Moderation of masked affective priming 1 Running head: MODERATION OF MASKED AFFECTIVE PRIMING Automatic evaluation isn’t that crude! Moderation of masked affective priming by type of valence

In two experiments, the automatic processing of evaluative information was investigated using a masked affective priming paradigm, varying valence (positive vs. negative) and relevance (other-relevant traits vs. possessor-relevant traits; Peeters, 1983) of prime and target stimuli. It was found that under specified conditions, valence-congruency effects were only found if prime and target matched with regard to relevance type (i.e., were both either of the other-relevant or possessor-relevant type). These results suggest that automatic processing of affective information conveys not only the positive-negative differentiation, but also the relevance type of valence. Consequences for research on automatic attitudes, especially prejudice are discussed. For current research on masked priming, it is important to highlight that the subliminal effect was found even for non-practiced prime stimuli. Moderation of masked affective priming 4 Automatic evaluation isn’t that crude! Moderation of masked affective priming by type of valence Since the seminal article on affective priming by Fazio, Sanbonmatsu, Powell, and Kardes (1986), a lot of research has been done on the automaticity of evaluation (for reviews see Fazio, 2001; Klauer & Musch, 2003; Wentura & Rothermund, 2003). Typically, affective priming is assessed with an evaluation task: participants have to decide for each target stimulus whether it is positive or negative. Shortly before each target, a prime stimulus is presented. In cases of affective congruence (i.e., prime and target share the same valence), response times (and/or number of errors) are expected to be lower than in cases of affective incongruence (i.e., prime and target are of different valence). Although there has been some debate about the underlying mechanisms (see, e.g., De Houwer, Hermans, Rothermund, & Wentura, 2002; Klauer & Musch, 2003; Spruyt et al., 2002), it is rather undisputed that this effect reflects the automatically – in the sense of nonintentionally – activated evaluation of the prime. Given this background, the affective priming procedure was proposed as a measurement tool for automatic attitude activation (Fazio, Jackson, Dunton, & Williams, 1995). The underlying rationale is that automatic evaluation of attitude-related stimuli can be inferred from affective priming effects if these stimuli are presented as primes. Moderation of masked affective priming 5 Meanwhile, it has been shown that affective priming effects can be found with subliminally presented prime stimuli (Abrams, Klinger, & Greenwald, 2002; Draine & Greenwald, 1998; Greenwald, Draine, & Abrams, 1996; Greenwald, Klinger, & Liu, 1989; Greenwald, Klinger, & Schuh, 1995; Klauer, Eder, Greenwald, & Abrams, 2007). Thus, even if prime stimuli are processed outside of awareness, their valence is activated. This feature renders the evaluation task the most unobtrusive measure of automatic evaluations. Given this backdrop, we were able to fruitfully use the affective priming technique in its masked version in several studies, tapping different domains of evaluation (Degner & Wentura, in press; Degner, Wentura, Gniewosz, & Noack, 2007; Frings & Wentura, 2003; Otten & Wentura, 1999; Wentura, Kulfanek, & Greve, 2005). The present studies are concerned with a question that is highly relevant for research on automatic processing of affective connotations in general, and research on indirect assessment of attitudes in specific: Is automatic evaluation a rather undifferentiated process that only conveys general positive or negative affective connotations? Or do we find more subtle differentiations at the level of fast, effortless, and unconscious valence processing? With the following studies we explore a candidate distinction for such a differentiation, that is, the othervs. possessor-relevance introduced by Peeters (1983; Peeters & Czapinski, 1990). Other vs. Possessor-Relevance Moderation of masked affective priming 6 Peeters (1983; Peeters & Czapinski, 1990) introduced a typology of valenced trait adjectives according to the kind of positivity or negativity they convey. That is, the evaluation of a given trait depends on the perspective of the evaluator – whether they evaluate the trait from the perspective of the trait-holder him/herself or from the perspective of someone who has to interact with the trait-holder. Thus, this typology is termed possessorvs. other-relevance (or selfvs. other-profitability, see Peeters, 1983). To give an example, being brutal is primarily bad for the social environment of the brutal person, but not necessarily for the brutal person him-/herself (he/she might be a sadist). On the contrary, being lonely is primarily bad for the lonely individual but not necessarily for his/her social environment. The same applies to positive adjectives: Being honest is primarily good for those who interact with the honest person but not necessarily for the honest person him-/herself (honesty might be abused), whereas being intelligent is primarily good for intelligent persons themselves but not necessarily for the social environment (she or he might have selfish motives). Adjectives like brutal or honest are called otherrelevant, whereas words like lonely or intelligent are called possessorrelevant. Wentura, Rothermund, and Bak (2000) provided first evidence that this typology is relevant for automatic evaluation processes. They found that other-relevant stimuli increased color-naming times in the “Emotional Moderation of masked affective priming 7 Stroop” task compared to possessor-relevant words. Thus, there is some evidence that our cognitive-affective system is tuned to distinguish between these kinds of positivity and negativity at a very basic level. If automatic evaluation differentiates into otherversus possessor-relevance, it can be hypothesized that affective priming will be susceptible to different types of valence as well. In fact, applying the affective priming paradigm to the indirect assessment of attitudes provided first indications that (a) automatic evaluation of socially relevant stimuli seem to comprise a differentiation of possessorand other-relevance and that (b) this can potentially be assessed with the affective priming task: Interindividual differences in social attitudes such as self-esteem or intergroup prejudice – as measured by questionnaires – were differentially and meaningfully related to priming effects based on attitude-related primes and either self-relevant or other relevant target words (Wentura et al., 2005; Degner et al., 2007). These results indicate that the potential of the affective priming paradigm for the assessment of different types of automatic evaluation should be systematically explored. Overview In the following two experiments, we test for the moderation of masked affective priming effects by relevance type of valence. We hypothesize that – possibly given some boundary conditions – masked affective priming effects are most pronounced if primes and targets are from the same relevance type. Moderation of masked affective priming 8 We used the masked version of affective priming to make strategic influences maximally unlikely. Masking, for example, precludes that participants acquire a clear picture of the trial structure which might be used by participants to successfully manipulate effects (Degner, in press; Klauer & Teige-Mocigemba, 2007). Although it is not our dominant concern to add to the debate on unconscious priming (see, e.g., Dehaene & Naccache, 2001), one might evaluate the experiments also in the light of this research field. Therefore we conducted a direct test of prime awareness. For both experiments, we added a further factor, that is, whether participants were instructed to emphasize speed or accuracy in responding. This was done for the following reason. It is known from the affective priming literature that masked priming effects are a rather robust finding if participants are urged to respond very quickly (see, e.g., Draine & Greenwald, 1998; Greenwald et al., 1996). An obvious explanation for affective priming effects is the assumption that the prime prepares either for the correct or the wrong response needed for the target. Given this explanation, it might be relevant that the targets are not yet fully processed when the response is given: it could then be argued that the weight of the task-relevant target attribute is relatively decreased compared to the corresponding prime attribute when the response is generated (Wentura & Rothermund, 2003). Accuracy instructions might counteract this tendency, because participants will give the target features considerably more weight Moderation of masked affective priming 9 and will thus be more susceptible to differences between prime and target (Wentura & Rothermund, 2003). However, accuracy instructions involve the risk of getting no priming effects at all. Thus, we can consider varying instructions as an “adjustment tool” that helps identifying conditions that make the masked affective priming paradigm sensitive to the relevance type of primes and targets. STUDY 1 We start our exploration with an experiment that should guarantee to establish masked affective priming, using our materials and apparatus. Previous studies found stable masked priming effects especially if primes and targets were drawn from the same set of stimuli, such that over the course of the experiment the primes were presented as targets too (Abrams et al., 2002; Abrams & Greenwald, 2000; Draine & Greenwald, 1998). Method Participants Nineteen students (12 women; 7 men) participated in Study 1a. Ten participants were instructed to emphasize speed; nine participants were instructed to emphasize accuracy. The median age was 22.0 years; all of them were native speakers of German. One participant of the emphasis-onaccuracy sample was replaced because his mean overall response time was more than 2.5 standard deviation units below the mean response time of the emphasis-on-speed sample. Study 1b was done to further clarify a detail in Moderation of masked affective priming 10 the results of the emphasis-on-speed sample of Study 1a (see Results section). Ten students (eight women; two men) participated in this replication. They all received the speed instruction. The median age was 22.0 years; all of them were native speakers of German. Design In detail, the experiment was based on a 2 (target valence: positive vs. negative) x 2 (target relevance: other vs. possessor) x 2 (prime valence: positive vs. negative) x 2 (prime relevance: other vs. possessor) withinparticipants design. However, we can reframe the design as essentially two factors that were varied within participants. First, the congruence vs. incongruence of prime and target valences was manipulated. Second, it was manipulated whether prime and target belonged to the same or different category of relevance. A neutral prime condition was added to get a hint as to whether priming effects (i.e., the difference of RTs to congruent and incongruent conditions) were caused by interference or facilitation processes. For Study 1a, instruction (emphasis on speed vs. emphasis on accuracy) was manipulated between participants. In Study 1b, all participants received the speed instruction. Materials The stimulus set comprised 10 positive and 10 negative German adjectives (see Appendix). Within each valence set, five adjectives were other-relevant and five adjectives were possessor-relevant according to Moderation of masked affective priming 11 norm data (Wentura, Rothermund, & Bak, 1998). The adjectives had a length of five to eight letters. Mean length of the four type sets ranged from 5.60 letters to 6.40 letters. Absolute pleasantness (on a scale ranging from 0 to +100 was between 52 and 89; Hager, Mecklenbräuker, Möller, & Westermann, 1985; Möller & Hager, 1991). Mean absolute pleasantness of the four type sets ranged from 56 to 72. A string of six x’s was used as a neutral prime condition. Procedure Participants were tested either alone or in pairs. They were seated in front of standard personal computers separated by partition walls. Instructions were given on the CRT screen. Participants were told that they had to classify words with regard to their valence. Participants in the emphasis-on-speed samples received the following instruction: “Decide as quickly as possible. Of course, you should not make errors permanently (otherwise your data would be worthless); but the emphasis on speed might be somewhat at the cost of accuracy. Your error rate can be within the range of 15 to 20 %.” Participants in the emphasis-on-accuracy sample received the following instruction: “Decide quickly, but be as accurate as possible. Of course, you should decide quickly (otherwise your data would be worthless) but the emphasis on accuracy might be somewhat at the cost of speed. On average, your error rate should be below 3 %.” For participants of Moderation of masked affective priming 12 the emphasis-on-accuracy sample, a feedback (“Error!”) appeared on the screen in the case of a false response. Parameters of masked prime presentation were the same as those used by Otten and Wentura (1999) and Wentura et al. (2005). A randomly generated string of nine consonant letters (e.g., mlsdzkhwd) marked the beginning of the trial and also served as a forward mask. It remained on the screen for 300 ms and was then immediately replaced by the prime. Thus, as the prime word was always shorter than the forward mask, it was embedded into the string of random consonants (e.g., mlhonestwd). The prime remained on the screen for 43 ms (three refresh cycles) and was immediately replaced by a different randomly generated string of nine consonant letters that served as a backward mask and was displayed for 14 ms (one cycle). It was replaced by the target stimulus that remained on the screen until a response was given. The ‘positive’ response was assigned to the right index finger, the ‘negative’ response to the left index finger. The inter-trial interval was 2500 ms. At the beginning of the priming task, participants worked through 20 practice trials (i.e., each target word was presented once) in order to become familiarized with the task. After the practice trials, a summarized performance feedback was given, indicating the number of errors, the percentage of errors, and the mean RT. In the emphasis-on-speed sample, the following messages were added depending on participants' performance: Moderation of masked affective priming 13 (a) if the number of errors was less than two: “You have made very few errors. You should try to be faster and less cautious”; (b) if the number of errors was between 2 and 4: “Your error rate is appropriate. Continue to respond quickly”; (c) if the number of errors was above 4: “As an average value across the experiment, the error rate is somewhat too high”. In the emphasis-on-accuracy sample, the following messages were added for the emphasis-on-accuracy participants: (a) if there were any errors: “Try to be more cautious!” (b) if they had made no errors: “Continue to be accurate!”. The main part of the experiment consisted of three blocks of 100 trials each. Within a block, each target word was presented once in each of the five priming conditions, with the sequence determined by a Latin-square design. Each block was further subdivided into five 20-trial sequences. Each 20-trial sequence comprised each target word once, and was followed by the summarized feedback (see above). In each word-prime trial, the prime was randomly chosen from the list of the five possible stimuli, with the restriction that in each 20-trial sequence a prime word appeared only once. In the condition ‘congruent pairs of same relevance type,’ the program precluded the selection of a prime that was identical to the target. Following the main blocks, a direct test of prime identification was administered. Forty more trials (eight trials per prime) were presented that were identical with the trials of the main phase, except for the following detail. Instead of a target word, a row of nine question marks appeared in the Moderation of masked affective priming 14 center of the screen accompanied by two stimuli, one to the right and one to the left of the question marks. One of these words was a repetition of the prime and one was a distractor, which was drawn from the stimulus list with the constraint that it was always from the same valence and relevance type as the prime. The location of the prime repetition was determined at random. Participants were informed that the sequence of flickers preceding the question marks (i.e., the mask-prime-mask sequence) contained either the left or right word. They were instructed to identify this word (i.e., the prime) by pressing the corresponding key (i.e., the key assigned to the right index finger if they decided for the right word, the key assigned to the left index finger if they decided for the left word). Results Unless otherwise noted, all effects referred to as statistically significant are associated with p values less than .05, two-tailed. All priming analyses were performed using both subjects (referred to by t1, F1) and item (t2, F2) means. Manipulation check Mean reaction times were derived from correct responses only. Reaction times that were three interquartile ranges above the third quartile with respect to the individual distribution were discarded (see Tukey, 1977), as were those above 1500 ms, or below 150 ms (0.96 %, 0.70 %, and 0.83 % of all values of the emphasis-on-accuracy sample and the emphasis-on-speed Moderation of masked affective priming 15 samples [Exp. 1a and 1b], respectively). The emphasis-on-speed sample (Exp. 1a) had a mean RT of M = 527 ms (SD = 40 ms), whereas the emphasis-on-accuracy sample had a mean RT of M = 587 ms (SD = 64 ms); this constitutes a significant difference, t(17) = 2.48. The emphasis-on-speed sample (Exp. 1a) had a mean error rate of M = 9.5 % (SD = 2.7 %), whereas the emphasis-on-accuracy sample had a mean error rate of M = 2.4 % (SD = 1.3 %); this again constitutes a significant difference, t(12.99) = 7.37. The emphasis-on-speed samples of Study 1a and 1b did not significantly differ neither with regard to mean RT (M[Exp. 1b] = 543 ms; SD = 43 ms), t(18) = 0.85, ns, and mean error rate (M[Exp. 1b] = 8.0 %; SD = 4.1 %), t(18) = 0.95, ns. Priming effects (Study 1a) The Block factor did not significantly moderate the priming results. Thus, this factor was discarded. To simplify the presentation of results, priming indices were calculated as the difference of incongruent and congruent mean RTs. Mean reaction times for the conditions of interest are shown in Table 1. A 2 (instruction: emphasis on speed vs. accuracy) x 2 (prime-target relevance: same vs. different) analysis of variance with the priming indices as the dependent variable yielded a significant interaction of instruction and relevance, F1(1,17) = 8.36, MSe = 179; F2(1,19) = 8.14, MSe = 481, all other Fs < 1. Moderation of masked affective priming 16 In the emphasis-on-speed sample, there no moderation of priming by relevance match was found, t1(9) = 1.73, p > .10; t2(19) = 1.46, p > .16. For both relevance types, a significant and marked priming effect emerged for prime-target pairs of the same relevance type, t1(9) = 2.93, d1 = 0.93; t2(19) = 2.19, d2 = 0.49, and for prime-target pairs of different relevance types, t1(9) = 4.57, d1 = 1.45; t2(19) = 5.86, d2 = 1.31, respectively. On the contrary, in the emphasis-on-accuracy sample, the priming effect was significantly moderated by relevance match, t1(8) = 2.34, d = 0.78; t2(19) = 2.37, d = 0.53. There was a marked and significant priming effect for primetarget pairs of the same relevance type, t1(8) = 4.25, d1 = 1.42; t2(19) = 4.11, d2 = 0.92, whereas no priming effect occurred for prime-target pairs of different relevance types, t1(8) = 1.12, ns, d1 = 0.37; t2(19) = 1.49, ns, d2 = 0.33. Mean error rates for the conditions of interest are shown in Table 1. In a 2 (instruction: emphasis on speed vs. accuracy) x 2 (prime-target relevance: same vs. different) analysis of variance with priming indices as the dependent variable, a significant main effect of instruction occured, F1(1,17) = 7.49, MSe = 24.2; F2(1,19) = 12.03, MSe = 31.75, all other Fs < 1. There was a significant priming effect for the emphasis-on-speed sample, M = 4.8 %, t1(9) = 3.38, d1 = 1.07; t2(19) = 4.78, d2 = 1.07, whereas there was no priming effect for the emphasis-on-accuracy sample, M = 0.5 %, Moderation of masked affective priming 17 both |t1|s < 1. Priming was not moderated by relevance match in either sample, all |t|s < 1. Priming effects (Study 1b) In the emphasis-on-speed sample of Study 1a, there was a slight numerical (but non-significant) difference between priming effects for same and different relevance in an unexpected direction (12 ms vs. 22 ms). To test whether this difference is replicable, we conducted Study 1b. Mean reaction times for the conditions of interest are shown in Table 1. In a 2 (prime-target relevance: same vs. different) x 2 (affective priming: congruent vs. incongruent) analysis of variance there was nothing but a significant main effect of affective priming F1(1,9) = 14.40, MSe = 297; F2(1,19) = 29.94, MSe = 350, indicating an affective congruency effect (all other Fs < 1). For both relevance types, a significant and marked priming effect emerged: t1(9) = 3.60, d1 = 1.14; t2(19) = 4.71, d2 = 1.05, for prime-target pairs of the same relevance type, t1(9) = 2.11, d1 = 0.67, p < .05 (one-tailed); t2(19) = 3.90, d2 = 0.78, for prime-target pairs of different relevance types. The results for the error rates are comparable to those for RTs (see Table 1). In a 2 (prime-target relevance: same vs. different) x 2 (affective priming: congruent vs. incongruent) analysis of variance, there was nothing but a significant main effect of Affective Priming F1(1,9) = 8.71, MSe = 22.4; F2(1,19) = 24.58, MSe = 15.9, indicating an affective congruency effect (both Fs < 1 for relevance and both Fs < 2.57, ns for the interaction). Moderation of masked affective priming 18 Direct effects (Study 1a) The signal detection sensitivity for the masked primes was d’ = .32 (SD = .50; range of -0.13 to 0.65 plus one outlier with d’ = 1.56) for the emphasis-on-speed sample and d’ = .17 (SD = .29; range of -0.29 to 0.63) for the emphasis-on-accuracy sample, t(17) = 0.77, ns, for the difference. Although the overall mean d’ of M = 0.25 (SD = .41) was significantly above zero, t(18) = 2.65, there was only one participant with a significant contingency between the position of the prime repetition (i.e., to the left or to the right of the question marks) and response (left or right key), χ2 > 11.31, all other χ2s < 3.28, p > .07. We followed the suggestion by Draine and Greenwald (1998) to regress the priming indices (i.e., the difference variable ‘incongruent minus congruent priming’) on d’. Since both variables have a meaningful neutral point, the intercept can be interpreted as the amount of priming in the absence of awareness (i.e., d’ = 0). We regressed the RT priming indices on d’, separately for the speed and the accuracy sample. Table 1 includes the intercepts (and their standard errors). For the emphasis-on-speed sample, both intercepts were significantly above zero, t(8) = 2.00, p < .05 (onetailed; t(8) = 0.62, ns, for the slope) for prime-target pairs of same relevance type, t(8) = 3.23 (t(8) = 1.03, ns, for the slope) for prime-target pairs of different relevance type. For the emphasis-on-accuracy sample, the intercept for prime-target pairs of same relevance type was significantly above zero, Moderation of masked affective priming 19 t(7) = 3.20 (t(7) = 0.41, ns, for the slope), whereas the intercept for primetarget pairs of different relevance type failed to be different from zero, t(7) = 0.30, ns (t(7) = 1.30, ns, for the slope). Direct effects (Study 1b) The mean signal detection sensitivity for the masked primes was d’ = .29 (SD = .53; range of –1.07 to 0.65), t(6) = -1.48, ns. For prime-target pairs of the same relevance type, the intercept was significantly above zero, t(5) = 2.20, p < .05 (one-tailed; t(5) = -0.77, ns, for the slope). Interestingly, the intercept for prime-target pairs of different relevance type failed to be significantly above zero, t(5) = 0.40, ns (t(5) = -1.22, ns, for the slope). Discussion Study 1 revealed two essential results. Given instructions to emphasize speed, a robust, replicable, and undifferentiated affective priming effect emerged. This result is in line with several other studies published during the last decade (e.g., Draine & Greenwald, 1998). Not unexpectedly, the priming effect was not moderated by a match or mismatch of prime and target relevance. However, given instructions to emphasize accuracy, a moderation by match of relevance type was observed. If prime and target belonged to the same type of relevance with regard to the distinction introduced by Peeters (1983), a significant positive priming effect was observed. If the relevance types of prime and target were different, priming was no longer found. These results at indicate that affective priming effects Moderation of masked affective priming 20 are susceptible to the kind of positivity or negativity of primes and targets. Thus, we can infer that the relevance type of valence is in fact a feature of valent stimuli that is activated under conditions of limited processing and limited awareness. However, one caveat remains with regard to this conclusion. As we have already noted, masked affective priming effects are especially observed if primes are practiced, for example, if primes and targets are drawn from the same set of stimuli (as in Study 1; see Abrams & Greenwald, 2000; Abrams & Grinspan, 2007). Several explanations of the masked affective priming effect focus on this detail. For example, Damian (2001) argued that masked affective priming effects are based on stimulusresponse mappings that become automatized during the course of the experiment. If we attribute our results to this process, we can easily explain the undifferentiated character of the priming effect under speed instructions: This instruction might not only increase the weight of the prime relatively to the target in quickly establishing a response, it might induce a strategy to simply respond on the first evidence available without paying attention to the source of this evidence, that is, whether it stems from the masked prime or the target (Wentura & Rothermund, 2003). Thus, this condition might quickly lead to stimulus-response associations that are blurred with regard to the relevance distinction, because ultimately only a simple positive or negative response has to be given. Moderation of masked affective priming 21 By contrast, accuracy instructions induce a strategy to respond only on the basis of the correct evidence – evidence unambiguously attributable to the target (Wentura & Rothermund, 2003). This can explain that primes that are congruent to the target with regard to valence but not with regard to relevance do not facilitate target responses because the sources of evaluative information can be disentangled on the basis of relevance. Similarly, primes and targets that are incongruent in valence can always be disentangled (irrespective of their relevance). Primes, however, that are congruent to the target with regard to both features facilitate the responding because there is no indication of a wrong source of evidence. This explains why there is affective priming only for pairs matching in relevance but no effect for nonmatching pairs. Essentially, stimulus-response associations evolving throughout the course of the experiment might become more sophisticated than simple positive-negative responses and thus contain connotations of relevance. If this explanation fits, we are faced by somewhat disappointing implications about the automatic processing of the relevance feature. The priming effect that is differentiated according to relevance might be a mere side effect of a deeper processing of the targets, given accuracy instructions: To disentangle sources of evidence, the target is more deeply processed than is needed for a superficial positive versus negative categorization. Eventually, stimulus-response associations are stored including valence and Moderation of masked affective priming 22 relevance and will later be easily retrieved if the same word is presented as a prime. Recently, however, Klauer, Eder, Greenwald, and Abrams (2007) showed masked affective priming with novel stimuli, that is, with stimuli never presented for classification during the course of the experiment. They explicitly argued for a two-component explanation of masked priming: One component is based on response-related stimulus-associations and depends on practice. Another smaller component is based on long-term stored features of the masked prime. Given these considerations, it is important to explore the meaning of the relevance distinction in an experiment that rules out automatized response mappings by using primes that are never openly presented for classification. In addition, this manipulation is important for another reason. Given our goal to corroborate our findings in the area of indirect assessment of attitudes, it is even more important to explore priming effects caused by novel stimuli, since it is a conditio sine qua non in this field of research that the attitude-related primes are never presented for evaluative categorization. STUDY 2 Study 2 is an almost exact copy of Study 1 except that primes and targets were now from different stimulus sets. That is, primes were neither categorized nor even consciously visible throughout the task. Again, we explored affective priming given speed versus accuracy instructions, Moderation of masked affective priming 23 although the reason for doing so was somewhat different than before. Given overall sparse and weak evidence of priming with novel masked primes, which, however, were always found with speed instructions, we used the speed instruction condition simply for establishing a condition that might be needed for obtaining priming at all. Method Participants Study 2 included two independent recruitments of participants. The only difference between Study 2a and 2b was that Study 2b lacked the direct test. A total of fifty-eight students (51 women; 7 men) participated in the experiments (29 participants for Study 2a and 2b, respectively). 30 of them were instructed to emphasize speed, 28 were instructed to emphasize accuracy. The median age was 21.0 years; all of them were native speakers of German. Two participants of the emphasis-on-accuracy sample were replaced because their mean overall response time was more than a standard deviation unit below the mean response time of the emphasis-on-speed sample. Design, Materials, and Procedure Design and Procedure were essentially the same as in Study 1. Now, primes and targets were taken from different stimulus sets. Both, the prime and the target set, comprised 10 positive and 10 negative German adjectives (see Appendix). Within each valence subset, five adjectives were otherModeration of masked affective priming 24 relevant and five adjectives were possessor-relevant according to norm data (Wentura et al., 1998). Absolute pleasantness (on a scale ranging from 0 to +100; Hager et al., 1985; Möller & Hager, 1991) was between 50 and 87 (prime list) and between 50 and 72 (target list). Mean absolute pleasantness of the four type sets ranged from 60 to 68 (prime list) and from 57 to 66 (target list). To bolster dissimilarity between primes and targets, we selected short adjectives (four to six letters) for the prime list and long adjectives (seven to nine letters) for the target list. Mean length of the four type sets ranged from 4.4 letters to 5.6 letters (prime list) and from 8.0 letters to 8.8 letters (target list). Since it is known that the response-related component of priming might already be triggered by fragments of practiced stimuli (Abrams & Greenwald, 2000), we made sure that there was no confound of matching letters between prime and target for the different conditions (see also Klauer et al., 2007). Therefore, we calculated the overlap of single letters, letter pairs, and letter triplets for each combination of prime and target. Mean overlap for those prime-target combinations matching in valence and relevance was M = 1.97 for single letters (compared to M = 1.94 for the remaining combinations; t[19] = .31, ns), M = 0.23 for letter pairs (compared to M = 0.21; t[19] = .29, ns), and M = 0.04 for triplets (compared to M = 0.02; t[19] = .85, ns).