Implicit Measures of Personality

Implicit personality; Implicit self-concept; Indirect measure of personality

Implicit measures of personality that for reasons addressed below could be better named indirect are used to reflect the relation between “self” and a given personality feature and allow to infer one’s implicit self-concept indirectly from behavioral responses in specifically designed computer tasks instead of relying on a direct self-assessment. Hence, they can provide information about personality aspects that people may be unable or unwilling to report when asked directly and have demonstrated predictive validity of relevant criteria, especially under certain conditions (De Cuyper et al. 2017; Greenwald et al. 2009; Perugini et al. 2010).

The most common way to measure personality is via self-report. Basic personality structures such as the Big Five (McCrae and Costa 2008) and the Big Six (Ashton et al. 2004) are based on direct self-report measures, that is, by asking people about some aspects of their personality such as how frequently they act in a certain way or what kind of reactions they have in a certain context. Despite the many advantages of personality self-reports (e.g., ease of administration, flexibility) and cumulated evidence concerning their validity (e.g., Ozer and Benet-Martínez 2006), there is shared awareness that people are sometimes unwilling or unable to provide accurate reports of their own personality. Another less common but still popular way to measure personality is by using peer reports, the main difference being that this time the source of information is someone else, typically one or more well-acquainted individuals. Although these measures suffer less from introspective biases, self-presentational tendencies, and social desirability (Vazire and Carlson 2011), their validity for personality assessment is not unequivocally superior to self-reports, nor should they be seen as substitute for the latter (Paunonen and O’Neill 2010) especially when trait observability is low and the focus is on self-concepts. During the last two decades, starting from the seminal work of Greenwald et al. (1998), another way of measuring the self-concept of personality has been developed that reduces participants’ ability to control their responses, and it is less reliant on introspection, hence reducing the potential biases recalled above. These types of measures are commonly referred to as implicit or indirect, in contrast with traditional self-reports that can be described as explicit or direct.

The implicit versus indirect issue has been frequently addressed (e.g., Gawronski and De Houwer 2014; Perugini et al. 2015), and a certain consensus seems to have emerged in recent years. On the one hand, implicit refers to the way by which a psychological attribute influences measurement outcomes if this influence is unintentional, resource-independent, unconscious, or uncontrollable. It does not refer to the measurement procedure itself or the underlying psychological construct. On the other hand, indirect refers to measurement procedures when the measurement outcome is not based on a direct assessment of the to-be-measured attribute (e.g., conscientiousness is inferred from reaction times to a categorization task containing stimuli related to the trait and to the self). Therefore, an indirect measure is a task that reflects the to-be-measured construct through automatic processes that operate in the relative absence of awareness, intention, cognitive resources, or time (Gawronski and De Houwer 2014). This type of measures have thus the potential to tap into aspects of the self that are not, fully or partially, accessible through introspection or willing to be disclosed.

Indirect measures for personality research rely on the self-concept. Self-concepts are defined at an abstract level and incorporate summary beliefs about oneself (e.g., outgoing) instead of contextualized specific behavioral episodes (e.g., being the first to start a conversation with strangers during a party). These beliefs become connected over time to one’s self-representation (see Back et al. 2009, for a model of how this can happen). The content of these personality-related beliefs can range from motivational to gender identity, among others. Although a compelling case could be made that they are all related to personality in a broader sense provided that they refer to relatively stable aspects of the individual (Baumert et al. 2017), for pragmatic reasons we will consider here only personality traits and self-esteem. With this restriction in mind, we can define therefore the self-concept of personality as a collection of attributes (i.e., stimuli) describing relatively stable personality traits (i.e., categories) that are paired with the concept of the self. Examples of stable personality traits are broad traits such as the Big Five or Big Six or narrower traits such as anxiety or aggressiveness. The most commonly used attributes are personality-descriptive adjectives (e.g., threatening), but they can also be other types of words, such as verbs (e.g., to threat). When the attributes refer to a general valence dimension (e.g., positive vs. negative), we talk about self-esteem. When the assessment method is direct and aims at capturing one’s knowledge about oneself, we refer to as a measure of explicit self-concept, whereas when it is indirect and aims at reflecting one’s automatic beliefs about oneself, we refer to as a measure of implicit self-concept. Note that we explicitly avoided using the term “associations” in contrast with standard definitions in the field. The reason is that recent theorizing and empirical evidence suggest that what are commonly called associations might in fact be also the result of the automatic activation and formation of propositions (see De Cuyper et al. 2017). It seems important therefore to use the more neutral term beliefs instead of associations and to leave as an empirical question whether automatically activated beliefs include also relational qualifiers (e.g., I am) instead of assuming that they are associations, hence by definition without relational qualifiers (De Houwer 2014).

Several indirect measurement tasks have been used to measure specific or broad implicit self-concepts of personality and implicit self-esteem. Table 1 reports a list of tasks with examples of research applications. We provide procedural and validity details for some of them.

The most frequently used paradigm to assess the implicit self-concept of personality is Greenwald et al.’s (1998) IAT (see De Cuyper et al. 2017). The IAT is a classification task during which words appearing one by one in the middle of a computer screen are assigned to four categories – two attribute categories (e.g., moral vs. immoral) and two target categories (e.g., self vs. others) – by pressing one of two keys, as fast and accurately as possible. There are usually five different tasks starting with a single classification block of attributes (i.e. moral vs. immoral) and a single classification block of targets (i.e. self vs. others), then one critical block, a reverse classification block of targets (i.e. others vs. self), and finally a second critical block. During the critical blocks, one target and one attribute category (e.g., moral and self) are classified with one key, and the other two categories (e.g., immoral and others) are assigned to the other response key. The key assignments of the categories vary between the two critical blocks. The underlying principle of the IAT is that if an individual’s one’s automatic beliefs about him-/herself have more to do with morality rather than with immorality, he/she will perform better (faster and more accurate responses) in the critical block in which moral and self-stimuli are classified with the same key than in the critical block in which immoral and self-stimuli are classified with the same key. The difference in performance (errors and reaction time) between the two critical categorization tasks is reflected in a D-score (Greenwald et al. 2003).

Although far from being systematic, evidences of validity for the self-concept IATs are present in many research. Schmukle et al. (2008) validated the five personality factor structure at the implicit level using the IAT, demonstrating a structural equivalence between implicit and explicit measures. Asendorpf et al. (2002) showed that a shyness IAT uniquely predicted spontaneous but not deliberate behavior, whereas self-reports uniquely predicted deliberate but not spontaneous behavior (double dissociation pattern; for other examples of patterns of prediction, see Perugini et al. (2010)). Few variants of the IAT have also been used to study aspects of the self-concept of personality. For example, the Single Category IAT (SC-IAT; Karpinski and Steinman 2006), in which only one attribute category is assessed, is useful for concepts that do not have a natural opposite such as perfectionism (e.g., De Cuyper et al. 2013). The Brief IAT (BIAT; Sriram and Greenwald 2009) that is a third shorter than the standard IAT in terms of trials and allows focusing on just two of the four IATs’ category-response mappings in each combined task has been used to assess implicit self-esteem (e.g., Krause et al. 2011).

Schnabel et al. (2006) developed the IAP to assess the implicit personality self-concept of shyness. Participants pushed a joystick toward or away from oneself dependent on whether a stimulus had to be paired with me or not me. Similar to the IAT, the IAP combined attribute discriminations (shy vs. non-shy) with target discriminations (me vs. not me). However, unlike in the IAT, only the me label was present on the screen, not me describing the non-self-relevant alternatives and allowing thus the assessment of beliefs with unipolar concepts. Participants first discriminated, as quickly and accurately as possible, me and not me words; then, with the addition of shy and non-shy words, they had to pull them toward or push them away from themselves, respectively (combined block 1); and, finally, shy and non-shy words responses were reversed (combined block 2). Error feedbacks were given in case of incorrect, too slow, or too fast responses. The IAP score was the difference in reaction times between the two combined blocks. The IAP showed satisfactory internal consistency and correlated with a shyness IAT and with explicit shyness. However, a double dissociation pattern where indirect measures predicted spontaneous and direct measures predicted controlled behavior (Asendorpf et al. 2002) could not be replicated.

Among the other measures, the EAST (De Houwer 2003) was adapted by Teige et al. (2004) to assess shyness, anxiousness, and angriness together with IATs and direct self-ratings. In a first block, participants had to discriminate, as quickly and accurately as possible, the white stimuli of the category Others vs. Me by pressing two keys, thus defining an “Others-key” and a “Me-key.” In a second block, participants had to discriminate colored attribute stimuli (shy vs. non-shy, anxious vs. secure, and angry vs. self-controlled) according to their color (blue vs. green) but by pressing the same left “Others- key” or right “Me-key.” The tasks were combined in two test blocks. The EAST showed low internal consistencies and correlated neither with the IATs nor with the direct measures.

Huntjens et al. (2014) elaborated the Irrelevant Feature Task used for assessing the implicit self-concept of self-control. Each trial was preceded by a 1,500 ms prime “I am” vs. “Others are.” After a fixation cross of 500 ms, the stimulus was displayed. The stimuli consisted of self-control congruent words (e.g., controlled, calm) and self-control incongruent words (e.g., chaotic, impulsive). Participants had to answer orally and as fast as possible “Yes” (vs. “No”) to uppercase letters and “No” (“Yes”) to lowercase letters. The relationship between relevant feature (i.e., upper or lower case) and relevant response (“Yes” or “No”) was counterbalanced across participants. Response latencies were recorded through a voice key and errors by a research assistant. Although the evidence was obtained on a small sample, the IFT showed predictive validity for spontaneous self-control behavior.

With the Semantic Misattribution Procedure, Sava et al. (2012) assessed the implicit personality self-concept related to conscientiousness, neuroticism, and extraversion. Each trial was structured with a sequence of a 200 ms prime, a 125 ms mask, a 200 ms neutral Chinese pictograph, and a gray rectangular noise pattern until the participant responded. Participants indicated whether they would like each Chinese character to be printed on a personalized T-shirt by responding “Fits me” or “Does not fit me.” Descriptors for each pole of each trait were used as primes. A SMP score was obtained by summing “fits” responses to high conscientiousness (neuroticism) primes and “does not fit” responses to low conscientiousness (neuroticism) primes. Results showed good convergent validity of the SMP scores with self-reports and predictive validity for both self-reported and objective behaviors.

Boldero et al. (2007) used the GNAT (Nosek and Banaji 2001) to assess Big Five dimensions (neuroticism, extraversion, openness, agreeableness, conscientiousness) using idiographically generated roles. The extraversion/introversion GNAT had two blocks, one in which the target categories were “self” and extraversion whereas in the other they were “self” and introversion. The distractor terms were “not self” roles and alternate personality terms (i.e., introversion terms for the self-extraversion block). For each of the single Big Five factors, there was only one block in which the target categories were “self” and the to-be-measured trait and the distractor categories were “not self” roles and the alternate personality traits (i.e., terms from the other four traits). Participants had to press the spacebar as quickly as possible (response window of 700 ms) if the term belonged to a target category or do nothing if it did not. They received performance feedback. Split-half reliabilities of the d’ and D indices were adequate. Implicit extraversion, openness, and agreeableness predicted their respective explicit measures.

Remue et al. (2014) adapted the IRAP (Barnes-Holmes et al. 2006) to assess the actual and the ideal selves. Different from the other measures presented here, the IRAP can allow differentiating between beliefs related to being bad and wanting to be good using different types of stimulus combinations. In the actual self-esteem IRAP, each block consisted of trials that presented one of two self-related stimuli (e.g., “I Am” or “I Am Not”) in the presence of one of two types of target stimuli (positive or negative words) and required participants to provide “True” or “False” response. In this way, there were four different types of trials (self-positive, self-not positive, self-negative, and self-not negative). In the ideal self-esteem IRAP, participants responded to the same stimuli in terms of “I Want To Be” or “I Don’t Want To Be.” Whereas the IAT was not able to distinguish between dysphoric and non-dysphoric participants, the two IRAPs were able to do so such that there was no difference between actual and ideal self-esteem for non-dysphorics, whereas dysphorics demonstrated lower actual than ideal self-esteem.

Focusing more specifically on the psychometric properties, De Cuyper et al. (2017) with their systematic review of implicit self-concept of personality measures showed that there is evidence for reliability and construct and predictive validity but not for all measures and all domains of personality. Reliability was higher for the IAT compared to Single Category and Brief IAT and compared to SMP. Concerning the IAT, the meta-analysis provides evidence of construct and predictive validity, especially in the extraversion and agreeableness domains. Current evidence suggest that the IAT is the best available indirect measure and that alternatives do not approach the level of the IAT in terms of psychometric properties (De Cuyper et al. 2017; Sava et al. 2012). Moreover, Bosson et al. (2000) tested the psychometric properties of seven measures of implicit self-esteem and showed that only three measures correlated with some criterion variables, with the IAT showing the best performances in terms of predictive validity. The IAT should therefore be considered as the most established indirect measurement task for assessing the self-concept of personality.

Each of the indirect measurement procedures described above can be implemented in very different ways, thereby a procedure such as the IAT or the IRAP should be conceived of as “class of measures” rather than as “a measure.” Knowledge that the IAT is in general a valid procedure from a psychometric standpoint does not imply that each specific implementation will be valid nor that all implementations are equally valid (Perugini et al. 2015). When developing an indirect measure, researchers face several choices that affect substantially the properties of the measure. First, one has to identify a measure that is suitable for assessing the construct at hand. Whereas the IAT seems to possess good psychometric qualities in general, other measures can be more suitable for specific purposes. For instance, whereas the IAT requires a contrast attribute category, other procedures such as the SC-IAT do not have this requirement. Furthermore, the IAT cannot easily assess propositions, a possibility that is instead straightforward with the IRAP (Remue et al. 2014).

Second, it is important to choose an appropriate level of abstractness for the construct. For example, in the case of the self-concept of personality, one could choose to assess a broad dimension, such as one of the Big Five (Schmukle et al. 2008) or its facets separately (Costantini et al. 2015). In the case of the IAT, it is important that the construct selected is sufficiently cohesive, such that the stimuli can be easily paired with the corresponding category (De Cuyper et al. 2017). It is also important to choose an appropriate set of stimuli to assess the construct. Whereas the importance of stimuli (i.e., items) is clearly recognized for direct measures, it is often underestimated when it comes to indirect measures (Perugini et al. 2015). One possibility is to identify a valid set of markers for each construct through a preliminary study, which gives also the possibility of validating an additional direct measure of the construct against known validated measures and establishing convergent and discriminant validity (e.g., Costantini et al. 2015). When examining candidate stimuli for self-concept measures, adjectives and nouns are often preferred; however, some empirical evidence suggests that verbs may perform better when assessing constructs that are strongly connected to action tendencies, such as aggressiveness (see De Cuyper et al. 2017). Furthermore, stimuli can also be personalized for each participant (e.g., Boldero et al. 2007). Apart from the choice of stimuli, several other details of the implementation of an indirect measure can affect its psychometric properties (Krause et al. 2011; Remue et al. 2014). Among these aspects, one that is particularly important is the choice of a scoring algorithm. Different ways of treating errors and extreme latencies can make a difference for the psychometric properties of a measure (Greenwald et al. 2003; Krause et al. 2011; Richetin et al. 2015). For instance, the IAT and its variants are often scored using the algorithm developed by Greenwald et al. (2003). However, incorporating robust statistical concepts in the scoring procedures, for instance, to handle extreme latencies, may improve reliability and validity (Richetin et al. 2015).

A recent meta-analysis on direct and indirect measures of self-concept (excluding self-esteem) did find an average correlation of around 0.20 (De Cuyper et al. 2017), which indicates a modest but not negligible amount of convergence. It should be noted however that research using indirect measures was initially focused on patterns of double dissociation (e.g., Asendorpf et al. 2002) as a way to show evidence of incremental validity of indirect measures as well as of divergence with direct measures such as self-reports. An initial mainstream assumption was that the more the two types of measures diverged, the better, both because this was creating better conditions to show evidence of validity of indirect measures and because it was an indication that they were actually tapping into different processes and mental contents. However, Greenwald et al. (2009) did find in their meta-analysis of the IAT that the strongest significant moderator of the IAT predictive validity was its correlation with direct measures, meaning that IATs were more predictive with increasing correlations with indirect measures. The explanation given by Greenwald et al. (2009) is that the constructs assessed by indirect and direct measures tap likely reinforce each other in determining behavior (see also Perugini et al. 2010). Conversely, discrepancies between measures have been taken to indicate unpleasant psychological states or the asymmetric effect of additional factors such as social desirability concerns, affective focus, and validation processes (Gawronski and Hahn in press). However, from a methodological point of view, discrepancies between measures could be due to lower reliability of indirect measures, to violations of the principle of correspondence (i.e., degree of matching content in the relevant target object), of the principle of structural fit (i.e., degree of methodological similarity between two different measures), and to a poor process of development of an indirect measure leading to mediocre psychometric properties, including their validity (Gawronski and De Houwer 2014; Perugini et al. 2010, 2015). One implication is that before interpreting divergence between direct and indirect measures in terms of theoretical implications, one should make sure that it is not a by-product of psychometrically deficient measures. More in general, the current practice of the process of development and validation of an indirect measure needs improvements to achieve the standards established over the years for the development of direct measures (e.g., self-reports for personality traits).

Indirect measures have started to be used in the personality field mainly during the last two decades following the seminal contribution of Greenwald et al. (1998) where the IAT was first introduced. Their use has been increasing over time although not as much as in the attitudinal fields. The evidence in terms of their validity accumulated so far is encouraging, especially for the IAT. Alternative paradigms of indirect measures have been developed during the years and some of them appear promising, although the evidence is still too limited and future studies are needed to establish their validity. Increasing attention to methodological issues can lead to improvements in the research area.

• Actual Self

• Big-Five Model

• Conscious Versus Unconscious Determinants of Behavior

• Conscious, Preconscious, and Unconscious

• Extrinsic Affective Simon Task

• Ideal Self

• Identity

• Implicit Self-esteem

• Self-Appraisals

• Self-Esteem