Transitive Relations Cause Indirect Association Formation between Concepts. Yoav Bar-Anan and Brian A. Nosek. University of Virginia

Transcription

1 1 Transitive Association Formation Running head: TRANSITIVE ASSOCIATION FORMATION Transitive Relations Cause Indirect Association Formation between Concets Yoav Bar-Anan and Brian A. Nosek University of Virginia Date of Submission: January, 1, 008 Author s Note: The authors thank Jeff Hansen for his technical exertise. This research was suorted by a grant from the National Institute of Mental Health (R01 MH68447) to Brian Nosek. Corresondence concerning this article should be sent to Yoav Bar-Anan, Deartment of Psychology, University of Virginia, PO Box , Charlottesville, VA 904. Electronic mail may be sent to baranan@virginia.edu.

2 Transitive Association Formation ABSTRACT Social knowledge is often inferred without direct exerience. The resent research shows that associations between concets, affective and nonaffective, can be formed without exerience through the rincile of transitivity. In four studies, strengthening the association between one concet and a second concet, and then between the second concet and a third concet, strengthened the association between the first and the third concets desite never resenting them together. Study and 3 showed that valence and identity, core features of social knowledge, were more effective association carriers (i.e., the shared concet in a transitive association formation) than animals and loudness. In Study 4, inducing the two direct associations without awareness created automatic transitive associations, but not self-reorted transitive associations. Self-reorted associations were formed transitively only if articiants were aware of the association formation airing. Word count = 133 Recent date of submission: 1/1/008 Keywords: Indirect Association Formation, Nonaffective Conditioning, Associative Network, Cognitive Balance, Automatic

3 3 Transitive Association Formation Imagine that you just moved and have two new neighbors, one olitically liberal and the other conservative. The conservative neighbor likes to relax and rest in the sun, whereas the liberal neighbor is more of the sorty tye who jogs and lays ballgames with her kids. If you were a liberal, would this lead you to identify yourself more with exercising over relaxing? The fact that your liberal neighbor is an exerciser does not have obvious imlications for your own identity. Even so, your olitical association with the liberal neighbor could lead to associating with other qualities of your neighbor. This is anticiated by theories of cognitive consistency (Greenwald et al., 00; Heider, 1958; Osgood & Tannenbaum, 1955). Cognitive balance would be maintained among the associations self with liberal, and liberal with exerciser, if self is also associated with exerciser. In the resent research, we examine whether the rinciles of cognitive consistency govern association formation and cause concets to become associated, even in the absence of direct exerience of the association. We test whether associations between semantic concets can sread through shared concets to reinforce cognitive balance via the rincile of transitivity. Transitivity is a fundamental relation in logic: A relation r is transitive if ArB and BrC imly that ArC. We aly this transitivity rule to associative relations: if A associates with B, and B associates with C, then A will associate with C, even if A and C have never been resented or conditioned together. Walther (00) reorted some evidence suorting this henomenon in an indirect evaluative conditioning aradigm for self-reorted attitudes toward individuals. We extend this work by forming self-reorted and automatic associations between

4 4 Transitive Association Formation concets across varieties of social knowledge attitudes, identity, and general social concets (e.g., ets, loudness). This rovides evidence that transitivity is a general means for association formation, even or erhas esecially in the absence of exerience. We also examined whether some concets serve as better transitive association carriers. If associations are formed between A and B and between B and C, then B is the carrier of the indirect association between A and C. B is the linking concet, such as the shared association of olitical identity in the oening examle, and might be seen as the key concet for transitivity effects. Finally, we examined whether awareness of the direct association formation (i.e., between A and B, and between B and C) is necessary for transitive effect on the unexerienced link (A and C) to occur. Evaluative Conditioning Association formation research in social sychology focuses almost exclusively on attitudes. Automatic attitudes, a target-valence association (Fazio 1990), can form when eole consciously rocess evaluative information about targets (Foroni & Mayr, 005; Rydell & McConnell, 006), but most research alications create automatic attitudes using evaluative conditioning (e.g., Mitchell, Anderson & Lovibond, 003; Gregg, Seibt & Banaji, 006). In evaluative conditioning, the valence of a neutral stimulus is changed after it is reeatedly aired with an affective stimulus (De Houwer, Thomas & Baeyens, 001; Krosnick, Betz, Jussim, & Lynn, 199). For instance, conditioning activities with ositive valence caused eole to refer those activities over those conditioned with neutral stimuli (Custers & Aarts, 005). Almost all evaluative conditioning aradigms directly air the attitude object with valence. A few rocedures, anticiate our indirect aroach with aradigms that shaed

5 5 Transitive Association Formation attitudes toward human targets (Walther, 00; Hebl & Mannix, 003; Gawronski, Walther & Blank, 005; Ranganath & Nosek, in ress). For instance, Walther (00) demonstrated that self-reorted liking of eole can form through reconditioning and second-order conditioning. Neutral human faces were conditioned to a second, different set of neutral faces. The second set of neutral faces was later conditioned to other faces that were reviously rated as ositive, neutral or negative. Eventually, when articiants evaluated the first set of neutral faces, they reorted the most liking of the face that was indirectly conditioned with a ositive face, and the least liking of the neutral face that was indirectly conditioned to a negative face. This shows the otential for comlex affective knowledge formation that integrates multile sources of information, (i.e., if A is related to B, and B is revealed as good, than A is now considered good as well). Association Formation between Nonevaluative Concets The resent framework construes evaluation as one tye of association the link between valence and a target concet. Social life involves many other associative relations beyond likes and dislikes: Is this food sour or bitter? What ast works are related to my research? Is she the kind of erson who reads Chekhov? What kind of erson reads Chekhov anyway? In the context of social knowledge, research on nonaffective association formation is scarce (Staats & Staats, 1961; Meersmans, De Houwer, Baeyens, Randell & Eelen, 005; see also Hayes & Bissett, 1998; Smyth, Barnes-Holmes & Forsyth, 006), and none has examined indirect nonaffective association formation between concets. This is surrising because nonaffective associations are comonents of imortant social henomena such as stereotyes and identity (Greenwald et al., 00), automatic behavior (Aarts & Dijksterhuis, 003), and

6 6 Transitive Association Formation construal level (Bar Anan, Liberman & Troe, 006). Nonaffective associations are also the focus of cognitive work on semantic riming 1. If indirect association formation is a general mechanism, it should account for many non-attitudinal effects. Consider for instance a research that found that eole who felt guilty about immoral deeds from their ast, showed increased accessibility of the words shower and soa (Zhong & Liljenquist, 006). Though the issue of direct versus indirect association was not considered by the authors, it seems unlikely that deliberate thinking formed an association between guilt and soa, but indirect association formation can exlain it. Guilt is associated with the need to come clean or wash away one s sins, and the actions cleaning and washing are also associated with soa and shower. Thus, the motivation to reduce feelings of guilt might be indirectly associated with soa and shower. In the resent article, we reort evidence that associations between concets can form without exeriencing them together. This indirect association formation occurs via transitive association that, as discussed above, may account for the acquisition of various tyes of social knowledge, affective and nonaffective, and to the attainment of cognitive balance. In our aradigms, articiants were not required to consider the combination of the two direct associations (Studies 1-3), or to even have awareness of the direct associations that were created (Study 4). Transitive association formation under these conditions would suggest that cognitive balance can be maintained not only by deliberately motivated logical coherence, but also by the rules that govern the associative network automatically.

7 7 Transitive Association Formation Overview of Studies We emloyed association formation rocedures that do not rovide exerience with a secific relation between two concets, so this relation could be roduced only by alication of transitive relations. In each study, we exerimentally maniulated two associative links (A with B; B with C) in an associative triangle comrised of three concets. Our deendent variable was change in associations in the third link (A and C). We measured both automatic association and a self-reort about their relationshi (Does A go with C?). In Studies 1-3, the directly exerienced associations (i.e., A with B and B with C) were modified using a novel association formation rocedure. For each airing, articiants learned that two concets go together, yet the arbitrariness of the relation was clear and did not imly a secific tye (e.g., similarity, oosition, hierarchy). The task never resented the two directly induced associations together, and did not encourage thinking about the two of them at the same time, leaving no reason to infer the transitive association. We used natural concets (e.g., gender, shaes) and we created a comarative context by airing, for examle, dogs with olygons and cats with ovals. This fits our stance that associations are context-sensitive and acquire meaning from that context (Mitchell, Nosek, & Banaji, 003; see also Gawronski & Bodenhausen, in ress; Rydell & Gawronski, 007). Using concet airs (e.g, dogs/cats) means that our triangle of relations contained six and not three concets. Figure 1 illustrates the triangle for Study 1. The three airs of concets were R/Z (the letters), dogs/cats and olygons/ovals. We maniulated the first link by airing R/Z with dogs/cats. For half of the articiants we induced a R+dogs,

8 8 Transitive Association Formation Z+cats association (the sign + means with ); the other half learned the oosite: R+cats, Z+dogs. Next, the second link was induced as either dogs+ovals, cats+olygons or dogs+olygons, cats+ovals. Then, we measured the third link between R/Z and ovals/olygons to see if it formed in a transitive manner. The Imlicit Association Test (IAT; Greenwald, McGhee, & Schwartz, 1998), a measure of comarative association strengths, assessed the indirect association formation. Also, articiants self-reorted which of two ossible transitive airings (consistent or inconsistent with training) feels more right to them. In four studies, indirect association formation occurred following the rules of transitivity with both automatic and self-reort measures. Studies and 3 extended indirect association formation to multile toics demonstrating its generality. These studies also found that the strength of the indirect, transitive association is affected by the concet that aears in both direct association inductions, but is not art of the indirect association (B in the A+B, B+C format). In the oening examle, olitics is the shared association, the carrier, that links self to exerciser. Study 3 also revealed that the transitive association could form even when articiants could not remember which concets were aired in the induction. Finally, Study 4 created associations very subtly with a different training rocedure, and found that automatic (but not self-reorted) associations can form indirectly via transitivity, even without awareness of the direct association formation. Studies 1-3 Method Overview

9 9 Transitive Association Formation Studies 1-3 used a similar rocedure and are reorted together. Study 1 established the aradigm and evidence for indirect association formation via transitivity. Study made two changes to Study 1: (1) the carrier concets were varied between articiants, and () the association formation task was longer (4 blocks instead of ). Study 3 made three modifications to Study : (1) new carriers were examined, () a maniulation check measured the association formation between both direct association airs, not just the first one, and (3) we tested memory of the association formation induction. Particiants Particiants volunteered at the Project Imlicit research website (htts://imlicit.harvard.edu; see Nosek, 005 and Nosek et al., 006 for more information) and were randomly assigned to this study from a ool of aroximately 100 studies. Of those that initiated a study session (996 is Study 1, 099 in Study, 196 in Study 3), we retained the articiants who comleted sufficient numbers of measures for analysis and had better than 70% accuracy in the association formation task (795 in Study 1, 195 in Study, 084 in Study 3). Materials Associations were formed among three airs of concets. Study 1 used letters (R/Z), animals (dogs/cats) and shaes (olygons/ovals). In Study, the first and third airs were the same as in Study 1, but the second air, the carrier, was maniulated between subjects: dogs/cats, male/female, or good/bad. In Study 3, the first air was always crossing lines/arallel lines. The third air was always trees/rocks. The carrier air was either: good/bad, self/other, or quiet/loud.

10 10 Transitive Association Formation The items were black and white ictures of dogs, cats, ovals, olygons, crossing lines, arallel lines, trees, rocks, ictures of the letters R and Z in different fonts and rotations, and words related to the concets good, bad, female, male, self, other, quiet and loud 3. Association formation task. We created the direct associations using a task that involved rehearsing airing rules. The rules were, for instance, dogs go with ovals and cats go with olygons. The rehearsal of this airing rule was intended to form an association of dogs with ovals and cats with olygons. Particiants categorized stimuli according to the airing rule. For examle, for the rule dogs go with ovals and cats go with olygons, every time an oval aeared on the screen, it would be categorized as dogs; cats would be categorized as olygons, ovals would be categorized as dogs, and olygons would be categorized as cats. Each trial resented one stimulus item (i.e., a dog, cat, oval, or olygon) at the center of the screen, and labels corresonding to two categories at the to left and right of the screen (see Figure for illustration of a trial). If the stimulus item was an oval or olygon, the category labels at the to were dogs and cats, and vice versa. Particiants categorized the stimulus item by ressing K if the item belonged to the category on the left, and S if the item belonged to the category on the right. The category labels always reresented the other dimension (i.e., dogs and cats if the stimulus was an oval). For each trial, the stimulus item was selected at random from the four categories, and the category labels were randomly assigned to be a left or right resonse (e.g., dogs on left, cats on right versus dogs on right, cats on left). This way, articiants could not use resonse location as a basis for forming associations.

11 11 Transitive Association Formation The interval between the articiant s resonse and the next trial was 50ms. A red X was dislayed for errant resonses and the trial did not end until the articiant hit the correct resonse key. The first block (50 trials) created the first direct associations (between R/Z and dogs/cats in Study 1). The second block (40 trials) created the second direct associations (between dogs/cats and ovals/olygons in Study 1). In Studies and 3, two more blocks (0 and 30 trials) reeated the first blocks to strengthen the association formation. To account for the influence of any reexisting associations, we randomized the 4 airing rules between articiants. In Study 1 the four association chains were: [1] R with dogs and dogs with ovals, [] R with dogs and dogs with olygons, [3] R with cats and cats with ovals, and [4] R with cats and cats with olygons. Studies and 3 also maniulated the carrier air between subjects: dogs/cats, male/female, good/bad in Study and good/bad, self/other, quiet/loud in Study 3. In Studies and 3, during the association formation task, a erformance score was dislayed. Particiants earned 7 oints for each correct resonse and lost 7 oints for each incorrect resonse. This increased articiants accuracy rate from 83.6% to 88.6%. Measures Imlicit Association Test. After the airing task, the articiants erformed a short version of the IAT (Greenwald et al., 1998). The IAT is resumed to measure associations between concets without requiring introsection (see Nosek, Greenwald, & Banaji, 007 for a review) 4. Particiants categorize stimuli reresenting four concets with two resonse keys two concets on each key. The core assumtion is that

12 1 Transitive Association Formation categorization will be faster when more strongly associated concets share a resonse key, comared to when they do not. The first two blocks (15 trials each) were ractice. In Block 1, articiants categorized items from just two categories (R and Z in Study 1). In Block, articiants categorized items from two other categories (olygons and ovals in Study 1). Together the four categories from these blocks form the basis of testing indirect association formation. In Block 3 and 4 (30 trials each), articiants categorized items from all four categories. In one block, R and olygons shared one resonse key, and Z and ovals shared the other; in the other block Z and olygons one key, and R and ovals shared the other. If articiants were faster and more accurate in the first of these, it suggests that the associations R+Polygons, Z+Ovals were stronger than the associations R+Ovals, Z+Polygons. The order of the airings was counterbalanced between Blocks 3 and 4. As a maniulation check, Blocks 5 and 6 (0 trials each) measured the first association that was created in the induction. In Study 1, the four categories in these blocks were R/Z and dogs/cats. R shared a key with dogs in one of the blocks and with cats in the other block, and the order of blocks was counterbalanced. Study 3 also measured the second directly created association in Blocks 7 and 8. Self-reort. At the end of the session, articiants were asked one question: Which of these two airings feels more right to you? The otions in Studies 1 and were R aired with ovals and Z aired with olygons, and Z aired with ovals and R aired with olygons. Study 3 included these two otions (with different concets), and also a no reference otion.

13 13 Transitive Association Formation Memory. In Study 3, items gauged articiants memory for the induction asking Which of these associations was used in the airing game? For each of the three ossible links (the actual two airings and the transitive airing) articiants could choose one of the association conditions (e.g., Crossing lines with rocks and arallel lines with trees ), or These items were not directly associated in the airing game. Procedure The study started with the direct association induction. Afterwards, the IAT instructions aeared exlaining its distinct format and rocedure. After the IAT, articiants answered the self-reort questionnaire. In Study 3, the order of the memory and feeling items was counterbalanced between articiants. Results IAT measures were comuted using the IAT D score (Greenwald, Nosek, & Banaji, 003). The association effects in Study 1 were tested with t-tests, with the airing condition redicting the IAT or self-reort. In Studies and 3, the same effects were tested with (airing condition) x 3 (carrier air: the three ossible second air concets) ANOVAs redicting the association measure (IAT or self-reort). Direct Association Formation We first confirmed that the association formation task effectively modified the directly maniulated associations. As exected, in Study 1, when the rules in the airing task associated R with dogs, R was more associated with dogs and Z was more associated with cats on the IAT (M = 0.03, SD = 0.5), than when the maniulation associated R with cats (M = -0.0, SD = 0.48), t(793) = 6.69, <.0001, d =.47. As illustrated in Figure 3,

14 14 Transitive Association Formation across studies, the induction rocedure effectively created associations, with effect sizes that ranged from d =.47, to d = 1.37, all s < Direct association formation differed between the carriers. The induction was more effective at creating some associations directly than others. As evident from Figures 3B and 3C, there was a main effect of airing condition showing the association formation effect discussed above, and a main effect of the carrier. This latter effect suggests a riori differences in associations of the carriers and the air R/Z (the difference between the mean heights of the bar airs in Figures 3B and 3C). This main effect is not relevant for our resent theoretical interest. In Study, there was also a significant interaction, F(, 1765) = 7.15, =.0008, η =.01, indicating that when the carrier air was dogs/cats, the training had a weaker effect than when the air was good/bad or male/female. 5 In Study 3, there was no interaction, F < 1, which means that all the carriers (self/other, good/bad, quiet/loud) were equally sensitive to the induction forming the direct associations with crossing/arallel lines. Study 3 included a measurement of the second direct association induction (Figure 3D). In addition to the significant main effects, there was also a significant interaction, F(, 186) = 5.49, =.004, η =.01. The direct association formation with trees/rocks was stronger for quiet/loud and self/others than good/bad. Note that this interaction is just showing that the induction rocedure is more effective for training some associations directly over others; it is not testing whether the carrier tye affects the strength of indirect association transfer. These results suggest that features of the toics

15 15 Transitive Association Formation make them more or less amenable to associative training. The cause of this is an intriguing question for future research. Indirect Association Formation via Transitivity in Automatic Associations Figure 4 illustrates the effect of the transitive airing condition in Studies 1-3. To test whether the association formation rocedure modified indirectly the transitive association, we comuted an IAT D score that reflected the association between the concet airs that never aeared together in training. In Study 1, this was the association between R/Z and ovals/olygons. Positive scores indicated a stronger association of R+ovals and Z+olygons than R+olygons and Z+ovals. As redicted, the score was higher when the indirect association formation transitively aired R+ovals, Z+olygons (M = 0., SD = 0.4) than when it aired R+olygons, Z+ovals (M = 0.14, SD = 0.45), t(793) =.65, =.0083, d =.19 (see Figure 4A). This indirect association formation effect was relicated in Studies and 3. The ANOVAs documented the redicted main effect of airing condition, F(1, 1753) = 98.49, <.0001, η =.05, in Study (Figure 4B), and F(1, 1468) = , <.0001, η =.18 in Study 3 (Figure 4C). This effect indicates that the association between the concets that were never linked in training was affected in a transitive manner by the direct airing of each of the airs with the carriers. Neither Study or 3 found a main effect of carrier, F (, 1753) = 0.7, =.77, η <.01, and F(, 1468) = 1.99, =.14, η <.01 resectively. The strength of indirect association formation varied as a function of the carrier toic. The magnitude of the indirect, transitive effect on the IAT varied across the

16 16 Transitive Association Formation carrier toics. Both Study and 3 showed significant interactions between the airing condition and the carrier toics, F(, 1753) = 3.6, =.03, η =.01, in Study, and F(, 1468) = 5.8, =.003, η =.01. This effect was decomosed with three additional x ANOVAs, each with only two of the carrier airs. In Study, as illustrated in Figure 4B, the carrier air good/bad yielded a stronger indirect association formation effect than the carrier air dogs/cats, F (1, 118) = 7.7, =.007, η =.01. Male/female was in between and not significantly different from either, s > 11. In Study 3, the carrier quiet/loud induced a weaker indirect association formation effect than self/others, F (1, 10) = 11.49, =.0007, η =.01, and good/bad F (1, 10) = 6.46, =.01, η =.01. The latter two did not differ, F < 1, η <.01. Indirect Association Formation via Transitivity in Self-Reort As illustrated in Figure 5, indirect formation of self-reorted associations occurred in all three studies. In Study 1, when the induction rocedure transitively linked R+ovals, Z+olygons, 8% ercent of the articiants reorted that it felt more right, whereas 7% referred that airing when the association formation transitively linked R+olygons, Z+ovals, t(774) = 3.36, =.0008, d =.13. Study and Study 3 relicated this indirect, transitive effect (Figures 5B and 5C). Particiants reorted references for associations were reliably influenced via transitive inference even though the associations were never resented together in the induction rocedure, F(1, 170) = 13.44, <.0001, η =.07, in Study, and F(1, 183) = 169, <.0001, η =.09, in Study 3.

17 17 Transitive Association Formation The carrier toic had no influence on indirect association formation in selfreort. Earlier, the magnitude of indirect association formation differed across carrier tyes with automatic associations as the deendent variable. The same did not occur in self-reort. Both Study and 3 found no main effect for carrier, Fs < 1 and no interaction between airing condition and carrier toic: F(, 170) = 1.87, =.15, η <.01, and F < 1. The indirect associations formed even if eole misremembered the airing rules In Study 3, we assessed memory for the airings that were resented during training to confirm that the association measures catured the associations that were created, and not just memory for the rules in the task. There was no effect of the order of asking about memory before or after self-reorted associations, ts < 1, suggesting that articiants were not treating the association self-reort (above) as a test of their memory for the training, because asking about memory first would have made clear that the reference item was measuring something different (Grice, 1975). Most articiants remembered the association formation conditions. 79.9% remembered the first induction, 7% remembered the second induction. However, articiants may have used the formed associations to recall the induction airings. In fact, only 38.4% of the articiants recalled correctly that indirectly formed associations were not associated directly in the airing game. The transitive association influenced this misrecollection. When crossing lines were transitively associated with rocks and arallel lines with trees, 50.6% of the articiants thought that this association was resented directly in the induction, but only 4.6% thought so when the oosite association would be derived from transitive relations, X (, N = 1819) = 185, <

18 18 Transitive Association Formation This shows that the transitive relations roduced false memories for revious exerience. We used the memory items to test whether eole who did not remember the induction airings would still show the exected transitive association formation. 33 articiants (7% of those who answered these items) misremembered one of the direct association formation links (meaning that they believed that the association formation in the training was oosite to what it actually was). Yet, when asked about their feelings, these articiants still showed the exected transitivity effect. When the airing imlied by transitivity was crossing+rocks, arallel+trees, misrecollecting articiants selfreorted more reference for this association (N = 140, M = -0.63, SD = 0.63) than misrecollecting articiants who underwent the oosite airing (N = 185, M = -0.3, SD = 0.8), t(33) = 4.07, <.0001, d =.47. These articiants also showed the exected transitivity effect with automatic associations, t(38) = 5.75, <.0001, d =.64. A logical, deliberate alication of transitivity based on their false memory would have roduced the oosite effects. These results suggest that the feelings reorted by articiants in general were gut feelings resumably influenced by the formed associations (Strack & Deutsch, 004), rather than a logical deliberate rocess, based on (mis)memory. Discussion Studies 1-3 showed that training relations among two airs of concets, A with B and B with C, changes the associations not only between those concets, but also between comonent concets that were never linked directly (A with C). The indirect

19 19 Transitive Association Formation association formation via transitivity was detected with both automatic association and self-reorts about which associations felt right. Studies and 3 found that the transitive effect of the two direct associative links over the third, indirect, link was moderated by the carrier toic. Valence was more effective than animals in forming both the direct associations and the indirect transitive association. The effectiveness of gender seemed better than animals and worse than valence in both tyes of associations, but usually not significantly. The airs self/other and good/bad were more effective than the air quiet/loud. This might suggest that some concets are rivileged either by exerience or by qualitative features in their ability to associate with other concets and shae the associative network. Study 4 In Studies 1-3, the transitive association formation was detected with both automatic measures and self-reort. This does not indicate whether the rocesses that roduced the indirect association formation occurred inside or outside of awareness, and whether these rocesses were deliberate or automatic (De Houwer, 006). The memory data suggest that indirect association formation may have occurred outside of awareness (and hence, control), and even in oosition to what was remembered (incorrectly) about the training. In the aradigm used in Studies 1-3, articiants had to be aware of the direct association formation rules in order to erform the task, so lack of memory may reflect forgetting and not unawareness (though this would require articiants to make an unlikely series of actions: deliberately conclude the transitive association without rovocation, refer this association in the self-reort measure, and then misremember a airing rule in the memory test). In Study 4, we ursued more definitive evidence that

20 0 Transitive Association Formation indirect association formation can occur outside of awareness and control by making the direct association training much more subtle. The induction for Study 4 resented all ossible airings between four concets aeared (e.g., quiet-rocks, quiet-trees, loud-rocks, loud-trees) and induced associations by having two (e.g., quiet-rocks, loud-trees) occur more frequently in task erformance. If the robabilistic differences are influential on association formation without articiants being aware of the contingencies, then indirect association formation would likely be the roduct of automatic rocesses. Method Particiants Of the 909 (600 women, M age = 31.9, SD = 10.5) articiants that were assigned to this study, we included the 813 eole who comleted some measures and had more than 70% accuracy in the association formation task. Materials Associations were formed among three airs of concets. Relicating one of the carrier conditions in Study 3, the airs were: crossing lines/arallel lines, quiet/loud, trees/rocks. The stimuli were the same as in Study 3. Association Formation Task For each trial, articiants categorized airs of stimuli to airs of categories. In each block, there were four ossible category airs resented in the to-left, to-right, bottom-left, and bottom-right corners of the screen (see Figure 6). In blocks 1 (50 trials) and 3 (30 trials) the airs were crossing lines-quiet, crossing lines-loud, arallel linesquiet, arallel lines-loud. In blocks (40 trials) and 4 (0 trials), the airs were trees-

21 1 Transitive Association Formation quiet, trees-loud, rocks-quiet, rocks -loud. Each category air corresonded with a resonse key ( q,, c, and m ). Each trial resented a stimuli air in the center of the screen e.g., a icture of crossing lines and the word noisy. Particiants categorized the items conjointly into one of the four category airs as quickly as ossible (e.g., icture of crossing lines+noisy was categorized to the category air crossing lines+loud). If a mistake was made, a red X aeared below the stimuli until the articiants corrected the error. Essentially, the rocedure was a modified version of the Sorting Paired Features task (SPF; Bar-Anan, Nosek, & Vianello, 007; Ranganath, Smith, & Nosek, in ress). The association formation was maniulated by controlling the number of resentations of each category air. For each block, of the airs (e.g., rocks-quiet and trees-loud) comrised 75% of the trials (each 37.5% trials), and the other two comrised 5% (each 1.5% trials). The two more frequent airs always included four different categories (i.e., the same category could not aear in both frequent airs). There was no indication to articiants that the contingencies were imbalanced, and knowing or noticing the contingencies was irrelevant for erforming the task correctly. The key assignments for the category airs rotated corners (clockwise) every 5 trials within a block to decrease the likelihood of noticing the contingencies based on frequency of resonse selection (i.e., no key was ressed more often than the others). Measures Associations. The IATs and self-reort measures were identical to Study 3, with the addition of self-reort measures of the two directly induced associations.

22 Transitive Association Formation Memory. Particiants were reminded of eight category airs that aeared in the induction. They reorted whether the airs aeared the same number of times: yes; no recollection or intuition about that; no, but I cannot remember which the frequent airs were; and no, and I remember which the frequent airs were. Then, all articiants selected the airs that they thought might have aeared more often in the induction. They could select as many of the airs they wished, or could resond: All airs aeared the same number of times, or I have no idea. 6 Results and Discussion Association Formation. The subtle induction created direct and indirect (transitive) automatic associations, but only direct, and no transitive, self-reorted associations. For each of the two direct associations, the relevant IAT was affected by the airing condition, t(695) = 5.13, d =.40, t(69) = 7.80, d =.58, resectively, both s < The IAT was also affected in the redicted direction by the transitive association formation, t(706) = 3.06, =.00, d =.4. The same was found with selfreorted associations for the direct association formations, t(61) = 6.65, d =.50, t(701) = 9.08, d =.68, resectively, both s < However, the self-reorted association for the transitive relation did not differ between the two conditions, t(707) = 1.1, =.6, d =.14. Few articiants could identify the association contingencies during training. Of the 680 (84%) articiants who resonded to the general question about the air frequencies, 13% said incorrectly that all airs aeared the same number of times, 47% reorted no recollection or intuition about the frequencies, 33% said that the airs differed in frequency but that they did not know which, and 7% claimed that they knew

23 3 Transitive Association Formation which airs were more frequent. Of the 655 articiants who resonded to the secific memory question (that included the otion to indicate the airs that were more frequent), 15% said incorrectly that none of the airs aeared more frequently than the other, 51% said that they had no idea, and 34% selected some airs. For the latter articiants, a memory score rovided indication of their memory accuracy (+1 for each correct resonse, -1 for each incorrect resonse, 0 for non-resonses; range -4 to +4). Of the minority of articiants who selected some airs, the average score was 0.86 (SD = 1.61), with just 15% of that subset scoring oints or more. In summary, about 10% of all articiants listed more correct answers than wrong answers; the large majority had incorrect or no knowledge of the induction contingencies. When we tested the relation between awareness and association formation, we analyzed two grous searately: A no-awareness memory grou (N = 33) consisted of those who reorted, in both questions, that the airs aeared equal number of times, or that they have no idea. To be conservative, the aware grou (N = 11) included anyone with score over 0 as indicating awareness of the induction (only 18 articiants correctly identified all four higher frequency contingencies). The Relation between Awareness and Association Formation To test the relation between awareness and association formation, for each of the three association formation effects ( direct and 1 transitive), we conducted (awareness grou) X (association formation condition) ANOVAs for each association measure. All ANOVAs revealed the redicted main effect of association formation condition. One found unexected main effect of awareness on the association, indicating that, across association formation conditions, unaware articiants referred the association

24 4 Transitive Association Formation Quiet+Trees, Loud+Rocks over the oosite association more than aware articiants did, in both self-reort and IAT ( η s =.01). We have no exlanation for this small effect, and it is inconsequential for the resent investigation. Of main interest, as detailed below, some interaction effects revealed that association formations were stronger in the awareness grou. Awareness had little effect on direct association formation. We conducted four ANOVAs, one for each measure (self-reort, IAT) for each of the two directly trained associations. All four revealed the exected main effect of association formation condition (s <.0001, η s >.04). A significant interaction aeared in only one of the four ANOVAs. The induction had a stronger influence on self-reorted associations between quiet/loud and trees/rocks for articiants with some awareness (d = 1.8), than those with none (d =.60), F(1, 447) = 1.49, <.01, η =.0. Notice that, even here, the induction occurred whether articiants were aware of the contingencies or not, s <.0001, suggesting direct association formation without awareness. Lack of any other interaction effect indicated that the aware and unaware grous did not differ in the strength of direct association induction on either of the directly induced automatic associations (s >.15 ; η s <.01), and on the other self-reorted association ( =.5; η s <.01). Overall, self-reorted and automatic associations were formed whether eole were aware or unaware of the contingencies, and awareness had no relation to the strength of the direct automatic association formation and inconsistent relation to the corresonding self-reorted association formation.

25 5 Transitive Association Formation Indirect association formation via transitivity. There was a main effect of indirect association formation by transitive relation on the IAT, F(1, 448) = 19.69, <.0001, η =.04. A lack of interaction effect, F(1, 448) =.11, =.14, η <.01, suggested that the automatic transitive association was formed in both the unaware d =.31 and aware grous, d =.6, all s <.001 (see Figure 7A). Self-reorted transitive association occurred only among the articiants aware of the contingencies. The ANOVA documented (small) main effect of association formation condition, F(1, 448) = 3.90, =.05, η =.01, a marginal main effect of awareness, F(1, 448) = 3.44, =.06, η =.01, and, imortantly, a marginal interaction, F(1, 448) = 3.9, =.07, η < 0.1. Though follow-u analysis of a such an effect must be interreted cautiously, it seems that the association formation maniulation affected the self-reort measure only among the aware, t(119) =.09, =.04, d =.38, and not at all among the unaware t < 1, =.87, d =.01 (see Figure 7B). With a new induction aradigm, Study 4 relicates the earlier studies showing that associations between concets can form without any direct exerience via transitivity. The new aradigm induced the associations more subtly 60% had no recognition of the contingency maniulation at all, and just 10% reorted even a ortion of it accurately. Even the unaware articiants showed both the direct association formations and the indirect, transitive association formation, in their automatic reactions measured with the IAT. However, erhas illustrating the distinctiveness of imlicit and exlicit measurement, the self-reort measure documented direct association formation without awareness of the induction contingencies only occasionally. Also, indirect,

26 6 Transitive Association Formation transitive self-reorted association formation occurred only when articiants had at least some awareness of the induction contingencies. General Discussion Four studies demonstrated that an association between two concets can be formed indirectly, via the rincile of transitivity, by airing each concet searately with a third concet. The indirect association formed desite the fact that the concets were never exerienced together. Social knowledge, stored in associative networks, forms directly through exerience, and indirectly through roagation of associations via transitivity. As a consequence, eole do not need to exerience concets together in order to conclude that they are related. Studies and 3 found that the induction rocedure formed direct and indirect associations of different strength deending on the concets involved. Some concets, such as good/bad and self/other, were articularly effective carriers of associations from directly exerienced to indirectly inferred, and others such as dogs/cats and quiet/loud were less effective. Study 4 found that even among articiants who did not notice the subtle association induction, automatic indirect associations were formed, by the rincile of transitivity. However, with no awareness, self-reorted associations were not formed transitively. This work extends ast research on indirect evaluative conditioning and cognitive balance to a general mechanism of association formation. The effects were observed with both affective and nonaffective stimuli showing that the association effects may be general across tyes of social knowledge attitudes, stereotyes, identity, and more. Our results suggest that cognitive balance is maintained not only by deliberate motivation to

27 7 Transitive Association Formation reserve coherence, but also by the rules of associations that oerate automatically (Greenwald et al., 00). Concets as well as associations between concets do not exist in isolation. They are integrated comonents in a large associative structure (Greenwald et al., 00). The idea of transitive associations highlights this focus on associative networks. The resent results suggest that individual associations are influenced by both direct exerience with the concets, and by their other relations in the broader associative network. Smith and Decoster (1998) suggested a connectionist model that allows for accumulation of information acquired across occasions. This accumulation can result in associations that were not directly learned. Our research is an examle of a simle concetual framework that may hel study such comlex accounts by forming direct associations and measuring other relations that might be affected. Mechanism We hyothesize that when concet C was aired with B, A was also activated because of its trained association with B. This caused coactivation of C with A, and formed the transitive A-C association. Based on evidence that associations (direct and indirect) sometimes formed in absence of, and even oosition with, memory for the task, we assume that this mechanism can oerate without awareness. Because awareness is a requisite of control (Wegner & Bargh, 1998), and because the rocedures did not give reason to deliberately validate the indirect association, we assume that automatic rocesses are sufficient to create transitive associations. In Study 4, self-reorted associations were affected by the transitivity rules of association formation only among articiants who reorted correctly at least some of the

28 8 Transitive Association Formation airings. These results may suggest that indirect formation of automatic associations haens without awareness, but self-reorted association can form indirectly only with awareness, and erhas control. However, success in the contingencies memory test might suggest usage of the formed associations to answer the memory questions, rather than awareness of the contingencies during the association formation task. Even awareness does not indubitably imly that deliberate (i.e., intentional and controlled) rocesses contributed to the transitive association formation. A deliberate rocess that causes transitive association formation could haen only if articiants (1) detected the frequent resentation of at least one air in each of the two direct association formations, () concluded the transitivity, and (3) validated this association as true. Because of this we seculate that self-reorted transitive association formation in Study 4 occurred with awareness because both memory and self-reorted associations were a roduct of the induction, not because the indirect association formation occurred deliberately. The debate on whether awareness of contingencies is required for association formation is still unsettled (Walther & Nagengast, 006; Dawson, Rissling, Schell & Wilcox, 007). Meersmans et. al (005) found that direct nonaffective associations were formed only when awareness was resent. However, comared to the resent studies, that research was underowered with small samles, and used a different rocedure with a less reliable automatic measure (affective riming, Studies 7 & 8). Our studies suggest that indirect association formation via transitivity may be useful for further investigation of awareness because it is difficult to notice all the contingencies. Further research with more stringent tests of awareness (e.g., trial-by-trial measurement), or more subtle

29 9 Transitive Association Formation maniulations (e.g., subliminal resentations), may further clarify the role of awareness in associative formation. Possible Effects of Transitive Association Formation Association formation is a form of learning. For instance, children may form an association between stove and burn, and learn to be afraid of stoves as they are afraid to burn. Associations hel eole to acquire automatic behavior that saves energy and frees resources for more tasks. Transitive association formation imroves learning, because it rovides new associations without requiring direct exerience. The association of a stove with barbeque grill may hel children associate the grill with burn and stay away. Imortantly, the children do not have to logically conclude that grill and stove roduce high heat and that this is why they burn and ose danger. Instead, the associations can be acquired by reeated airing without logic or semantic overla, and rovide new rules for behavior and thought. This is a owerful tool to acquire knowledge efficiently without using substantial cognitive resources (see Hayes, Barnes-Holmes & Roche, 001, for more about the learning benefits of derived relations). Association formation may also cause biases, because associations sometimes reflect co-occurrence that coy ersonal exerience but do not deict a valid icture of reality. For instance, if a erson hears about immigrants only with relation to violence and crime, an association between immigrants and violence may form. In that context, immigrants might easily become associated with other associates of violence such as hate, enemy, fear, danger, snakes, dark, strong, and so on. If eole later learn that immigrants-violence is an incorrect association, they might be able to suress or alter that association. However, they might fail to notice, or have no way of knowing,

30 30 Transitive Association Formation that they also ossess the associations, such as immigrant-enemy, that formed indirectly as a consequence of the initial errant data. In this sense, transitive associations may act as a viral agent for sreading social biases in memory. Indirect formation may create associations, or social biases, that are articularly challenging to extinguish because their source, creation, and resence may be unknown. Transitive association formation may also generate new associations that are oosite in their validity to the direct associations that formed them. For instance, even if the associations immigrants+oor education and immigrants+crime were both false, they could still strengthen a valid association between oor education and crime. In contrast, two valid associations, (e.g., rhinoceros+fierce and carnivore+fierce), may strengthen an invalid association (e.g., rhinoceros+carnivore). Not All Transitive Association Formations Are Equally Strong In Studies and 3 we tested different carries of the association from the first to the third concet. In Study, we found that good/bad erformed better as a carrier of association than dogs/cats. In Study 3, we found that good/bad and self/other erformed better than quiet/loud. This is in line with sychological theory that has long singled out valence (Osgood, Suci & Tannenbaum, 1957; Zajnoc, 1980; Lang, Bradley & Cuthbert, 1990; Duckworth, Bargh, Garcia & Chaiken, 00) and the self (Greenwald & Pratkanis, 1984; Kihlstrom & Cantor, 1984; Nowak, Vallacher, Tesser & Borkowski, 000) as imortant cognitive constructs that organize social knowledge and may be activated automatically and chronically. Further research could investigate which kinds of concets have more effect than others on associative links, and how different contexts, or eole (i.e., their associative

31 31 Transitive Association Formation networks; Coronges, Stacy & Valente, 007), may rivilege some concets as effective association conductors. A rogram of research, for examle, could examine whether the three basic dimensions of meaning valence, arousal and otency, suggested by Osgood, Suci and Tannebaum (1957) also ossess rivileged influence in associative networks. Uncovering such central concets in the context of association formation may guide a theoretical framework for understanding how associative networks roduce thought and behavior.

32 3 Transitive Association Formation References Aarts, H., & Dijksterhuis, A. (003). The silence of the library: Environment, situational norms and social behavior. Journal of Personality and Social Psychology, 78, Bar-Anan, Y., Liberman, N., & Troe, Y. (006). The association between sychological distance and construal level: An imlicit association test. Journal of Exerimental Psychology: General, 135, Bar-Anan, Y., Nosek, B. A., & Vianello, M. (007). The sorting aired features task: A measure of association strengths. Unublished Manuscrit. Bargh, J. A., Chaiken, S., Govender, R., & Pratto, F. (199). The generality of the automatic attitude effect. Journal of Personality and Social Psychology, 6, Coronges, K., A., Stacy, A., W., & Valente, T., W. (007). Structural comarison of cognitive associative networks in two oulations. Journal of Alied Social Psychology, 37, Custers, R., & Aarts, H. (005). Positive affect as imlicit motivator: On the nonconscious oeration of behavioral goals. Journal of Personality and Social Psychology, 89, Dawson, M., E., Rissling, A., J., Schell, A., M., & Wilcox, R. (007). Under what conditions can human affective conditioning occur without contingency awareness? Test of the evaluative conditioning aradigm. Emotion, 7, De Houwer, J. (006). Using the imlicit association test does not rule out an imact of conscious roositional knowledge on evaluative conditioning. Learning and Motivation, 37,

33 33 Transitive Association Formation De Houwer, J., Thomas, S., & Baeyens, F. (001). Associative learning of likes and dislikes: A review of 5 years of research on human evaluative conditioning. Psychological Bulletin, 17, Duckworth, K., L., Bargh, J., A., Garcia, M., & Chaiken, S. (00). The automatic evaluation of novel stimuli. Psychological Science 13, Fazio, R. H. (1990). Multile rocesses by which attitudes guide behavior: The MODE model as an integrative framework. In M. P. Zanna (Ed.), Advances in exerimental social sychology (Vol. 3, ). San Diego, CA: Academic Press. Fesitnger, L. (1957). A theory of cognitive dissonance. Stanford, CA: Stanford University Press. Foroni, F., & Mayr, U. (005). The ower of a story: New, automatic associations from a single reading of a short scenario. Psychonomic Bulletin & Review, 1, Gawronski, B., & Bodenhausen, G. V. (006). Associative and roositional rocesses in evaluation: an integrative review of imlicit and exlicit attitude change. Psychological Bulletin, 13, Gawronski, B., & Bodenhausen, G. V. (in ress). Unraveling the rocesses underlying evaluation: Attitudes from the ersective of the APE Model. Social Cognition. Gawronski, B., Walther, E., & Blank, H. (005). Cognitive consistency and the formation of interersonal attitudes: Cognitive balance affects the encoding of social information. Journal of Exerimental Social Psychology, 41, Greenwald, A. G., Banaji, M. R., Rudman, L. A., Farnham, S. D., Nosek, B. A., & Mellott, D. S. (00). A unified theory of imlicit attitudes, stereotyes, self-esteem, and self-concet. Psychological Review, 109, 3-5.

34 34 Transitive Association Formation Greenwald, A. G., & Pratkanis, A. R. (1984). The self. In R. S. Wyer & T. K. Srull (Eds.), Handbook of social cognition ( ). Hillsdale, NJ: Erlbaum. Greenwald, A. G., Nosek, B. A., & Banaji, M. R. (003). Understanding and using the Imlicit Association Test: I. An imroved scoring algorithm. Journal of Personality and Social Psychology, 85(), Greenwald, A., G., Nosek, B., A., Banaji, M., R., & Klauer, K., C. (005). Validity of the Salience Asymmetry Interretation of the Imlicit Association Test: Comment on Rothermund and Wentura (004). Journal of Exerimental Psychology: General, 134(3), Gregg, A. P., Seibt, B., & Banaji, M. R. (006). Easier done than undone: Asymmetry in the malleability of imlicit references. Journal of Personality and Social Psychology, 90, 1-0. Grice, H. P. (1975). Logic and conversation. In P. Cole & J. L. Morgan (Eds.), Syntax and semantics, 3: Seech acts ( ). New York: Academic Press. Hayes, S. C., & Bissett, R. T. (1998). Derived stimulus relations roduce mediated and eisodic riming. Psychological Record, 48, Hayes, S. C., Barnes-Holmes, D., & Roche, B. (001). Relational frame theory: A ost- Skinnerian account of human language and cognition. New York: Kluwer Academic/Plenum Publishers. Hebl, M. R., & Mannix, L. M. (003). The weight of obesity in evaluative others: A mere roximity effect. Personality and Social Psychology Bulletin, 9, Heider, F. (1958). The sychology of interersonal relations. New York: Wiley.

35 35 Transitive Association Formation Hutchison, K. A. (003). Is semantic riming due to association strength or feature overla? A microanalytic review. Psychonomic Bulletin & Review, 10, Kihlstrom, J.F., & Cantor, N. (1984). Mental reresentations of the self. In L. Berkowitz (Ed.), Advances in exerimental social sychology. 17. New York: Academic Press. Krosnick, J. A., Betz, A. L., Jussim, L. J., & Lynn, A. R. (199). Subliminal conditioning of attitudes. Personality and Social Psychology Bulletin, 18, Lang, P. J., Bradley, M. M., & Cuthbert, B. N. (1990). Emotion, attention, and the startle reflex. Psychological Review, 97, Livesay, K., & Burgess, C. (1998). Mediated riming in high-dimensional semantic sace: No effect of direct semantic relationshis or cooccurrence. Brain & Cognition, 37, McKoon, G., & Ratcliff, R. (199). Sreading activation versus comound cue accounts of riming: Mediated riming revisited. Journal of Exerimental Psychology: Learning, Memory, & Cognition, 18, Meersmans, T., De Houwer, J., Baeyens, F., Randell, T., & Eelen, P. (005). Beyond evaluative conditioning? Searching for associative transfer of nonevaluative stimulus roerties. Cognition and Emotion, 19, Mitchell, C. J., Anderson, N. E., & Lovibond, P. F. (003). Measuring evaluative conditioning using the Imlicit Association Test. Learning and Motivation, 3, Mitchell, J. P., Nosek, B. A., & Banaji, M. R. (003). Contextual variations in imlicit evaluation. Journal of Exerimental Psychology: General, 13(3),

36 36 Transitive Association Formation Nosek, B. A. (005). Moderators of the relationshi between imlicit and exlicit evaluation. Journal of Exerimental Psychology: General, 134, Nosek, B. A., Greenwald, A. G., & Banaji, M. R. (007). The Imlicit Association Test at age 7: A methodological and concetual review. In J. A. Bargh (Ed.), Social Psychology and the Unconscious: The Automaticity of Higher Mental Processes (. 65-9). Psychology Press. Nosek, B. A., & Hansen, J. J. (in ress). The associations in our heads belong to us: Searching for attitudes and knowledge in imlicit evaluation. Cognition and Emotion. Nosek, B. A., Sutin, E., Hansen, J. J., Wu, L., Sriram, N., Smyth, F. L., Banaji, M. R., & Greenwald, A. G. (006). Project Imlicit: Advancing theory and evidence with technical and methodological innovation. Unublished manuscrit: University of Virginia. Nowak A., Vallacher R.R., Tesser A., & Borkowski W. (000). Society of Self: The emergence of collective roerties in self-structure. Psychological Review, 107, Osgood, C. E., Suci, G. J., & Tannebaum, P. H. (1957). The Measurement of Meaning. University of Illinois Press, Urbana, IL. Osgood, C. E., & Tannenbaum, P. H. (1955). The rincile of congruity in the rediction of attitude change. Psychological Review, 6, Ranganath, K. A. & Nosek, B. A. (in ress). Imlicit attitude generalization occurs immediately, exlicit attitude generalization takes time. Psychological Science.

37 37 Transitive Association Formation Ranganath, K. A., Smith, C. T., & Nosek, B. A. (in ress). Distinguishing automatic and controlled comonents of attitudes from direct and indirect measurement. Journal of Exerimental Social Psychology. Rothermund, K., & Wentura, D. (004). Underlying Processes in the Imlicit Association Test: Dissociating Salience From Associations. Journal of Exerimental Psychology: General, 133, Rydell, R., J., & Gawronski, B. (007). I Like You, I Like You Not: Understanding the Context Deendency of Automatic Evaluation. Unublished manuscrit. Rydell, R., J. & McConnell, A., R. (006). Understanding Imlicit and Exlicit Attitude Change: A Systems of Reasoning Analysis. Journal of Personality and Social Psychology, 91, Sayette, M. A., Hufford, M. R., & Thorson, G. M. (1996). Develoment of a brief measure of semantic riming. Journal of Clinical & Exerimental Neurosychology, 18, Smith, E. R., & DeCoster, J. (1998). Knowledge acquisition, accessibility, and use with a recurrent connectionist network. Journal of Personality and Social Psychology, 74 (1), Staats, A. W., & Staats, C. K. (1961). Denotative meaning established by classical conditioning. Journal of Exerimental Psychology, 61, Smyth, S., Barnes-Holmes, D., & Forsyth, J. P. (006). A derived transfer of simle discrimination and self-reorted arousal functions in sider fearful and non-siderfearful articiants. Journal of the Exerimental Analysis of Behavior, 85, 3-46.

38 38 Transitive Association Formation Strack, F., & Deutsch, R. (004). Reflective and reflexive determinants of social behavior. Personality and Social Psychology Review, 8, Walther, E. (00). Guilty by mere association: evaluative conditioning and the sreading attitude effect. Journal of Personality and Social Psychology, 8: Walther, E., & Nagengast, B. (006). Evaluative conditioning and the awareness issue: Assessing contingency awareness with the four-icture recognition test. Journal of Exerimental Psychology: Animal Behavioral Processes, 3, Wegner, D. M., & Bargh, J. A. (1998). Control and automaticity in social life. In D.Gilbert, S. T.Fiske, & G.Lindzey (Eds.), Handbook of social sychology (4th ed., Vol. 1, ). New York: McGraw-Hill. Zajonc, R. B. (1980). Feeling and thinking: Preferences need no inferences. American Psychologist, 35, Zhong C., & Liljenquist, K. (006), Washing Away Your Sins: Threatened Morality and Physical Cleansing. Science,

39 39 Transitive Association Formation Footnotes 1 Notably, findings of mediated riming (Balota & Lorch, 1986; McKoon & Ractliff, 199; Sayette, Hufford & Thorson, 1996, Livesay & Burgess, 1998) suggest transitive association, because words that do not have any semantic or exeriential relationshi show riming relations when they are both associated with the same word. For instance, Lion rimes Stries, because of their shared association with Tiger. However, some have argued that this may also reflect direct riming effect of weakly related concets (McKoon & Ratcliff, 199; see Hutchison, 003 for a review). All reorted analyses in all studies remain significant if we include all articiants. However, when we include articiants that did not erform well in the association formation task, unsurrisingly, the effects sizes tend to be weaker. 3 The concets were selected because they did not have obvious re-existing associations among them. Eventually, almost all airs of concets did show a certain re-existing associations common among most eole. In Study 1, for instance, there was an overall re-existing reference for the association R+Cats, Z+Dogs (over the oosite R+Dogs, Z+Cats) and the association R+Ovals, Z+Polygons over the oosite. The re-existing associations between the three concets should not disturb our research because we measured changes in associations as a result of the maniulation, not absolute associations, and we fully counterbalanced the research design so that re-existing associations would not be an alternative exlanation.

40 40 Transitive Association Formation 4 Rothermund and Wentura (004) argue that the IAT is redominately influenced by salience asymmetries and not associative relations. While we agree that salience asymmetries can influence IAT effects (Greenwald, Nosek, Banaji, & Klauer, 006), they are unlikely to account for the exerimental effects roduced here. Exerimental maniulations rovided equivalent exosure to concets and are unlikely to have differentially affected concet salience. 5 Illustrating this differential effect, using similar ANOVAs, but testing only of 3 airs at a time, the interaction was significant between dogs/cats and good/bad, F(1, 1174) = 5.71, =.0, η =.01, and between dogs/cats and male/female, F(1, 1180) = 13.55, =.00, η =.01, but not between good/bad and male/female, F(1, 114) = 1.85, =.17, η < Particiant also rated crossing lines and rocks on the three basic dimensions suggested by Osgood, Suci, & Tannenbaum (1957): Valence (good bad), Potency (strong weak) and Action (assive active). We tested whether the association between rocks and crossing lines would affect the relationshi between the ratings of these two concets on the different dimensions, but none occurred.

41 41 Transitive Association Formation Figure Cations Figure 1. Illustration of one out of six airing conditions in Study 1. The solid lines illustrate the direct associations formed in this condition: R with Dogs, Z with Cats, Dogs with Polygons, and Cats with Ovals. The dotted lines illustrate the redicted indirect association that would form transitively: R with Polygons, Z with Cats. Figure. Illustration of a trial in the airing rocedure (Studies 1-3). The target stimulus is an oval, and should be categorized as either cats (with the left key) or dogs (with the right key), deending on the task s airing rule ( ovals go with cats and olygons go with dogs, or ovals go with dogs and olygons go with cats ). Figure 3. The effect of the direct association formation on the relevant automatic association. The association in Panels A-C is between the first two concet airs in the association formation chain, in Studies 1-3, resectively. Panel D illustrates the same effect for the second and third concet airs in the association formation chain of Study 3. In Studies -3, the effects are resented searately for each carrier concets air (the air that was directly aired with the two other airs). The measure of automatic associations is the IAT. Higher values indicate stronger reference for the association secified in the Y axis over the oosite association. Error bars indicate the standard errors of the means.

42 4 Transitive Association Formation Figure 4. The effect of the indirect association formation on the relevant automatic association, in Studies 1 (Panel A), ( Panel B), and 3 (Panel C). In Studies -3, the effects are resented searately for each carrier concets air (the air that was directly aired with the two other concets). Larger values indicate stronger reference for the association secified in the Y axis over the oosite association. Error bars indicate the standard errors of the means. Figure 5. The effect of the indirect association formation on the relevant self-reorted association, in Studies 1 (Panel A), ( Panel B), and 3 (Panel C). In Studies -3, the effects are resented searately for each carrier concets air (the one that was directly aired with the two other concets). The ossible resonses in Studies 1 & were -1 (reference for one airing) and 1 (reference for the other airing). Study 3 also included the resonse 0 (no airing is referred). Figure 6. Illustration of two trials in the subtle airing rocedure (Study 4). Panel (A) illustrates a trial in Blocks 1 and 3. The target stimuli are icture of crossing lines and the word Noisy, and should be categorized with the key O to the to-right corner of the screen, where the categories crossing and loud aear. Panel (B) illustrates a trial in Blocks and 4, in which the category airs were different. The ositions of the category airs on the screen were rotated every 5 trials. Figure 7. The effect of the indirect association formation (Study 4) on the measured indirect association Automatic (Panel A), and self-reorted (Panel B). The results are

43 43 Transitive Association Formation searated by awareness grous. The measure of automatic associations is the IAT. The ossible resonses for the self-reort association measure were 1 (reference for crossing lines+rocks, arallel+trees), -1 (reference for the other airing), and 0 (no reference). Larger values indicate stronger reference for the association crossing lines+rocks, arallel+trees. Error bars indicate the standard errors of the means.

44 44 Transitive Association Formation Figure 1

45 45 Transitive Association Formation Figure

46 46 Transitive Association Formation Figure 3 Automatic Association of R+Dogs R+Dogs R+Cats Pairing condition A Automatic Association of R+Ovals Paired R+Polygons Paired R+Ovals Dogs/Cats Good/Bad Male/Female Carrier Pair B

47 47 Transitive Association Formation Automatic Association of Crossing+Loud/Bad/Other Paired Crossing+Quiet/Good/Self Paired Crossing+Loud/Bad/Other Quiet/Loud Good/Bad Self/Other Carrier Pair C Automatic Association of Quiet/Good/Self+Trees Paired Quiet/Good/Self+Rocks Paired Quiet/Good/Self+Trees Quiet/Loud Good/Bad Self/Other Carrier Pair D

48 48 Transitive Association Formation Figure 4 Automatic Association of R+Ovals R+Ovals R+Polygons Pairing condition A Automatic Association of R+Dogs/Good/Male Paired R+Dogs/Good/Male Paired Z+Dogs/Good/Male Dogs/Cats Good/Bad Male/Female Carrier Pair B

49 49 Transitive Association Formation 0.8 Automatic Association of Parallel Lines+Trees Paired Parallel Lines+Rocks Paired Parallel Lines+Trees Quiet/Loud Good/Bad Self/Other Carrier Pair C

50 50 Transitive Association Formation Figure 5 Self-Reorted Association of R+Ovals of answers R+Ovals R+Polygons Pairing condition A Self-Reorted Association of R+Ovals Paired R+Polygons Paired R+Ovals Dogs/Cats Good/Bad Male/Female Carrier Pair B

51 51 Transitive Association Formation Self-Reorted Association of Parallel Lines+Trees Paired Parallel Lines+Trees Paired Parallel Lines+Rocks Quiet/Loud Good/Bad Self/Other Carrier Pair C

52 5 Transitive Association Formation Figure 6 A B