-dependency Frontiers in Psychology | www.frontiersin.org 1 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures Why does he  act like this? Why does she not do what she In this regard, dual process or dual system models posit that intended to do? In our everyday life, we  oen ft try to find parts of human behavior can only be  explained with processes explanations for the behavior of others, and of ourselves, that operate below the threshold of personal control and respectively. Explaining and predicting behavior is also of key awareness (e.g., Strack and Deutsch, 2004; Hofmann et  al., interest across all fields of scientific psychology, especially when 2009; Kahneman, 2011), a view that fueled the interest in the it comes to deviations between individuals’ actual behavior “sub”-personal level of behavior control. and the attitudes, goals, or values held by these very individuals. Over the last decades, a number of new attitude measurement Why do people discriminate although they report to hold procedures were introduced that aimed to tap into these processes egalitarian values? Why do they not quit smoking although since they do not rely on introspection (e.g., the IAT, Greenwald they know that smoking is bad? Why is there a gap between et  al., 1998; the Ae ff ctive Priming Paradigm, Fazio et  al., 1986; people’s self-reported attitudes and actual behavior? the Aeff ct Misattribution Procedure, Payne et  al., 2005; for Dual-process or dual-system models attribute seemingly overviews, see Teige-Mocigemba et  al., 2010; Wentura and inconsistent behavior to the triumph of an impulsive system Degner, 2010; Gawronski and De Houwer, 2014; Gawronski over a reflective system of behavior control (e.g., Strack and and Hahn, 2019). Although differing in their procedural details, Deutsch, 2004; Hofmann et  al., 2009; Kahneman, 2011). The all of these measurement procedures involve computerized tasks notion that the prediction of behavior could be  improved requiring individuals to quickly execute a specific response to considerably if one succeeds in measuring the processes of the a set of stimuli. The performance in these tasks is then influenced impulsive system (Hofmann et  al., 2007; Friese et  al., 2008; by stimulus-response compatibility due to the automatic Hofmann and Friese, 2008) fueled research applying so-called evaluations of these stimuli (De Houwer, 2001, 2003a). Hence, implicit measures of attitudes. The most popular of these measures, the scores obtained from the observed performance are the Implicit Association Test (IAT, Greenwald et al., 1998) evoked interpreted in terms of attitude strength. Compared to self- enthusiastic hopes regarding its predictive value. Unfortunately, report measures, these measurement procedures were assumed however, the IAT and its derivatives have not met these expectations. to provide little opportunity to control responding, preventing In this article, we  review findings illustrating reasons for the an influence of deliberate manipulation attempts and self- IAT’s unsatisfying predictive value, as well as promising ways presentational concerns (e.g., Fazio et  al., 1986; Greenwald forward. We  will outline that in order to improve the predictive et  al., 1998). Some even argued that these procedures succeed power of implicit measures, differentiation is key. We  will argue in measuring a unique construct (implicit attitude) that is that future research should put more emphasis on the underlying introspectively less accessible and thus distinct from the construct processes and concepts behind these measures. We  begin with captured in self-report measures (explicit attitude; Greenwald sketching the discrepancy between individuals’ behaviors and and Banaji, 1995; Wilson et  al., 2000; but see Fazio, 2007, for their self-expressed attitudes. We  then summarize the (mostly a different view). Accordingly, researchers oen u ft se implicit unsatisfying) attempts to close this attitude-behavior gap with measures and explicit measures as labels for these measurement the help of implicit measures. In the main part of this article, procedures. Not surprisingly, implicit measures, first and foremost we  identify features of implicit measures that are responsible the IAT (Greenwald et al., 1998), were embraced by the scientific for their weak predictive validity. We  review findings illustrating community since they came along with the potential to measure each of these problematic aspects along with specific, sophisticated the hidden forces of behavior. The hope was that they would solutions providing promising directions for future research. finally enable researchers to understand and to predict individual behavior over and above self-report measures. Unfortunately, the predictive validity of the IAT fell short THE ATTITUDE-BEHAVIOR GAP AND of these expectations. Meta-analytic findings ( Greenwald et  al., 2009; Oswald et  al., 2013; Kurdi et  al., 2019) suggest that the IMPLICIT MEASURES implicit-criterion correlation (ICC) is unsatisfactorily low (average Attitudes and values that people express are oen in co ft nflict r   =  0.27, Greenwald et  al., 2009; average r   =  0.14, Oswald ICC ICC with their actual behavior. Indeed, although widely postulated et  al., 2013; 90-percent prediction interval for ICCs from to be  associated with cognitive processes, judgments, and most r  =  −0.14 to r  =  0.32; Kurdi et  al., 2019). Equally upsetting importantly, behavior (e.g., Katz, 1960; Fazio et al., 1983; Ajzen, is the fact that the incremental predictive validity over and 1991), self-reported attitudes show weak predictive validity at above self-report measures is obviously negligible (i.e., ranging best (with correlation coefficients being “rarely” above r  =  0.30, between 1 and 5%; Greenwald et  al., 2009; Oswald et  al., 2013; Wicker, 1969; see also Kraus, 1995, who found an average Kurdi et al., 2019). Such a disappointingly low predictive validity r  =  0.38). How can we  close this attitude-behavior gap? A is a frustrating state of affairs, especially because it was the prominent way forward relied on the assumption that people low predictive value of self-reported attitudes that initiated the might not be  able to report on their mental processes in an development of implicit measures like the IAT in the first place. accurate fashion (e.g., Nisbett and Wilson, 1977), implying What are the reasons for the weak relationship between that self-reports can never achieve convincing predictive value. implicit measures and behavioral criteria? An obvious candidate Instead, “introspectively unidentified (or inaccurately identified) is a potential lack of internal consistency in the predictor variables. traces of past experience” (Greenwald and Banaji, 1995, p.  5) Unfortunately, reporting reliability coefficients is by no means were proposed to be  more crucial precursors of behavior. the rule for studies on predictive validity. Nevertheless, over time, Frontiers in Psychology | www.frontiersin.org 2 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures the picture emerged that implicit measures oen s ft uffer from them could be  responsible for the IAT’s weak predictive power. low internal consistency (for overviews, see Gawronski and De This however does not exclude the possibility that researchers Houwer, 2014; Gawronski and Hahn, 2019). High amounts of might have to address several (if not all) of these features in measurement error in the resulting scores, however, shuffle the order to achieve the desired results. In the remainder of this rank order of individuals, and thus constitute a serious problem article, we  explain all of these potentially problematic features when it comes to predicting relevant criteria like behavior (for in detail, along with promising ways forward and specific an elaboration on further consequences of low reliability, see recommendations for future research. LeBel and Paunonen, 2011; but see also De Schryver et  al., 2016). Reliability, however, seems to be  less of an issue for the most popular implicit measure, the IAT (Greenwald et al., 1998). ISSUE 1: EXTRANEOUS INFLUENCES On the contrary, IAT scores typically achieve acceptable levels ON IMPLICIT MEASURES of reliability, and outperform other implicit measures in terms of internal consistency and test-retest reliability (e.g., Nosek Implicit measures (just like explicit ones) should not et  al., 2007; Gawronski and De Houwer, 2014; Gawronski and be  understood as process-pure measures of attitudes. They are Hahn, 2019). Note, however, that it has also been suggested characterized by additional, non-attitudinal influences, and this that the comparatively high internal consistency of the IAT kind of error variance reduces their predictive validity. This might be  due to systematic error variance (so-called method also applies to the IAT (Greenwald et  al., 1998), one of the variance; see below) rather than construct-related variance (e.g., most popular implicit measures. Teige-Mocigemba et  al., 2010; Kraus and Scholderer, 2015). If e I Th AT involves two binary classification tasks, a target this holds true, given that method-related variance is unlikely task and an attribute task, that have to be  performed with to explain behavior, it is not surprising that the IAT’s predictive two response keys. Importantly, the key assignment varies across validity turned out to be  bounded. So, even for the IAT, the two IAT test blocks. In the compatible block, participants (a lack of) reliability might be  part of the problem. are instructed to press one key for the positively evaluated For the remainder of this article, however, we  put reliability target category (e.g., flower) as well as the positive pole of issues aside, and instead focus on four potentially problematic the attribute dimension (e.g., positive), and to press the other features of implicit measures and, in particular, of the IAT. key for the more negatively evaluated target category (e.g., We  will review relevant findings as well as theoretical insect) as well as the negative pole of the attribute dimension considerations, and we  will outline that each of these features (e.g., negative). In the incompatible block, negative targets and might be  responsible for the IAT’s weak predictive validity: positive attributes are assigned to the same key (and positive First, the IAT turned out not to be  a process-pure measure targets and negative attributes to the other key, respectively). of attitudes. Instead, non-attitudinal influences also play a role Participants typically respond faster and more accurate in (for an overview of these and other shortcomings of the IAT compatible compared to incompatible IAT blocks. The and its derivatives, see Fiedler et  al., 2006; Teige-Mocigemba performance difference between compatible and incompatible et  al., 2010; Gawronski and Hahn, 2019). If we  want to predict blocks (compatibility effect , IAT effect, or IAT score) is then individual’s behavior, we have to filter out this construct-irrelevant interpreted as a measure for the strength of associations between variance. Second, the IAT (just as most implicit measures) the respective categories (Greenwald et  al., 1998) . focuses on evaluation rather than motivation. However, people During the 20 years since its introduction, however, numerous do not always want what they like (and vice versa). We  should findings challenged the IAT’s construct validity (for an overview, thus not confuse liking with wanting (e.g., Tibboel et  al., see Teige-Mocigemba et  al., 2010). An illustrative example is 2015b), and in many situations, the latter might actually be more the finding that content-unrelated IATs (i.e., two IATs that relevant in driving behavior than the former. Third, as disclosed involve non-overlapping target concepts) share a considerable by its very name, the IAT was introduced to quantify associations. amount of variance (so-called method variance; e.g., Greenwald Associations, however, might be too unspecific to unambiguously et  al., 1998; McFarland and Crouch, 2002; Mierke and Klauer, relate to and account for a particular behavior in a specific 2003; Back et  al., 2005; Klauer et  al., 2010). In search for an situation. Instead, (implicit) propositional beliefs could be  a explanation for this shared method variance, several groups of more plausible precursor of behavior (e.g., Hughes et al., 2011). researchers proposed attitude-unrelated processes that ae ff ct IAT Finally, when applying the IAT researchers typically aim at responding, such as general processing speed (McFarland and assessing attitudes or stereotypes globally, that is, in a context- Crouch, 2002; Blanton et  al., 2006) or executive functions like independent fashion. Mental representations of attitudes and task-switching ability (Klauer et  al., 2010; Ito et  al., 2015). stereotypes, however, are highly context-dependent. Similarly, Another potential flaw of the IAT is the fact that it suffers real-life behavior does not occur in a situational vacuum. The from usually unwanted block order effects: IAT scores turn out predictive validity of implicit measures like the IAT might We are aware that a couple of researchers actually exercise due caution when thus be  improved by increasing the match between predictor interpreting IAT scores, understanding them as response time differences in and criterion (i.e., overcoming the lack of specificity in the a computerized categorization task – no more, no less. However, the majority predictor by incorporating contextual information). of researchers do interpret IAT scores as reflecting associative strength or Note that we  do not want to imply any order or priority implicit bias. Aer a ft ll, the IATs very name suggests such an interpretation. with regard to these four issues. We  will outline that each of In this paper, we  therefore proceed from this more common viewpoint. Frontiers in Psychology | www.frontiersin.org 3 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures to be  larger if participants started with the compatible block Whether it is valence, salience or some other feature, if the task (e.g., Greenwald et  al., 1998; Nosek et  al., 2005; for a possible was recoded, responses are based on the shared feature, and explanation, see Klauer and Mierke, 2005). Finally, IAT scores thus necessarily unrelated to the (attitudes toward the) nominal do not only reflect the valence of the target categories but can categories (e.g., faces in a Black-White IAT are no longer processed also be  influenced by stimulus effects (e.g., Steffens and Plewe, as Black vs. White but rather as more vs. less salient, Kinoshita 2001; Mitchell et  al., 2003; Govan and Williams, 2004; and Peek-O’Leary, 2005). Even more important, recoding should Bluemke and Friese, 2006; Gast and Rothermund, 2010). not be  understood as a more or less constant error that boosts Summing up, numerous studies revealed that IAT scores do IAT scores equally for everyone. Instead, there might be  inter- not reflect pure attitude strength but also contain systematic individual differences in recoding [e.g., due to individual differences error variance. The mere amount and variety of different findings in familiarity, Greenwald et al., 1998 (Exp. 2), salience, Rothermund (for an overview, see Teige-Mocigemba et  al., 2010) is not and Wentura, 2004 (Exp.’s 2A and 2B), or fluid intelligence, von particularly easy to grasp. In the following, however, we  outline Stülpnagel and Steffens, 2010 ] that can be  unrelated to the to-be- that there is a common core behind these additional processes: measured attitudes. In this sense, recoding represents a source recoding (e.g., De Houwer, 2003b; Wentura and Rothermund, 2007; of variance that might distort the predictive validity of the IAT Rothermund et  al., 2009). score for behavioral criteria. For more detailed elaborations on this issue, and for findings of recoding being unrelated to the construct of interest (i.e., attitudes), we  refer to the work of The Role of Recoding in the Implicit Meissner and Rothermund (2013, 2015a,b). Association Test Recoding can be  understood as the most crucial extraneous Although instructed to perform a double categorization task, influence in the IAT because it can account for other extraneous participants can oen e ft asily simplify the IAT through so-called influences that were identified throughout the last couple of task recoding. Recoding refers to a combination of targets and years. As an example, consider the negative correlation of IAT attributes to superordinate categories. It is based on some scores with task-switching ability (e.g., Klauer et al., 2010). Task- degree of similarity in the IAT’s stimulus material, that is, switching ability, that is, high cognitive flexibility, enables fast some feature that targets and attributes share. In a flower-insect and effortless switches between two tasks. Therefore, high switching IAT, for example, participants can profoundly simplify the task ability reduces switch costs between the two classification tasks in the compatible block by categorizing each stimulus according in the IAT (i.e., between target and attribute classification). This to its valence, and ignoring the fact that some stimuli should is especially helpful in the incompatible block of the IAT, where actually be  categorized according to their identity (i.e., flowers participants have to perform the double categorization task. In vs. insects). If the task is recoded in this sense, the compatible the compatible block, on the other hand, the task can be simplified block involves only one and the same binary decision (i.e., is by recoding. If they engage in recoding, people no longer switch the current stimulus positive or negative?). In the incompatible between the two tasks: By combining pairs of targets and attributes block, on the other hand, the incongruent response assignment into superordinate categories, they now perform only a single prevents recoding. Here, participants have no choice but to binary decision. Consequently, people with high vs. low switching follow the instructed, rather difficult double categorization task ability will perform equally well in the compatible IAT block. (i.e., flowers vs. insects, and positive vs. negative). Recoding thus results in a negative correlation of switching Recoding thus results in a substantial block difference in ability and IAT scores. Similarly, the relationship between IAT task difficulty, and therefore accounts for the observed block scores and general processing speed (e.g., McFarland and Crouch, difference in response times and error rates (e.g., Rothermund 2002) can be  explained with recoding as well. Finally, it has et  al., 2009). Remarkably, it has been shown that even in the been shown that task recoding can also account for stimulus absence of any category-based associations, recoding processes effects in the IAT (e.g., Gast and Rothermund, 2010). produce significant IAT scores (e.g., Mierke and Klauer, 2003; At this point, it should be  clear that the IAT score should Rothermund and Wentura, 2004; De Houwer et  al., 2005). be  understood as a mixture of both relevant influences (e.g., Note that recoding must not be  based on stimulus valence. associations) and irrelevant influences, first and foremost, recoding. Instead, every feature that is shared by targets and attributes If researchers want to increase the IAT’s predictive validity, they might be  used to form superordinate categories (e.g., salience, should thus try to separate effects of associations from the familiarity, valence, or even perceptual features like color or influence of recoding. In the past few years, two different shape; Rothermund et  al., 2009; see also Mierke and Klauer, approaches were introduced that claim to do so: The first approach 2003; Rothermund and Wentura, 2004; De Houwer et  al., 2005; aims at minimizing recoding processes by modifying the IAT Kinoshita and Peek-O’Leary, 2006; Chang and Mitchell, 2009) . procedure. The second approach disentangles associations and recoding processes with the help of multinomial modeling. In the following, we will present a short overview of these suggestions. e Th recoding account subsumes two earlier process models for the IAT: the so-called figure-ground account ( Rothermund and Wentura, 2001, 2004; Rothermund et  al., 2005; see also Chang and Mitchell, 2009; Kinoshita & Peak- A Solution: Dropping the Block Structure O’Leary, 2006; Mitchell, 2004) and the task-switching account (Mierke and Klauer, As outlined above, recoding effects in the IAT can be  traced 2001,2003; Klauer and Mierke, 2005). For an overview of these and other process back to its characteristic structure: the arrangement of trials accounts for the IAT we  refer to the work of Teige-Mocigemba and colleagues in (compatible vs. incompatible) blocks. When it comes to (Teige-Mocigemba et  al., 2010; Teige-Mocigemba and Klauer, 2015). Frontiers in Psychology | www.frontiersin.org 4 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures reducing the influence of recoding, an obvious possible solution e s Th econd approach also dealing with the problem of is thus to simply omit this structure. In this regard, several recoding follows a different rationale. Instead of trying to variants of the IAT have been introduced that dropped the reduce the influence of recoding, it assumes that IAT scores characteristic block structure, and varied response compatibility result from a mixture of different processes. As summarized within one test block instead: the Single-Block IAT (SB-IAT, in the following section, this approach then relies on mathematical Teige-Mocigemba et  al., 2008) and the Recoding-Free IAT modeling to measure each of these processes. This allows (IAT-RF, Rothermund et al., 2009) . While the category-response researchers to separately examine the predictive power of both assignment is constant throughout a block of trials in the construct-related and method-related variance due to recoding. standard IAT, it varies randomly from trial to trial in the Another Solution: Adopting newer IAT variants. Consequently, scores in those procedures are obtained by computing performance differences between a Modeling Approach compatible and incompatible trials rather than between Recently, a multinomial processing tree model has been compatible and incompatible blocks. introduced that enables a remarkably fine-grained analysis of In these IAT variants, participants are informed about the the IAT: The ReAL model ( Meissner and Rothermund, 2013). current category-response assignment either by simply showing Most importantly, this model successfully disentangles the effects it shortly before the stimulus appears (IAT-RF) or by using of evaluative associations from the distorting influence of task stimulus position as a cue (with an appearance in the upper recoding. In this section, we  provide a brief overview of the half of the screen signaling a compatible assignment, and an ReAL model’s basic idea, and we  review relevant findings appearance in the lower half of the screen indicating an concerning (improvements on) the IAT’s validity. incompatible assignment; SB-IAT). Crucially, the upcoming e R Th eAL model assumes that the observable responses in category-response assignment is not predictable. Consequently, the IAT (i.e., correct and incorrect responses in compatible and a stable and efficient recoding strategy specifically for the incompatible blocks) result from the interplay of specific compatible assignment becomes much harder than in the standard unobservable processes (e.g., associations and recoding; see below). IAT. This reasoning was supported by Rothermund et  al. (2009) es Th e processes are represented by separate model parameters; who found that dropping the IAT’s block structure successfully their assumed interplay is displayed in a tree architecture (i.e., reduces switch cost asymmetries, a marker of recoding processes. the multinomial processing tree). Based on observed response Besides reducing the effects of recoding, the block-free IAT patterns, algorithms estimate values for the model parameters variants come with some further advantages. For example, which are then interpreted as measures for the respective cognitive block order effects which usually influence conclusions in the processes (for mathematical details on multinomial processing standard IAT (e.g., Greenwald et  al., 1998) are no longer an tree models, see Riefer and Batchelder, 1988; Hu and Batchelder, issue. Furthermore, the newer IAT variants eliminate method- 1994; Batchelder and Riefer, 1999; for reviews of applications, related variance (Teige-Mocigemba et  al., 2008) and stimulus see Erdfelder et  al., 2009; Klauer et  al., 2012). effects ( Gast and Rothermund, 2010). These findings also support e Th ReAL model distinguishes three different processes: the assumption that recoding is one of the most crucial validity recoding (Re), evaluative associations (A) and the resource- threats of the IAT. Finally, the block-free IAT variants are not consuming label-based identification of the correct response ( L). only correlated with behavioral criteria (Teige-Mocigemba et al., e t Th ree structure incorporates theoretical assumptions concerning 2008; Houben et al., 2009), there is also evidence that dropping these processes. For example, the ReAL model assumes that the block structure of the IAT can actually improve its predictive task recoding determines responding for both targets and attributes validity (Kraus and Scholderer, 2015). but only in one of the IAT blocks (i.e., in the compatible block) . Despite these strengths of SB-IAT and IAT-RF, the strategy Evaluative associations, on the other hand, are assumed to to minimize recoding effects by dropping the IAT’s block structure influence responding in both compatible and incompatible bears the risk to miss potentially interesting effects. Although blocks but they should be  triggered only in target trials, not recoding processes do not represent the construct that researchers in attribute trials (reflecting the understanding of attitudes as typically attempt to measure when employing the IAT, they evaluative associations triggered by an attitude object, not vice might nevertheless represent variance that is related to criteria versa; Fazio et  al., 1986; see also Anderson, 1983). of interest. It has been proposed, for example, that recoding As a multinomial model, the ReAL model is able to disentangle could reflect explicit attitudes ( Rothermund et  al., 2009) and multiple cognitive processes accounting for the same observable that occasionally, it might be  related to relevant criteria (e.g., response (Batchelder and Riefer, 1999). First and foremost, the behavior; Rothermund et al., 2005; Teige-Mocigemba et al., 2008). ReAL model controls for the effects of recoding by measuring 3 4 Note that there is another procedure that dropped the IAT’s block structure, Note that for many IATs, we  do not know a priori which of the two blocks namely, the Extrinsic Aeff ctive Simon Task (EAST, De Houwer, 2003b; see will be  simplified by recoding. Even within one sample, some participants also its close cousin, the Identification EAST, De Houwer and De Bruycker, might recode the task in one IAT block (e.g., in the Black/positive block), 2007). Importantly, however, the EAST does not contain classification responses others will do so in the other block (i.e., the White/positive block). The ReAL based on the target categories and is thus strongly susceptible to stimulus model accounts for these differences by making use of the task switch cost effects (Gast and Rothermund, 2010). Furthermore, it suffers from low reliability effect as a marker for recoding processes. More precisely, the sign of the (De Houwer, 2003b). We  therefore consider the EAST a less recommendable individual switch cost effect determines the block in which the Re parameter approach to account for the problem of recoding. is modeled (for more details, see Meissner and Rothermund, 2013, 2015b). Frontiers in Psychology | www.frontiersin.org 5 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures them in a separate model parameter (which clearly represents ISSUE 2: DISTINGUISHING BETWEEN a unique feature as compared to other mathematical models LIKING AND WANTING for the IAT; e.g., the quad model, Conrey et  al., 2005; or the diffusion model, Klauer et  al., 2007). Besides addressing the Insights from recent neuropsychological research raise the question problem of recoding, the ReAL model comes with another whether evaluations are indeed the driving force behind behavioral advantage: While IAT scores only reflect relative preferences impulses. According to the incentive salience hypothesis (Robinson (which could be  problematic; for an overview, see Teige- and Berridge, 1993, 2001; Berridge and Robinson, 2003; Berridge, Mocigemba et  al., 2010), the ReAL model provides separate 2009), liking an object and wanting it are separable processes association parameters for each of the two target categories. that are mediated by different brain substrates and are differentially Consequently, the model can successfully handle situations where ae ff cted by various factors. Whereas “liking” refers to the hedonic both attitude objects trigger equally strong positive, negative, aspects of a stimulus (i.e., the pleasure or positive aeff ct it or neutral associations. Note that the conventional IAT score causes), “wanting” is the result of the attribution of incentive would only yield a null effect in these cases (i.e., no preference). salience. The latter describes a particular quality that, when Numerous studies revealed that the ReAL model parameters added to the mental representation of a given stimulus, transforms are valid measures of the processes they stand for (Meissner the mere sensory percept of this stimulus to become attention- and Rothermund, 2013; Meissner and Rothermund, 2015a,b; see grabbing, attractive, and potent to elicit behavioral impulses of also Koranyi and Meissner, 2015; Jin, 2016). Most importantly, approach or consumption, which is the very essence of behavioral the ReAL model’s association parameters reflect the direction motivation (Berridge and Robinson, 2003; Berridge, 2009). and the strength of evaluative associations for each of the two Importantly, while “wanting” and “liking” should generally target concepts (Meissner and Rothermund, 2013). This holds covary (i.e., the strength of “wanting” experienced for a specific true even in IAT applications where recoding processes pushed object should be proportional to the hedonic “liking” it produces), the overall IAT score in the opposite direction (Meissner and there are specific classes of stimuli and situations where the Rothermund, 2015a). The association parameters turned out to two processes can become uncoupled. The most prominent be  sensitive to manipulations of evaluation (Meissner and example for such a dissociation is the case of addiction, where Rothermund, 2013) but immune against artificial, non-evaluative “wanting” for the addictive drug is extremely enhanced long influences (i.e., salience asymmetries and modality match effects; aer ft it ceases to evoke hedonic experiences (i.e., “liking”), and Meissner and Rothermund, 2015a,b). Additionally, and in line even despite the addict’s recognition of its harmful effects with theoretical considerations (e.g., Fazio and Towles-Schwen, (Robinson and Berridge, 1993; Stacy and Wiers, 2010). Even 1999), association parameters correlated with self-reported attitudes though momentary dissociations of “wanting” and “liking” are in non-sensitive attitude domains (consumer preferences; Meissner at the heart of many chronic clinical psychological conditions and Rothermund, 2013). Finally, Meissner and Rothermund (e.g., Rømer Thomsen et  al., 2015 ; Olney et  al., 2018), they (2013) also tested the predictive validity of the model’s association are not in themselves pathological (Dill and Holton, 2014). parameters. As expected, the amount of chocolate consumed Rather, the closeness of the relationship between “wanting” and while watching a video was successfully predicted by the ReAL “liking” fluctuates in healthy individuals ( Epstein et  al., 2003; model’s association parameter (estimated from the response Hobbs et  al., 2005; Dai et  al., 2010, 2014; Litt et  al., 2010). pattern in a fruit-chocolate IAT). Note that the behavior was An illustrative example is the moment aer fini ft shing a delicious unrelated to the recoding parameter and also unrelated to the meal. While “liking” for the food will be unaltered, being satiated conventional IAT score (i.e., the D score; Meissner and Rothermund, will reduce “wanting” more of it (Kraus and Piqueras-Fiszman, 2013). When it comes to increasing the IAT’s predictive validity, 2016; Stevenson et al., 2017). However, not only states of satiation an application of the ReAL model thus constitutes a promising and deprivation have differential effects on “wanting” and “liking.” step forward. Given the recent developments in the field of It has also been shown, for instance, that stress increases multinomial processing tree models (i.e., allowing the incorporation “wanting” but not ‘liking’ for sweet rewards (Pool et  al., 2015). of response time data, Heck and Erdfelder, 2016; Klauer and To sum up, “wanting” and “liking,” though typically highly Kellen, 2018; and a sophisticated treatment of possible parameter correlated, can diverge. Whenever they do, “wanting” is much heterogeneity, e.g., Klauer, 2010; Matzke et  al., 2015) further more likely to guide behavior than “liking” (Berridge et  al., improvements are to be  expected. Given that the ReAL model 1989; Peciña et  al., 2003). Researchers interested in predicting has already outperformed the IAT score with regard to construct behavior are therefore well advised to incorporate measures validity in a number of studies (e.g., Meissner and Rothermund, of “wanting” (Lades, 2012). 2013, 2015a,b), we  recommend researchers to consider an application of the ReAL model as an alternative, or at least as Initial Attempts in Assessing “Wanting” an additional analysis tool for the IAT in their studies. How do we measure “wanting”? Self-reports are not an advisable So, we cannot deny that extraneous influences on IAT scores option. Obviously, they involve the risk of potential distortions like recoding do exist. However, there are promising approaches due to self-presentational concerns, especially when it comes to to address this problem. With procedural modifications or sensitive topics. Apart from that, however, disentangling “wanting” mathematical modeling, we  can measure more validly what and “liking” on a semantic level is complicated. Participants people actually like. But what if it is irrelevant what people might fail to grasp the distinction or simply confuse the two like? Maybe it is more important what people want? processes since the consideration of wanting as independent Frontiers in Psychology | www.frontiersin.org 6 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures from liking violates laymen’s intuition. Furthermore, as pointed To achieve this, several adjustments to the conventional out by Pool et  al. (2016), it is likely that self-reported “wanting” IAT procedure are necessary. First, instead of valence (as in primarily reflects expected pleasantness, and is inferred from traditional IATs), or purely semantic meaning (as in previous past hedonic experiences (i.e., “liking”). Actual implicit “wanting,” attempts at creating a “wanting” IAT), the relevant criterion on the other hand, should in principle be  independent from for the categorization of attribute stimuli in the W-IAT must any hedonic aspects of reward (Robinson and Berridge, 2013). consist in participants’ “wanting” for them, or lack thereof, Several researchers have therefore turned to established respectively. This entails the need for a set of attribute stimuli implicit measurement procedures, most oen t ft he IAT, in order that is potent to trigger acute bursts of “wanting,” and another to develop a measure of implicit “wanting” as distinct from that is not. Second, execution of the required response for implicit “liking” (for an overview, see Tibboel et  al., 2015b). wanted stimuli must acquire the quality of an actual “wanting”- By now, several IAT variants have been introduced that aim triggered consummatory response. to measure implicit “wanting” for a given target dimension As for the first requirement, it must be  considered that of interest (e.g., alcohol vs. no alcohol, smoking vs. no smoking, being “wanted” is not an inherent property of any specific attractive vs. unattractive persons). All of these approaches stimulus, but instead hinges on its interaction with the individual’s share one basic idea. That is, in order to transform the IAT current psychological or physiological state (Zhang et  al., 2009; into a measure of implicit “wanting” the category labels of Robinson and Berridge, 2013). Thus, to ensure “wanting” for the evaluative attribute dimension have to be  replaced with one set of attribute stimuli in the W-IAT, a physiological need concepts representing some aspect of “wanting.” Based on the state is induced in participants before completion of the W-IAT, notion that “wanting” entails the urge to approach the object and one set of attribute stimuli is selected to be  highly relevant in question, Palfai and Ostafin (2003) for instance, introduced for satisfying this very need. Specifically, before starting the an IAT that employs the attribute categories “approach” and W-IAT, participants are made thirsty with salty snacks. Attribute “avoidance,” with semantically related words (e.g., advance, stimuli in the following W-IAT then consist of images of drinks withdraw) as stimulus material (see Kraus and Scholderer, 2015, (need-relevant) and neutral objects (need-irrelevant). The attribute for a similar approach using the IAT-RF). In a similar vein, task in the W-IAT is then to sort these stimuli into the categories Wiers et  al. (2002) developed an IAT employing the attribute “I want” (for drinks) and “I don’t want” (for neutral objects). categories “active” and “passive” represented by arousal and Executing this categorization is then transformed into a sedation-related words. Tibboel et  al. (2011, 2015a), on the consummatory response by making “I want”-key presses other hand, used “I want” vs. “I do not want” as attribute instrumental for need satisfaction. More precisely, whenever categories in their IAT with positive vs. negative (e.g., holiday, participants correctly press the “I want”-key in response to pain; Tibboel et  al., 2011), or motivational words (e.g., gain pictures of drinks, they gain a small amount of water for later vs. deprivation; Tibboel et  al., 2015a) as stimulus material. consumption. To further increase the consummatory character However, there are reasons to doubt the validity of these of the “I want” response, this gain is signaled by immediate attempts at creating a measure of implicit “wanting.” For example, visual and auditory action effects: a small glass appears in the in situations that should actually reveal a dissociation of “wanting” lower part of the screen, and a drinking-related sound (e.g., and “liking,” these IAT variants designed to measure “wanting” cork popping and/or gurgling water) is presented via headphones. typically reveal a high overlap with “liking” measures (for an e p Th otential of this new W-IAT was illustrated in a study overview, see Tibboel et  al., 2015b). Obviously, changing the on attraction in a mating context (Koranyi et  al., 2017). attribute categorization task on a merely semantic level by simply Heterosexual male participants completed the previously replacing the category labels cannot transform the IAT into an described W-IAT procedure as well as a standard valence IAT implicit measure of “wanting.” If anything, these IATs most likely (i.e., positive vs. negative attribute dimension). Target stimuli reflect semantic associations, or a “cognitive form of wanting” in both IATs were very attractive vs. less attractive faces. IAT (Tibboel et  al., 2015b, p.  189). Recently, however, a new scores should therefore reflect participants’ implicit “wanting” Wanting-IAT was introduced (Koranyi et  al., 2017) that can and “liking” for those faces. Importantly, however, half of the be considered a more promising way forward in multiple respects. target faces were male, while the other half was female. The study revealed the expected dissociation of “wanting” and A Solution: The Wanting Implicit “liking”: Both attractive male and attractive female stimuli Association Test elicited “liking” (as measured by the standard valence IAT) e b Th asic idea of the Wanting-IAT (W-IAT, Koranyi et  al., but only attractive female (not male) faces triggered “wanting” 2017) consists in endowing the attribute discrimination task (as measured by the W-IAT). In other words, the results show with motivational character. More precisely, execution of one a general positive evaluation of attractiveness, irrespective of of the attribute responses should come to equal execution of gender, while an implicit wanting can only be found for attractive a “wanting”-triggered consummatory response. Relative “wanting” opposite-sex targets (Dai et  al., 2010). for a pair of target concepts could then be  assessed in the Note that this study additionally employed another version form of stimulus–response-compatibility effects ( De Houwer, of the wanting IAT, namely a variant that used only the semantic 2001, 2003a) by comparing the speed and accuracy of responses labels “I want” and “I do not want” without bestowing any when either of the two target categories is mapped onto the additional motivational meaning onto the attribute discrimination established “wanting” response key. task. This variant yielded the same effects as the standard Frontiers in Psychology | www.frontiersin.org 7 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures valence IAT. This detail in the results underpins the assumption together will lead to a negative evaluation of the CS (Fiedler that purely semantic “wanting” measures fail to dissociate and Unkelbach, 2011; see also Peters and Gawronski, 2011; Förderer themselves from comparable measures of “liking” (c.f., Tibboel and Unkelbach, 2012; Zanon et  al., 2014; Van Dessel et  al., 2018). et  al., 2011, 2015a). The findings of Koranyi et  al. (2017) thus Associations as they should be  measured by implicit suggest that an implicit measure of “wanting” should establish measurement procedures do not contain qualitative relational the motivational quality of relevant responses. information. Therefore, a given association between two concepts The validity of the W-IAT was further corroborated in can reflect different, sometimes even opposite beliefs. For example, a study that compared smokers’ and nonsmokers’ “wanting” “I” and “good” may be  associated either because I  believe that and “liking” for smoking cues (Grigutsch et  al., 2019). This I  am  good, or because I  believe that I  am  no good, or because study revealed that the W-IAT is better suited to discriminate I  would desperately like to be  good, or because I  know that between smokers and nonsmokers than a standard valence others would like me to be  good (see also De Houwer, 2014; IAT tapping “liking.” Specifically, W-IAT scores were positive De Houwer et  al., 2015). This raises the question whether the for smokers but negative for nonsmokers, while “liking”-IAT weak predictive validity of implicit measures of associations (e.g., scores were negative for both groups. Furthermore, in line Greenwald et  al., 2009; Oswald et  al., 2013) is due to the fact with the notion of an addiction-related decoupling of “wanting” that associations are simply unspecific. Some researchers even and “liking,” the correlation of W-IAT and “liking”-IAT was argued that the attempt to predict behavior with associations significantly weaker for smokers than for nonsmokers. In must fail because all information stored in memory is inherently contrast to previous attempts at this matter, the W-IAT propositional (e.g., Hughes et  al., 2011; De Houwer, 2014). The thus proved to measure actual “wanting” instead of purely latter, however, is part of an ongoing debate in the literature, semantic associations (c.f., Palfai and Ostafin, 2003; Tibboel and we  will not address it in detail in this overview. Still, what et  al., 2011, 2015a) both in situations where “liking” is high remains is that (measures of) associations are ambiguous with (Koranyi et  al., 2017) and in situations where “liking” is regard to the qualitative relation between the concepts involved, low (Grigutsch et  al., 2019). and that this could be responsible for the weak predictive validity So, when behavior is not in line with attitudes or values, of implicit measures. The attitude-behavior gap might be addressed this might be  due to a dissociation of “wanting” and “liking.” more convincingly with implicit measures of propositional beliefs Implicit measures of “wanting,” first and foremost those that instead of associations. actually realize a wanting quality (i.e., the W-IAT), are a promising alternative to existing measures of implicit “liking” A Solution: Implicit Measures of Beliefs when it comes to closing the attitude-behavior gap. e n Th otion of implicit measures of beliefs represents a relatively recent development (Barnes-Holmes et  al., 2010; De Houwer et  al., 2015; Müller and Rothermund, 2019). Although the ISSUE 3: FOCUS ON ASSOCIATIONS procedural details of these different measures vary, they all capitalize on the finding that during an evaluative processing of VERSUS BELIEFS propositions (e.g., “Milk is not white.”) beliefs about the truth Interestingly, when researchers started to engage in the of these propositions (i.e., “False”) are activated automatically development of implicit measurement procedures, many also (e.g., Wiswede et  al., 2013). In contrast to established implicit changed the focus with regard to the construct they attempted measures of attitudes that do not take into account the specific to measure. Self-report measures assessed complex personal semantic relationship between concepts, implicit measures of beliefs that can be  expressed in propositional statements. With beliefs allow for the assessment of complex propositions. They the development of the IAT and other implicit measures (e.g., naturally employ more complex stimuli than traditional attitude Aeff ctive Priming, Fazio et  al., 1986), the concept of beliefs measures, that is, combinations of stimuli including their semantic took a backseat in many studies. A lot of researchers now relationship, or even whole sentences. This common basis focused on measuring associations, that is, the mental connection notwithstanding, these measures utilize different approaches to between an object and a given attribute (e.g., positive or negative assess implicit beliefs, each entailing unique advantages as well valence). Such an associative link, however, is unspecific in as shortcomings. In the following, we  provide a brief overview. its nature, and admits several meanings. Implicit Relational Assessment Procedure Ambiguity of Associations In each trial of the Implicit Relational Assessment Procedure From the literature on evaluative learning, we  know that it is (IRAP, Barnes-Holmes et  al., 2010; see also Remue et  al., 2013, not only mere associative co-occurrence that determines valence 2014), participants are presented with two concepts that are transfer from an unconditioned stimulus (US) to a conditioned simultaneously displayed in the top and bottom half of the screen stimulus (CS). Instead, relational qualifiers moderate this (e.g., “I” and “nice” or “I” and “worthless”). Additionally, the relationship. For example, experiencing a neutral person (CS) IRAP highlights the propositional relationship between the two together with a positively evaluated person (US) will result in concepts by presenting a relational qualifier (e.g., “I am nice.” positive evaluations of the CS if the relationship between the or “I am not worthless.”). Participants are instructed to respond two persons is framed as friendship. If the relation between the to these stimuli in a specific manner across the two blocks of two is described as being antagonistic, however, presenting them the task. In a first block they are to classify these stimuli as Frontiers in Psychology | www.frontiersin.org 8 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures true or false (by pressing one of two keys labeled “true” and preferences: On average, they showed better performance if “false”) depending on whether they are in line with a specific they should respond as if they held pro-Flemish beliefs. belief (e.g., the belief “I am  good.”). In the second block of the As pointed out by De Houwer et al. (2015), the RRT’s structure task, this reference belief is reversed (i.e., stimuli in line with is similar to that of the IAT. For instance, the RRT employs two the belief “I am  no good” would require a “true” response). binary classification tasks sharing a set of two response keys. Additionally, in order to prevent confounding the physical location Furthermore, it consists of two critical blocks differing with regard of a response key (i.e., left vs. right) and its meaning (e.g., true to the specific response rules, and its resulting global score is vs. false) key assignment is varied on a trial by trial basis. based on the performance difference between these blocks. Mirroring Attesting to the fact that beliefs drive responding in the findings for the IAT, the RRT is reliable ( De Houwer et  al., 2015; IRAP, task performance differs between both blocks. Specifically, Tibboel et  al., 2017) while being less demanding on participants responding in the IRAP is faster and more accurate if the as indicated by markedly reduced dropout over the IRAP (4% response rule is in line with personal beliefs (Barnes-Holmes vs. 20%, De Houwer et  al., 2015). On the other hand, by virtue et  al., 2010). Additionally, these effects are sensitive to changes of these shared structural properties, the RRT runs the risk to in the relational qualifier, such as from “I am” to “I want to be  subject to similar flaws as the IAT (e.g., recoding). Last, but be” allowing for dissociation of different kinds of beliefs (e.g., not least, the necessity to instruct participants to react to statements uncovering differences between actual and ideal self, Remue in line with a block specific reference belief effectively limits the et  al., 2013, 2014) that are impervious to traditional implicit RRT to the assessment of a single belief for a given measurement measures like the IAT. session (similar to the IRAP). However, due to its block-based nature, the IRAP is limited to assessing implicit beliefs toward a single set of beliefs at a Propositional Evaluation Paradigm time (i.e., for a given pair of blocks with their associated A final implicit measure of beliefs employs a completely different reference beliefs). In addition, IRAP scores have been shown rationale. Whereas the previously discussed procedures resemble to be  susceptible to faking attempts (Hughes et  al., 2016) and the basic structure of the IAT, the so-called Propositional Evaluation oen exhi ft bit moderate reliability only (e.g., Remue et  al., 2013, Paradigm (PEP, Müller and Rothermund, 2019; see also Wiswede 2014; see also Gawronski and De Houwer, 2014). Finally, the et  al., 2013) is similar in design to classic priming procedures. IRAP is also held back by substantial dropout rates in participants Each PEP trial starts with a simple sentence that is presented in (more than 20% dropout is reported among university students a word-by-word fashion (e.g., “Milk is red.”) to participants in in Remue et  al., 2013; for a discussion, see De Houwer et  al., the center of the screen. Depending on the type of trial, this is 2015) – an issue that is thought to be  attributable to the followed by a specific response prompt. On measurement trials, trial-by-trial response key reassignment. the response prompt (either “true” or “false”) signals to participants which of two response keys (“true”-key or “false”-key) is to Relational Responding Task be  pressed. Note that the prime sentence is completely irrelevant e s Th o-called Relational Responding Task (RRT, De Houwer et al., for participants’ decision – the task is to react to the response 2015) directly addresses the issue of dropouts in the IRAP by prompt only. In contrast, on inducer trials the response prompt avoiding the trial-by-trial response key reassignment. To this end, “? true  - false?” signals participants to indicate whether the prime inducer trials require participants to classify synonyms of the sentence they just saw was orthographically correct (i.e., whether concepts “true” and “false” by button press as either “true” or or not it contained a spelling error). As in the RRT, inducer “not true” thereby constantly reinforcing the intended key meaning trials thus reinforce the intended key meaning. (De Houwer et  al., 2015). On the other hand, target trials present e Th irrelevance of the prime sentence for participants’ reactions participants with whole sentences stating certain kinds of beliefs in the measurement trials notwithstanding, compatibility effects (e.g., regarding immigrants, De Houwer et  al., 2015; or smoking, between the validity of the prime sentence and the required Tibboel et  al., 2017). Mirroring the design of the IRAP discussed response emerge. For example, the prime sentence “Milk is above, a block specific reference belief governs which of two red” is (obviously) false, hence, “false” is automatically activated. responses (i.e., “true” vs. “not true”) participants should give. One This in turn facilitates responding if the response prompt block requires participants to respond “as if ” they held a specific requires a congruent response (i.e., “false”) but interferes with belief (e.g., as if they believed that immigrants were smarter than responding if it requires an incongruent response (i.e., “true”) natives). A second block then requires participants to respond instead. Similarly, in the case of a valid (i.e., true) prime sentence “as if” they held the opposite belief (e.g., as if they believe that faster and more accurate responding would be expected following natives are smarter than immigrants). Consequently, the correct a “true” response prompt, compared to a “false” response prompt. response to a particular target sentence is “true” in one block However, whereas the PEP’s ability to measure beliefs concerning but “not true” in the other block. objectively true or false statements has been demonstrated previously If implicit beliefs drive responding in the RRT, task performance (Wiswede et  al., 2013) the true potential of an implicit measure should differ between the two blocks. Consequently, a relative of beliefs is its ability to tap into inter-individual differences in performance increase of one RRT block over the other is beliefs. This is especially true for beliefs related to more sensitive assumed to indicate that the individual’s beliefs are more in domains, such as beliefs concerning different social groups. As line with this block’s reference belief. De Houwer et  al. (2015) a case in point, Müller and Rothermund (2019) employed the found that implicit beliefs of Flemish participants reflect ingroup PEP to assess individuals’ implicit beliefs concerning racism Frontiers in Psychology | www.frontiersin.org 9 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures against immigrants. Therefore the items of established self-report that is due to salience asymmetries) or to individual attitudes measures of classic and modern racism (e.g., Akrami et  al., 2000) (e.g., extrapersonal associations; Karpinski and Hilton, 2001). served as prime sentences in the PEP. On the sample level the Of course, we  do not want to deny that an effect in an implicit PEP indicated the endorsement of tolerant and welcoming beliefs measure can provide strong evidence for inferring racial bias; about minorities and a rejection of racist beliefs. More precisely, however, we  want to emphasize that such a claim rests on responding with “true” was facilitated when positive beliefs about the assumption that the effect is driven by (implicit) evaluations minorities were shown as primes (e.g., “A multicultural Germany of the categories in question. To bolster this claim, alternative would be good.”). In contrast, responding with “false” was facilitated explanations r fi st have to be identie fi d and ruled out convincingly. when negative beliefs about minorities were shown as primes In this section, however, we  do not want to discuss studies (e.g., “Racist groups are no longer a threat toward immigrants.”). that did not even assess discriminatory behaviors. Instead, Going beyond characteristic patterns at the sample level, the we  want to focus on the lack of fit between predictor and PEP proved to be sensitive to inter-individual differences in these criterion as an explanation for the low predictive validity of beliefs. Specifically, more endorsement of racist attitudes on the implicit measures with regard to behavioral outcomes. More PEP predicted (1) explicit endorsement of these statements, (2) precisely, we argue that the predictive validity of implicit measures political orientation, and (3) behavioral efforts aimed at raising suffers from the fact that (1) studies oen do n ft ot assess behavior money for refugees (see Müller and Rothermund, 2019, for similar proper but rather employ self-report measures as a criterion, findings concerning hiring discrimination and endorsement of and (2) implicit measures typically do not provide contextual gender stereotypes). information; details that are crucial for real-life behavior. To summarize, processing and evaluation of complex propositional content can occur in a rapid and automatic (i.e., Behavioral Intentions Versus implicit) fashion. Recently, a number of promising implicit measures Behavior Proper of beliefs have been introduced. Their strength lies in their ability Although the obvious criterion variable for a study on the to measure complex, propositional relationships among different predictive validity of implicit measures is behavior (e.g., actual concepts. This allows for more fine-grained insights as compared discrimination), the assessment of behavior proper is by no to measures of simple associations that have become a hallmark means the rule. As has been prominently argued by Baumeister of established implicit measures. In our efforts at bridging the et  al. (2007), measurement of actual behavior (a dominant attitude-behavior gap, we should thus not rely solely on associations. approach during the 70s) in the field of social psychology has We  should get beliefs back on board. largely been superseded by “pseudo”-behavioral measures such as rating scale measures assessing behavioral intentions or past behavior. It is thus not surprising, that the same applies to ISSUE 4: LACK OF FIT BETWEEN studies assessing the predictive validity of implicit measures: Behavioral criteria in IAT studies oen co ft nsist of self-report PREDICTOR AND CRITERION measures or similarly indirect indicators (e.g., Oswald et  al., e p Th revious sections discussed shortcomings of the IAT and 2013; Carlsson and Agerström, 2016). Unfortunately, opting similar implicit measures and highlighted possible solutions. for self-report measures of behavior entails a number of Note though that improving the measurement of implicit shortcomings that are especially troublesome for testing the attitudes and beliefs solves only parts of the equation. It is relationship of implicit measures and behavioral outcomes. equally important to ensure adequate measurement of the First, it has long been known that self-reported behavioral respective criterion variable. intentions are not an adequate proxy for actual behavior. For In this section, we  argue that findings of low predictive example, West and Brown (1975; for a detailed elaboration, validity of implicit measures require careful consideration. If see Baumeister et  al., 2007) demonstrated a striking difference the criterion was not properly assessed, then the absence of between participants’ intention to donate money for someone a relation between an implicit measure and a criterion should in need (participants were more than willing to help) and actual not be  understood as evidence against the measure’s validity. helping behavior (donations were close to zero). Second, indirect On the other hand, some of the reported evidence for the measures were conceived to overcome self-presentational concerns validity of implicit measures in predicting behavior must that typically aeff ct self-report measures and/or to measure be  discounted based on the fact that the behavior of interest introspectively less accessible traces of experience. Consequently, was simply not assessed in the first place. Some researchers relying on these very self-reports as the major criterion for interpreted the mere presence of IAT effects as sufficient evidence predictive validity may have contributed to the heterogeneous for discrimination, which it is not. An IAT effect is just a landscape of findings on the validity of implicit measures. response time difference in a computerized categorization task, What is more, we  should probably refrain from referring to not discriminatory behavior (e.g., Arkes and Tetlock, 2004). behavior as if it were a unitary construct. Instead, researchers In our view, an effect in an implicit measure like the IAT should put forward specific hypotheses concerning the relationship might not even count as sufficient evidence for inferring the of implicit measures, different types of behavior, and specific existence of racial biases. As the previous paragraphs have situational conditions. Dual-process or dual-systems models (e.g., shown, these effects might be  driven by various influences Metcalfe and Mischel, 1999; Smith and DeCoster, 2000; Strack that can be unrelated to the categories in question (e.g., recoding and Deutsch, 2004; Friese et  al., 2008; Hofmann et  al., 2009; Frontiers in Psychology | www.frontiersin.org 10 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures Kahneman, 2011) provide a fine-grained view on this question that implicit measures tap into processes operating outside of and have frequently formed the basis for differentiation. These cognitive control, they should relate to impulsive behavior. Thus, models essentially assume that there are different kinds of although some assumptions of these models might have been processes competing for behavior control. The processes differ too strict, dual-process or dual-systems models have enriched with respect to the form in which information is stored and the literature with inspiring hypotheses and findings. They have accessed, as well as the degree of conscious awareness and proven successful in integrating and organizing a large part of cognitive control involved. o Th ugh details and labels vary (e.g., the literature on implicit and explicit measures and their relation automatic vs. controlled: Friese et al., 2008; hot vs. cool: Metcalfe to behavior. Indisputably, an important strength of these models and Mischel, 1999; associative vs. rule-based: Smith and DeCoster, lies in their differentiation between various forms of behavior. 2000; impulsive vs. reflective: Strack and Deutsch, 2004), the It is reasonable to assume different predictive power depending common idea in these models is the distinction between two on the degree of cognitive control involved. So, when it comes cognitive players. On the one hand, there is a system in which to improving the predictive power of implicit measures, our call information is usually assumed to be  stored and accessed in for differentiation also applies to the criterion variable: not all an associative manner. This system should operate fast, effortlessly forms of behavior should be treated equal, and cognitive resources and with little or no awareness and control. On the other hand, should be  taken into account. Researchers are well advised not there is a second system in which information is assumed to to simply explore whether an implicit measure predicts behavior, be  stored and accessed propositionally and which should drive or whether it outperforms explicit measures in doing so. They controlled, slow and effortful deliberation. Both systems are should rather specify more sophisticated hypotheses on the kind hypothesized to compete for behavioral control, in a tug-of-war of behavior that should be predicted (e.g., spontaneous behavior), fashion, with motivation and opportunity for control as crucial or under which conditions (e.g., depleted self-control resources) moderators (e.g., Fazio and Towles-Schwen, 1999; Friese et  al., such a relationship is to be  expected. 2008; Hofmann et  al., 2009). While the first system is assumed To sum up, we  want to highlight the notion that a robust to prompt spontaneous and impulsive behavior, the second estimation of implicit measures’ predictive validity critically should  allow for reasoned action  - but only if people are both hinges on the quality of the criterion. We therefore recommend motivated and able to spare the necessary cognitive resources to drop self-report measures and other indirect criterion variables (e.g., Hofmann et  al., 2007; Friese et  al., 2008). As a case in in favor of actual, rather spontaneous forms of behavior. point, Pearson  et  al. (2009) summarize: Context Dependency of “Whereas explicit attitudes typically shape deliberative, Attitudes and Beliefs well-considered responses for which people have the Finally, it is important to realize that behavior is enacted in motivation and opportunity to weigh the costs and a specific situation or context (e.g., we  react to someone at benefits of various courses of action, implicit attitudes work vs. in the family). Therefore, behavior is inherently context- typically influence responses that are more difficult to specific . In contrast, implicit measures in general do not specify monitor or control […] or responses that people do not contextual information and assess attitudes, stereotypes, or view as diagnostic of their attitude and thus do not try beliefs in a context-independent, global fashion. Aiming for to control.” (p. 322). such an assessment of “the” attitude (e.g., toward Blacks, women, gays, or the elderly) is also at odds with the finding that more A comprehensive overview of the more nuanced theoretical or less all attitudes, beliefs, and stereotypes are context-specific views on conditions under which implicit vs. explicit measures (Blair, 2002; Wigboldus et  al., 2003; Casper et  al., 2010, 2011; predict behavior is beyond the scope of this paper. For an overview Kornadt and Rothermund, 2011, 2015; Müller and Rothermund, of different models, we  refer readers to Perugini et  al. (2010). 2012; Gawronski and Cesario, 2013). Consequently, assessing As for now, however, it is important to note that dual-systems attitudes or beliefs in situational vacuum will oen n ft ot be specific enough to predict a particular behavior toward a specific attitude models are not without criticism (e.g., Rothermund, 2011; Gawronski and Creighton, 2013). Some of their assumptions have object in a specific situation ( Blanton and Jaccard, 2015). even set confining boundaries and require revision. Especially A Solution: Introducing the Context Into the frequently deduced notion that implicit measures like the IAT would reflect associations and therefore predict impulsive Implicit Measures behavior while explicit measures like self-reports would reflect One possibility to address this gap in “level of detail” is to propositional reasoning and therefore explain deliberate acts (e.g., aggregate behavioral outcomes across different situations, time Friese et  al., 2008) is probably an oversimplification. As we  noted points, and target objects yielding a context-independent in the section on implicit beliefs, some features of automaticity behavioral indicator in line with the context independent nature that had previously been reserved exclusively for associative of implicit measures (e.g., of discriminatory behavior; Ajzen, processes also apply to propositional information. At the same 1991). Another and more economic possibility would be  to time, ostensibly implicit measures like the IAT do not necessarily increase the “structural fit” ( Payne et al., 2008) between implicit reflect purely automatic processes, as also outlined before. Instead, measures of attitudes and the to-be-predicted situation-specific it might prove more useful to distinguish between the different behaviors by introducing context-specici fi ty also on the level processes that might be  involved. In other words, to the extent of implicit measures of attitudes. This allows us to capture Frontiers in Psychology | www.frontiersin.org 11 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures the heterogeneity of evaluations that an individual can harbor sophisticated analysis tools (i.e., the ReAL model, Meissner and with regard to the same object (Gawronski et  al., 2018), and Rothermund, 2013) that separate relevant processes from those it increases the chances to predict matching context-specific extraneous influences. Second, we  presented an overview of behaviors (e.g., Blanton and Jaccard, 2015). In this regard, different implicit measures that go beyond the measurement of measures employing dual primes incorporating both category evaluative associations, and instead quantify actual implicit wanting and context information (Casper et al., 2010, 2011) or specifying (e.g., the W-IAT, Koranyi et  al., 2017). Third, we  pointed to context-dependent evaluative meanings when choosing attribute implicit measures of beliefs (e.g., the PEP, Müller and Rothermund, categories in the IAT (Kornadt et al., 2016) represent promising 2019) that allow a more nuanced view on individual attitudes approaches for future research. Implicit measures of propositional and values than measures that tap into associations. Finally, beliefs (see Issue 3 above) are also well-suited in this regard we  emphasized the importance of measuring behavior proper since they allow researchers to clearly specify contextualized and outlined that implicit measures incorporating contextual meanings in the stimulus materials. Similarly, the strength of information might be  more adequate in assessing the structure the motivational drive to pursue specific incentives typically of implicit attitudes or beliefs and their implications for behavior depends on context cues signaling their (un-)availability. For (Casper et  al., 2011; Kornadt et  al., 2016). Each of the recent instance, individual differences in the strengths of motivational developments presented in the current paper has the potential approach (or avoidance) tendencies regarding relationship to increase the predictive power of implicit measures. Future initiation will be  triggered in a dating context (Nikitin et  al., research will also have to clarify whether a combination of these 2019) but probably will not influence behavior toward men approaches may lead to further improvement. Inspired by the and women in the work context. Incorporating this context- fruitful research on dual-process or dual-systems models, we further specificity into implicit measures of wanting (see Issue 2 above) suggest to invest in theoretical considerations: Which forms or will thus be  an important step to capture the determinants aspects of behavior should be related to which processes involved of our desires and to better explain and predict social behavior. in which implicit measures? Differentiation is key, with regard To summarize, assessing the potential of implicit measures to both the predictor and the criterion. for explaining and closing the attitude-behavior gap requires We strongly argue not to take the validity of implicit measures both predictors (implicit attitudes and beliefs) and criterion like the IAT for granted. Instead, we  should take into account variables (e.g., discriminatory behaviors) to be  assessed in a the complexity of these measures, especially when it comes reliable, valid, and contextualized way. This necessitates both to the predictive value for real-life behavior. As outlined in changes in implicit measures (to address the context-specificity the current review, the past 20  years of research have provided of the to-be-measured constructs) as well as rigorous theorizing us with a number of good reasons for why the IAT and its about which aspects of which type of behavior are to derivatives did not succeed in closing the attitude-behavior be  influenced by (context-specific) attitudes and beliefs. gap, and enriched our toolbox with promising, sophisticated improvements. Future research will benefit from harnessing the power of such a more differentiated view on implicit measures. CLOSING THOUGHTS In this article, we  presented an overview of possible reasons for AUTHOR CONTRIBUTIONS the weak relationship between implicit measures like the IAT and behavioral criteria. We outlined that the unsatisfying predictive FMe and KR wrote the first draft of the manuscript. LG, NK, value of the IAT is due to (1) extraneous influences like recoding, and FMü wrote sections of the manuscript. All authors (2) the measurement of liking instead of wanting, (3) the contributed to manuscript revision, read and approved the measurement of associations instead of complex beliefs, and/or submitted version. (4) a conceptual mismatch of predictor and criterion. We presented precise solutions for each of these problems. More precisely, we  suggested to switch to procedural variations that minimize FUNDING extraneous influences (i.e., the SB-IAT, Teige-Mocigemba et  al., 2008; and the IAT-RF; Rothermund et  al., 2009), and to apply This work was funded by grant RO 1272/11-1 to KR. REFERENCES Arkes, H. R., and Tetlock, P. E. (2004). Attributions of implicit prejudice, or “Would Jesse Jackson ‘fail’ the implicit association test?”. Psychol. Inq. 15, Ajzen, I. (1991). The theory of planned behavior. Organ. Behav. Hum. Decis. 257–278. doi: 10.1207/s15327965pli1504_01 Back, M. D., Schmukle, S. C., and Eglo, B ff . (2005). Measuring task-switching Process. 50, 179–211. doi: 10.1016/0749-5978(91)90020-T Akrami, N., Ekehammar, B., and Araya, T. (2000). Classical and modern racial ability in the implicit association test. Exp. Psychol. 52, 167–179. doi: 10.1027/1618-3169.52.3.167 prejudice: a study of attitudes toward immigrants in Sweden. Eur. J. Soc. Psychol. 30, 521–532. doi: 10.1002/1099-0992(200007/08)30:4<521::AID- Barnes-Holmes, D., Barnes-Holmes, Y., Stewart, I., and Boles, S. (2010). A sketch of the implicit relational assessment procedure (IRAP) and the relational EJSP5>3.0.CO;2-N Anderson, J. R. (1983). A spreading activation theory of memory. J. Verbal elaboration and coherence (REC) model. Psychol. Rec. 60, 527–542. doi: 10.1007/BF03395726 Learn. Verbal Behav. 22, 261–295. doi: 10.1016/S0022-5371(83)90201-3 Frontiers in Psychology | www.frontiersin.org 12 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures Batchelder, W. H., and Riefer, D. M. (1999). Theoretical and empirical review Epstein, L. H., Truesdale, R., Wojcik, A., Paluch, R. A., and Raynor, H. A. of multinomial process tree modeling. Psychon. Bull. Rev. 6, 57–86. doi: (2003). Effects of deprivation on hedonics and reinforcing value of food. 10.3758/BF03210812 Physiol. Behav. 78, 221–227. doi: 10.1016/S0031-9384(02)00978-2 Baumeister, R. F., Vohs, K. D., and Funder, D. C. (2007). Psychology as the Erdfelder, E., Auer, T. S., Hilbig, B. E., Aßfalg, A., Moshagen, M., and Nadarevic, L. science of self-reports and finger movements. Whatever happened to actual (2009). Multinomial processing tree models: a review of the literature. behavior? Perspect. Psychol. Sci. 2, 396–403. doi: 10.1111/j.1745-6916.2007.00051.x J. Psychol. 217, 108–124. doi: 10.1027/0044-3409.217.3.108 Berridge, K. C. (2009). Wanting and liking: observations from the neuroscience Fazio, R. H. (2007). Attitudes as object-evaluation associations of varying and psychology laboratory. Inquiry 52:378. doi: 10.1080/00201740903087359 strength. Soc. Cogn. 25, 603–637. doi: 10.1521/soco.2007.25.5.603 Berridge, K. C., and Robinson, T. E. (2003). Parsing reward. Trends Neurosci. Fazio, R. H., Powell, M. C., and Herr, P. M. (1983). Toward a process model of 26, 507–513. doi: 10.1016/S0166-2236(03)00233-9 the attitude-behavior relation: accessing one’s attitude upon mere observation of Berridge, K. C., Venier, I. L., and Robinson, T. E. (1989). Taste reactivity the attitude object. J. Pers. Soc. Psychol. 44, 723–735. doi: 10.1037/0022-3514.44.4.723 analysis of 6-hydrocydopamine-induced aphagia: implications for arousal Fazio, R. H., Sanbonmatsu, D. M., Powell, M. C., and Kardes, F. R. (1986). and anhedonia hypotheses of dopamine function. Behav. Neurosci. 103, On the automatic activation of attitudes. J. Pers. Soc. Psychol. 50, 229–238. 36–45. doi: 10.1037/0735-7044.103.1.36 doi: 10.1037/0022-3514.50.2.229 Blair, I. V. (2002). e Th malleability of automatic stereotypes and prejudice. Fazio, R. H., and Towles-Schwen, T. (1999). “e M Th ODE model of attitude- Personal. Soc. Psychol. Rev. 6, 242–261. doi: 10.1207/S15327957PSPR0603_8 behavior processes” in Dual-process theories in social psychology. eds. S. Chaiken and Y. Trope (New York, NY: Guilford Press), 97–116. Blanton, H., and Jaccard, J. (2015). Not so fast: ten challenges to importing implicit attitude measures to media psychology. Media Psychol. 18, 338–369. Fiedler, K., Messner, C., and Bluemke, M. (2006). Unresolved problems with the “I”, the “A”, and the “T”: a logical and psychometric critique of the doi: 10.1080/15213269.2015.1008102 Blanton, H., Jaccard, J., Gonzales, P. M., and Christie, C. (2006). Decoding implicit association test (IAT). Eur. Rev. Soc. Psychol. 17, 74–147. doi: 10.1080/10463280600681248 the implicit association test: implications for criterion prediction. J. Exp. Fiedler, K., and Unkelbach, C. (2011). Evaluative conditioning depends on Soc. Psychol. 42, 192–212. doi: 10.1016/j.jesp.2005.07.003 higher order encoding processes. Cognit. Emot. 25, 639–656. doi: Bluemke, M., and Friese, M. (2006). Do features of stimuli influence IAT 10.1080/02699931.2010.513497 effects? J. Exp. Soc. Psychol. 42, 163–176. doi: 10.1016/j.jesp.2005.03.004 Förderer, S., and Unkelbach, C. (2012). Hating the cute kitten or loving the Carlsson, R., and Agerström, J. (2016). A closer look at the discrimination aggressive pit-bull: EC effects depend on CS-US relations. Cognit. Emot. outcomes in the IAT literature. Scand. J. Psychol. 57, 278–287. doi: 10.1111/ 26, 534–540. doi: 10.1080/02699931.2011.588687 sjop.12288 Friese, M., Hofmann, W., and Schmitt, M. (2008). When and why do implicit Casper, C., Rothermund, K., and Wentura, D. (2010). Automatic stereotype measures predict behaviour? Empirical evidence for the moderating role of activation is context dependent. Soc. Psychol. 41, 131–136. doi: 10.1027/1864-9335/ opportunity, motivation, and process reliance. Eur. Rev. Soc. Psychol. 19, a000019 285–338. doi: 10.1080/10463280802556958 Casper, C., Rothermund, K., and Wentura, D. (2011). The activation of specific Friese, M., Hofmann, W., and Wänke, M. (2008). When impulses take over: facets of age stereotypes depends on individuating information. Soc. Cogn. moderated predictive validity of explicit and implicit attitude measures in 29, 393–414. doi: 10.1521/soco.2011.29.4.393 Chang, B. P. I., and Mitchell, C. J. (2009). Processing fluency as a predictor predicting food choice and consumption behaviour. Br. J. Soc. Psychol. 47, 397–419. doi: 10.1348/014466607X241540 of salience asymmetries in the implicit association test. Q. J. Exp. Psychol. 62, 2030–2054. doi: 10.1080/17470210802651737 Gast, A., and Rothermund, K. (2010). When old and frail is not the same. Dissociating category-based and stimulus-based influences on compatibility Conrey, F. R., Sherman, J. W., Gawronski, B., Hugenberg, K., and Groom, C. J. (2005). Separating multiple processes in implicit social cognition: the quad effects in four implicit measurement methods. Q. J. Exp. Psychol. 63, 479–498. model of implicit task performance. J. Pers. Soc. Psychol. 89, 469–487. doi: doi: 10.1080/17470210903049963 10.1037/0022-3514.89.4.469 Gawronski, B., and Cesario, J. (2013). Of mice and men: what animal research Dai, X. C., Brendl, C. M., and Ariely, D. (2010). Wanting, liking, and preference can tell us about context effects on automatic responses in humans. Personal. construction. Emotion 10, 324–334. doi: 10.1037/a0017987 Soc. Psychol. Rev. 17, 187–215. doi: 10.1177/1088868313480096 Dai, X. C., Dong, P., and Jia, J. S. (2014). When does playing hard to get Gawronski, B., and Creighton, L. A. (2013). “Dual-process theories” in The increase romantic attraction? J. Exp. Psychol. Gen. 143, 521–526. doi: 10.1037/ Oxford handbook of social cognition. ed. D. E. Carlston (New York, NY: a0032989 Oxford University Press), 282–312. De Houwer, J. (2001). A structural and process analysis of the implicit association Gawronski, B., and De Houwer, J. (2014). “Implicit measures in social and test. J. Exp. Soc. Psychol. 37, 443–451. doi: 10.1006/jesp.2000.1464 personality psychology” in Handbook of research methods in social and De Houwer, J. (2003a). “A structural analysis of indirect measures of attitudes” personality psychology. 2nd Edn. eds. H. T. Reis and C. M. Judd (New York, NY: in e p Th sychology of evaluation: Ae ff ctive processes in cognition and emotion . Cambridge University Press), 283–310. eds. J. Musch and K. C. Klauer (Mahwah, NJ: Erlbaum), 219–244. Gawronski, B., and Hahn, A. (2019). “Implicit measures: procedures, use, and De Houwer, J. (2003b). The extrinsic aeff ctive Simon task. Exp. Psychol. 50, interpretation” in Measurement in social psychology. eds. H. Blanton, J. M. 77–85. doi: 10.1026//1618-3169.50.2.77 LaCroix, and G. D. Webster (New York: Taylor & Francis), 29–55. De Houwer, J. (2014). A propositional model of implicit evaluation. Soc. Psychol. Gawronski, B., Rydell, R. J., De Houwer, J., Brannon, S. M., Ye, Y., Vervliet, B., Person. Compass 8, 342–353. doi: 10.1111/spc3.12111 et al. (2018). Contextualized attitude change. Adv. Exp. Soc. Psychol. 57, De Houwer, J., and De Bruycker, E. (2007). The identification-EAST as a valid 1–52. doi: 10.1016/bs.aesp.2017.06.001 measure of implicit attitudes toward alcohol-related stimuli. J. Behav. Ther. Govan, C. L., and Williams, K. D. (2004). Changing the aeff ctive valence of Exp. Psychiatry 38, 133–143. doi: 10.1016/j.jbtep.2006.10.004 the stimulus items influences the IAT by re-defining the category labels. De Houwer, J., Geldof, T., and De Bruycker, E. (2005). The implicit association J. Exp. Soc. Psychol. 40, 357–365. doi: 10.1016/j.jesp.2003.07.002 test as a general measure of similarity. Can. J. Exp. Psychol. 59, 228–239. Greenwald, A. G., and Banaji, M. R. (1995). Implicit social cognition: attitudes, doi: 10.1037/h0087478 self-esteem, and stereotypes. Psychol. Rev. 102, 4–27. doi: 10.1037/0033-295X.102.1.4 De Houwer, J., Heider, N., Spruyt, A., Roets, A., and Hughes, S. (2015). The Greenwald, A. G., McGhee, D. E., and Schwartz, J. L. K. (1998). Measuring relational responding task: toward a new implicit measure of beliefs. Front. individual differences in implicit cognition: the implicit association test. Psychol. 6, 1–9. doi: 10.3389/fpsyg.2015.00319 J. Pers. Soc. Psychol. 74, 1464–1480. doi: 10.1037/0022-3514.74.6.1464 De Schryver, M., Hughes, S., Rosseel, Y., and De Houwer, J. (2016). Unreliable Greenwald, A. G., Poehlman, T. A., Uhlmann, E. L., and Banaji, M. R. (2009). yet still replicable: a comment on LeBel and Paunonen (2011). Front. Psychol. Understanding and using the implicit association test: III. Meta-analysis of 6:2039. doi: 10.3389/fpsyg.2015.02039 predictive validity. J. Pers. Soc. Psychol. 97, 17–41. doi: 10.1037/a0015575 Dill, B., and Holton, R. (2014). The addict in us all. Front. Psych. 5:139. doi: Grigutsch, L. A., Lewe, G., Rothermund, K., and Koranyi, N. (2019). Implicit 10.3389/fpsyt.2014.00139 ‘wanting’ without implicit ‘liking’: a test of incentive-sensitization-theory in Frontiers in Psychology | www.frontiersin.org 13 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures the context of smoking addiction using the wanting-implicit-association-test Koranyi, N., Grigutsch, L. A., Algermissen, J., and Rothermund, K. (2017). (W-IAT). J. Behav. Ther. Exp. Psychiatry 64, 9–14. doi: 10.1016/j.jbtep.2019.01.002 Dissociating implicit wanting from implicit liking: development and validation Heck, D. W., and Erdfelder, E. (2016). Extending multinomial processing tree of the wanting-implicit-association-test (W-IAT). J. Behav. Ther. Exp. Psychiatry models to measure the relative speed of cognitive processes. Psychon. Bull. 54, 165–169. doi: 10.1016/j.jbtep.2016.08.008 Rev. 23, 1440–1465. doi: 10.3758/s13423-016-1025-6 Koranyi, N., and Meissner, F. (2015). Handing over the reins: neutralizing Hobbs, M., Remington, B., and Glautier, S. (2005). Dissociation of wanting negative attitudes toward dependence in response to reciprocal romantic and liking for alcohol in humans: a test of the incentive-sensitisation theory. liking. Soc. Psychol. Personal. Sci. 6, 685–691. doi: 10.1177/1948550615580169 Psychopharmacology 178, 493–499. doi: 10.1007/s00213-004-2026-0 Kornadt, A. E., Meissner, F., and Rothermund, K. (2016). Implicit and explicit Hofmann, W., and Friese, M. (2008). Impulses got the better of me: alcohol age stereotypes for specific life domains across the life span: distinct patterns moderates the influence of implicit attitudes toward food cues on eating and age group differences. Exp. Aging Res. 42, 195–211. doi: 10.1080/ behavior. J. Abnorm. Psychol. 117, 420–427. doi: 10.1037/0021-843X.117.2.420 0361073X.2016.1132899 Hofmann, W., Friese, M., and Strack, F. (2009). Impulse and self-control from Kornadt, A. E., and Rothermund, K. (2011). Contexts of aging: assessing a dual-systems perspective. Perspect. Psychol. Sci. 4, 162–176. doi: 10.1111/j. evaluative age stereotypes in different life domains. J. Gerontol. Psychol. Sci. 1745-6924.2009.01116.x 66, 547–556. doi: 10.1093/geronb/gbr036 Hofmann, W., Rauch, W., and Gawronski, B. (2007). And deplete us not into Kornadt, A. E., and Rothermund, K. (2015). Views on aging: domain-specific temptation: automatic attitudes, dietary restraint, and self-regulatory resources approaches and implications for developmental regulation. Annu. Rev. Gerontol. as determinants of eating behavior. J. Exp. Soc. Psychol. 43, 497–504. doi: Geriatr. 35, 121–144. doi: 10.1891/0198-8794.35.121 10.1016/j.jesp.2006.05.004 Kraus, S. J. (1995). Attitudes and the prediction of behavior: a meta-analysis Houben, K., Rothermund, K., and Wiers, R. W. (2009). Predicting alcohol use of the empirical literature. Personal. Soc. Psychol. Bull. 21, 58–75. doi: with a recoding-free variant of the implicit association test. Addict. Behav. 10.1177/0146167295211007 34, 487–489. doi: 10.1016/j.addbeh.2008.12.012 Kraus, A. A., and Piqueras-Fiszman, B. (2016). Sandwich or sweets? An assessment Hu, X., and Batchelder, W. H. (1994). The statistical analysis of general processing of two novel implicit association tasks to capture dynamic motivational tree models with the EM algorithm. Psychometrika 59, 21–47. doi: 10.1007/ tendencies and stable evaluations towards foods. Food Qual. Prefer. 49, BF02294263 11–19. doi: 10.1016/j.foodqual.2015.11.005 Hughes, S., Barnes-Holmes, D., and De Houwer, J. (2011). The dominance of Kraus, A. A., and Scholderer, J. (2015). Indirect measurement of motivation: associative theorizing in implicit attitude research: propositional and behavioral developing and testing a motivational recoding-free implicit association test alternatives. Psychol. Rec. 61, 465–496. doi: 10.1007/BF03395772 (m-IAT-RF). Soc. Psychol. 46, 142–156. doi: 10.1027/1864-9335/a000234 Hughes, S., Hussey, I., Corrigan, B., Jolie, K., Murphy, C., and Barnes-Holmes, D. Kurdi, B., Seitchik, A. E., Axt, J. R., Carroll, T. J., Karapetyan, A., Kaushik, N., (2016). Faking revisited: exerting strategic control over performance on the et al. (2019). Relationship between the implicit association test and intergroup implicit relational assessment procedure. Eur. J. Soc. Psychol. 46, 632–648. behavior: a meta-analysis. Am. Psychol. 74, 569–586. doi: 10.1037/amp0000364 doi: 10.1002/ejsp.2207 Lades, L. K. (2012). Towards an incentive salience model of intertemporal Ito, T. A., Friedman, N. P., Bartholow, B. D., Correll, J., Loersch, C., Altamirano, L. J., choice. J. Econ. Psychol. 33, 833–841. doi: 10.1016/j.joep.2012.03.007 et al. (2015). Toward a comprehensive understanding of executive cognitive LeBel, E. P., and Paunonen, S. V. (2011). Sexy but oen unr ft eliable: the impact function in implicit racial bias. J. Pers. Soc. Psychol. 108, 187–218. doi: of unreliability on the replicability of experimental findings with implicit 10.1037/a0038557 measures. Personal. Soc. Psychol. Bull. 37, 570–583. doi: 10.1177/ Jin, Z. (2016). Disentangling recoding processes and evaluative associations in 0146167211400619 a gender attitude implicit association test among adult males. Q. J. Exp. Litt, A., Khan, U., and Shiv, B. (2010). Lusting while loathing: parallel counterdriving Psychol. 69, 2276–2284. doi: 10.1080/17470218.2015.1126290 of wanting and liking. Psychol. Sci. 21, 118–125. doi: 10.1177/0956797609355633 Kahneman, D. (2011). in Th king, fast and slow . New York: Farrar, Straus, Matzke, D., Dolan, C. V., Batchelder, W. H., and Wagenmakers, E.-J. (2015). and Giroux. Bayesian estimation of multinomial processing tree models with heterogeneity Karpinski, A., and Hilton, J. L. (2001). Attitudes and the implicit association in participants and items. Psychometrika 80, 205–235. doi: 10.1007/ test. J. Pers. Soc. Psychol. 81, 774–788. doi: 10.1037/0022-3514.81.5.774 s11336-013-9374-9 Katz, D. (1960). The functional approach to the study of attitudes. Public Opin. McFarland, S. G., and Crouch, Z. (2002). A cognitive skill confound on the Q. 24, 163–204. doi: 10.1086/266945 implicit association test. Soc. Cogn. 20, 483–510. doi: 10.1521/soco.20.6.483.22977 Kinoshita, S., and Peek-O’Leary, M. (2005). Does the compatibility effect in Meissner, F., and Rothermund, K. (2013). Estimating the contributions of the race implicit association test (IAT) reflect familiarity or ae ff ct? Psychon. associations and recoding in the implicit association test: the ReAL model Bull. Rev. 12, 442–452. doi: 10.3758/BF03193786 for the IAT. J. Pers. Soc. Psychol. 104, 45–69. doi: 10.1037/a0030734 Kinoshita, S., and Peek-O’Leary, M. (2006). Two bases of the compatibility Meissner, F., and Rothermund, K. (2015a). The insect-nonword IAT revisited: effect in the implicit association test (IAT). Q. J. Exp. Psychol. 59, 2102–2120. dissociating between evaluative associations and recoding. Soc. Psychol. 46, doi: 10.1080/17470210500451141 46–54. doi: 10.1027/1864-9335/a000220 Klauer, K. C. (2010). Hierarchical multinomial processing tree models: a latent- Meissner, F., and Rothermund, K. (2015b). A thousand words are worth more trait approach. Psychometrika 75, 70–98. doi: 10.1007/s11336-009-9141-0 than a picture? The effects of stimulus modality on the implicit association Klauer, K. C., and Kellen, D. (2018). RT-MPTs: process models for response-time test. Soc. Psychol. Personal. Sci. 6, 740–748. doi: 10.1177/1948550615580381 distributions based on multinomial processing trees with applications to Metcalfe, J., and Mischel, W. (1999). A hot/cool-system analysis of delay of recognition memory. J. Math. Psychol. 82, 111–130. doi: 10.1016/j.jmp.2017.12.003 gratification: dynamics of willpower. Psychol. Rev. 106, 3–19. doi: Klauer, K. C., and Mierke, J. (2005). Task-set inertia, attitude accessibility, and 10.1037/0033-295X.106.1.3 compatibility-order effects: new evidence for a task-set switching account Mierke, J., and Klauer, K. C. (2001). Implicit association measurement with of the implicit association test effect. Personal. Soc. Psychol. Bull. 31, 208–217. the IAT: evidence for effects of executive control processes. Z. Exp. Psychol. doi: 10.1177/0146167204271416 48, 107–122. doi: 10.1026//0949-3946.48.2.107 Klauer, K. C., Schmitz, F., Teige-Mocigemba, S., and Voss, A. (2010). Understanding Mierke, J., and Klauer, K. C. (2003). Method-specific variance in the implicit the role of executive control in the implicit association test: why flexible association test. J. Pers. Soc. Psychol. 85, 1180–1192. doi: 10.1037/0022-3514.85.6.1180 people have small IAT effects. Q. J. Exp. Psychol. 63, 595–619. doi: Mitchell, C. J. (2004). Mere acceptance produces apparent attitude in the implicit 10.1080/17470210903076826 association test. J. Exp. Soc. Psychol. 40, 366–373. doi: 10.1016/j.jesp.2003.07.003 Klauer, K. C., Stahl, C., and Voss, A. (2012). “Multinomial models and diffusion Mitchell, J. P., Nosek, B. A., and Banaji, M. R. (2003). Contextual variations models” in Cognitive methods in social psychology. Abridged edition. eds. in implicit evaluation. J. Exp. Psychol. Gen. 132, 455–469. doi: K. C. Klauer, A. Voss, and C. Stahl (New York, NY: Guilford Press), 331–354. 10.1037/0096-3445.132.3.455 Klauer, K. C., Voss, A., Schmitz, F., and Teige-Mocigemba, S. (2007). Process Müller, F., and Rothermund, K. (2012). Talking loudly but lazing at work – components of the implicit association test: a diffusion-model analysis. behavioral effects of stereotypes are context dependent. Eur. J. Soc. Psychol. J. Pers. Soc. Psychol. 93, 353–368. doi: 10.1037/0022-3514.93.3.353 42, 557–563. doi: 10.1002/ejsp.1869 Frontiers in Psychology | www.frontiersin.org 14 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures Müller, F., and Rothermund, K. (2019). e Th Propositional Evaluation Paradigm: Rømer Thomsen, K., Whybrow, P. C., and Kringelbach, M. L. (2015). Reconceptualizing Indirect Assessment of Personal Beliefs and Attitudes. Front. Psychol. 10:2385. anhedonia: novel perspectives on balancing the pleasure networks in the human doi: 10.3389/fpsyg.2019.02385 brain. Front. Behav. Neurosci. 9:49. doi: 10.3389/fnbeh.2015.00049 Nikitin, J., Gong, X., Schoch, S., and Freund, A. M. (2019). Social motives, Rothermund, K. (2011). Selbstregulation statt Selbstkontrolle: Intentionalität, attributions and expectations as predictors of the decision to participate in Ae ff kt und Kognition als Teamgefährten im Dienste der Handlungssteuerung. a speed-dating event. Motiv. Emot. 43, 610–624. doi: 10.1007/s11031-019-09762-0 Psychol. Rundsch. 62, 167–173. doi: 10.1026/0033-3042/a000087 Nisbett, R. E., and Wilson, T. D. (1977). Telling more than we  can know: Rothermund, K., Teige-Mocigemba, S., Gast, A., and Wentura, D. (2009). verbal reports on mental processes. Psychol. Rev. 84, 231–259. doi: Minimizing the influence of recoding in the implicit association test: the 10.1037/0033-295X.84.3.231 recoding-free implicit association test (IAT-RF). Q. J. Exp. Psychol. 62, Nosek, B. A., Greenwald, A. G., and Banaji, M. R. (2005). Understanding and 84–98. doi: 10.1080/17470210701822975 using the implicit association test: II. Method variables and construct validity. Rothermund, K., and Wentura, D. (2001). Figure-ground asymmetries in the Personal. Soc. Psychol. Bull. 31, 166–180. doi: 10.1177/0146167204271418 implicit association test (IAT). Z. Exp. Psychol. 48, 94–106. doi: Nosek, B. A., Greenwald, A. G., and Banaji, M. R. (2007). “e im Th plicit association 10.1026//0949-3946.48.2.94 test at age 7: a methodological and conceptual review” in Frontiers of social Rothermund, K., and Wentura, D. (2004). Underlying processes in the implicit psychology: Social psychology and the unconscious. The automaticity of higher association test: dissociating salience from associations. J. Exp. Psychol. Gen. mental processes. ed. J. A. Bargh (New York, NY: Psychology Press), 265–292. 133, 139–165. doi: 10.1037/0096-3445.133.2.139 Olney, J. J., Warlow, S. M., Naffziger, E. E., and Berridge, K. C. (2018). Current Rothermund, K., Wentura, D., and De Houwer, J. (2005). Validity of the salience perspectives on incentive salience and applications to clinical disorders. asymmetry account of the implicit association test: reply to Greenwald, Curr. Opin. Behav. Sci. 22, 59–69. doi: 10.1016/j.cobeha.2018.01.007 Nosek, Banaji, and Klauer (2005). J. Exp. Psychol. Gen. 134, 426–430. doi: Oswald, F. L., Mitchell, G., Blanton, H., Jaccard, J., and Tetlock, P. E. (2013). 10.1037/0096-3445.134.3.426 Predicting ethnic and racial discrimination: a meta-analysis of IAT criterion Smith, E. R., and DeCoster, J. (2000). Dual-process models in social and cognitive studies. J. Pers. Soc. Psychol. 105, 171–192. doi: 10.1037/a0032734 psychology: conceptual integration and links to underlying memory systems. Palfai, T. P., and Ostafin, B. D. (2003). Alcohol-related motivational tendencies Personal. Soc. Psychol. Rev. 4, 108–131. doi: 10.1207/S15327957PSPR0402_01 in hazardous drinkers: assessing implicit response tendencies using the modified- Stacy, A. W., and Wiers, R. W. (2010). Implicit cognition and addiction: a IAT. Behav. Res. Ther. 41, 1149–1162. doi: 10.1016/S0005-7967(03)00018-4 tool for explaining paradoxical behavior. Annu. Rev. Clin. Psychol. 6, 551–575. Payne, B. K., Burkley, M. A., and Stokes, M. B. (2008). Why do implicit and doi: 10.1146/annurev.clinpsy.121208.131444 explicit attitude tests diverge? The role of structural fit. J. Pers. Soc. Psychol. Steffens, M. C., and Plewe, I. (2001). Items’ cross-category associations as a 94, 16–31. doi: 10.1037/0022-3514.94.1.16 confounding factor in the implicit association test. Z. Exp. Psychol. 48, Payne, B. K., Cheng, C. M., Govorun, O., and Stewart, B. D. (2005). An 123–134. doi: 10.1026//0949-3946.48.2.123 inkblot for attitudes: aeff ct misattribution as implicit measurement. J. Pers. Stevenson, R. J., Francis, H. M., Attuquayefio, T., and Ockert, C. (2017). Explicit Soc. Psychol. 89, 277–293. doi: 10.1037/0022-3514.89.3.277 wanting and liking for palatable snacks are differentially ae ff cted by change Pearson, A. R., Dovidio, J. F., and Gaertner, S. L. (2009). The nature of in physiological state, and differentially related to salivation and hunger. contemporary prejudice: insights from aversive racism. Soc. Personal. Psychol. Physiol. Behav. 182, 101–106. doi: 10.1016/j.physbeh.2017.10.007 Compass 3, 314–338. doi: 10.1111/j.1751-9004.2009.00183.x Strack, F., and Deutsch, R. (2004). Reflective and impulsive determinants of Peciña, S., Cagniard, B., Berridge, K. C., Aldridge, J. W., and Zhuang, X. social behavior. Personal. Soc. Psychol. Rev. 8, 220–247. doi: 10.1207/ (2003). Hyperdopaminergic mutant mice have higher “wanting” but not s15327957pspr0803_1 “liking” for sweet rewards. J. Neurosci. 23, 9395–9402. doi: 10.1523/ Teige-Mocigemba, S., and Klauer, K.C. (2015). Implicit association test. In J. D. JNEUROSCI.23-28-09395.2003 Wright (editor-in-chief ), International encyclopedia of the social and behavioral Perugini, M., Richetin, J., and Zogmaister, C. (2010). “Prediction of behavior” sciences. Vol. 11, 2nd Edn (Oxford, UK: Elsevier), 703–708. in Handbook of implicit social cognition: Measurement, theory, and applications. Teige-Mocigemba, S., Klauer, K. C., and Rothermund, K. (2008). Minimizing eds. B. Gawronski and B. K. Payne (New York, NY: Guilford Press), 255–277. method-specific variance in the IAT: a single block IAT. Eur. J. Psychol. Peters, K. R., and Gawronski, B. (2011). Are we puppets on a string? Comparing Assess. 24, 237–245. doi: 10.1027/1015-5759.24.4.237 the impact of contingency and validity on implicit and explicit evaluations. Teige-Mocigemba, S., Klauer, K. C., and Sherman, J. W. (2010). “Practical guide Personal. Soc. Psychol. Bull. 37, 557–569. doi: 10.1177/0146167211400423 to implicit association tests and related tasks” in Handbook of implicit social Pool, E., Brosch, T., Delplanque, S., and Sander, D. (2015). Stress increases cognition: Measurement, theory, and applications. eds. B. Gawronski and cue-triggered “wanting” for sweet reward in humans. J. Exp. Psychol Anim. B. K. Payne (New York, NY: Guilford Press), 117–139. Learn. Cogn. 41, 128–136. doi: 10.1037/xan0000052 Tibboel, H., De Houwer, J., Dirix, N., and Spruyt, A. (2017). Beyond associations: Pool, E., Sennwald, V., Delplanque, S., Brosch, T., and Sander, D. (2016). do implicit beliefs play a role in smoking addiction? J. Psychopharmacol. Measuring wanting and liking from animals to humans: a systematic review. 31, 43–53. doi: 10.1177/0269881116665327 Neurosci. Biobehav. Rev. 63, 124–142. doi: 10.1016/j.neubiorev.2016.01.006 Tibboel, H., De Houwer, J., Spruyt, A., Brevers, D., Roy, E., and Noël, X. Remue, J., De Houwer, J., Barnes-Holmes, D., Vanderhasselt, M. A., and (2015a). Heavy social drinkers score higher on implicit wanting and liking De Raedt, R. (2013). Self-esteem revisited: performance on the implicit for alcohol than alcohol-dependent patients and light social drinkers. relational assessment procedure as a measure of self-versus ideal self-related J. Behav. er Th . Exp. Psychiatry 48, 185–191. doi: 10.1016/j.jbtep.2015.04.003 cognitions in dysphoria. Cognit. Emot. 27, 1441–1449. doi: 10.1080/ Tibboel, H., De Houwer, J., Spruyt, A., Field, M., Kemps, E., and Crombez, G. 02699931.2013.786681 (2011). Testing the validity of implicit measures of wanting and liking. Remue, J., Hughes, S., De Houwer, J., and De Raedt, R. (2014). To be  or want J. Behav. er Th . Exp. Psychiatry 42, 284–292. doi: 10.1016/j.jbtep.2011.01.002 to be: disentangling the role of actual versus ideal self in implicit self- Tibboel, H., De Houwer, J., and Van Bockstaele, B. (2015b). Implicit measures esteem. PLoS One 9:e108837. doi: 10.1371/journal.pone.0108837 of “wanting” and “liking” in humans. Neurosci. Biobehav. Rev. 57, 350–364. Riefer, D. M., and Batchelder, W. H. (1988). Multinomial modeling and the doi: 10.1016/j.neubiorev.2015.09.015 measurement of cognitive processes. Psychol. Rev. 95, 318–339. doi: Van Dessel, P., De Houwer, J., and Smith, C. T. (2018). Relational information 10.1037/0033-295X.95.3.318 moderates approach-avoidance instruction effects on implicit evaluation. Acta Robinson, T. E., and Berridge, K. C. (1993). The neural basis of drug craving – an Psychol. 184, 137–143. doi: 10.1016/j.actpsy.2017.03.016 incentive-sensitization theory of addiction. Brain Res. Rev. 18, 247–291. doi: von Stülpnagel, R., and Steffens, M. C. (2010). Prejudiced or just smart? 10.1016/0165-0173(93)90013-P Intelligence as a confounding factor in the IAT effect. J. Psychol. 218, Robinson, T. E., and Berridge, K. C. (2001). Incentive-sensitization and addiction. 51–53. doi: 10.1027/0044-3409/a000008 Addiction 96, 103–114. doi: 10.1080/09652140020016996 Wentura, D., and Degner, J. (2010). “A practical guide to sequential priming Robinson, M. J. F., and Berridge, K. C. (2013). Instant transformation of learned and related tasks” in Handbook of implicit social cognition: Measurement, repulsion into motivational "wanting". Curr. Biol. 23, 282–289. doi: 10.1016/j. theory, and applications. eds. B. Gawronski and B. K. Payne (New York, cub.2013.01.016 NY: Guilford Press), 95–116. Frontiers in Psychology | www.frontiersin.org 15 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures Wentura, D., and Rothermund, K. (2007). “Paradigms we  live by: a plea for evaluation processes. Soc. Cogn. Aeff ct. Neurosci. 8, 647–653. doi: 10.1093/ more basic research on the implicit association test” in Implicit measures scan/nss042 of attitudes: Procedures and controversies. eds. B. Wittenbrink and N. S. Zanon, R., De Houwer, J., Gast, A., and Smith, C. T. (2014). When does Schwarz (New York, NY: Guilford Press), 195–215. relational information influence evaluative conditioning? Q. J. Exp. Psychol. West, S. G., and Brown, T. J. (1975). Physical attractiveness, the severity of 67, 2105–2122. doi: 10.1080/17470218.2014.907324 the emergency and helping: a field experiment and interpersonal simulation. Zhang, J., Berridge, K. C., Tindell, A. J., Smith, K. S., and Aldridge, J. W. J. Exp. Soc. Psychol. 11, 531–538. doi: 10.1016/0022-1031(75)90004-9 (2009). A neural computational model of incentive salience. PLoS Comput. Wicker, A. W. (1969). Attitudes versus actions: the relationship of verbal and Biol. 5:e1000437. doi: 10.1371/journal.pcbi.1000437 overt behavioral responses to attitude objects. J. Soc. Issues 25, 41–78. doi: 10.1111/j.1540-4560.1969.tb00619.x Conflict of Interest: The authors declare that the research was conducted in Wiers, R. W., Van Woerden, N., Smulders, F. T. Y., and De Jong, P. J. (2002). the absence of any commercial or financial relationships that could be  construed Implicit and explicit alcohol-related cognitions in heavy and light drinkers. as a potential conflict of interest. J. Abnorm. Psychol. 111, 648–658. doi: 10.1037//0021-843X.111.4.648 Wigboldus, D. H. J., Dijksterhuis, A., and van Knippenberg, A. (2003). When Copyright © 2019 Meissner, Grigutsch, Koranyi, Müller and Rothermund. This is stereotypes get in the way: stereotypes obstruct stereotype-inconsistent trait an open-access article distributed under the terms of the Creative Commons Attribution inferences. J. Pers. Soc. Psychol. 84, 470–484. doi: 10.1037/0022-3514.84.3.470 License (CC BY). The use, distribution or reproduction in other forums is permitted, Wilson, T. D., Lindsey, S., and Schooler, T. Y. (2000). A model of dual attitudes. provided the original author(s) and the copyright owner(s) are credited and that Psychol. Rev. 107, 101–126. doi: 10.1037/0033-295X.107.1.101 the original publication in this journal is cited, in accordance with accepted academic Wiswede, D., Koranyi, N., Müller, F., Langner, O., and Rothermund, K. (2013). practice. No use, distribution or reproduction is permitted which does not comply Validating the truth of propositions: behavioral and ERP indicators of truth with these terms. Frontiers in Psychology | www.frontiersin.org 16 November 2019 | Volume 10 | Article 2483 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Frontiers in Psychology Pubmed Central

Predicting Behavior With Implicit Measures: Disillusioning Findings, Reasonable Explanations, and Sophisticated Solutions

Frontiers in Psychology , Volume 10 – Nov 8, 2019

Loading next page...
 
/lp/pubmed-central/predicting-behavior-with-implicit-measures-disillusioning-findings-RRX360y101

References (172)

Publisher
Pubmed Central
Copyright
Copyright © 2019 Meissner, Grigutsch, Koranyi, Müller and Rothermund.
ISSN
1664-1078
eISSN
1664-1078
DOI
10.3389/fpsyg.2019.02483
Publisher site
See Article on Publisher Site

Abstract

REVIEW published: 08 November 2019 doi: 10.3389/fpsyg.2019.02483 Predicting Behavior With Implicit Measures: Disillusioning Findings, Reasonable Explanations, and Sophisticated Solutions 1 1 1 2 Franziska Meissner *, Laura Anne Grigutsch , Nicolas Koranyi , Florian Müller and Klaus Rothermund 1 2 General Psychology II, Institute of Psychology, Friedrich Schiller University Jena, Jena, Germany, Department for the Psychology of Human Movement and Sport, Institute for Sports Science, Friedrich Schiller University Jena, Jena, Germany Two decades ago, the introduction of the Implicit Association Test (IAT) sparked enthusiastic reactions. With implicit measures like the IAT, researchers hoped to finally be able to bridge the gap between self-reported attitudes on one hand and behavior on the other. Twenty years of research and several meta-analyses later, however, we have to conclude that neither the IAT nor its derivatives have fulfilled these expectations. Their predictive value Edited by: for behavioral criteria is weak and their incremental validity over and above self-report Zheng Jin, measures is negligible. In our review, we present an overview of explanations for these Zhengzhou Normal University, China unsatisfactory findings and delineate promising ways forward. Over the years, several Reviewed by: reasons for the IAT’s weak predictive validity have been proposed. They point to four Xiaoming Wang, potentially problematic features: First, the IAT is by no means a pure measure of individual Qufu Normal University, China differences in associations but suffers from extraneous influences like recoding. Hence, Colin Smith, the predictive validity of IAT-scores should not be confused with the predictive validity of University of Florida, associations. Second, with the IAT, we usually aim to measure evaluation (“liking”) instead United States of motivation (“wanting”). Yet, behavior might be determined much more often by the latter *Correspondence: Franziska Meissner than the former. Third, the IAT focuses on measuring associations instead of propositional franziska.meissner@uni-jena.de beliefs and thus taps into a construct that might be too unspecific to account for behavior. Finally, studies on predictive validity are often characterized by a mismatch between Specialty section: This article was submitted to predictor and criterion (e.g., while behavior is highly context-specific, the IAT usually takes Cognitive Science, into account neither the situation nor the domain). Recent research, however, also revealed a section of the journal Frontiers in Psychology advances addressing each of these problems, namely (1) procedural and analytical Received: 05 June 2019 advances to control for recoding in the IAT, (2) measurement procedures to assess implicit Accepted: 21 October 2019 wanting, (3) measurement procedures to assess implicit beliefs, and (4) approaches to Published: 08 November 2019 increase the fit between implicit measures and behavioral criteria (e.g., by incorporating Citation: contextual information). Implicit measures like the IAT hold an enormous potential. In order Meissner F, Grigutsch LA, Koranyi N, Müller F and Rothermund K (2019) to allow them to fulfill this potential, however, we have to refine our understanding of these Predicting Behavior With Implicit measures, and we should incorporate recent conceptual and methodological advancements. Measures: Disillusioning Findings, Reasonable Explanations, and This review provides specific recommendations on how to do so. Sophisticated Solutions. Front. Psychol. 10:2483. Keywords: implicit measures, predictive validity, IAT, attitude-behavior gap, multinomial processing tree models, doi: 10.3389/fpsyg.2019.02483 wanting vs. liking, propositions vs. associations, context-dependency Frontiers in Psychology | www.frontiersin.org 1 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures Why does he  act like this? Why does she not do what she In this regard, dual process or dual system models posit that intended to do? In our everyday life, we  oen ft try to find parts of human behavior can only be  explained with processes explanations for the behavior of others, and of ourselves, that operate below the threshold of personal control and respectively. Explaining and predicting behavior is also of key awareness (e.g., Strack and Deutsch, 2004; Hofmann et  al., interest across all fields of scientific psychology, especially when 2009; Kahneman, 2011), a view that fueled the interest in the it comes to deviations between individuals’ actual behavior “sub”-personal level of behavior control. and the attitudes, goals, or values held by these very individuals. Over the last decades, a number of new attitude measurement Why do people discriminate although they report to hold procedures were introduced that aimed to tap into these processes egalitarian values? Why do they not quit smoking although since they do not rely on introspection (e.g., the IAT, Greenwald they know that smoking is bad? Why is there a gap between et  al., 1998; the Ae ff ctive Priming Paradigm, Fazio et  al., 1986; people’s self-reported attitudes and actual behavior? the Aeff ct Misattribution Procedure, Payne et  al., 2005; for Dual-process or dual-system models attribute seemingly overviews, see Teige-Mocigemba et  al., 2010; Wentura and inconsistent behavior to the triumph of an impulsive system Degner, 2010; Gawronski and De Houwer, 2014; Gawronski over a reflective system of behavior control (e.g., Strack and and Hahn, 2019). Although differing in their procedural details, Deutsch, 2004; Hofmann et  al., 2009; Kahneman, 2011). The all of these measurement procedures involve computerized tasks notion that the prediction of behavior could be  improved requiring individuals to quickly execute a specific response to considerably if one succeeds in measuring the processes of the a set of stimuli. The performance in these tasks is then influenced impulsive system (Hofmann et  al., 2007; Friese et  al., 2008; by stimulus-response compatibility due to the automatic Hofmann and Friese, 2008) fueled research applying so-called evaluations of these stimuli (De Houwer, 2001, 2003a). Hence, implicit measures of attitudes. The most popular of these measures, the scores obtained from the observed performance are the Implicit Association Test (IAT, Greenwald et al., 1998) evoked interpreted in terms of attitude strength. Compared to self- enthusiastic hopes regarding its predictive value. Unfortunately, report measures, these measurement procedures were assumed however, the IAT and its derivatives have not met these expectations. to provide little opportunity to control responding, preventing In this article, we  review findings illustrating reasons for the an influence of deliberate manipulation attempts and self- IAT’s unsatisfying predictive value, as well as promising ways presentational concerns (e.g., Fazio et  al., 1986; Greenwald forward. We  will outline that in order to improve the predictive et  al., 1998). Some even argued that these procedures succeed power of implicit measures, differentiation is key. We  will argue in measuring a unique construct (implicit attitude) that is that future research should put more emphasis on the underlying introspectively less accessible and thus distinct from the construct processes and concepts behind these measures. We  begin with captured in self-report measures (explicit attitude; Greenwald sketching the discrepancy between individuals’ behaviors and and Banaji, 1995; Wilson et  al., 2000; but see Fazio, 2007, for their self-expressed attitudes. We  then summarize the (mostly a different view). Accordingly, researchers oen u ft se implicit unsatisfying) attempts to close this attitude-behavior gap with measures and explicit measures as labels for these measurement the help of implicit measures. In the main part of this article, procedures. Not surprisingly, implicit measures, first and foremost we  identify features of implicit measures that are responsible the IAT (Greenwald et al., 1998), were embraced by the scientific for their weak predictive validity. We  review findings illustrating community since they came along with the potential to measure each of these problematic aspects along with specific, sophisticated the hidden forces of behavior. The hope was that they would solutions providing promising directions for future research. finally enable researchers to understand and to predict individual behavior over and above self-report measures. Unfortunately, the predictive validity of the IAT fell short THE ATTITUDE-BEHAVIOR GAP AND of these expectations. Meta-analytic findings ( Greenwald et  al., 2009; Oswald et  al., 2013; Kurdi et  al., 2019) suggest that the IMPLICIT MEASURES implicit-criterion correlation (ICC) is unsatisfactorily low (average Attitudes and values that people express are oen in co ft nflict r   =  0.27, Greenwald et  al., 2009; average r   =  0.14, Oswald ICC ICC with their actual behavior. Indeed, although widely postulated et  al., 2013; 90-percent prediction interval for ICCs from to be  associated with cognitive processes, judgments, and most r  =  −0.14 to r  =  0.32; Kurdi et  al., 2019). Equally upsetting importantly, behavior (e.g., Katz, 1960; Fazio et al., 1983; Ajzen, is the fact that the incremental predictive validity over and 1991), self-reported attitudes show weak predictive validity at above self-report measures is obviously negligible (i.e., ranging best (with correlation coefficients being “rarely” above r  =  0.30, between 1 and 5%; Greenwald et  al., 2009; Oswald et  al., 2013; Wicker, 1969; see also Kraus, 1995, who found an average Kurdi et al., 2019). Such a disappointingly low predictive validity r  =  0.38). How can we  close this attitude-behavior gap? A is a frustrating state of affairs, especially because it was the prominent way forward relied on the assumption that people low predictive value of self-reported attitudes that initiated the might not be  able to report on their mental processes in an development of implicit measures like the IAT in the first place. accurate fashion (e.g., Nisbett and Wilson, 1977), implying What are the reasons for the weak relationship between that self-reports can never achieve convincing predictive value. implicit measures and behavioral criteria? An obvious candidate Instead, “introspectively unidentified (or inaccurately identified) is a potential lack of internal consistency in the predictor variables. traces of past experience” (Greenwald and Banaji, 1995, p.  5) Unfortunately, reporting reliability coefficients is by no means were proposed to be  more crucial precursors of behavior. the rule for studies on predictive validity. Nevertheless, over time, Frontiers in Psychology | www.frontiersin.org 2 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures the picture emerged that implicit measures oen s ft uffer from them could be  responsible for the IAT’s weak predictive power. low internal consistency (for overviews, see Gawronski and De This however does not exclude the possibility that researchers Houwer, 2014; Gawronski and Hahn, 2019). High amounts of might have to address several (if not all) of these features in measurement error in the resulting scores, however, shuffle the order to achieve the desired results. In the remainder of this rank order of individuals, and thus constitute a serious problem article, we  explain all of these potentially problematic features when it comes to predicting relevant criteria like behavior (for in detail, along with promising ways forward and specific an elaboration on further consequences of low reliability, see recommendations for future research. LeBel and Paunonen, 2011; but see also De Schryver et  al., 2016). Reliability, however, seems to be  less of an issue for the most popular implicit measure, the IAT (Greenwald et al., 1998). ISSUE 1: EXTRANEOUS INFLUENCES On the contrary, IAT scores typically achieve acceptable levels ON IMPLICIT MEASURES of reliability, and outperform other implicit measures in terms of internal consistency and test-retest reliability (e.g., Nosek Implicit measures (just like explicit ones) should not et  al., 2007; Gawronski and De Houwer, 2014; Gawronski and be  understood as process-pure measures of attitudes. They are Hahn, 2019). Note, however, that it has also been suggested characterized by additional, non-attitudinal influences, and this that the comparatively high internal consistency of the IAT kind of error variance reduces their predictive validity. This might be  due to systematic error variance (so-called method also applies to the IAT (Greenwald et  al., 1998), one of the variance; see below) rather than construct-related variance (e.g., most popular implicit measures. Teige-Mocigemba et  al., 2010; Kraus and Scholderer, 2015). If e I Th AT involves two binary classification tasks, a target this holds true, given that method-related variance is unlikely task and an attribute task, that have to be  performed with to explain behavior, it is not surprising that the IAT’s predictive two response keys. Importantly, the key assignment varies across validity turned out to be  bounded. So, even for the IAT, the two IAT test blocks. In the compatible block, participants (a lack of) reliability might be  part of the problem. are instructed to press one key for the positively evaluated For the remainder of this article, however, we  put reliability target category (e.g., flower) as well as the positive pole of issues aside, and instead focus on four potentially problematic the attribute dimension (e.g., positive), and to press the other features of implicit measures and, in particular, of the IAT. key for the more negatively evaluated target category (e.g., We  will review relevant findings as well as theoretical insect) as well as the negative pole of the attribute dimension considerations, and we  will outline that each of these features (e.g., negative). In the incompatible block, negative targets and might be  responsible for the IAT’s weak predictive validity: positive attributes are assigned to the same key (and positive First, the IAT turned out not to be  a process-pure measure targets and negative attributes to the other key, respectively). of attitudes. Instead, non-attitudinal influences also play a role Participants typically respond faster and more accurate in (for an overview of these and other shortcomings of the IAT compatible compared to incompatible IAT blocks. The and its derivatives, see Fiedler et  al., 2006; Teige-Mocigemba performance difference between compatible and incompatible et  al., 2010; Gawronski and Hahn, 2019). If we  want to predict blocks (compatibility effect , IAT effect, or IAT score) is then individual’s behavior, we have to filter out this construct-irrelevant interpreted as a measure for the strength of associations between variance. Second, the IAT (just as most implicit measures) the respective categories (Greenwald et  al., 1998) . focuses on evaluation rather than motivation. However, people During the 20 years since its introduction, however, numerous do not always want what they like (and vice versa). We  should findings challenged the IAT’s construct validity (for an overview, thus not confuse liking with wanting (e.g., Tibboel et  al., see Teige-Mocigemba et  al., 2010). An illustrative example is 2015b), and in many situations, the latter might actually be more the finding that content-unrelated IATs (i.e., two IATs that relevant in driving behavior than the former. Third, as disclosed involve non-overlapping target concepts) share a considerable by its very name, the IAT was introduced to quantify associations. amount of variance (so-called method variance; e.g., Greenwald Associations, however, might be too unspecific to unambiguously et  al., 1998; McFarland and Crouch, 2002; Mierke and Klauer, relate to and account for a particular behavior in a specific 2003; Back et  al., 2005; Klauer et  al., 2010). In search for an situation. Instead, (implicit) propositional beliefs could be  a explanation for this shared method variance, several groups of more plausible precursor of behavior (e.g., Hughes et al., 2011). researchers proposed attitude-unrelated processes that ae ff ct IAT Finally, when applying the IAT researchers typically aim at responding, such as general processing speed (McFarland and assessing attitudes or stereotypes globally, that is, in a context- Crouch, 2002; Blanton et  al., 2006) or executive functions like independent fashion. Mental representations of attitudes and task-switching ability (Klauer et  al., 2010; Ito et  al., 2015). stereotypes, however, are highly context-dependent. Similarly, Another potential flaw of the IAT is the fact that it suffers real-life behavior does not occur in a situational vacuum. The from usually unwanted block order effects: IAT scores turn out predictive validity of implicit measures like the IAT might We are aware that a couple of researchers actually exercise due caution when thus be  improved by increasing the match between predictor interpreting IAT scores, understanding them as response time differences in and criterion (i.e., overcoming the lack of specificity in the a computerized categorization task – no more, no less. However, the majority predictor by incorporating contextual information). of researchers do interpret IAT scores as reflecting associative strength or Note that we  do not want to imply any order or priority implicit bias. Aer a ft ll, the IATs very name suggests such an interpretation. with regard to these four issues. We  will outline that each of In this paper, we  therefore proceed from this more common viewpoint. Frontiers in Psychology | www.frontiersin.org 3 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures to be  larger if participants started with the compatible block Whether it is valence, salience or some other feature, if the task (e.g., Greenwald et  al., 1998; Nosek et  al., 2005; for a possible was recoded, responses are based on the shared feature, and explanation, see Klauer and Mierke, 2005). Finally, IAT scores thus necessarily unrelated to the (attitudes toward the) nominal do not only reflect the valence of the target categories but can categories (e.g., faces in a Black-White IAT are no longer processed also be  influenced by stimulus effects (e.g., Steffens and Plewe, as Black vs. White but rather as more vs. less salient, Kinoshita 2001; Mitchell et  al., 2003; Govan and Williams, 2004; and Peek-O’Leary, 2005). Even more important, recoding should Bluemke and Friese, 2006; Gast and Rothermund, 2010). not be  understood as a more or less constant error that boosts Summing up, numerous studies revealed that IAT scores do IAT scores equally for everyone. Instead, there might be  inter- not reflect pure attitude strength but also contain systematic individual differences in recoding [e.g., due to individual differences error variance. The mere amount and variety of different findings in familiarity, Greenwald et al., 1998 (Exp. 2), salience, Rothermund (for an overview, see Teige-Mocigemba et  al., 2010) is not and Wentura, 2004 (Exp.’s 2A and 2B), or fluid intelligence, von particularly easy to grasp. In the following, however, we  outline Stülpnagel and Steffens, 2010 ] that can be  unrelated to the to-be- that there is a common core behind these additional processes: measured attitudes. In this sense, recoding represents a source recoding (e.g., De Houwer, 2003b; Wentura and Rothermund, 2007; of variance that might distort the predictive validity of the IAT Rothermund et  al., 2009). score for behavioral criteria. For more detailed elaborations on this issue, and for findings of recoding being unrelated to the construct of interest (i.e., attitudes), we  refer to the work of The Role of Recoding in the Implicit Meissner and Rothermund (2013, 2015a,b). Association Test Recoding can be  understood as the most crucial extraneous Although instructed to perform a double categorization task, influence in the IAT because it can account for other extraneous participants can oen e ft asily simplify the IAT through so-called influences that were identified throughout the last couple of task recoding. Recoding refers to a combination of targets and years. As an example, consider the negative correlation of IAT attributes to superordinate categories. It is based on some scores with task-switching ability (e.g., Klauer et al., 2010). Task- degree of similarity in the IAT’s stimulus material, that is, switching ability, that is, high cognitive flexibility, enables fast some feature that targets and attributes share. In a flower-insect and effortless switches between two tasks. Therefore, high switching IAT, for example, participants can profoundly simplify the task ability reduces switch costs between the two classification tasks in the compatible block by categorizing each stimulus according in the IAT (i.e., between target and attribute classification). This to its valence, and ignoring the fact that some stimuli should is especially helpful in the incompatible block of the IAT, where actually be  categorized according to their identity (i.e., flowers participants have to perform the double categorization task. In vs. insects). If the task is recoded in this sense, the compatible the compatible block, on the other hand, the task can be simplified block involves only one and the same binary decision (i.e., is by recoding. If they engage in recoding, people no longer switch the current stimulus positive or negative?). In the incompatible between the two tasks: By combining pairs of targets and attributes block, on the other hand, the incongruent response assignment into superordinate categories, they now perform only a single prevents recoding. Here, participants have no choice but to binary decision. Consequently, people with high vs. low switching follow the instructed, rather difficult double categorization task ability will perform equally well in the compatible IAT block. (i.e., flowers vs. insects, and positive vs. negative). Recoding thus results in a negative correlation of switching Recoding thus results in a substantial block difference in ability and IAT scores. Similarly, the relationship between IAT task difficulty, and therefore accounts for the observed block scores and general processing speed (e.g., McFarland and Crouch, difference in response times and error rates (e.g., Rothermund 2002) can be  explained with recoding as well. Finally, it has et  al., 2009). Remarkably, it has been shown that even in the been shown that task recoding can also account for stimulus absence of any category-based associations, recoding processes effects in the IAT (e.g., Gast and Rothermund, 2010). produce significant IAT scores (e.g., Mierke and Klauer, 2003; At this point, it should be  clear that the IAT score should Rothermund and Wentura, 2004; De Houwer et  al., 2005). be  understood as a mixture of both relevant influences (e.g., Note that recoding must not be  based on stimulus valence. associations) and irrelevant influences, first and foremost, recoding. Instead, every feature that is shared by targets and attributes If researchers want to increase the IAT’s predictive validity, they might be  used to form superordinate categories (e.g., salience, should thus try to separate effects of associations from the familiarity, valence, or even perceptual features like color or influence of recoding. In the past few years, two different shape; Rothermund et  al., 2009; see also Mierke and Klauer, approaches were introduced that claim to do so: The first approach 2003; Rothermund and Wentura, 2004; De Houwer et  al., 2005; aims at minimizing recoding processes by modifying the IAT Kinoshita and Peek-O’Leary, 2006; Chang and Mitchell, 2009) . procedure. The second approach disentangles associations and recoding processes with the help of multinomial modeling. In the following, we will present a short overview of these suggestions. e Th recoding account subsumes two earlier process models for the IAT: the so-called figure-ground account ( Rothermund and Wentura, 2001, 2004; Rothermund et  al., 2005; see also Chang and Mitchell, 2009; Kinoshita & Peak- A Solution: Dropping the Block Structure O’Leary, 2006; Mitchell, 2004) and the task-switching account (Mierke and Klauer, As outlined above, recoding effects in the IAT can be  traced 2001,2003; Klauer and Mierke, 2005). For an overview of these and other process back to its characteristic structure: the arrangement of trials accounts for the IAT we  refer to the work of Teige-Mocigemba and colleagues in (compatible vs. incompatible) blocks. When it comes to (Teige-Mocigemba et  al., 2010; Teige-Mocigemba and Klauer, 2015). Frontiers in Psychology | www.frontiersin.org 4 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures reducing the influence of recoding, an obvious possible solution e s Th econd approach also dealing with the problem of is thus to simply omit this structure. In this regard, several recoding follows a different rationale. Instead of trying to variants of the IAT have been introduced that dropped the reduce the influence of recoding, it assumes that IAT scores characteristic block structure, and varied response compatibility result from a mixture of different processes. As summarized within one test block instead: the Single-Block IAT (SB-IAT, in the following section, this approach then relies on mathematical Teige-Mocigemba et  al., 2008) and the Recoding-Free IAT modeling to measure each of these processes. This allows (IAT-RF, Rothermund et al., 2009) . While the category-response researchers to separately examine the predictive power of both assignment is constant throughout a block of trials in the construct-related and method-related variance due to recoding. standard IAT, it varies randomly from trial to trial in the Another Solution: Adopting newer IAT variants. Consequently, scores in those procedures are obtained by computing performance differences between a Modeling Approach compatible and incompatible trials rather than between Recently, a multinomial processing tree model has been compatible and incompatible blocks. introduced that enables a remarkably fine-grained analysis of In these IAT variants, participants are informed about the the IAT: The ReAL model ( Meissner and Rothermund, 2013). current category-response assignment either by simply showing Most importantly, this model successfully disentangles the effects it shortly before the stimulus appears (IAT-RF) or by using of evaluative associations from the distorting influence of task stimulus position as a cue (with an appearance in the upper recoding. In this section, we  provide a brief overview of the half of the screen signaling a compatible assignment, and an ReAL model’s basic idea, and we  review relevant findings appearance in the lower half of the screen indicating an concerning (improvements on) the IAT’s validity. incompatible assignment; SB-IAT). Crucially, the upcoming e R Th eAL model assumes that the observable responses in category-response assignment is not predictable. Consequently, the IAT (i.e., correct and incorrect responses in compatible and a stable and efficient recoding strategy specifically for the incompatible blocks) result from the interplay of specific compatible assignment becomes much harder than in the standard unobservable processes (e.g., associations and recoding; see below). IAT. This reasoning was supported by Rothermund et  al. (2009) es Th e processes are represented by separate model parameters; who found that dropping the IAT’s block structure successfully their assumed interplay is displayed in a tree architecture (i.e., reduces switch cost asymmetries, a marker of recoding processes. the multinomial processing tree). Based on observed response Besides reducing the effects of recoding, the block-free IAT patterns, algorithms estimate values for the model parameters variants come with some further advantages. For example, which are then interpreted as measures for the respective cognitive block order effects which usually influence conclusions in the processes (for mathematical details on multinomial processing standard IAT (e.g., Greenwald et  al., 1998) are no longer an tree models, see Riefer and Batchelder, 1988; Hu and Batchelder, issue. Furthermore, the newer IAT variants eliminate method- 1994; Batchelder and Riefer, 1999; for reviews of applications, related variance (Teige-Mocigemba et  al., 2008) and stimulus see Erdfelder et  al., 2009; Klauer et  al., 2012). effects ( Gast and Rothermund, 2010). These findings also support e Th ReAL model distinguishes three different processes: the assumption that recoding is one of the most crucial validity recoding (Re), evaluative associations (A) and the resource- threats of the IAT. Finally, the block-free IAT variants are not consuming label-based identification of the correct response ( L). only correlated with behavioral criteria (Teige-Mocigemba et al., e t Th ree structure incorporates theoretical assumptions concerning 2008; Houben et al., 2009), there is also evidence that dropping these processes. For example, the ReAL model assumes that the block structure of the IAT can actually improve its predictive task recoding determines responding for both targets and attributes validity (Kraus and Scholderer, 2015). but only in one of the IAT blocks (i.e., in the compatible block) . Despite these strengths of SB-IAT and IAT-RF, the strategy Evaluative associations, on the other hand, are assumed to to minimize recoding effects by dropping the IAT’s block structure influence responding in both compatible and incompatible bears the risk to miss potentially interesting effects. Although blocks but they should be  triggered only in target trials, not recoding processes do not represent the construct that researchers in attribute trials (reflecting the understanding of attitudes as typically attempt to measure when employing the IAT, they evaluative associations triggered by an attitude object, not vice might nevertheless represent variance that is related to criteria versa; Fazio et  al., 1986; see also Anderson, 1983). of interest. It has been proposed, for example, that recoding As a multinomial model, the ReAL model is able to disentangle could reflect explicit attitudes ( Rothermund et  al., 2009) and multiple cognitive processes accounting for the same observable that occasionally, it might be  related to relevant criteria (e.g., response (Batchelder and Riefer, 1999). First and foremost, the behavior; Rothermund et al., 2005; Teige-Mocigemba et al., 2008). ReAL model controls for the effects of recoding by measuring 3 4 Note that there is another procedure that dropped the IAT’s block structure, Note that for many IATs, we  do not know a priori which of the two blocks namely, the Extrinsic Aeff ctive Simon Task (EAST, De Houwer, 2003b; see will be  simplified by recoding. Even within one sample, some participants also its close cousin, the Identification EAST, De Houwer and De Bruycker, might recode the task in one IAT block (e.g., in the Black/positive block), 2007). Importantly, however, the EAST does not contain classification responses others will do so in the other block (i.e., the White/positive block). The ReAL based on the target categories and is thus strongly susceptible to stimulus model accounts for these differences by making use of the task switch cost effects (Gast and Rothermund, 2010). Furthermore, it suffers from low reliability effect as a marker for recoding processes. More precisely, the sign of the (De Houwer, 2003b). We  therefore consider the EAST a less recommendable individual switch cost effect determines the block in which the Re parameter approach to account for the problem of recoding. is modeled (for more details, see Meissner and Rothermund, 2013, 2015b). Frontiers in Psychology | www.frontiersin.org 5 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures them in a separate model parameter (which clearly represents ISSUE 2: DISTINGUISHING BETWEEN a unique feature as compared to other mathematical models LIKING AND WANTING for the IAT; e.g., the quad model, Conrey et  al., 2005; or the diffusion model, Klauer et  al., 2007). Besides addressing the Insights from recent neuropsychological research raise the question problem of recoding, the ReAL model comes with another whether evaluations are indeed the driving force behind behavioral advantage: While IAT scores only reflect relative preferences impulses. According to the incentive salience hypothesis (Robinson (which could be  problematic; for an overview, see Teige- and Berridge, 1993, 2001; Berridge and Robinson, 2003; Berridge, Mocigemba et  al., 2010), the ReAL model provides separate 2009), liking an object and wanting it are separable processes association parameters for each of the two target categories. that are mediated by different brain substrates and are differentially Consequently, the model can successfully handle situations where ae ff cted by various factors. Whereas “liking” refers to the hedonic both attitude objects trigger equally strong positive, negative, aspects of a stimulus (i.e., the pleasure or positive aeff ct it or neutral associations. Note that the conventional IAT score causes), “wanting” is the result of the attribution of incentive would only yield a null effect in these cases (i.e., no preference). salience. The latter describes a particular quality that, when Numerous studies revealed that the ReAL model parameters added to the mental representation of a given stimulus, transforms are valid measures of the processes they stand for (Meissner the mere sensory percept of this stimulus to become attention- and Rothermund, 2013; Meissner and Rothermund, 2015a,b; see grabbing, attractive, and potent to elicit behavioral impulses of also Koranyi and Meissner, 2015; Jin, 2016). Most importantly, approach or consumption, which is the very essence of behavioral the ReAL model’s association parameters reflect the direction motivation (Berridge and Robinson, 2003; Berridge, 2009). and the strength of evaluative associations for each of the two Importantly, while “wanting” and “liking” should generally target concepts (Meissner and Rothermund, 2013). This holds covary (i.e., the strength of “wanting” experienced for a specific true even in IAT applications where recoding processes pushed object should be proportional to the hedonic “liking” it produces), the overall IAT score in the opposite direction (Meissner and there are specific classes of stimuli and situations where the Rothermund, 2015a). The association parameters turned out to two processes can become uncoupled. The most prominent be  sensitive to manipulations of evaluation (Meissner and example for such a dissociation is the case of addiction, where Rothermund, 2013) but immune against artificial, non-evaluative “wanting” for the addictive drug is extremely enhanced long influences (i.e., salience asymmetries and modality match effects; aer ft it ceases to evoke hedonic experiences (i.e., “liking”), and Meissner and Rothermund, 2015a,b). Additionally, and in line even despite the addict’s recognition of its harmful effects with theoretical considerations (e.g., Fazio and Towles-Schwen, (Robinson and Berridge, 1993; Stacy and Wiers, 2010). Even 1999), association parameters correlated with self-reported attitudes though momentary dissociations of “wanting” and “liking” are in non-sensitive attitude domains (consumer preferences; Meissner at the heart of many chronic clinical psychological conditions and Rothermund, 2013). Finally, Meissner and Rothermund (e.g., Rømer Thomsen et  al., 2015 ; Olney et  al., 2018), they (2013) also tested the predictive validity of the model’s association are not in themselves pathological (Dill and Holton, 2014). parameters. As expected, the amount of chocolate consumed Rather, the closeness of the relationship between “wanting” and while watching a video was successfully predicted by the ReAL “liking” fluctuates in healthy individuals ( Epstein et  al., 2003; model’s association parameter (estimated from the response Hobbs et  al., 2005; Dai et  al., 2010, 2014; Litt et  al., 2010). pattern in a fruit-chocolate IAT). Note that the behavior was An illustrative example is the moment aer fini ft shing a delicious unrelated to the recoding parameter and also unrelated to the meal. While “liking” for the food will be unaltered, being satiated conventional IAT score (i.e., the D score; Meissner and Rothermund, will reduce “wanting” more of it (Kraus and Piqueras-Fiszman, 2013). When it comes to increasing the IAT’s predictive validity, 2016; Stevenson et al., 2017). However, not only states of satiation an application of the ReAL model thus constitutes a promising and deprivation have differential effects on “wanting” and “liking.” step forward. Given the recent developments in the field of It has also been shown, for instance, that stress increases multinomial processing tree models (i.e., allowing the incorporation “wanting” but not ‘liking’ for sweet rewards (Pool et  al., 2015). of response time data, Heck and Erdfelder, 2016; Klauer and To sum up, “wanting” and “liking,” though typically highly Kellen, 2018; and a sophisticated treatment of possible parameter correlated, can diverge. Whenever they do, “wanting” is much heterogeneity, e.g., Klauer, 2010; Matzke et  al., 2015) further more likely to guide behavior than “liking” (Berridge et  al., improvements are to be  expected. Given that the ReAL model 1989; Peciña et  al., 2003). Researchers interested in predicting has already outperformed the IAT score with regard to construct behavior are therefore well advised to incorporate measures validity in a number of studies (e.g., Meissner and Rothermund, of “wanting” (Lades, 2012). 2013, 2015a,b), we  recommend researchers to consider an application of the ReAL model as an alternative, or at least as Initial Attempts in Assessing “Wanting” an additional analysis tool for the IAT in their studies. How do we measure “wanting”? Self-reports are not an advisable So, we cannot deny that extraneous influences on IAT scores option. Obviously, they involve the risk of potential distortions like recoding do exist. However, there are promising approaches due to self-presentational concerns, especially when it comes to to address this problem. With procedural modifications or sensitive topics. Apart from that, however, disentangling “wanting” mathematical modeling, we  can measure more validly what and “liking” on a semantic level is complicated. Participants people actually like. But what if it is irrelevant what people might fail to grasp the distinction or simply confuse the two like? Maybe it is more important what people want? processes since the consideration of wanting as independent Frontiers in Psychology | www.frontiersin.org 6 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures from liking violates laymen’s intuition. Furthermore, as pointed To achieve this, several adjustments to the conventional out by Pool et  al. (2016), it is likely that self-reported “wanting” IAT procedure are necessary. First, instead of valence (as in primarily reflects expected pleasantness, and is inferred from traditional IATs), or purely semantic meaning (as in previous past hedonic experiences (i.e., “liking”). Actual implicit “wanting,” attempts at creating a “wanting” IAT), the relevant criterion on the other hand, should in principle be  independent from for the categorization of attribute stimuli in the W-IAT must any hedonic aspects of reward (Robinson and Berridge, 2013). consist in participants’ “wanting” for them, or lack thereof, Several researchers have therefore turned to established respectively. This entails the need for a set of attribute stimuli implicit measurement procedures, most oen t ft he IAT, in order that is potent to trigger acute bursts of “wanting,” and another to develop a measure of implicit “wanting” as distinct from that is not. Second, execution of the required response for implicit “liking” (for an overview, see Tibboel et  al., 2015b). wanted stimuli must acquire the quality of an actual “wanting”- By now, several IAT variants have been introduced that aim triggered consummatory response. to measure implicit “wanting” for a given target dimension As for the first requirement, it must be  considered that of interest (e.g., alcohol vs. no alcohol, smoking vs. no smoking, being “wanted” is not an inherent property of any specific attractive vs. unattractive persons). All of these approaches stimulus, but instead hinges on its interaction with the individual’s share one basic idea. That is, in order to transform the IAT current psychological or physiological state (Zhang et  al., 2009; into a measure of implicit “wanting” the category labels of Robinson and Berridge, 2013). Thus, to ensure “wanting” for the evaluative attribute dimension have to be  replaced with one set of attribute stimuli in the W-IAT, a physiological need concepts representing some aspect of “wanting.” Based on the state is induced in participants before completion of the W-IAT, notion that “wanting” entails the urge to approach the object and one set of attribute stimuli is selected to be  highly relevant in question, Palfai and Ostafin (2003) for instance, introduced for satisfying this very need. Specifically, before starting the an IAT that employs the attribute categories “approach” and W-IAT, participants are made thirsty with salty snacks. Attribute “avoidance,” with semantically related words (e.g., advance, stimuli in the following W-IAT then consist of images of drinks withdraw) as stimulus material (see Kraus and Scholderer, 2015, (need-relevant) and neutral objects (need-irrelevant). The attribute for a similar approach using the IAT-RF). In a similar vein, task in the W-IAT is then to sort these stimuli into the categories Wiers et  al. (2002) developed an IAT employing the attribute “I want” (for drinks) and “I don’t want” (for neutral objects). categories “active” and “passive” represented by arousal and Executing this categorization is then transformed into a sedation-related words. Tibboel et  al. (2011, 2015a), on the consummatory response by making “I want”-key presses other hand, used “I want” vs. “I do not want” as attribute instrumental for need satisfaction. More precisely, whenever categories in their IAT with positive vs. negative (e.g., holiday, participants correctly press the “I want”-key in response to pain; Tibboel et  al., 2011), or motivational words (e.g., gain pictures of drinks, they gain a small amount of water for later vs. deprivation; Tibboel et  al., 2015a) as stimulus material. consumption. To further increase the consummatory character However, there are reasons to doubt the validity of these of the “I want” response, this gain is signaled by immediate attempts at creating a measure of implicit “wanting.” For example, visual and auditory action effects: a small glass appears in the in situations that should actually reveal a dissociation of “wanting” lower part of the screen, and a drinking-related sound (e.g., and “liking,” these IAT variants designed to measure “wanting” cork popping and/or gurgling water) is presented via headphones. typically reveal a high overlap with “liking” measures (for an e p Th otential of this new W-IAT was illustrated in a study overview, see Tibboel et  al., 2015b). Obviously, changing the on attraction in a mating context (Koranyi et  al., 2017). attribute categorization task on a merely semantic level by simply Heterosexual male participants completed the previously replacing the category labels cannot transform the IAT into an described W-IAT procedure as well as a standard valence IAT implicit measure of “wanting.” If anything, these IATs most likely (i.e., positive vs. negative attribute dimension). Target stimuli reflect semantic associations, or a “cognitive form of wanting” in both IATs were very attractive vs. less attractive faces. IAT (Tibboel et  al., 2015b, p.  189). Recently, however, a new scores should therefore reflect participants’ implicit “wanting” Wanting-IAT was introduced (Koranyi et  al., 2017) that can and “liking” for those faces. Importantly, however, half of the be considered a more promising way forward in multiple respects. target faces were male, while the other half was female. The study revealed the expected dissociation of “wanting” and A Solution: The Wanting Implicit “liking”: Both attractive male and attractive female stimuli Association Test elicited “liking” (as measured by the standard valence IAT) e b Th asic idea of the Wanting-IAT (W-IAT, Koranyi et  al., but only attractive female (not male) faces triggered “wanting” 2017) consists in endowing the attribute discrimination task (as measured by the W-IAT). In other words, the results show with motivational character. More precisely, execution of one a general positive evaluation of attractiveness, irrespective of of the attribute responses should come to equal execution of gender, while an implicit wanting can only be found for attractive a “wanting”-triggered consummatory response. Relative “wanting” opposite-sex targets (Dai et  al., 2010). for a pair of target concepts could then be  assessed in the Note that this study additionally employed another version form of stimulus–response-compatibility effects ( De Houwer, of the wanting IAT, namely a variant that used only the semantic 2001, 2003a) by comparing the speed and accuracy of responses labels “I want” and “I do not want” without bestowing any when either of the two target categories is mapped onto the additional motivational meaning onto the attribute discrimination established “wanting” response key. task. This variant yielded the same effects as the standard Frontiers in Psychology | www.frontiersin.org 7 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures valence IAT. This detail in the results underpins the assumption together will lead to a negative evaluation of the CS (Fiedler that purely semantic “wanting” measures fail to dissociate and Unkelbach, 2011; see also Peters and Gawronski, 2011; Förderer themselves from comparable measures of “liking” (c.f., Tibboel and Unkelbach, 2012; Zanon et  al., 2014; Van Dessel et  al., 2018). et  al., 2011, 2015a). The findings of Koranyi et  al. (2017) thus Associations as they should be  measured by implicit suggest that an implicit measure of “wanting” should establish measurement procedures do not contain qualitative relational the motivational quality of relevant responses. information. Therefore, a given association between two concepts The validity of the W-IAT was further corroborated in can reflect different, sometimes even opposite beliefs. For example, a study that compared smokers’ and nonsmokers’ “wanting” “I” and “good” may be  associated either because I  believe that and “liking” for smoking cues (Grigutsch et  al., 2019). This I  am  good, or because I  believe that I  am  no good, or because study revealed that the W-IAT is better suited to discriminate I  would desperately like to be  good, or because I  know that between smokers and nonsmokers than a standard valence others would like me to be  good (see also De Houwer, 2014; IAT tapping “liking.” Specifically, W-IAT scores were positive De Houwer et  al., 2015). This raises the question whether the for smokers but negative for nonsmokers, while “liking”-IAT weak predictive validity of implicit measures of associations (e.g., scores were negative for both groups. Furthermore, in line Greenwald et  al., 2009; Oswald et  al., 2013) is due to the fact with the notion of an addiction-related decoupling of “wanting” that associations are simply unspecific. Some researchers even and “liking,” the correlation of W-IAT and “liking”-IAT was argued that the attempt to predict behavior with associations significantly weaker for smokers than for nonsmokers. In must fail because all information stored in memory is inherently contrast to previous attempts at this matter, the W-IAT propositional (e.g., Hughes et  al., 2011; De Houwer, 2014). The thus proved to measure actual “wanting” instead of purely latter, however, is part of an ongoing debate in the literature, semantic associations (c.f., Palfai and Ostafin, 2003; Tibboel and we  will not address it in detail in this overview. Still, what et  al., 2011, 2015a) both in situations where “liking” is high remains is that (measures of) associations are ambiguous with (Koranyi et  al., 2017) and in situations where “liking” is regard to the qualitative relation between the concepts involved, low (Grigutsch et  al., 2019). and that this could be responsible for the weak predictive validity So, when behavior is not in line with attitudes or values, of implicit measures. The attitude-behavior gap might be addressed this might be  due to a dissociation of “wanting” and “liking.” more convincingly with implicit measures of propositional beliefs Implicit measures of “wanting,” first and foremost those that instead of associations. actually realize a wanting quality (i.e., the W-IAT), are a promising alternative to existing measures of implicit “liking” A Solution: Implicit Measures of Beliefs when it comes to closing the attitude-behavior gap. e n Th otion of implicit measures of beliefs represents a relatively recent development (Barnes-Holmes et  al., 2010; De Houwer et  al., 2015; Müller and Rothermund, 2019). Although the ISSUE 3: FOCUS ON ASSOCIATIONS procedural details of these different measures vary, they all capitalize on the finding that during an evaluative processing of VERSUS BELIEFS propositions (e.g., “Milk is not white.”) beliefs about the truth Interestingly, when researchers started to engage in the of these propositions (i.e., “False”) are activated automatically development of implicit measurement procedures, many also (e.g., Wiswede et  al., 2013). In contrast to established implicit changed the focus with regard to the construct they attempted measures of attitudes that do not take into account the specific to measure. Self-report measures assessed complex personal semantic relationship between concepts, implicit measures of beliefs that can be  expressed in propositional statements. With beliefs allow for the assessment of complex propositions. They the development of the IAT and other implicit measures (e.g., naturally employ more complex stimuli than traditional attitude Aeff ctive Priming, Fazio et  al., 1986), the concept of beliefs measures, that is, combinations of stimuli including their semantic took a backseat in many studies. A lot of researchers now relationship, or even whole sentences. This common basis focused on measuring associations, that is, the mental connection notwithstanding, these measures utilize different approaches to between an object and a given attribute (e.g., positive or negative assess implicit beliefs, each entailing unique advantages as well valence). Such an associative link, however, is unspecific in as shortcomings. In the following, we  provide a brief overview. its nature, and admits several meanings. Implicit Relational Assessment Procedure Ambiguity of Associations In each trial of the Implicit Relational Assessment Procedure From the literature on evaluative learning, we  know that it is (IRAP, Barnes-Holmes et  al., 2010; see also Remue et  al., 2013, not only mere associative co-occurrence that determines valence 2014), participants are presented with two concepts that are transfer from an unconditioned stimulus (US) to a conditioned simultaneously displayed in the top and bottom half of the screen stimulus (CS). Instead, relational qualifiers moderate this (e.g., “I” and “nice” or “I” and “worthless”). Additionally, the relationship. For example, experiencing a neutral person (CS) IRAP highlights the propositional relationship between the two together with a positively evaluated person (US) will result in concepts by presenting a relational qualifier (e.g., “I am nice.” positive evaluations of the CS if the relationship between the or “I am not worthless.”). Participants are instructed to respond two persons is framed as friendship. If the relation between the to these stimuli in a specific manner across the two blocks of two is described as being antagonistic, however, presenting them the task. In a first block they are to classify these stimuli as Frontiers in Psychology | www.frontiersin.org 8 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures true or false (by pressing one of two keys labeled “true” and preferences: On average, they showed better performance if “false”) depending on whether they are in line with a specific they should respond as if they held pro-Flemish beliefs. belief (e.g., the belief “I am  good.”). In the second block of the As pointed out by De Houwer et al. (2015), the RRT’s structure task, this reference belief is reversed (i.e., stimuli in line with is similar to that of the IAT. For instance, the RRT employs two the belief “I am  no good” would require a “true” response). binary classification tasks sharing a set of two response keys. Additionally, in order to prevent confounding the physical location Furthermore, it consists of two critical blocks differing with regard of a response key (i.e., left vs. right) and its meaning (e.g., true to the specific response rules, and its resulting global score is vs. false) key assignment is varied on a trial by trial basis. based on the performance difference between these blocks. Mirroring Attesting to the fact that beliefs drive responding in the findings for the IAT, the RRT is reliable ( De Houwer et  al., 2015; IRAP, task performance differs between both blocks. Specifically, Tibboel et  al., 2017) while being less demanding on participants responding in the IRAP is faster and more accurate if the as indicated by markedly reduced dropout over the IRAP (4% response rule is in line with personal beliefs (Barnes-Holmes vs. 20%, De Houwer et  al., 2015). On the other hand, by virtue et  al., 2010). Additionally, these effects are sensitive to changes of these shared structural properties, the RRT runs the risk to in the relational qualifier, such as from “I am” to “I want to be  subject to similar flaws as the IAT (e.g., recoding). Last, but be” allowing for dissociation of different kinds of beliefs (e.g., not least, the necessity to instruct participants to react to statements uncovering differences between actual and ideal self, Remue in line with a block specific reference belief effectively limits the et  al., 2013, 2014) that are impervious to traditional implicit RRT to the assessment of a single belief for a given measurement measures like the IAT. session (similar to the IRAP). However, due to its block-based nature, the IRAP is limited to assessing implicit beliefs toward a single set of beliefs at a Propositional Evaluation Paradigm time (i.e., for a given pair of blocks with their associated A final implicit measure of beliefs employs a completely different reference beliefs). In addition, IRAP scores have been shown rationale. Whereas the previously discussed procedures resemble to be  susceptible to faking attempts (Hughes et  al., 2016) and the basic structure of the IAT, the so-called Propositional Evaluation oen exhi ft bit moderate reliability only (e.g., Remue et  al., 2013, Paradigm (PEP, Müller and Rothermund, 2019; see also Wiswede 2014; see also Gawronski and De Houwer, 2014). Finally, the et  al., 2013) is similar in design to classic priming procedures. IRAP is also held back by substantial dropout rates in participants Each PEP trial starts with a simple sentence that is presented in (more than 20% dropout is reported among university students a word-by-word fashion (e.g., “Milk is red.”) to participants in in Remue et  al., 2013; for a discussion, see De Houwer et  al., the center of the screen. Depending on the type of trial, this is 2015) – an issue that is thought to be  attributable to the followed by a specific response prompt. On measurement trials, trial-by-trial response key reassignment. the response prompt (either “true” or “false”) signals to participants which of two response keys (“true”-key or “false”-key) is to Relational Responding Task be  pressed. Note that the prime sentence is completely irrelevant e s Th o-called Relational Responding Task (RRT, De Houwer et al., for participants’ decision – the task is to react to the response 2015) directly addresses the issue of dropouts in the IRAP by prompt only. In contrast, on inducer trials the response prompt avoiding the trial-by-trial response key reassignment. To this end, “? true  - false?” signals participants to indicate whether the prime inducer trials require participants to classify synonyms of the sentence they just saw was orthographically correct (i.e., whether concepts “true” and “false” by button press as either “true” or or not it contained a spelling error). As in the RRT, inducer “not true” thereby constantly reinforcing the intended key meaning trials thus reinforce the intended key meaning. (De Houwer et  al., 2015). On the other hand, target trials present e Th irrelevance of the prime sentence for participants’ reactions participants with whole sentences stating certain kinds of beliefs in the measurement trials notwithstanding, compatibility effects (e.g., regarding immigrants, De Houwer et  al., 2015; or smoking, between the validity of the prime sentence and the required Tibboel et  al., 2017). Mirroring the design of the IRAP discussed response emerge. For example, the prime sentence “Milk is above, a block specific reference belief governs which of two red” is (obviously) false, hence, “false” is automatically activated. responses (i.e., “true” vs. “not true”) participants should give. One This in turn facilitates responding if the response prompt block requires participants to respond “as if ” they held a specific requires a congruent response (i.e., “false”) but interferes with belief (e.g., as if they believed that immigrants were smarter than responding if it requires an incongruent response (i.e., “true”) natives). A second block then requires participants to respond instead. Similarly, in the case of a valid (i.e., true) prime sentence “as if” they held the opposite belief (e.g., as if they believe that faster and more accurate responding would be expected following natives are smarter than immigrants). Consequently, the correct a “true” response prompt, compared to a “false” response prompt. response to a particular target sentence is “true” in one block However, whereas the PEP’s ability to measure beliefs concerning but “not true” in the other block. objectively true or false statements has been demonstrated previously If implicit beliefs drive responding in the RRT, task performance (Wiswede et  al., 2013) the true potential of an implicit measure should differ between the two blocks. Consequently, a relative of beliefs is its ability to tap into inter-individual differences in performance increase of one RRT block over the other is beliefs. This is especially true for beliefs related to more sensitive assumed to indicate that the individual’s beliefs are more in domains, such as beliefs concerning different social groups. As line with this block’s reference belief. De Houwer et  al. (2015) a case in point, Müller and Rothermund (2019) employed the found that implicit beliefs of Flemish participants reflect ingroup PEP to assess individuals’ implicit beliefs concerning racism Frontiers in Psychology | www.frontiersin.org 9 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures against immigrants. Therefore the items of established self-report that is due to salience asymmetries) or to individual attitudes measures of classic and modern racism (e.g., Akrami et  al., 2000) (e.g., extrapersonal associations; Karpinski and Hilton, 2001). served as prime sentences in the PEP. On the sample level the Of course, we  do not want to deny that an effect in an implicit PEP indicated the endorsement of tolerant and welcoming beliefs measure can provide strong evidence for inferring racial bias; about minorities and a rejection of racist beliefs. More precisely, however, we  want to emphasize that such a claim rests on responding with “true” was facilitated when positive beliefs about the assumption that the effect is driven by (implicit) evaluations minorities were shown as primes (e.g., “A multicultural Germany of the categories in question. To bolster this claim, alternative would be good.”). In contrast, responding with “false” was facilitated explanations r fi st have to be identie fi d and ruled out convincingly. when negative beliefs about minorities were shown as primes In this section, however, we  do not want to discuss studies (e.g., “Racist groups are no longer a threat toward immigrants.”). that did not even assess discriminatory behaviors. Instead, Going beyond characteristic patterns at the sample level, the we  want to focus on the lack of fit between predictor and PEP proved to be sensitive to inter-individual differences in these criterion as an explanation for the low predictive validity of beliefs. Specifically, more endorsement of racist attitudes on the implicit measures with regard to behavioral outcomes. More PEP predicted (1) explicit endorsement of these statements, (2) precisely, we argue that the predictive validity of implicit measures political orientation, and (3) behavioral efforts aimed at raising suffers from the fact that (1) studies oen do n ft ot assess behavior money for refugees (see Müller and Rothermund, 2019, for similar proper but rather employ self-report measures as a criterion, findings concerning hiring discrimination and endorsement of and (2) implicit measures typically do not provide contextual gender stereotypes). information; details that are crucial for real-life behavior. To summarize, processing and evaluation of complex propositional content can occur in a rapid and automatic (i.e., Behavioral Intentions Versus implicit) fashion. Recently, a number of promising implicit measures Behavior Proper of beliefs have been introduced. Their strength lies in their ability Although the obvious criterion variable for a study on the to measure complex, propositional relationships among different predictive validity of implicit measures is behavior (e.g., actual concepts. This allows for more fine-grained insights as compared discrimination), the assessment of behavior proper is by no to measures of simple associations that have become a hallmark means the rule. As has been prominently argued by Baumeister of established implicit measures. In our efforts at bridging the et  al. (2007), measurement of actual behavior (a dominant attitude-behavior gap, we should thus not rely solely on associations. approach during the 70s) in the field of social psychology has We  should get beliefs back on board. largely been superseded by “pseudo”-behavioral measures such as rating scale measures assessing behavioral intentions or past behavior. It is thus not surprising, that the same applies to ISSUE 4: LACK OF FIT BETWEEN studies assessing the predictive validity of implicit measures: Behavioral criteria in IAT studies oen co ft nsist of self-report PREDICTOR AND CRITERION measures or similarly indirect indicators (e.g., Oswald et  al., e p Th revious sections discussed shortcomings of the IAT and 2013; Carlsson and Agerström, 2016). Unfortunately, opting similar implicit measures and highlighted possible solutions. for self-report measures of behavior entails a number of Note though that improving the measurement of implicit shortcomings that are especially troublesome for testing the attitudes and beliefs solves only parts of the equation. It is relationship of implicit measures and behavioral outcomes. equally important to ensure adequate measurement of the First, it has long been known that self-reported behavioral respective criterion variable. intentions are not an adequate proxy for actual behavior. For In this section, we  argue that findings of low predictive example, West and Brown (1975; for a detailed elaboration, validity of implicit measures require careful consideration. If see Baumeister et  al., 2007) demonstrated a striking difference the criterion was not properly assessed, then the absence of between participants’ intention to donate money for someone a relation between an implicit measure and a criterion should in need (participants were more than willing to help) and actual not be  understood as evidence against the measure’s validity. helping behavior (donations were close to zero). Second, indirect On the other hand, some of the reported evidence for the measures were conceived to overcome self-presentational concerns validity of implicit measures in predicting behavior must that typically aeff ct self-report measures and/or to measure be  discounted based on the fact that the behavior of interest introspectively less accessible traces of experience. Consequently, was simply not assessed in the first place. Some researchers relying on these very self-reports as the major criterion for interpreted the mere presence of IAT effects as sufficient evidence predictive validity may have contributed to the heterogeneous for discrimination, which it is not. An IAT effect is just a landscape of findings on the validity of implicit measures. response time difference in a computerized categorization task, What is more, we  should probably refrain from referring to not discriminatory behavior (e.g., Arkes and Tetlock, 2004). behavior as if it were a unitary construct. Instead, researchers In our view, an effect in an implicit measure like the IAT should put forward specific hypotheses concerning the relationship might not even count as sufficient evidence for inferring the of implicit measures, different types of behavior, and specific existence of racial biases. As the previous paragraphs have situational conditions. Dual-process or dual-systems models (e.g., shown, these effects might be  driven by various influences Metcalfe and Mischel, 1999; Smith and DeCoster, 2000; Strack that can be unrelated to the categories in question (e.g., recoding and Deutsch, 2004; Friese et  al., 2008; Hofmann et  al., 2009; Frontiers in Psychology | www.frontiersin.org 10 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures Kahneman, 2011) provide a fine-grained view on this question that implicit measures tap into processes operating outside of and have frequently formed the basis for differentiation. These cognitive control, they should relate to impulsive behavior. Thus, models essentially assume that there are different kinds of although some assumptions of these models might have been processes competing for behavior control. The processes differ too strict, dual-process or dual-systems models have enriched with respect to the form in which information is stored and the literature with inspiring hypotheses and findings. They have accessed, as well as the degree of conscious awareness and proven successful in integrating and organizing a large part of cognitive control involved. o Th ugh details and labels vary (e.g., the literature on implicit and explicit measures and their relation automatic vs. controlled: Friese et al., 2008; hot vs. cool: Metcalfe to behavior. Indisputably, an important strength of these models and Mischel, 1999; associative vs. rule-based: Smith and DeCoster, lies in their differentiation between various forms of behavior. 2000; impulsive vs. reflective: Strack and Deutsch, 2004), the It is reasonable to assume different predictive power depending common idea in these models is the distinction between two on the degree of cognitive control involved. So, when it comes cognitive players. On the one hand, there is a system in which to improving the predictive power of implicit measures, our call information is usually assumed to be  stored and accessed in for differentiation also applies to the criterion variable: not all an associative manner. This system should operate fast, effortlessly forms of behavior should be treated equal, and cognitive resources and with little or no awareness and control. On the other hand, should be  taken into account. Researchers are well advised not there is a second system in which information is assumed to to simply explore whether an implicit measure predicts behavior, be  stored and accessed propositionally and which should drive or whether it outperforms explicit measures in doing so. They controlled, slow and effortful deliberation. Both systems are should rather specify more sophisticated hypotheses on the kind hypothesized to compete for behavioral control, in a tug-of-war of behavior that should be predicted (e.g., spontaneous behavior), fashion, with motivation and opportunity for control as crucial or under which conditions (e.g., depleted self-control resources) moderators (e.g., Fazio and Towles-Schwen, 1999; Friese et  al., such a relationship is to be  expected. 2008; Hofmann et  al., 2009). While the first system is assumed To sum up, we  want to highlight the notion that a robust to prompt spontaneous and impulsive behavior, the second estimation of implicit measures’ predictive validity critically should  allow for reasoned action  - but only if people are both hinges on the quality of the criterion. We therefore recommend motivated and able to spare the necessary cognitive resources to drop self-report measures and other indirect criterion variables (e.g., Hofmann et  al., 2007; Friese et  al., 2008). As a case in in favor of actual, rather spontaneous forms of behavior. point, Pearson  et  al. (2009) summarize: Context Dependency of “Whereas explicit attitudes typically shape deliberative, Attitudes and Beliefs well-considered responses for which people have the Finally, it is important to realize that behavior is enacted in motivation and opportunity to weigh the costs and a specific situation or context (e.g., we  react to someone at benefits of various courses of action, implicit attitudes work vs. in the family). Therefore, behavior is inherently context- typically influence responses that are more difficult to specific . In contrast, implicit measures in general do not specify monitor or control […] or responses that people do not contextual information and assess attitudes, stereotypes, or view as diagnostic of their attitude and thus do not try beliefs in a context-independent, global fashion. Aiming for to control.” (p. 322). such an assessment of “the” attitude (e.g., toward Blacks, women, gays, or the elderly) is also at odds with the finding that more A comprehensive overview of the more nuanced theoretical or less all attitudes, beliefs, and stereotypes are context-specific views on conditions under which implicit vs. explicit measures (Blair, 2002; Wigboldus et  al., 2003; Casper et  al., 2010, 2011; predict behavior is beyond the scope of this paper. For an overview Kornadt and Rothermund, 2011, 2015; Müller and Rothermund, of different models, we  refer readers to Perugini et  al. (2010). 2012; Gawronski and Cesario, 2013). Consequently, assessing As for now, however, it is important to note that dual-systems attitudes or beliefs in situational vacuum will oen n ft ot be specific enough to predict a particular behavior toward a specific attitude models are not without criticism (e.g., Rothermund, 2011; Gawronski and Creighton, 2013). Some of their assumptions have object in a specific situation ( Blanton and Jaccard, 2015). even set confining boundaries and require revision. Especially A Solution: Introducing the Context Into the frequently deduced notion that implicit measures like the IAT would reflect associations and therefore predict impulsive Implicit Measures behavior while explicit measures like self-reports would reflect One possibility to address this gap in “level of detail” is to propositional reasoning and therefore explain deliberate acts (e.g., aggregate behavioral outcomes across different situations, time Friese et  al., 2008) is probably an oversimplification. As we  noted points, and target objects yielding a context-independent in the section on implicit beliefs, some features of automaticity behavioral indicator in line with the context independent nature that had previously been reserved exclusively for associative of implicit measures (e.g., of discriminatory behavior; Ajzen, processes also apply to propositional information. At the same 1991). Another and more economic possibility would be  to time, ostensibly implicit measures like the IAT do not necessarily increase the “structural fit” ( Payne et al., 2008) between implicit reflect purely automatic processes, as also outlined before. Instead, measures of attitudes and the to-be-predicted situation-specific it might prove more useful to distinguish between the different behaviors by introducing context-specici fi ty also on the level processes that might be  involved. In other words, to the extent of implicit measures of attitudes. This allows us to capture Frontiers in Psychology | www.frontiersin.org 11 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures the heterogeneity of evaluations that an individual can harbor sophisticated analysis tools (i.e., the ReAL model, Meissner and with regard to the same object (Gawronski et  al., 2018), and Rothermund, 2013) that separate relevant processes from those it increases the chances to predict matching context-specific extraneous influences. Second, we  presented an overview of behaviors (e.g., Blanton and Jaccard, 2015). In this regard, different implicit measures that go beyond the measurement of measures employing dual primes incorporating both category evaluative associations, and instead quantify actual implicit wanting and context information (Casper et al., 2010, 2011) or specifying (e.g., the W-IAT, Koranyi et  al., 2017). Third, we  pointed to context-dependent evaluative meanings when choosing attribute implicit measures of beliefs (e.g., the PEP, Müller and Rothermund, categories in the IAT (Kornadt et al., 2016) represent promising 2019) that allow a more nuanced view on individual attitudes approaches for future research. Implicit measures of propositional and values than measures that tap into associations. Finally, beliefs (see Issue 3 above) are also well-suited in this regard we  emphasized the importance of measuring behavior proper since they allow researchers to clearly specify contextualized and outlined that implicit measures incorporating contextual meanings in the stimulus materials. Similarly, the strength of information might be  more adequate in assessing the structure the motivational drive to pursue specific incentives typically of implicit attitudes or beliefs and their implications for behavior depends on context cues signaling their (un-)availability. For (Casper et  al., 2011; Kornadt et  al., 2016). Each of the recent instance, individual differences in the strengths of motivational developments presented in the current paper has the potential approach (or avoidance) tendencies regarding relationship to increase the predictive power of implicit measures. Future initiation will be  triggered in a dating context (Nikitin et  al., research will also have to clarify whether a combination of these 2019) but probably will not influence behavior toward men approaches may lead to further improvement. Inspired by the and women in the work context. Incorporating this context- fruitful research on dual-process or dual-systems models, we further specificity into implicit measures of wanting (see Issue 2 above) suggest to invest in theoretical considerations: Which forms or will thus be  an important step to capture the determinants aspects of behavior should be related to which processes involved of our desires and to better explain and predict social behavior. in which implicit measures? Differentiation is key, with regard To summarize, assessing the potential of implicit measures to both the predictor and the criterion. for explaining and closing the attitude-behavior gap requires We strongly argue not to take the validity of implicit measures both predictors (implicit attitudes and beliefs) and criterion like the IAT for granted. Instead, we  should take into account variables (e.g., discriminatory behaviors) to be  assessed in a the complexity of these measures, especially when it comes reliable, valid, and contextualized way. This necessitates both to the predictive value for real-life behavior. As outlined in changes in implicit measures (to address the context-specificity the current review, the past 20  years of research have provided of the to-be-measured constructs) as well as rigorous theorizing us with a number of good reasons for why the IAT and its about which aspects of which type of behavior are to derivatives did not succeed in closing the attitude-behavior be  influenced by (context-specific) attitudes and beliefs. gap, and enriched our toolbox with promising, sophisticated improvements. Future research will benefit from harnessing the power of such a more differentiated view on implicit measures. CLOSING THOUGHTS In this article, we  presented an overview of possible reasons for AUTHOR CONTRIBUTIONS the weak relationship between implicit measures like the IAT and behavioral criteria. We outlined that the unsatisfying predictive FMe and KR wrote the first draft of the manuscript. LG, NK, value of the IAT is due to (1) extraneous influences like recoding, and FMü wrote sections of the manuscript. All authors (2) the measurement of liking instead of wanting, (3) the contributed to manuscript revision, read and approved the measurement of associations instead of complex beliefs, and/or submitted version. (4) a conceptual mismatch of predictor and criterion. We presented precise solutions for each of these problems. More precisely, we  suggested to switch to procedural variations that minimize FUNDING extraneous influences (i.e., the SB-IAT, Teige-Mocigemba et  al., 2008; and the IAT-RF; Rothermund et  al., 2009), and to apply This work was funded by grant RO 1272/11-1 to KR. REFERENCES Arkes, H. R., and Tetlock, P. E. (2004). Attributions of implicit prejudice, or “Would Jesse Jackson ‘fail’ the implicit association test?”. Psychol. Inq. 15, Ajzen, I. (1991). The theory of planned behavior. Organ. Behav. Hum. Decis. 257–278. doi: 10.1207/s15327965pli1504_01 Back, M. D., Schmukle, S. C., and Eglo, B ff . (2005). Measuring task-switching Process. 50, 179–211. doi: 10.1016/0749-5978(91)90020-T Akrami, N., Ekehammar, B., and Araya, T. (2000). Classical and modern racial ability in the implicit association test. Exp. Psychol. 52, 167–179. doi: 10.1027/1618-3169.52.3.167 prejudice: a study of attitudes toward immigrants in Sweden. Eur. J. Soc. Psychol. 30, 521–532. doi: 10.1002/1099-0992(200007/08)30:4<521::AID- Barnes-Holmes, D., Barnes-Holmes, Y., Stewart, I., and Boles, S. (2010). A sketch of the implicit relational assessment procedure (IRAP) and the relational EJSP5>3.0.CO;2-N Anderson, J. R. (1983). A spreading activation theory of memory. J. Verbal elaboration and coherence (REC) model. Psychol. Rec. 60, 527–542. doi: 10.1007/BF03395726 Learn. Verbal Behav. 22, 261–295. doi: 10.1016/S0022-5371(83)90201-3 Frontiers in Psychology | www.frontiersin.org 12 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures Batchelder, W. H., and Riefer, D. M. (1999). Theoretical and empirical review Epstein, L. H., Truesdale, R., Wojcik, A., Paluch, R. A., and Raynor, H. A. of multinomial process tree modeling. Psychon. Bull. Rev. 6, 57–86. doi: (2003). Effects of deprivation on hedonics and reinforcing value of food. 10.3758/BF03210812 Physiol. Behav. 78, 221–227. doi: 10.1016/S0031-9384(02)00978-2 Baumeister, R. F., Vohs, K. D., and Funder, D. C. (2007). Psychology as the Erdfelder, E., Auer, T. S., Hilbig, B. E., Aßfalg, A., Moshagen, M., and Nadarevic, L. science of self-reports and finger movements. Whatever happened to actual (2009). Multinomial processing tree models: a review of the literature. behavior? Perspect. Psychol. Sci. 2, 396–403. doi: 10.1111/j.1745-6916.2007.00051.x J. Psychol. 217, 108–124. doi: 10.1027/0044-3409.217.3.108 Berridge, K. C. (2009). Wanting and liking: observations from the neuroscience Fazio, R. H. (2007). Attitudes as object-evaluation associations of varying and psychology laboratory. Inquiry 52:378. doi: 10.1080/00201740903087359 strength. Soc. Cogn. 25, 603–637. doi: 10.1521/soco.2007.25.5.603 Berridge, K. C., and Robinson, T. E. (2003). Parsing reward. Trends Neurosci. Fazio, R. H., Powell, M. C., and Herr, P. M. (1983). Toward a process model of 26, 507–513. doi: 10.1016/S0166-2236(03)00233-9 the attitude-behavior relation: accessing one’s attitude upon mere observation of Berridge, K. C., Venier, I. L., and Robinson, T. E. (1989). Taste reactivity the attitude object. J. Pers. Soc. Psychol. 44, 723–735. doi: 10.1037/0022-3514.44.4.723 analysis of 6-hydrocydopamine-induced aphagia: implications for arousal Fazio, R. H., Sanbonmatsu, D. M., Powell, M. C., and Kardes, F. R. (1986). and anhedonia hypotheses of dopamine function. Behav. Neurosci. 103, On the automatic activation of attitudes. J. Pers. Soc. Psychol. 50, 229–238. 36–45. doi: 10.1037/0735-7044.103.1.36 doi: 10.1037/0022-3514.50.2.229 Blair, I. V. (2002). e Th malleability of automatic stereotypes and prejudice. Fazio, R. H., and Towles-Schwen, T. (1999). “e M Th ODE model of attitude- Personal. Soc. Psychol. Rev. 6, 242–261. doi: 10.1207/S15327957PSPR0603_8 behavior processes” in Dual-process theories in social psychology. eds. S. Chaiken and Y. Trope (New York, NY: Guilford Press), 97–116. Blanton, H., and Jaccard, J. (2015). Not so fast: ten challenges to importing implicit attitude measures to media psychology. Media Psychol. 18, 338–369. Fiedler, K., Messner, C., and Bluemke, M. (2006). Unresolved problems with the “I”, the “A”, and the “T”: a logical and psychometric critique of the doi: 10.1080/15213269.2015.1008102 Blanton, H., Jaccard, J., Gonzales, P. M., and Christie, C. (2006). Decoding implicit association test (IAT). Eur. Rev. Soc. Psychol. 17, 74–147. doi: 10.1080/10463280600681248 the implicit association test: implications for criterion prediction. J. Exp. Fiedler, K., and Unkelbach, C. (2011). Evaluative conditioning depends on Soc. Psychol. 42, 192–212. doi: 10.1016/j.jesp.2005.07.003 higher order encoding processes. Cognit. Emot. 25, 639–656. doi: Bluemke, M., and Friese, M. (2006). Do features of stimuli influence IAT 10.1080/02699931.2010.513497 effects? J. Exp. Soc. Psychol. 42, 163–176. doi: 10.1016/j.jesp.2005.03.004 Förderer, S., and Unkelbach, C. (2012). Hating the cute kitten or loving the Carlsson, R., and Agerström, J. (2016). A closer look at the discrimination aggressive pit-bull: EC effects depend on CS-US relations. Cognit. Emot. outcomes in the IAT literature. Scand. J. Psychol. 57, 278–287. doi: 10.1111/ 26, 534–540. doi: 10.1080/02699931.2011.588687 sjop.12288 Friese, M., Hofmann, W., and Schmitt, M. (2008). When and why do implicit Casper, C., Rothermund, K., and Wentura, D. (2010). Automatic stereotype measures predict behaviour? Empirical evidence for the moderating role of activation is context dependent. Soc. Psychol. 41, 131–136. doi: 10.1027/1864-9335/ opportunity, motivation, and process reliance. Eur. Rev. Soc. Psychol. 19, a000019 285–338. doi: 10.1080/10463280802556958 Casper, C., Rothermund, K., and Wentura, D. (2011). The activation of specific Friese, M., Hofmann, W., and Wänke, M. (2008). When impulses take over: facets of age stereotypes depends on individuating information. Soc. Cogn. moderated predictive validity of explicit and implicit attitude measures in 29, 393–414. doi: 10.1521/soco.2011.29.4.393 Chang, B. P. I., and Mitchell, C. J. (2009). Processing fluency as a predictor predicting food choice and consumption behaviour. Br. J. Soc. Psychol. 47, 397–419. doi: 10.1348/014466607X241540 of salience asymmetries in the implicit association test. Q. J. Exp. Psychol. 62, 2030–2054. doi: 10.1080/17470210802651737 Gast, A., and Rothermund, K. (2010). When old and frail is not the same. Dissociating category-based and stimulus-based influences on compatibility Conrey, F. R., Sherman, J. W., Gawronski, B., Hugenberg, K., and Groom, C. J. (2005). Separating multiple processes in implicit social cognition: the quad effects in four implicit measurement methods. Q. J. Exp. Psychol. 63, 479–498. model of implicit task performance. J. Pers. Soc. Psychol. 89, 469–487. doi: doi: 10.1080/17470210903049963 10.1037/0022-3514.89.4.469 Gawronski, B., and Cesario, J. (2013). Of mice and men: what animal research Dai, X. C., Brendl, C. M., and Ariely, D. (2010). Wanting, liking, and preference can tell us about context effects on automatic responses in humans. Personal. construction. Emotion 10, 324–334. doi: 10.1037/a0017987 Soc. Psychol. Rev. 17, 187–215. doi: 10.1177/1088868313480096 Dai, X. C., Dong, P., and Jia, J. S. (2014). When does playing hard to get Gawronski, B., and Creighton, L. A. (2013). “Dual-process theories” in The increase romantic attraction? J. Exp. Psychol. Gen. 143, 521–526. doi: 10.1037/ Oxford handbook of social cognition. ed. D. E. Carlston (New York, NY: a0032989 Oxford University Press), 282–312. De Houwer, J. (2001). A structural and process analysis of the implicit association Gawronski, B., and De Houwer, J. (2014). “Implicit measures in social and test. J. Exp. Soc. Psychol. 37, 443–451. doi: 10.1006/jesp.2000.1464 personality psychology” in Handbook of research methods in social and De Houwer, J. (2003a). “A structural analysis of indirect measures of attitudes” personality psychology. 2nd Edn. eds. H. T. Reis and C. M. Judd (New York, NY: in e p Th sychology of evaluation: Ae ff ctive processes in cognition and emotion . Cambridge University Press), 283–310. eds. J. Musch and K. C. Klauer (Mahwah, NJ: Erlbaum), 219–244. Gawronski, B., and Hahn, A. (2019). “Implicit measures: procedures, use, and De Houwer, J. (2003b). The extrinsic aeff ctive Simon task. Exp. Psychol. 50, interpretation” in Measurement in social psychology. eds. H. Blanton, J. M. 77–85. doi: 10.1026//1618-3169.50.2.77 LaCroix, and G. D. Webster (New York: Taylor & Francis), 29–55. De Houwer, J. (2014). A propositional model of implicit evaluation. Soc. Psychol. Gawronski, B., Rydell, R. J., De Houwer, J., Brannon, S. M., Ye, Y., Vervliet, B., Person. Compass 8, 342–353. doi: 10.1111/spc3.12111 et al. (2018). Contextualized attitude change. Adv. Exp. Soc. Psychol. 57, De Houwer, J., and De Bruycker, E. (2007). The identification-EAST as a valid 1–52. doi: 10.1016/bs.aesp.2017.06.001 measure of implicit attitudes toward alcohol-related stimuli. J. Behav. Ther. Govan, C. L., and Williams, K. D. (2004). Changing the aeff ctive valence of Exp. Psychiatry 38, 133–143. doi: 10.1016/j.jbtep.2006.10.004 the stimulus items influences the IAT by re-defining the category labels. De Houwer, J., Geldof, T., and De Bruycker, E. (2005). The implicit association J. Exp. Soc. Psychol. 40, 357–365. doi: 10.1016/j.jesp.2003.07.002 test as a general measure of similarity. Can. J. Exp. Psychol. 59, 228–239. Greenwald, A. G., and Banaji, M. R. (1995). Implicit social cognition: attitudes, doi: 10.1037/h0087478 self-esteem, and stereotypes. Psychol. Rev. 102, 4–27. doi: 10.1037/0033-295X.102.1.4 De Houwer, J., Heider, N., Spruyt, A., Roets, A., and Hughes, S. (2015). The Greenwald, A. G., McGhee, D. E., and Schwartz, J. L. K. (1998). Measuring relational responding task: toward a new implicit measure of beliefs. Front. individual differences in implicit cognition: the implicit association test. Psychol. 6, 1–9. doi: 10.3389/fpsyg.2015.00319 J. Pers. Soc. Psychol. 74, 1464–1480. doi: 10.1037/0022-3514.74.6.1464 De Schryver, M., Hughes, S., Rosseel, Y., and De Houwer, J. (2016). Unreliable Greenwald, A. G., Poehlman, T. A., Uhlmann, E. L., and Banaji, M. R. (2009). yet still replicable: a comment on LeBel and Paunonen (2011). Front. Psychol. Understanding and using the implicit association test: III. Meta-analysis of 6:2039. doi: 10.3389/fpsyg.2015.02039 predictive validity. J. Pers. Soc. Psychol. 97, 17–41. doi: 10.1037/a0015575 Dill, B., and Holton, R. (2014). The addict in us all. Front. Psych. 5:139. doi: Grigutsch, L. A., Lewe, G., Rothermund, K., and Koranyi, N. (2019). Implicit 10.3389/fpsyt.2014.00139 ‘wanting’ without implicit ‘liking’: a test of incentive-sensitization-theory in Frontiers in Psychology | www.frontiersin.org 13 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures the context of smoking addiction using the wanting-implicit-association-test Koranyi, N., Grigutsch, L. A., Algermissen, J., and Rothermund, K. (2017). (W-IAT). J. Behav. Ther. Exp. Psychiatry 64, 9–14. doi: 10.1016/j.jbtep.2019.01.002 Dissociating implicit wanting from implicit liking: development and validation Heck, D. W., and Erdfelder, E. (2016). Extending multinomial processing tree of the wanting-implicit-association-test (W-IAT). J. Behav. Ther. Exp. Psychiatry models to measure the relative speed of cognitive processes. Psychon. Bull. 54, 165–169. doi: 10.1016/j.jbtep.2016.08.008 Rev. 23, 1440–1465. doi: 10.3758/s13423-016-1025-6 Koranyi, N., and Meissner, F. (2015). Handing over the reins: neutralizing Hobbs, M., Remington, B., and Glautier, S. (2005). Dissociation of wanting negative attitudes toward dependence in response to reciprocal romantic and liking for alcohol in humans: a test of the incentive-sensitisation theory. liking. Soc. Psychol. Personal. Sci. 6, 685–691. doi: 10.1177/1948550615580169 Psychopharmacology 178, 493–499. doi: 10.1007/s00213-004-2026-0 Kornadt, A. E., Meissner, F., and Rothermund, K. (2016). Implicit and explicit Hofmann, W., and Friese, M. (2008). Impulses got the better of me: alcohol age stereotypes for specific life domains across the life span: distinct patterns moderates the influence of implicit attitudes toward food cues on eating and age group differences. Exp. Aging Res. 42, 195–211. doi: 10.1080/ behavior. J. Abnorm. Psychol. 117, 420–427. doi: 10.1037/0021-843X.117.2.420 0361073X.2016.1132899 Hofmann, W., Friese, M., and Strack, F. (2009). Impulse and self-control from Kornadt, A. E., and Rothermund, K. (2011). Contexts of aging: assessing a dual-systems perspective. Perspect. Psychol. Sci. 4, 162–176. doi: 10.1111/j. evaluative age stereotypes in different life domains. J. Gerontol. Psychol. Sci. 1745-6924.2009.01116.x 66, 547–556. doi: 10.1093/geronb/gbr036 Hofmann, W., Rauch, W., and Gawronski, B. (2007). And deplete us not into Kornadt, A. E., and Rothermund, K. (2015). Views on aging: domain-specific temptation: automatic attitudes, dietary restraint, and self-regulatory resources approaches and implications for developmental regulation. Annu. Rev. Gerontol. as determinants of eating behavior. J. Exp. Soc. Psychol. 43, 497–504. doi: Geriatr. 35, 121–144. doi: 10.1891/0198-8794.35.121 10.1016/j.jesp.2006.05.004 Kraus, S. J. (1995). Attitudes and the prediction of behavior: a meta-analysis Houben, K., Rothermund, K., and Wiers, R. W. (2009). Predicting alcohol use of the empirical literature. Personal. Soc. Psychol. Bull. 21, 58–75. doi: with a recoding-free variant of the implicit association test. Addict. Behav. 10.1177/0146167295211007 34, 487–489. doi: 10.1016/j.addbeh.2008.12.012 Kraus, A. A., and Piqueras-Fiszman, B. (2016). Sandwich or sweets? An assessment Hu, X., and Batchelder, W. H. (1994). The statistical analysis of general processing of two novel implicit association tasks to capture dynamic motivational tree models with the EM algorithm. Psychometrika 59, 21–47. doi: 10.1007/ tendencies and stable evaluations towards foods. Food Qual. Prefer. 49, BF02294263 11–19. doi: 10.1016/j.foodqual.2015.11.005 Hughes, S., Barnes-Holmes, D., and De Houwer, J. (2011). The dominance of Kraus, A. A., and Scholderer, J. (2015). Indirect measurement of motivation: associative theorizing in implicit attitude research: propositional and behavioral developing and testing a motivational recoding-free implicit association test alternatives. Psychol. Rec. 61, 465–496. doi: 10.1007/BF03395772 (m-IAT-RF). Soc. Psychol. 46, 142–156. doi: 10.1027/1864-9335/a000234 Hughes, S., Hussey, I., Corrigan, B., Jolie, K., Murphy, C., and Barnes-Holmes, D. Kurdi, B., Seitchik, A. E., Axt, J. R., Carroll, T. J., Karapetyan, A., Kaushik, N., (2016). Faking revisited: exerting strategic control over performance on the et al. (2019). Relationship between the implicit association test and intergroup implicit relational assessment procedure. Eur. J. Soc. Psychol. 46, 632–648. behavior: a meta-analysis. Am. Psychol. 74, 569–586. doi: 10.1037/amp0000364 doi: 10.1002/ejsp.2207 Lades, L. K. (2012). Towards an incentive salience model of intertemporal Ito, T. A., Friedman, N. P., Bartholow, B. D., Correll, J., Loersch, C., Altamirano, L. J., choice. J. Econ. Psychol. 33, 833–841. doi: 10.1016/j.joep.2012.03.007 et al. (2015). Toward a comprehensive understanding of executive cognitive LeBel, E. P., and Paunonen, S. V. (2011). Sexy but oen unr ft eliable: the impact function in implicit racial bias. J. Pers. Soc. Psychol. 108, 187–218. doi: of unreliability on the replicability of experimental findings with implicit 10.1037/a0038557 measures. Personal. Soc. Psychol. Bull. 37, 570–583. doi: 10.1177/ Jin, Z. (2016). Disentangling recoding processes and evaluative associations in 0146167211400619 a gender attitude implicit association test among adult males. Q. J. Exp. Litt, A., Khan, U., and Shiv, B. (2010). Lusting while loathing: parallel counterdriving Psychol. 69, 2276–2284. doi: 10.1080/17470218.2015.1126290 of wanting and liking. Psychol. Sci. 21, 118–125. doi: 10.1177/0956797609355633 Kahneman, D. (2011). in Th king, fast and slow . New York: Farrar, Straus, Matzke, D., Dolan, C. V., Batchelder, W. H., and Wagenmakers, E.-J. (2015). and Giroux. Bayesian estimation of multinomial processing tree models with heterogeneity Karpinski, A., and Hilton, J. L. (2001). Attitudes and the implicit association in participants and items. Psychometrika 80, 205–235. doi: 10.1007/ test. J. Pers. Soc. Psychol. 81, 774–788. doi: 10.1037/0022-3514.81.5.774 s11336-013-9374-9 Katz, D. (1960). The functional approach to the study of attitudes. Public Opin. McFarland, S. G., and Crouch, Z. (2002). A cognitive skill confound on the Q. 24, 163–204. doi: 10.1086/266945 implicit association test. Soc. Cogn. 20, 483–510. doi: 10.1521/soco.20.6.483.22977 Kinoshita, S., and Peek-O’Leary, M. (2005). Does the compatibility effect in Meissner, F., and Rothermund, K. (2013). Estimating the contributions of the race implicit association test (IAT) reflect familiarity or ae ff ct? Psychon. associations and recoding in the implicit association test: the ReAL model Bull. Rev. 12, 442–452. doi: 10.3758/BF03193786 for the IAT. J. Pers. Soc. Psychol. 104, 45–69. doi: 10.1037/a0030734 Kinoshita, S., and Peek-O’Leary, M. (2006). Two bases of the compatibility Meissner, F., and Rothermund, K. (2015a). The insect-nonword IAT revisited: effect in the implicit association test (IAT). Q. J. Exp. Psychol. 59, 2102–2120. dissociating between evaluative associations and recoding. Soc. Psychol. 46, doi: 10.1080/17470210500451141 46–54. doi: 10.1027/1864-9335/a000220 Klauer, K. C. (2010). Hierarchical multinomial processing tree models: a latent- Meissner, F., and Rothermund, K. (2015b). A thousand words are worth more trait approach. Psychometrika 75, 70–98. doi: 10.1007/s11336-009-9141-0 than a picture? The effects of stimulus modality on the implicit association Klauer, K. C., and Kellen, D. (2018). RT-MPTs: process models for response-time test. Soc. Psychol. Personal. Sci. 6, 740–748. doi: 10.1177/1948550615580381 distributions based on multinomial processing trees with applications to Metcalfe, J., and Mischel, W. (1999). A hot/cool-system analysis of delay of recognition memory. J. Math. Psychol. 82, 111–130. doi: 10.1016/j.jmp.2017.12.003 gratification: dynamics of willpower. Psychol. Rev. 106, 3–19. doi: Klauer, K. C., and Mierke, J. (2005). Task-set inertia, attitude accessibility, and 10.1037/0033-295X.106.1.3 compatibility-order effects: new evidence for a task-set switching account Mierke, J., and Klauer, K. C. (2001). Implicit association measurement with of the implicit association test effect. Personal. Soc. Psychol. Bull. 31, 208–217. the IAT: evidence for effects of executive control processes. Z. Exp. Psychol. doi: 10.1177/0146167204271416 48, 107–122. doi: 10.1026//0949-3946.48.2.107 Klauer, K. C., Schmitz, F., Teige-Mocigemba, S., and Voss, A. (2010). Understanding Mierke, J., and Klauer, K. C. (2003). Method-specific variance in the implicit the role of executive control in the implicit association test: why flexible association test. J. Pers. Soc. Psychol. 85, 1180–1192. doi: 10.1037/0022-3514.85.6.1180 people have small IAT effects. Q. J. Exp. Psychol. 63, 595–619. doi: Mitchell, C. J. (2004). Mere acceptance produces apparent attitude in the implicit 10.1080/17470210903076826 association test. J. Exp. Soc. Psychol. 40, 366–373. doi: 10.1016/j.jesp.2003.07.003 Klauer, K. C., Stahl, C., and Voss, A. (2012). “Multinomial models and diffusion Mitchell, J. P., Nosek, B. A., and Banaji, M. R. (2003). Contextual variations models” in Cognitive methods in social psychology. Abridged edition. eds. in implicit evaluation. J. Exp. Psychol. Gen. 132, 455–469. doi: K. C. Klauer, A. Voss, and C. Stahl (New York, NY: Guilford Press), 331–354. 10.1037/0096-3445.132.3.455 Klauer, K. C., Voss, A., Schmitz, F., and Teige-Mocigemba, S. (2007). Process Müller, F., and Rothermund, K. (2012). Talking loudly but lazing at work – components of the implicit association test: a diffusion-model analysis. behavioral effects of stereotypes are context dependent. Eur. J. Soc. Psychol. J. Pers. Soc. Psychol. 93, 353–368. doi: 10.1037/0022-3514.93.3.353 42, 557–563. doi: 10.1002/ejsp.1869 Frontiers in Psychology | www.frontiersin.org 14 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures Müller, F., and Rothermund, K. (2019). e Th Propositional Evaluation Paradigm: Rømer Thomsen, K., Whybrow, P. C., and Kringelbach, M. L. (2015). Reconceptualizing Indirect Assessment of Personal Beliefs and Attitudes. Front. Psychol. 10:2385. anhedonia: novel perspectives on balancing the pleasure networks in the human doi: 10.3389/fpsyg.2019.02385 brain. Front. Behav. Neurosci. 9:49. doi: 10.3389/fnbeh.2015.00049 Nikitin, J., Gong, X., Schoch, S., and Freund, A. M. (2019). Social motives, Rothermund, K. (2011). Selbstregulation statt Selbstkontrolle: Intentionalität, attributions and expectations as predictors of the decision to participate in Ae ff kt und Kognition als Teamgefährten im Dienste der Handlungssteuerung. a speed-dating event. Motiv. Emot. 43, 610–624. doi: 10.1007/s11031-019-09762-0 Psychol. Rundsch. 62, 167–173. doi: 10.1026/0033-3042/a000087 Nisbett, R. E., and Wilson, T. D. (1977). Telling more than we  can know: Rothermund, K., Teige-Mocigemba, S., Gast, A., and Wentura, D. (2009). verbal reports on mental processes. Psychol. Rev. 84, 231–259. doi: Minimizing the influence of recoding in the implicit association test: the 10.1037/0033-295X.84.3.231 recoding-free implicit association test (IAT-RF). Q. J. Exp. Psychol. 62, Nosek, B. A., Greenwald, A. G., and Banaji, M. R. (2005). Understanding and 84–98. doi: 10.1080/17470210701822975 using the implicit association test: II. Method variables and construct validity. Rothermund, K., and Wentura, D. (2001). Figure-ground asymmetries in the Personal. Soc. Psychol. Bull. 31, 166–180. doi: 10.1177/0146167204271418 implicit association test (IAT). Z. Exp. Psychol. 48, 94–106. doi: Nosek, B. A., Greenwald, A. G., and Banaji, M. R. (2007). “e im Th plicit association 10.1026//0949-3946.48.2.94 test at age 7: a methodological and conceptual review” in Frontiers of social Rothermund, K., and Wentura, D. (2004). Underlying processes in the implicit psychology: Social psychology and the unconscious. The automaticity of higher association test: dissociating salience from associations. J. Exp. Psychol. Gen. mental processes. ed. J. A. Bargh (New York, NY: Psychology Press), 265–292. 133, 139–165. doi: 10.1037/0096-3445.133.2.139 Olney, J. J., Warlow, S. M., Naffziger, E. E., and Berridge, K. C. (2018). Current Rothermund, K., Wentura, D., and De Houwer, J. (2005). Validity of the salience perspectives on incentive salience and applications to clinical disorders. asymmetry account of the implicit association test: reply to Greenwald, Curr. Opin. Behav. Sci. 22, 59–69. doi: 10.1016/j.cobeha.2018.01.007 Nosek, Banaji, and Klauer (2005). J. Exp. Psychol. Gen. 134, 426–430. doi: Oswald, F. L., Mitchell, G., Blanton, H., Jaccard, J., and Tetlock, P. E. (2013). 10.1037/0096-3445.134.3.426 Predicting ethnic and racial discrimination: a meta-analysis of IAT criterion Smith, E. R., and DeCoster, J. (2000). Dual-process models in social and cognitive studies. J. Pers. Soc. Psychol. 105, 171–192. doi: 10.1037/a0032734 psychology: conceptual integration and links to underlying memory systems. Palfai, T. P., and Ostafin, B. D. (2003). Alcohol-related motivational tendencies Personal. Soc. Psychol. Rev. 4, 108–131. doi: 10.1207/S15327957PSPR0402_01 in hazardous drinkers: assessing implicit response tendencies using the modified- Stacy, A. W., and Wiers, R. W. (2010). Implicit cognition and addiction: a IAT. Behav. Res. Ther. 41, 1149–1162. doi: 10.1016/S0005-7967(03)00018-4 tool for explaining paradoxical behavior. Annu. Rev. Clin. Psychol. 6, 551–575. Payne, B. K., Burkley, M. A., and Stokes, M. B. (2008). Why do implicit and doi: 10.1146/annurev.clinpsy.121208.131444 explicit attitude tests diverge? The role of structural fit. J. Pers. Soc. Psychol. Steffens, M. C., and Plewe, I. (2001). Items’ cross-category associations as a 94, 16–31. doi: 10.1037/0022-3514.94.1.16 confounding factor in the implicit association test. Z. Exp. Psychol. 48, Payne, B. K., Cheng, C. M., Govorun, O., and Stewart, B. D. (2005). An 123–134. doi: 10.1026//0949-3946.48.2.123 inkblot for attitudes: aeff ct misattribution as implicit measurement. J. Pers. Stevenson, R. J., Francis, H. M., Attuquayefio, T., and Ockert, C. (2017). Explicit Soc. Psychol. 89, 277–293. doi: 10.1037/0022-3514.89.3.277 wanting and liking for palatable snacks are differentially ae ff cted by change Pearson, A. R., Dovidio, J. F., and Gaertner, S. L. (2009). The nature of in physiological state, and differentially related to salivation and hunger. contemporary prejudice: insights from aversive racism. Soc. Personal. Psychol. Physiol. Behav. 182, 101–106. doi: 10.1016/j.physbeh.2017.10.007 Compass 3, 314–338. doi: 10.1111/j.1751-9004.2009.00183.x Strack, F., and Deutsch, R. (2004). Reflective and impulsive determinants of Peciña, S., Cagniard, B., Berridge, K. C., Aldridge, J. W., and Zhuang, X. social behavior. Personal. Soc. Psychol. Rev. 8, 220–247. doi: 10.1207/ (2003). Hyperdopaminergic mutant mice have higher “wanting” but not s15327957pspr0803_1 “liking” for sweet rewards. J. Neurosci. 23, 9395–9402. doi: 10.1523/ Teige-Mocigemba, S., and Klauer, K.C. (2015). Implicit association test. In J. D. JNEUROSCI.23-28-09395.2003 Wright (editor-in-chief ), International encyclopedia of the social and behavioral Perugini, M., Richetin, J., and Zogmaister, C. (2010). “Prediction of behavior” sciences. Vol. 11, 2nd Edn (Oxford, UK: Elsevier), 703–708. in Handbook of implicit social cognition: Measurement, theory, and applications. Teige-Mocigemba, S., Klauer, K. C., and Rothermund, K. (2008). Minimizing eds. B. Gawronski and B. K. Payne (New York, NY: Guilford Press), 255–277. method-specific variance in the IAT: a single block IAT. Eur. J. Psychol. Peters, K. R., and Gawronski, B. (2011). Are we puppets on a string? Comparing Assess. 24, 237–245. doi: 10.1027/1015-5759.24.4.237 the impact of contingency and validity on implicit and explicit evaluations. Teige-Mocigemba, S., Klauer, K. C., and Sherman, J. W. (2010). “Practical guide Personal. Soc. Psychol. Bull. 37, 557–569. doi: 10.1177/0146167211400423 to implicit association tests and related tasks” in Handbook of implicit social Pool, E., Brosch, T., Delplanque, S., and Sander, D. (2015). Stress increases cognition: Measurement, theory, and applications. eds. B. Gawronski and cue-triggered “wanting” for sweet reward in humans. J. Exp. Psychol Anim. B. K. Payne (New York, NY: Guilford Press), 117–139. Learn. Cogn. 41, 128–136. doi: 10.1037/xan0000052 Tibboel, H., De Houwer, J., Dirix, N., and Spruyt, A. (2017). Beyond associations: Pool, E., Sennwald, V., Delplanque, S., Brosch, T., and Sander, D. (2016). do implicit beliefs play a role in smoking addiction? J. Psychopharmacol. Measuring wanting and liking from animals to humans: a systematic review. 31, 43–53. doi: 10.1177/0269881116665327 Neurosci. Biobehav. Rev. 63, 124–142. doi: 10.1016/j.neubiorev.2016.01.006 Tibboel, H., De Houwer, J., Spruyt, A., Brevers, D., Roy, E., and Noël, X. Remue, J., De Houwer, J., Barnes-Holmes, D., Vanderhasselt, M. A., and (2015a). Heavy social drinkers score higher on implicit wanting and liking De Raedt, R. (2013). Self-esteem revisited: performance on the implicit for alcohol than alcohol-dependent patients and light social drinkers. relational assessment procedure as a measure of self-versus ideal self-related J. Behav. er Th . Exp. Psychiatry 48, 185–191. doi: 10.1016/j.jbtep.2015.04.003 cognitions in dysphoria. Cognit. Emot. 27, 1441–1449. doi: 10.1080/ Tibboel, H., De Houwer, J., Spruyt, A., Field, M., Kemps, E., and Crombez, G. 02699931.2013.786681 (2011). Testing the validity of implicit measures of wanting and liking. Remue, J., Hughes, S., De Houwer, J., and De Raedt, R. (2014). To be  or want J. Behav. er Th . Exp. Psychiatry 42, 284–292. doi: 10.1016/j.jbtep.2011.01.002 to be: disentangling the role of actual versus ideal self in implicit self- Tibboel, H., De Houwer, J., and Van Bockstaele, B. (2015b). Implicit measures esteem. PLoS One 9:e108837. doi: 10.1371/journal.pone.0108837 of “wanting” and “liking” in humans. Neurosci. Biobehav. Rev. 57, 350–364. Riefer, D. M., and Batchelder, W. H. (1988). Multinomial modeling and the doi: 10.1016/j.neubiorev.2015.09.015 measurement of cognitive processes. Psychol. Rev. 95, 318–339. doi: Van Dessel, P., De Houwer, J., and Smith, C. T. (2018). Relational information 10.1037/0033-295X.95.3.318 moderates approach-avoidance instruction effects on implicit evaluation. Acta Robinson, T. E., and Berridge, K. C. (1993). The neural basis of drug craving – an Psychol. 184, 137–143. doi: 10.1016/j.actpsy.2017.03.016 incentive-sensitization theory of addiction. Brain Res. Rev. 18, 247–291. doi: von Stülpnagel, R., and Steffens, M. C. (2010). Prejudiced or just smart? 10.1016/0165-0173(93)90013-P Intelligence as a confounding factor in the IAT effect. J. Psychol. 218, Robinson, T. E., and Berridge, K. C. (2001). Incentive-sensitization and addiction. 51–53. doi: 10.1027/0044-3409/a000008 Addiction 96, 103–114. doi: 10.1080/09652140020016996 Wentura, D., and Degner, J. (2010). “A practical guide to sequential priming Robinson, M. J. F., and Berridge, K. C. (2013). Instant transformation of learned and related tasks” in Handbook of implicit social cognition: Measurement, repulsion into motivational "wanting". Curr. Biol. 23, 282–289. doi: 10.1016/j. theory, and applications. eds. B. Gawronski and B. K. Payne (New York, cub.2013.01.016 NY: Guilford Press), 95–116. Frontiers in Psychology | www.frontiersin.org 15 November 2019 | Volume 10 | Article 2483 Meissner et al. Predicting Behavior With Implicit Measures Wentura, D., and Rothermund, K. (2007). “Paradigms we  live by: a plea for evaluation processes. Soc. Cogn. Aeff ct. Neurosci. 8, 647–653. doi: 10.1093/ more basic research on the implicit association test” in Implicit measures scan/nss042 of attitudes: Procedures and controversies. eds. B. Wittenbrink and N. S. Zanon, R., De Houwer, J., Gast, A., and Smith, C. T. (2014). When does Schwarz (New York, NY: Guilford Press), 195–215. relational information influence evaluative conditioning? Q. J. Exp. Psychol. West, S. G., and Brown, T. J. (1975). Physical attractiveness, the severity of 67, 2105–2122. doi: 10.1080/17470218.2014.907324 the emergency and helping: a field experiment and interpersonal simulation. Zhang, J., Berridge, K. C., Tindell, A. J., Smith, K. S., and Aldridge, J. W. J. Exp. Soc. Psychol. 11, 531–538. doi: 10.1016/0022-1031(75)90004-9 (2009). A neural computational model of incentive salience. PLoS Comput. Wicker, A. W. (1969). Attitudes versus actions: the relationship of verbal and Biol. 5:e1000437. doi: 10.1371/journal.pcbi.1000437 overt behavioral responses to attitude objects. J. Soc. Issues 25, 41–78. doi: 10.1111/j.1540-4560.1969.tb00619.x Conflict of Interest: The authors declare that the research was conducted in Wiers, R. W., Van Woerden, N., Smulders, F. T. Y., and De Jong, P. J. (2002). the absence of any commercial or financial relationships that could be  construed Implicit and explicit alcohol-related cognitions in heavy and light drinkers. as a potential conflict of interest. J. Abnorm. Psychol. 111, 648–658. doi: 10.1037//0021-843X.111.4.648 Wigboldus, D. H. J., Dijksterhuis, A., and van Knippenberg, A. (2003). When Copyright © 2019 Meissner, Grigutsch, Koranyi, Müller and Rothermund. This is stereotypes get in the way: stereotypes obstruct stereotype-inconsistent trait an open-access article distributed under the terms of the Creative Commons Attribution inferences. J. Pers. Soc. Psychol. 84, 470–484. doi: 10.1037/0022-3514.84.3.470 License (CC BY). The use, distribution or reproduction in other forums is permitted, Wilson, T. D., Lindsey, S., and Schooler, T. Y. (2000). A model of dual attitudes. provided the original author(s) and the copyright owner(s) are credited and that Psychol. Rev. 107, 101–126. doi: 10.1037/0033-295X.107.1.101 the original publication in this journal is cited, in accordance with accepted academic Wiswede, D., Koranyi, N., Müller, F., Langner, O., and Rothermund, K. (2013). practice. No use, distribution or reproduction is permitted which does not comply Validating the truth of propositions: behavioral and ERP indicators of truth with these terms. Frontiers in Psychology | www.frontiersin.org 16 November 2019 | Volume 10 | Article 2483

Journal

Frontiers in PsychologyPubmed Central

Published: Nov 8, 2019

There are no references for this article.