7 General Discussion


The present research comprised 3 studies. In Study 1, the shyness IAT and a new indirect procedure, the shyness IAP, showed considerable convergent validity. Both of the indirect measures were much less susceptible to faking instructions than the direct self-ratings. Additionally, under faking instructions, the correlations of direct and indirect measures with shy behavior decreased more strongly for the direct rather than for the indirect measures. However, there was a lack of valid behavioral codings for controlled and spontaneous shy behavior. Therefore, the double dissociation pattern of Asendorpf et al.’s (2002) Study 1 was not replicated. In Study 2, the anxiousness IAT added incremental validity over direct anxiousness measures to the prediction of anxious behavior. However, the angriness IAT was affected by a transfer effect from the anxiousness IAT. Study 3 provided further evidence that this transfer effect was due to the salience of a positive-negative self-dimension.

The General Discussion refers to three aspects of these findings. First, to what extent are indirect measures influenced by the semantic meaning or by the positive and negative valence of the stimuli? Second, why are direct and indirect measures of the personality self-concept different from each other? Third, what can be learned from the present and other findings for the assessment of the implicit personality self-concept?

7.1  Semantic Meaning versus Valence


The findings of Study 2 and Study 3 suggested that the positive correlation between the anxiousness IAT and the angriness IAT was attributable to a task-recoding in terms of a positive-negative self-dimension. This raises a question about the extent to which IAT measures are driven by the semantic meaning as opposed to the positive or negative valence of the stimuli. If IATs mainly reflect the ease with which one combines positive versus negative stimuli with Me, then the IATs represent self-esteem IATs (e.g., Greenwald & Farnham, 2000) rather than indirect measures of different personality traits. Can the empirical findings of different self-concept IATs be re-interpreted in terms of implicit measures of self-esteem?

Concerning the anxiousness IAT, the answer might be ‘yes’. In the studies by Egloff and Schmukle (2002), the anxiousness IAT predicted performance decrement due to failures in a concentration test, and anxious behavior during an evaluative speech task. Both behaviors may also be predicted by a ‘pure’ self-esteem IAT that does not directly refer to anxiousness (cf. Greenwald & Farnham, 2000). The results from the anxiousness IAT in Study 2 of the present research may be re-interpreted using the same logic. The same reasoning can be applied to the shyness IAT (Asendorpf et al., 2002; Study 1 of the present research) as well, such that shy behavior could be predicted by low self-esteem. Already at the level of direct measures, shyness and anxiousness are negatively correlated with self-esteem (Cheek & Melchior, 1990; Judge, Erez, Bono, & Thoresen, 2002). Thus, it is difficult to disentangle valence and semantic meaning in anxiousness and shyness because a valid portion of these traits already contains negative valence.

There are at least three indicators that the semantic meaning of the stimuli may influence IAT scores. First, self-esteem IATs show that most people are quicker in combining Me with positive attributes than in combining Me with negative attributes. Thus, most individuals have positive implicit self-esteem (Greenwald & Farnham, 2000). This pattern did not hold to be true in the shyness IAT of Asendorpf et al.’s (2002) Study 1. The shyness IAT indicated that about 40 % of the participants are more shy than non-shy. If the shyness IAT could be re-interpreted as self-esteem IAT, one would expect fewer participants to show positive scores in the shyness IAT.


Second, the research on priming procedures provides evidence that semantic and affective priming procedures show effects even with very short SOA (stimulus onset asynchrony, i.e., the interval between start of prime and start of target stimulus) (Greenwald, Draine, & Abrams, 1996; Klauer & Musch, 2003). Thus, both semantic and evaluative information is processed very quickly. Importantly, affective priming seems to depend on whether respondents have to identify evaluative or non-evaluative target attributes. Affective priming does usually not occur when targets are classified on the basis of non-evaluative features. (De Houwer et al., 2002, Klauer & Musch, 2003). To my knowledge, the present evidence of affective and semantic priming does not indicate that priming effects are more influenced by either valence or semantic meaning of stimuli. In contrast, the characteristics of the priming task seem to influence whether valence information or semantic information causes priming effects.

Third, in two experimental groups of Study 3, the angriness IAT showed convergent validity for direct angriness measures (up to r = .52) in addition to discriminant validity for direct anxiousness measures (all r < .14). This finding could not have been explained had the participants classified stimuli within the angriness IAT only in terms of a positive-negative dimension.

On the other hand, task-recoding in terms of a positive-negative self-dimension cannot be completely ruled out in self-concept IATs due to a relatively strong connection between the concept of self and positive valence (Greenwald et al., 2002). Thus, the categorization of Me versus Others may automatically activate the positive-negative dimension. Self-concept IATs with two positive or two negative traits as target attributes could block the positive-negative dimension. For instance, one could employ a self-concept IAT with anxiousness versus angriness as attribute categories. However, this IAT should, then, show identical results for those individuals who score high on anxiousness and angriness and those who score low on both traits. The results of recent studies provide evidence that it is difficult to separate the IATs’ measure of relative associations into two independent measures (Nosek et al., 2003).


The positive-negative self-dimension may alternatively be weakened by the following factors. First, the positive or negative valence of IAT stimuli and IAT attribute categories should not be too extreme (cf. Footnote 4 from De Houwer, 2001; Steffens et al., 2003). For instance, although shyness is a negatively valenced trait, it is only so to a moderate extent. Likewise, with regard to anxiousness and angriness, one should consider stimuli with more positive valence, for example, cautious and resistant, respectively. Second, one should make an effort to highlight the semantic meaning of stimuli and to block a positive-negative self-dimension. The presentation of evaluatively neutral anagrams seemed to be successful in Study 3.

7.2 Dissociations between Direct and Indirect Measures of the
Personality Self-Concept

There would be no interest in researching indirect measures if indirect and direct assessment procedures measured identical constructs. In order to differentiate between operationalizations and constructs, in this work, the terms direct and indirect measures were used to label procedures, and the terms explicit and implicit representations were used to label the constructs. Similarly, there may be two sources for dissociations between direct and indirect measures of the personality self-concept: (a) theoretically-based dissociations between explicit and implicit representations at the construct level; (b) method factors in direct and indirect measures at the assessment level.

Concerning the construct level, explicit representations were regarded as propositional categorizations within the Reflective System, and implicit representations as associative clusters within the Impulsive System of the Reflective-Impulsive Model from Strack and Deutsch (in press) (see Chapter 2.2). Thus, explicit representations should be better predictors of controlled behavior, and implicit representations should be better predictors of spontaneous behavior. Recently, Asendorpf et al. (2002) carried out a double dissociation procedure between the explicit and implicit personality self-concept of shyness. A direct shyness questionnaire uniquely predicted controlled (but not spontaneous) shy behavior, whereas a shyness IAT uniquely predicted spontaneous (but not controlled) shy behavior. However, the results of the present studies, Study 1 and Study 2, showed that it is difficult to differentiate between indicators of spontaneous and controlled behavior.


Attempts to show predictive validity of indirect measures often follows an incremental validation strategy. This means that the studies usually explore whether indirect measures predict variance in relevant criteria in addition to direct measures of the same construct (for a review, cf. Fazio & Olson, 2003). In Study 2 of the present research, the anxiousness IAT added incremental validity over direct self-ratings to the prediction of the observer anxiety judgments. Conceptually, the incremental validity of the indirect measures might be attributed to two differences between explicit and implicit representations, (a) implicit representations have more direct access to the associative store than explicit representations, (b) explicit representations might be biased due to social desirability concerns (cf. Chapter 2.3).

Biases based on social desirability also affect method factors of direct and indirect measures. For instance, whereas direct self-ratings are certainly fakable (cf. Study 1), there is a controversy about IATs being fakable or not (cf. Chapter 2.4.1). Study 1 provided evidence that the shyness IAT was to some extent fakable by. However, effects were much smaller than for direct self-ratings.

Another methodical issue could refer to the question of whether the indirect procedure employs negation or not. For instance, the typical target categories of self-concept IATs are Me versus Others. For the IAP in Study 1, target categories were Me versus Notme. According to Strack and Deutsch (in press), the Impulsive System is not able to negate information. More precisely, the Impulsive System is not able to assign a true or a false value to the relation between two concepts. Instead, the Impulsive System only connects or does not connect concepts using episodic and semantic links that are available within the associative store. Therefore, it may be an interesting topic for further research to explore whether indirect measures that employ negation (e.g., the GNATs “Go/No-Go Association Tasks”, Nosek & Banaji, 2001; the shyness IAP of Study 1) are influenced by the Reflective System more than indirect measures that do not employ negation. However, the shyness IAP employed in Study 1 did not seem to be more controllable than the shyness IAT since the IAP was even less susceptible to faking instructions. Additionally, the IAT and the IAP contained a negation for the attribute category, that is, Shy versus Nonshy.


Another characteristic of indirect measures is that they can be influenced by the category frame of the categorization task and by individual stimulus features (cf. Fazio & Olson, 2003). In contrast, direct self-ratings, for example, bipolar adjectives, are judged individually, that is, they are only influenced by individual stimulus features. To obtain the mean scale, the bipolar items are combined such that every item is weighed equally. Concerning IATs, there is some evidence that individual stimulus features have an effect on the IAT score, while the category frame is more influential (cf. Chapter 2.4.2).

Finally, dissociations between direct and indirect measures may also be caused by the context dependency in indirect measures. Although indirect measures seem to be not affected by emotion inductions (Schmukle & Egloff, 2003; cf. Study 1 of the present research), they were shown to be influenced by other contextual variables (cf. Mitchell et al., 2003). Importantly enough, the present research revealed evidence that self-concept IATs are affected by the salience of a positive-negative self-dimension (see Chapter 7.2).

In summary, dissociations between direct and indirect self-concept measures might be attributed to differences at the construct and at the measurement level. At the construct level, implicit representations differ from explicit representations because of their more direct access to the associative store and a more direct effect on spontaneous behavior. At the measurement level, there are method factors that are characteristic of indirect procedures rather than of direct procedures. Indirect procedures are less fakable, presumably less apt to assess negated concepts, are influenced by the category frame and stimulus features, and are susceptible to contextual variables, particularly to the salience of a positive-negative self-dimension.

7.3  Recommendations for Future Research


This section summarizes some aspects of indirect measures, particularly of IATs, that may help future research on the implicit personality self-concept to be more successful. First, IAT D measures (Greenwald et al., 2003) were shown to control for method-specific variance due to task-switching (cf. Hypothesis 3 from Study 3). Therefore, future data reduction procedures should employ D measures.

Second, it would be most useful to standardize the IAT procedure, that is, using a standardized number of trials in the five different IAT tasks, and even using a standardized number of stimulus items per IAT category. Although the effects of such variations may be insignificant (Nosek et al, 2003), it could make research on IATs much more comparable. Particularly, it was shown that the effect of task order (IAT effects are larger if the compatible pairing is completed before the incompatible pairing) was reduced by increasing trials in the reversed target discrimination task (see Table 3). A number of 40 trials (like in the present studies) seems to be optimal for the reversed target discrimination (Nosek et al., 2003). This finding is important for the assessment of the personality self-concept because the task order effect seems to maximize interindividual differences for participants with positive IAT scores (see Chapter 2.4.2)

Third, the order of the compatible and incompatible IAT pairing should not be counterbalanced for the assessment of the personality self-concept. Otherwise, order variance is confounded with interindividual variance. Fourth, correlations between direct measures and the IAT seem to be higher if the direct measures are completed before the IATs (Hofmann, Gawronski, et al., 2003; cf. Chapter 6.5.2; for findings revealing no order effect, cf. Nosek et al., 2003). Thus, if one aims to maximize consistency between direct and indirect measures, one should apply direct measures first.


Fifth, results of the present studies indicate that IATs may be influenced by the semantic meaning and by the positive or negative valence of IAT stimuli. Therefore, personality self-concept IATs should try to use moderately valenced stimuli and to block the salience of a positive-negative self-dimension.

Sixth, there are promising results for self-concept IATs that use stimuli that describe behaviors (e.g., fistcuff) rather than personality traits (e.g., aggressive) (Banse & Fischer, 2002). Within the Reflective-Impulsive Model from Strack and Deutsch (in press), behavioral motor schemata are subsumed into the Impulsive System. Thus, it seems plausible that stimuli describing behavior are more strongly represented in the associative clusters of the Impulsive System rather than abstract personality traits. Therefore, IATs with behavior describing items may provide a more direct access to the associative store.

Seventh, one should include the stimuli of the indirect measures as direct self-ratings. This allows for a fair comparison between direct and indirect measures. Additionally, the internal consistency of the stimuli that represent one trait can also be checked at the direct level. This could indicate whether the stimuli fit well in the superordinate category.


Eighth, if one employs more than one IAT in order to assess different traits within one study, one should take into consideration that transfer effects between the IATs might distort the IAT effects. Study 2 and Study 3 provided evidence that the position effect on the angriness IAT was due to a task-recoding in terms of a positive-negative self-dimension that was transferred from the anxiousness IAT. A positive-negative task-recoding might be quite salient in self-concept IATs due to the strong connection of Me with positive valence (Greenwald et al., 2002). In addition, there may be other superordinate categories for other traits, for example, male-female, young-old, intelligent-unintelligent, that can bias the IAT scores.

© Die inhaltliche Zusammenstellung und Aufmachung dieser Publikation sowie die elektronische Verarbeitung sind urheberrechtlich geschützt. Jede Verwertung, die nicht ausdrücklich vom Urheberrechtsgesetz zugelassen ist, bedarf der vorherigen Zustimmung. Das gilt insbesondere für die Vervielfältigung, die Bearbeitung und Einspeicherung und Verarbeitung in elektronische Systeme.
DiML DTD Version 4.0Zertifizierter Dokumentenserver
der Humboldt-Universität zu Berlin
HTML-Version erstellt am: