Changing US Attributes After CS-US Pairings Changes CS-Attribute-Assessments

Abstract

Attribute Conditioning (AC) refers to people’s changed assessments of stimuli’s (CSs) attributes due to repeated pairing with stimuli (USs) possessing these attributes; for example, when an athletic person (US) is paired with a neutral person (CS), the neutral person is judged to be more athletic after the pairing. We hypothesize that this AC effect is due to CSs’ associations with USs rather than direct associations with attributes. Three experiments test this hypothesis by changing US attributes after CS-US pairings. Experiments 1 and 2 conditioned athleticism by pairing neutral men (CSs) with athletic and non-athletic USs. Post-conditioning, USs’ athleticism was reversed, which systematically influenced participants’ assessment of CS athleticism. Experiment 3 conditioned athleticism and changed USs’ musicality after CS-US pairings. This post-conditioning change affected musicality assessments of CSs but did not influence athleticism-assessments. The results indicate that AC effects are based on an associative CS-US-attribute structure.

Keywords

attribute conditioning US-attribute-reversal associative learning spontaneous trait transference

Basic learning phenomena have become attractive explanatory constructs within social cognitive research; this attraction is due to their power to explain complex social phenomena with a basic set of principles (Walther, Nagengast, & Trasselli, 2005). One such complex phenomenon is the process of social perception when people ascribe traits and attributes to people. There is much research about inferring attributes and traits from behavior (e.g., Carlston & Skowronski, 2005; Uleman, 1987; Winter & Uleman, 1984), from facial and bodily cues (e.g., Todorov, Said, Engel, & Oosterhof, 2008; Willis & Todorov, 2006), or verbal descriptions (e.g., Bazińska & Wojciszke, 1996).

Recent evidence suggests that social perceivers may ascribe attributes like intelligence, attractiveness, or athleticism simply due to the pairing of a person with another person possessing these attributes (Förderer & Unkelbach, 2014). For example, when a neutral man is repeatedly paired with an athlete, people assess this man as being more athletic than another man that was paired with non-athletic others (Förderer & Unkelbach, 2011). Förderer and Unkelbach (2015) termed this effect Attribute Conditioning (AC), referring to changes in people’s assessment of initially neutral stimuli’s (CSs) attributes due to their repeated pairing with other stimuli possessing these attributes (USs), beyond pure evaluative effects of stimulus pairings as in evaluative conditioning (EC) research (see Förderer & Unkelbach, 2011). Thereby, AC constitutes a simple learning phenomenon that contributes to attitude formation and person perception, with applied implications in areas such as consumer research or interventions aiming at stereotype change.

AC effects have been found, for example, for people’s assessment of CSs’ softness or size (Kim, Allen, & Kardes, 1996), gender (Meersmans, De Houwer, Baeyens, Randell, & Eelen, 2005), or color (Galli & Gorn, 2011). Further research examined boundary conditions like attribute accessibility, feature similarity, or contingency (Förderer & Unkelbach, 2014, 2015; Olson, Kendrick, & Fazio, 2009). However, at present, there is not yet a theoretical model for AC effects. None of these studies tested the possible processes underlying these changes in attribute assessments. The current study makes a first step toward a model of AC effects and examines what is learned when stimuli are paired.

What May Underlie AC Effects?

AC is based on observing stimuli’s pairings in the environment. As such, CS-US pairings may induce the formation or consolidation of associations (e.g., Gawronski & Bodenhausen, 2006) and consequently, CS-attribute-assessments change. Assuming that associations form when people observe CS-US pairings, a critical question is whether associations form between CS and US representations (i.e., indirect CS-attribute associations) or between CS and attribute representations (i.e., direct CS-attribute associations). Figure 1 depicts these two possible structures. An indirect association (see Figure 1’s left part) indicates that all US attributes and changes within these attributes should affect CS assessment. For example, if an athlete (US) is not athletic anymore, a formerly paired man (CS) should also be assessed as less athletic, although he does not co-occur with the non-athletic US anymore. A direct association (see Figure 1’s right part) indicates that the CS-attribute-assessment remains constant (i.e., man rated athletic after pairing with an athlete) even when US attributes change (i.e., the former athlete is not athletic anymore).

Figure 1.

The left part illustrates indirect CS-attribute associations via a CS-US association. The right part illustrates a direct CS-attribute 1 association.

The association between CS and US (Figure 1’s left part) is akin to the model underlying the procedurally very similar EC effects; in EC, a neutral stimulus (CS) takes on valence due to pairings with a positively or negatively evaluated stimulus (US). There is an ongoing debate about what processes may underlie EC; propositional processes (e.g., Stahl, Unkelbach, & Corneille, 2009) as well as associative processes are discussed (e.g., Hütter, Sweldens, Stahl, Unkelbach, & Klauer, 2012; see Sweldens, Corneille, & Yzerbyt, 2014). However, there is little doubt that the relevant processes happen between the CS and the US (please refer to Förderer & Unkelbach, 2015, for an elaborate discussion on the similarities and differences of AC and EC).

The association between CS and attribute (Figure 1’s right part) is akin to the model underlying spontaneous trait transference (STT); in STT experiments, a communicator takes on the attribute she/he is ascribing to another person (Skowronski, Carlston, Mae, & Crawford, 1998). For example, if Jeremy describes how Ian ran 100 m under 11 s, people transfer the trait “athletic” to Jeremy, albeit the trait is inferred from behavior that describes Ian. Brown and Bassili (2002) suggested that such STT effects

may result primarily from automatic associations formed in working memory between trait constructs [ . . . ] and other stimuli such as actors, bystanders, or even inanimate objects that happen to be part of the context that led to the activation of the traits. (p. 91)

Similarly, Skowronski and colleagues (Carlston & Skowronski, 2005; Crawford, Skowronski, Stiff, & Scherer, 2007) also postulated associative processes for STT effects, in contrast to inferential processes in spontaneous trait inferences (STIs; Winter & Uleman, 1984). In sum, there is little doubt that the most relevant processes happen between the communicator and the trait.

In other words, the empirical question whether AC procedures associate the CS with the US or the CS with the attribute locates the phenomenon theoretically closer to typical learning phenomena such as EC effects or to typical person perception phenomena such as STT effects.

The Present Research

To answer the question which associations form (see Figure 1), we use an US-attribute-reversal paradigm, similar to US-revaluation paradigms used in EC research (Mackintosh, 1983). The paradigm starts with a standard AC procedure: A neutral CS is paired with a US possessing an attribute. The typical AC effect is a change in participants’ CS-attribute assessment. After the CS-US pairing, the US attribute is reversed; for example, an initially athletic person turns into a couch potato. Then, the CS-attribute is assessed again. If participants’ CS assessments change based on attribute reversal (e.g., in comparison with a prior measurement or a no-reversal condition), it would support an indirect CS-attribute association. If the assessment does not change, direct CS-attribute associations would be supported, albeit such support would rest on a null result. The current study thereby examines whether changing US attributes post-conditioning changes CS assessment without further pairings. While this would be an interesting effect on its own with many practical implications, it allows for theoretical conclusions as well, namely, if indirect CS-US-attribute forms during conditioning.

Such a US-attribute-reversal paradigm has its caveats; participants’ explicit memory may guide their judgments, or awareness of the experiments’ hypotheses, or other effects that might change CS assessments independent of any US attribute changes or reversals. For example, CS assessments might change due to regression-to-the-mean or unsystematic variations in repeated measurements. The following three experiments, however, implement experimental settings that make such alternatives unlikely and, to anticipate the results, will provide evidence for indirect CS-attribute associations.

We condition athleticism to pictures of neutral men (CSs) by pairing them with athletic or non-athletic comic characters (USs). In the following US-attribute-reversal phase, participants either learn that USs change over time so that their original attribute is reversed (Experiments 1 and 2), or that some USs possess an additional unrelated attribute (i.e., musicality; Experiment 3). Introducing information about a new US attribute (i.e., Attribute 2 in Figure 1) allows examining whether changes in USs’ attributes only affect assessments of the initially conditioned attribute or whether all attributes associated with a US affect CS assessment. Given CS-US associations, all US attributes should influence CS assessment independent of when these associations formed.

The reversal procedure necessitates multiple CS assessments, which suggests a possible moderating factor from EC research, namely, assessment consolidation (cf. Förderer & Unkelbach, 2013). If participants express their CS-attribute assessment by rating CSs after conditioning, they may consolidate their CS assessment. Participants might then strive to be consistent with their initial assessment (e.g., Festinger, 1957; Heider, 1946) or they may simply recall their initial judgment (Hastie & Park, 1986). To examine this possible moderator, we manipulated whether participants rated CSs twice—after conditioning and after US-attribute-reversal—or only-after US-attribute-reversal. In addition to direct attribute ratings, a semantic misattribution procedure (SMP; Imhoff, Schmidt, Bernhardt, Dierksmeier, & Banse, 2011) served as an indirect attribute measure. Finally, participants’ demand awareness and demand compliance were assessed in Experiments 2 and 3.

Experiment 1

Experiment 1 paired photographs of male faces (CSs) with drawings of athletic or non-athletic comic characters (USs). Afterwards, the US characters’ athleticism was reversed using short comic strips (see Supplementary Materials SM_A1, for examples). The critical question was whether US-attribute-reversals change following CS assessments.

Method

Participants, design, and materials

Thirty-nine university students (27 women, 12 men) participated for 3€ or course credit. The sample size was based on prior experience within AC research (e.g., Förderer & Unkelbach, 2011; Olson et al., 2009). Within-participants, we manipulated the factor conditioning by pairing neutral men (CSs) with athletic or non-athletic comic characters (USs), leading to athletic or non-athletic CSs. Between-participants, we manipulated the factor measurement (before-and-after vs. only-after US-attribute-reversal). Participants either assessed CSs after conditioning and after US-attribute-reversal or they assessed CSs only-after US-attribute-reversal. This allowed comparing AC effects after conditioning but before US-attribute-reversal with AC effects after US-attribute-reversal. Furthermore, we could examine whether assessing CSs before US-attribute-reversal influences CS assessments afterwards. As a control factor, we manipulated the order of direct and indirect measurements (SMP order: SMP before vs. SMP after direct ratings).

Dependent variables were participants CS assessments of athleticism and likeability on 8-point scales (1 = non-athletic/not likeable to 8 = athletic/very likeable), as well as the SMP (see below for details). We always collected data on both CSs’ attributes and their likeability to distinguish between genuine AC effects and spillovers from possible EC effects (see Förderer & Unkelbach, 2011).

Six black-and-white portrait photographs of men used by Förderer and Unkelbach (2011) served as neutral CSs. Six self-sketched comic characters served as USs (see Supplementary Materials SM_A1). There were three different sketches for each character showing three gradations of athleticism. One sketch showed the character being non-athletic (e.g., corpulent man sitting on a couch and eating crisps), one being athletic (e.g., well-trained man doing speed-walking), and one showing a medium level of athleticism (e.g., man doing Nordic-walking but still having a belly). Sixty-eight participants rated athleticism and valence of all 18 sketches on 10-point scales in an independent pretest; results showed that participants perceived athletic USs as more athletic than neutral ones and these were more athletic than non-athletic ones (see Table A1 in the Supplementary Materials SM_A1). For each participant anew, the program randomly chose three US comic characters to be shown in an athletic state (i.e., athletic USs); the remaining three characters were used in non-athletic state (i.e., non-athletic USs). Each CS was randomly assigned to one US for each participant.

The US-attribute-reversal phase presented each US in its three athleticism gradations organized in a timeline ranging from “past” to “today.” The US’s athleticism state during conditioning constituted the past, then the medium state sketch followed, and the remaining extreme state constituted the current state. In addition, below the pictures, a textbox provided a short story about USs’ change in exercising, pointing out the change from athletic (non-athletic) behaviors to non-athletic (athletic) behaviors in this character (see Table A2 in the Supplementary Materials SM_A1). A VisualBasic (VB) program controlled instructions, stimulus presentation, and recorded dependent variables.

Procedure

Experimental sessions included up to six participants. Experimenters seated them in front of computers and after giving informed consent to participate, the VB program instructed them to closely observe a series of men’s pictures paired with comic characters. During conditioning, CSs were paired eight times with their assigned USs resulting in 48 pairing trials. CSs appeared in the upper screen half for 1 s, then the respective US appeared in the lower screen half for another 2.5 s with the CS still on the screen. After a pause of 1.5 s, the next trial started. After conditioning, participants in the measurement group before-and-after rated CSs’ athleticism and likeability and completed the SMP for athleticism; explicit ratings and SMP order was counterbalanced.

The SMP was introduced as an orientation task; participants would see an orientation stimulus (i.e., CS as prime) flashed on the screen followed by a Chinese character (i.e., Kanji as target). Their task was to decide whether the Kanji stands for a word with an athletic or non-athletic meaning as fast as possible and without reacting to the orientation stimulus. A trial consisted of a CS visible for 75 ms, followed by a blank screen for 125 ms. Then a Kanji appeared for 100 ms which was replaced by a black-and-white noise picture until participants made their decision by pressing one of two marked keys on the keyboard. Each CS served as prime 10 times resulting in 60 trials with 60 different Kanji. Participants in measurement group only-after completed an unrelated filler task rating women’s likeability on a track-bar.

Next, US-attribute-reversal began. The program informed participants that they would see some comic strips. After participants clicked a start button, the first strip appeared. A strip consisted of the three sketches of one comic character, a past-today-timeline below them, and a textbox including the exercising change story at the bottom. To ensure that participants took a close look at the strips and read the stories properly, the button to go on with the next strip was enabled only after 8 s.

Next, participants completed CSs ratings and the SMP (a second time for measurement group before-and-after). In addition, they rated USs’ athleticism on the same 8-point scales. As we were interested in participants’ assessment of US athleticism at that time, only comic characters’ faces were presented to safeguard against inferring athleticism from the characters’ body shapes. Upon completing these ratings, participants were thanked, debriefed, and if applicable, paid.

Results

To analyze direct athleticism ratings and SMP data, we used two mixed analyses of variance (ANOVAs): First, a before-after-reversal analysis compared the before-and-after group’s data after conditioning but before US-attribute-reversal with the only-after group’s CS ratings after US-attribute-reversal. This analysis tested whether AC effects emerged before US-attribute-reversal and whether they changed due to US-attribute-reversal. The model had conditioning (athletic vs. non-athletic) as a within-participants factor and measurement time (before vs. after US-attribute-reversal) as a between-participants factor (which is identical to “measurement group”). In addition, it only considers initial CSs ratings, thereby ensuring against repeated measurement effects.

Second, an after-reversal analysis compared all participants’ data after the reversal phase with conditioning (athletic vs. non-athletic) as within-participants factor and measurement group (before-and-after vs. only-after) as between-participants factor. This analysis tested whether rating CSs prior to US-attribute-reversal influenced AC effects after US-attribute-reversal (e.g., such as memory for prior ratings or assessment consolidation).

Measurement order (direct ratings vs. SMP) was included in all analyses as between-participants factor. If not stated otherwise, it did not have any significant or meaningful effects and, therefore, was not reported in this and all following experiments. To keep the report concise, we report non-significant effects only if they are relevant for hypotheses testing.

Manipulation check

After US-attribute-reversal, participants rated the faces of initially non-athletic USs more athletic (M = 4.81, SD = 1.27) than initially athletic USs (M = 4.03, SD = 1.52), F(1, 35) = 3.76, p = .061. This indicates that US-attribute-reversal influenced peoples’ athleticism-assessment although the reversal did not reach standard levels of significance when participants rated only the faces without the body being visible.

Direct ratings

We calculated participants’ mean athleticism ratings for athletic and non-athletic CSs. Figure 2’s upper half shows the relevant means. The before-after-reversal ANOVA showed—as expected—no conditioning main effect, F(1, 35) = 1.88, p = .179. Yet, there was a significant Conditioning × Measurement Time interaction, F(1, 35) = 7.28, p = .011, d = 0.83, CI_95% = [0.16, 1.50], resulting from standard AC effects after conditioning and no AC effects after US-attribute-reversal. This indicates that US-attribute-reversal significantly influenced CSs assessments. The only-after-reversal ANOVA showed no AC effect, F(1, 35) = 1.05, p = .313, indicating that US-attribute-reversal neutralized athleticism-assessments of the CSs. There was no conditioning by measurement group interaction, F(1, 35) = 0.05, p = .823, indicating that assessment consolidation did not moderate the effect of US-attribute-reversal.

Figure 2.

Experiment 1: Mean athleticism ratings of athletic and non-athletic CSs (upper part) and probability to respond “athletic” to Kanji following athletic and non-athletic CSs (lower part) before and after US-attribute-reversal split for measurement groups before-and-after and only-after.

To further examine the data pattern, we subtracted athleticism ratings after US-attribute-reversal from ratings after conditioning for initially athletic and non-athletic CSs (i.e., data from before-and-after group) and tested these two reversal effects against zero. These showed that the athleticism-assessment of initially athletic CSs significantly decreased (M = −1.15, SD = 1.88), t(19) = −2.74, p = .013, d = 1.26, CI_95% = [0.65, 1.87], whereas it significantly increased for initially non-athletic CSs (M = 0.88, SD = 1.86), t(19) = 2.13, p = .047, d = 0.98, CI_95% = [0.43, 1.53]. That is, for both athletic and non-athletic CSs post-conditioning, US-attribute-reversal significantly influenced CSs assessments.¹

SMP

We intended the SMP to serve as an indirect measure that does not involve substantial inferential processes or exhaustive memory searches. We thus deleted trials with response latencies slower than 2 s (about 8% of all trials). On average, participants responded within 928 ms (SD = 188 ms). For each participant, we calculated probabilities to respond “athletic” to Kanji following athletic and non-athletic CS primes. Figure 2’s lower half shows the relevant mean probabilities. We analyzed these probabilities using the same mixed ANOVAs as for direct ratings. The before-after-reversal ANOVA showed a trend for an overall AC effect, F(1, 35) = 2.96, p = .094, showing that participants rated athletic CSs more athletic than non-athletic ones. This main effect was due to a significant AC effect after conditioning but before US-attribute-reversal, shown in a separate analysis including only the before-and-after group’s ratings after conditioning, F(1, 18) = 6.26, p = .022, d = 1.04, CI_95% = [0.48, 1.60]. There was no conditioning by measurement time interaction, F(1, 35) = 2.26, p = .142; yet, the comparison of Figure 2’s upper and lower half showed that ratings changed in the expected direction. As expected, after-reversal showed no significant effects (Fs < 1.40), indicating that US-attribute-reversal neutralized conditioning effects and that assessment consolidation had no moderating effect.

Additional t tests—comparing the difference between before and after US-attribute-reversal—showed that the probability to respond “athletic” decreased significantly following initially athletic CSs (M = −0.13, SD = 0.21), t(19) = −2.71, p = .014, d = 1.24, CI_95% = [0.63, 1.85], whereas it increased for initially non-athletic CSs (M = 0.07, SD = 0.17), albeit not significantly, t(19) = 1.74, p = .098. Thus, US-attribute-reversal influenced CS assessments though this was stronger for initially athletic CSs.

Discussion

Experiment 1 found standard AC effects for neutral men (CSs) immediately after pairings with athletic or non-athletic comic characters (USs). These CSs athleticism-assessments were significantly influenced by US-attribute-reversal after the CS-US pairings. Participants rated athletic CSs less athletic after their paired US was reversed to be non-athletic and vice versa for non-athletic CSs. This pattern was also reflected, although less pronounced, in the SMP data, which speak against explicit inferential processes during measurement. The SMP asks participants to rate Kanji instead of CSs, and even if participants inferred and complied with the reversal hypothesis, they would have had to recognize the CS, remember the paired US, remember that the US’s attribute has changed, and judge the Kanji in accordance with the reversed attribute within less than a second on average. Assuming such inferences to be unlikely, these data provide initial evidence that CSs are indirectly associated with relevant attributes via the US (see Figure 1 left half). If CSs would be associated with the relevant attributes only, US-attribute-reversal would not have changed CS assessments.

As Experiment 1 was the first to examine US-attribute-reversal effects, we aimed to replicate it with small changes. First, US-attribute-reversal significantly influenced CS assessments in Experiment 1, but no fully reversed AC effects emerged. This is comparable with findings within EC research, where US-revaluations also only neutralized valence-assessments (cf. Baeyens, Eelen, Van den Bergh, & Crombez, 1992; Walther, Gawronski, Blank, & Langer, 2009). One possible explanation is that the US-attribute-reversal was not strong enough as the manipulation check of the US faces also showed no significant reversal. Thus, Experiment 2 included stronger US-attribute-reversal manipulations. Second, although the SMP makes demand effects and explicit reasoning during measurement less likely, the implicitness of the measure has been called into question (Bar-Anan & Nosek, 2012; but see Payne et al., 2013). Thus, Experiment 2 included systematic funneled debriefings to identify participants who might be aware of the experimental hypotheses.

Experiment 2

Experiment 2 replicated Experiment 1 with minor changes. Pretest ratings showed that participants rated supposedly non-athletic but slim comic characters less non-athletic compared with corpulent characters. Hence, we changed these three USs sketches so that all non-athletic USs gradations were corpulent.

Experiment 1 used the SMP to prevent substantial inferential processes and demand effects. Experiment 2 included additional measures of demand awareness to test whether participants only show AC and US-attribute-reversal effects if they recognize the influence of CS-US pairings on their assessments (i.e., if they are aware of the experimental demands).

Method

Participants, design, and materials

Sample size was comparable with Experiment 1 for comparable power. Forty university students (30 women, 10 men) participated for 3€ or course credit. Design and materials were identical to Experiment 1 except for stronger manipulation of US-attribute-reversal by changing three body shapes in the non-athletic gradations so that they were corpulent instead of slim.

Procedure

Procedures were identical to Experiment 1 except for additional funneled debriefings at the end of the experiment measuring participants’ demand awareness. Participants answered the following questions: Question 1 (Q1): “In your opinion, what did we study in this experiment?” Question 2 (Q2): “You’ve rated several male portraits two times. Do you think your second rating changed compared with the first?”; if they answered “yes,” the following question was asked. Question 2.1 (Q2.1): “Why did your second rating change?” Question 3 (Q3): “Men were always presented with comic characters. Did these comic characters influence your assessment of the men?”; if participants clicked “yes,” they were asked Question 3.1 (Q3.1): “How did the comic characters influence your assessment of the men?” Q1, Q2.1, and Q3.1 had an open-response format; others questions were answered by clicking on yes or no buttons.

Results

The data analysis followed Experiment 1, and we used the same mixed ANOVAs (i.e., before-after-reversal and after-reversal) and t tests as in Experiment 1. Again, we restrict the report to significant effects; we report non-significant effects only if they are relevant for hypotheses testing.

Manipulation check

After US-attribute-reversal, participants rated the faces of initially non-athletic USs more athletic (M = 4.98, SD = 1.36) than initially athletic USs (M = 4.34, SD = 1.33); yet, the reversal was again not significant, F(1, 36) = 2.48, p = .124. Making USs’ athleticism gradations more extreme did not reinforce US-attribute-reversal in participants’ ratings when only faces without the body were shown.

Direct ratings

Figure 3’s top half shows participants mean athleticism ratings of athletic and non-athletic CSs. The before-after-reversal ANOVA showed a conditioning main effect, F(1, 36) = 6.21, p = .018, d = 0.75, CI_95% = [0.40, 1.10], due to a significant AC effect before US-attribute-reversal, F(1, 18) = 13.97, p = .002, d = 1.65, CI_95% = [0.95, 2.35]. Importantly, there was a significant conditioning by measurement time interaction, F(1, 36) = 7.03, p = .012, d = 0.81, CI_95% = [0.45, 1.17], resulting from standard AC effects after conditioning, which were eliminated and even slightly reversed after US-attribute-reversal. This indicated that US-attribute-reversal significantly influenced CS assessments. The after-reversal ANOVA showed no AC effect, F(1, 36) = 0.08, ns, indicating that US-attribute-reversal neutralized CS athleticism-assessments. There was no conditioning by measurement group interaction, F(1, 36) = 0.02, ns, indicating that assessment consolidation did not moderate the US-attribute-reversal effect.

Figure 3.

Experiment 2: Mean athleticism ratings of athletic and non-athletic CSs (upper part) and probability to respond “athletic” to Kanji following athletic and non-athletic CSs (lower part) before and after US-attribute-reversal split for measurement groups before- and-after and only-after.

Testing the difference of assessments before and after US-attribute-reversal showed that the athleticism-assessments of initially athletic CSs significantly decreased (M = −1.23, SD = 2.16), t(19) = −2.56, p = .019, d = 1.17, CI_95% = [0.58, 1.76], whereas they increased for initially non-athletic CSs (M = 0.83, SD = 1.90), although this increase was not significant on a standard level of significance, t(19) = 1.97, p = .064. That is, US-attribute-reversal influenced CSs assessments in both CSs types while this was stronger for initially athletic CSs. In sum, Experiment 2 fully replicated results of Experiment 1 on direct ratings.

SMP

We again deleted trials with response latencies slower than 2 s (about 5% of all trials). On average, participants responded within 782 ms (SD = 206 ms). Figure 3’s lower half shows participants’ probabilities to respond “athletic” to Kanji following athletic and non-athletic CS primes. The before-after-reversal ANOVA showed a significant overall AC effect, F(1, 36) = 14.13, p < .001, d = 1.19, CI_95% = [0.48, 1.90], indicating that participants rated Kanji following athletic CSs to represent athletic words more often than Kanji following non-athletic CSs. This effect was due to a significant AC effect after conditioning (only before-and-after group’s responses), F(1, 18) = 12.01, p = .002, d = 1.52, CI_95% = [0.85, 2.19]. Importantly, the analysis revealed a significant conditioning by measurement time interaction F(1, 36) = 6.10, p = .018, d = 0.74, CI_95% = [0.06, 1.42] showing the expected AC effect before but not after US-attribute-reversal. The after-reversal ANOVA showed no significant effects (Fs < 1); that is, US-attribute-reversal neutralized conditioning effects and assessment consolidation had no moderating effect.

Additional t tests showed that athletic responses significantly decreased for Kanji following initially athletic CSs (M = −0.15, SD = 0.30), t(19) = −2.26, p = .036, d = 1.04, CI_95% = [0.48, 1.60], whereas they increased for Kanji following initially non-athletic CSs (M = 0.15, SD = 0.31), t(19) = 2.15, p = .044, d = 0.99, CI_95% = [0.44, 1.54]. In sum, SMP data showed all expected effects; that is, US-attribute-reversal significantly influenced CS assessments.

Demand awareness

Two independent raters who were blind for experimental conditions coded answers to Q1, Q2.1, and Q3.1. If coding diverged, a third rater decided between the two raters’ coding. Statements were coded “unaware” when participants had no or a wrong idea about study hypotheses and “aware” when participants partly or fully inferred the experiment’s hypotheses. Q2 or Q3 served as entry questions and were not included in any analyses. For example, participants might have answered that pairings influenced their ratings (“yes” in Q3), but might have given a wrong explanation of this influence (unaware in Q3.1). To keep the report concise, we only report demand analyses including data provided by all participants (i.e., data based on Q1 and Q3.1). Based on Q1, 15 participants were classified as “demand aware,” and 25 as “demand unaware.” Based on Q3.1, 23 participants were classified as “demand aware,” and 17 as “demand unaware.”

Based on these classifications, we included demand awareness as a factor for the theoretically relevant effects (i.e., the conditioning main effect and conditioning by measurement time interaction). If awareness of the experimental demands plays a crucial role for the observed effects to emerge, we would expect differences between the effects for the levels of awareness (i.e., classified aware vs. classified unaware); that is, an interaction of the relevant effects with demand awareness. The Supplementary Materials (SM_E2) report all the relevant means and analyses for the demand awareness data, separately for demand aware and unaware participants.

For explicit ratings, there was no interaction with demand awareness defined according to Q1, Q2.1, all Fs < 1.79, ps > .191. For Q3.1, the main effect for athleticism interacted significantly with awareness; the main effect was stronger for participants who reported that the pairings influenced their athleticism ratings. The critical interaction of conditioning by measurement time remained stable, F(1, 32) = 6.07, p = .019, d = 0.87, CI_95% = [0.46, 1.28] and did not interact with the awareness classification, F(1, 32) = 1.25, p = .273, ns.

For the SMP, there was no such interaction for Q1 and Q2.1, all Fs < 1.00, ps > .324. For Q3.1, there was an interaction of classified awareness and the conditioning main effect, F(1, 32) = 12.25, p < .005, d = 1.24, CI_95% = [0.78, 1.70]. Participants who reported that the pairings influenced their athleticism ratings showed stronger SMP effects (M_diff = .29, SD = .20) than participants who reported no such influence (M_diff = .04, SD = .18). In addition, there was a significant conditioning by measurement time interaction with classified awareness, F(1, 32) = 4.98, p < .005, d = 0.79, CI_95% = [0.39, 1.19]. Participants who reported an influence showed a clear reduction of the conditioning effect from first (M_diff = .51, SD = .22) to second measurement (M_diff = .10, SD = .14). For participants who reported no influence, the reduction was much weaker from first (M_diff = .05, SD = .21) to second measurement (M_diff = .03, SD = .14). While these interactions might indicate that the SMP effects depend on awareness, Q3.1 of the funneled debriefing is akin to an output bound sampling. That is, only participants who show strong conditioning effects might also be able to report them. We will get back to this point in the discussion.

Discussion

Replicating Experiment 1’s results, Experiment 2 showed that participants changed their athleticism-assessment of neutral men in the direction of paired athletic or non-athletic comic characters after conditioning. Furthermore, it replicated that changes in these characters’ athleticism after conditioning (i.e., US-attribute-reversal) also induced changes in men’s athleticism-assessment although these were not paired again; this is in line with indirect CS-US-attribute associations.

We found conditioning and reversal effects on direct ratings and on SMP data. Especially, finding effects on the SMP is striking as it is less susceptible to participants’ strategic rating control (cf. Payne et al., 2013). Nevertheless, we examine the possible influence of demand effects by including measures of demand awareness as a factor for the relevant effects. These analyses yielded a single possibly important influence. For Q3.1, only participants coded as demand aware showed the expected pattern; that is, participants correctly stating how CS-US pairings influenced their CSs ratings. This seems problematic, as participants might remember the pairing and rate CSs accordingly to comply with our hypotheses. However, Bar-Anan, De Houwer, and Nosek (2010) argued that participants might simply use the only information about CSs available—which is their pairing with specific USs and report this influence. Furthermore, participants might not have recognized this influence during CSs-attribute assessments; they might have inferred it only when we highlighted the systematic pairings and a possible influence during the funneled debriefing. Nevertheless, these data are consistent with the possibility that demand awareness might play a role; we hence further address the problem of demand awareness in Experiment 3.

These effects emerged although the US manipulation check did not reveal a significant effect. Yet, US athleticism-assessments went into the expected direction (i.e., former athletic USs were less athletic than former non-athletic USs). In addition, participants only rated the faces of these characters without the body, which should obviously weaken the influence of attribute reversal. In addition, we believe that these ratings would significantly differ from USs ratings before US-attribute-reversal; yet, due to missing baseline ratings for the same participants, we could not test changes in USs athleticism-assessments.

Together, Experiments 1 and 2 showed that changes in an US attribute after CS-US pairings changed CS-attribute-assessments. This indicates an associative CS-US-attribute structure. However, the conclusion has two obvious alternative explanations. First, the data patterns in Figures 2 and 3 are fully compatible with a regression-to-the-mean explanation (Fiedler & Unkelbach, 2014). Second, AC effects might have simply diminished over time (i.e., during the US-attribute-reversal phase), which would also imitate US-attribute-reversal effects.

To address these two alternatives, Experiment 3 varies the paradigm; instead of reversing an attribute, we introduced a new US attribute after CS-US pairings. If changes in CSs’ athleticism-assessments are indeed caused by US-attribute-reversal instead of non-conditioning related processes, athleticism-assessments before and after introducing a new unrelated attribute should be comparable. Conversely, if the AC effect in athleticism-assessments vanishes while the US attribute musicality is introduced, the data from Experiments 1 and 2 would likely be due to non-conditioning related processes (e.g., regression-to-the-mean).

In addition, CS-US-attribute associations (cf. Figure 1, left half) predict that all US attributes should influence CS assessments, even after CS-US pairings (cf., Förderer & Unkelbach, 2015). Experiment 3 provides the first empirical test of this prediction, which directly follows from the proposed model; that is, whether newly established US attributes (i.e., after conditioning) also influence CS assessment on this new attribute.

Experiment 3

Experiments 3 tested whether AC effects diminish over time (i.e., while learning a new US attribute) and whether providing participants with information about a new US attribute after conditioning influences assessments of this CS-attribute. We conditioned athleticism by pairing neutral men (CSs) with athletic or non-athletic comic characters (USs). Afterwards, some of these characters were introduced to be musical while others were presented in their current fitness level (i.e., same as in the conditioning phase).

If the AC effect changes in Experiments 1 and 2 are due to US-revaluation, CS athleticism-assessments should remain constant after introducing the new US attribute “musicality.” In addition, participants should rate CSs paired with USs that were shown to be musical as more musical than CSs paired with USs neutral in musicality. As the first prediction is based on the non-significant change of an effect, we used a larger sample than in Experiments 1 and 2 to allow detecting small changes in AC effects before and after introducing the new US attribute.

As in Experiments 1 and 2, we included SMPs measuring athleticism and musicality and added a measure of demand compliance in addition to measures of demand awareness. Being demand aware per se is no problem; participants might understand that pairings influenced their ratings. Yet, demand compliance would be a problem as participants might only show AC and reversal effects if they strategically control their ratings to comply with the study’s hypotheses.

Method

Participants, design, and materials

Seventy university students (55 women, 15 men) participated for 3€ payment or course credit. The within-participants factor conditioning (athletic vs. non-athletic) and the between-participants factor measurement (before-and-after vs. only-after) were identical to Experiments 1 and 2. In addition, we included another within-participants factor differentiating between USs introduced to be musical and USs with no new attribute (musicality: musical vs. neutral). For each participant, the program randomly chose three USs to be musical while the remaining ones stayed neutral in musicality. Finally, the program randomly manipulated between-participants whether participants completed direct athleticism ratings or musicality ratings first (rating-order: athletic-first vs. musical-first).

CSs and USs were identical to Experiments 1 and 2. The medium athleticism gradation of each comic character was dismissed while new sketches were drawn showing the musicality of the comic character (e.g., US as singer, US playing guitar, US playing saxophone). Musical USs sketches were drawn for both extreme athleticism gradations (i.e., for slim and corpulent characters) so that the body shape was identical to the conditioning phase. CS-US assignment was randomized for each participant.

To introduce the new US attribute, participants saw a US’s new musical sketch with a short sentence describing the musicality: For example, “George is an exceptionally gifted saxophone player.” For the three neutral US, participants saw the identical pictures of three USs with a short sentence describing their current fitness level. For example, the comic character sitting on the couch was described as, “George always sits on the couch and eats sweets.” The program presented USs in randomized order.

Demand awareness measures were identical to Experiment 2. To measure demand compliance, participants indicated whether they strategically controlled their CS assessments. A VB program controlled instructions, stimulus presentation, and recorded dependent variables.

Procedure

Instructions, conditioning, and first measurement of AC effects for measurement group before-and-after were identical to Experiments 1 and 2. Participants in measurement group only-after completed an unrelated filler task rating emoticons’ valence instead of assessing CSs. The introduction of a new US attribute showed a single US picture and below a textbox describing the musicality or current fitness level (i.e., three USs each, depending on randomly selected musicality-condition). The button to continue with the next comic character was enabled after 3 s.

Afterward, participants completed the same CSs ratings and an athleticism-SMP (a second time—measurement group: before-and-after) as in Experiments 1 and 2. In addition, all participants rated CSs musicality (1 = “not musical” to 8 = “very musical”) and completed a musicality-SMP. In the musicality-SMP, participants decided whether Kanji named musical instruments or furniture. Participants also assessed USs athleticism and musicality on 8-point scales. They completed both SMPs before direct ratings; order of direct athleticism and musicality ratings was counterbalanced (cf. rating-order).

Finally, participants answered the questions used in Experiment 2 to measure demand awareness (cf. Q1-Q3.1). In addition, they answered questions regarding their demand compliance. Question 4 (Q4): “Did you try to systematically control your ratings of the men?” (answer: yes or no). If they answered “yes,” Q4.1 followed: “How did you control your ratings?” (open-response format). Afterwards, participants answered whether they already participated in a comparable study to exclude participants who might have participated already in Experiments 1 and 2. Finally, experimenters thanked, paid, and debriefed participants.

Results

Prior to analyses, we excluded one participant as she reported participating in a previous AC study. To analyze direct ratings and SMP data of athleticism, we used comparable mixed ANOVAs (i.e., before-after and only-after) to Experiments 1 and 2. We included conditioning (athletic vs. non-athletic) as within-factor and measurement (before-and-after vs. only-after) and rating-order (athletic-first vs. musical-first) as between-factors.

To analyze direct ratings and SMP data of musicality, we used mixed ANOVAs including musicality (musical vs. neutral) as within-factor and measurement (of athleticism before-and-after vs. only-after US-attribute-reversal) and rating-order as between-factors. Again, we restrict the report to significant effects and report non-significant effects only if they are relevant for hypothesis testing.

Manipulation check

Participants rated athletic USs more athletic (M = 5.32, SD = 1.51) than non-athletic USs (M = 3.59, SD = 1.22), F(1, 67) = 52.48, p < .001, d = 1.74, CI_95% = [1.18, 2.03]. This effect was stronger for participants who did not rate CSs two times (i.e., measurement group only-after), F(1, 67) = 10.13, p = .002, d = 0.73, CI_95% = [0.24, 1.22]. Participants rated musical USs more musical (M = 5.35, SD = 1.17) than neutral USs (M = 4.11, SD = 1.11), F(1, 67) = 42.92, p < .001, d = 1.57, CI_95% = [1.02, 2.12]. This effect was slightly stronger for participants who did not rate CSs two times (i.e., measurement group only-after), F(1, 67) = 3.21, p = .077. In sum, athleticism- and musicality-manipulations were successful, and in comparison, these data show the reversal effectiveness in Experiments 1 and 2.

Athleticism—Direct Ratings

Figure 4’s top half shows participants mean athleticism ratings of athletic and non-athletic CSs. In contrast to Experiments 1 and 2, there is a consistent AC effect both before and after introducing the musicality attribute. The respective before-after analysis (as in Experiments 1 and 2) showed this AC main effect, F(1, 65) = 85.05, p < .001, d = 2.26, CI_95% = [1.81, 2.71], which was not influenced by measurement time, F(1, 65) = 1.90, p = .173. Similarly, again different from Experiments 1 and 2, the only-after analysis showed a strong AC effect, F(1, 65) = 71.04, p < .001, d = 2.09, CI_95% = [1.67, 2.51]. The analysis also yielded a significant conditioning by measurement group interaction, F(1, 65) = 4.52, p = .037, d = 0.53, CI_95% = [0.04, 1.02]. As Figure 4 shows, participants who only rated athleticism after introducing the musicality attribute showed a somewhat stronger AC effect.

Figure 4.

Experiment 3: Mean athleticism ratings of athletic and non-athletic CSs (upper part) and probability to respond “athletic” to Kanji following athletic and non-athletic CSs (lower part) before and after US-attribute-reversal split for measurement groups before- and-after and only-after.

Athleticism—SMP

Data conditioning was identical to the previous experiments. Figure 4’s lower half shows participants’ mean probability to respond “athletic” to a given Kanji following athletic and non-athletic CS primes. Participants responded on average within 862 ms (SD = 225 ms). As in the direct ratings, there is a consistent AC effect both before and after introducing the musicality attribute. The respective before-after analysis (as in Experiments 1 and 2) showed this AC main effect, F(1, 65) = 21.27, p < .001, d = 1.11, CI_95% = [0.60, 1.62], which was not influenced by measurement time, F(1, 65) = 0.07, ns. In addition, the only-after analysis showed also a clear AC effect, F(1, 65) = 23.43, p < .001, d = 1.20, CI_95% = [0.89, 1.51], an no other interactions or main effects, all Fs < 1.

In sum, conditioning of athleticism was successful and stable on both direct and indirect measures; introducing musicality did not affect athleticism-assessments. These data show that changes in attribute assessments in Experiments 1 and 2 were not due to diminished AC effects over time. The interaction for participants who rated the CSs twice indicates a small regression effect from the first to the second rating. This regression effect was not apparent for the SMP data.

Musicality—Direct ratings and SMP

Figure 5’s left part shows participants’ ratings of CSs when they learned that the respective USs were musical or neutral on musicality, after the actual CS-US pairings. When participants learned that a USs was musical, they rated the initially paired CSs more musical than CSs that were paired with USs for which participants did not receive musicality information, F(1, 65) = 9.20, p = .004, d = 0.70, CI_95% = [0.44, 0.96] (all other Fs < 1.80, ps > .187).

Figure 5.

Experiment 3: Mean musicality ratings of athletic and non-athletic CSs (left part) and probability to respond “musical” to Kanji following musical and musicality-neutral CSs (left part).

Figure 5’s right part shows participants’ mean probability to respond “musical instrument” to a given Kanji following musical and neutral CS primes. The musicality effect was not significant (F < 1) and there were no other main or interaction effects (Fs < 1), except for a significant Musicality × Measurement × Rating-Order interaction, F(1, 65) = 4.84, p = .031, d = 0.49, which could not be interpreted in a meaningful way (i.e., rating-order only concerned order of direct ratings and should not have influenced SMP data which were collected before ratings). In sum, we found an effect of introducing a new attribute after CS-US pairings on direct CS musicality assessments, but not on the SMP.

Demand awareness and demand compliance

For Q1 to Q4.1 data, coding procedure and analyses were comparable with Experiment 2. To keep the report concise, again we only report demand analyses including data provided by all participants (i.e., data based on Q1, Q3.1, and Q4.1). Based on Q1, 30 participants were classified as “demand aware,” and 39 as “demand unaware.” Based on Q3.1, 32 participants were classified as “demand aware,” and 37 as “demand unaware”; finally, based on Q4.1, only eight participants reported strategic control of their ratings while 61 participants reported no strategic control.

Similar to Experiment 2, we included demand awareness and demand compliance as a factor for all the theoretically relevant effects (i.e., the athleticism conditioning main effect and the musicality conditioning main effect). If demand awareness or demand compliance plays a role, we would again expect interactions of these variables with the relevant effects. The Supplementary Materials (SM_E3) report all the relevant means and analyses for the demand awareness data, separately for demand aware/controlled ratings and unaware/uncontrolled ratings participants.

For explicit athleticism ratings, there was no significant interaction involving any of the relevant effects for any demand or compliance measure, all Fs < 2.81, ps > 0.10. The explicit musicality ratings also did not interact with any demand and compliance measures, all Fs < 2.08, ps > 0.15. For SMP athleticism effects, there was no significant interaction involving any of the relevant effects for any demand or compliance measure, all Fs < 1. As we did not find SMP musicality effects, we omit the demand awareness and compliance analyses ( Supplementary SM_E3).

Together, these analyses suggest that demand control and demand compliance did not systematically influence the present results.

Discussion

Experiment 3 addresses three alternative explanations, regression-to-the-mean explanation (Fiedler & Unkelbach, 2014), diminishment over time (i.e., during the US-attribute-reversal), and possible demand compliance. The AC effects were fully stable before and after a procedurally similar (i.e., new information about the US post-conditioning) but functionally different phase, both on direct ratings and the SMP measure. In addition, the athleticism effects did not interact with participants’ awareness of the experiment’s hypotheses and their possible compliance with experimental demands. This strongly suggests that the CS assessment changes in Experiments 1 and 2 were indeed caused by US-attribute-reversals and the hypothesized CS-US association.

The most striking aspect is the influence of information about a completely new US attribute provided after conditioning on CS assessment. USs’ musicality affected CSs musicality assessments although CSs and musical USs were not paired again. This provides support for the model depicted in Figure 1’s left half. Only if CSs and USs are associated, additional US information provided after conditioning can influence CS assessment. In addition, this differentiated pattern also shows that two distinct attributes were conditioned instead of mere positive valence (see also Förderer & Unkelbach, 2014).

The effect of USs’ new attribute on CSs’ musicality was found on direct ratings but not in SMP data. In hindsight, this null effect is most likely due to procedural details. First, asking participants to decide whether Kanji represent musical instruments or furniture is not optimal; the categories “musical instruments” versus “furniture” are one step farther removed from the to-be-measured construct (i.e., CS musicality) than the categories athletic and non-athletic. Second, we introduced only musical US, but not “unmusical” US—that is, the contrast category was only neutral, which diminishes effects in binary forced choice tasks such as the SMP.

A more problematic explanation for this null effect would again be that participants strategically controlled their ratings to comply with experimental demands, which is more difficult in the SMP. Yet, participants’ responses in the funneled questions make such an explanation highly unlikely. Thus, we are confident to conclude that we found AC effects for athleticism and musicality independent of demand effects.

General Discussion

The current study investigated whether CSs are associated with paired USs and thereby indirectly with US attributes (i.e., via CS-US-attribute structures), or whether CSs are only directly associated with the conditioned attribute (cf. Figure 1). To do this, we reversed existing and introduced new US attributes post-conditioning and examined the influence of these changes on CS assessment.

On an effect level, US-attribute-reversal affected CS assessment on the originally conditioned attribute (i.e., athleticism) and on a new and unrelated attribute (i.e., musicality). When a US attribute changed after CS-US pairings, CS-attribute-assessments changed, too. We found standard AC effects on athleticism after conditioning, whereas these effects changed but did not reverse after US-attribute-reversal (Experiments 1 and 2). Assessment consolidation did not play any role; that is, US-attribute-reversal effects emerged independent of participants assessing CSs athleticism after conditioning or not.

Furthermore, when information about another US attribute was provided, CS assessment of this new attribute changed, too. We found AC effects on direct musicality ratings although CSs were not paired with now musical USs (Experiment 3). Importantly, CS athleticism-assessments did not change due to changes in musicality; that is, we found AC effects for athleticism before and after introducing US musicality. Thereby, we showed that CS changes in Experiments 1 and 2 were not a result of effects unrelated to AC effects (e.g., diminishment over time, regression-to-the-mean). Otherwise, we would not have found AC effects for athleticism in Experiment 3 as introducing musicality took as long as the reversal phase used in Experiments 1 and 2. Furthermore, Experiment 3 provided evidence for the specificity of AC effects beyond general positive or negative evaluations (see also Förderer & Unkelbach, 2014); that is, if attribute ratings would merely reflect participants general positive (or negative) evaluation of a given CS, this CS should have high athleticism and musicality ratings, as both are positive.

Finally, showing that US-attribute-reversal influences CS assessment implies that US attributes (e.g., initial attribute changes or information about new attributes) influence CS assessments even after CS-US pairings, as long as CS and US have once been paired (Förderer & Unkelbach, 2015). However, we would not assume that all USs attributes are equal in their influence on CSs-attribute-assessments. As Förderer and Unkelbach (2014) showed, attributes’ accessibility (e.g., manipulated by priming, context, or attributes’ dominance) determines which attributes are conditioned and influence CS assessments.

On this effect level, the experiments also addressed participants demand awareness and demand compliance, which is a prominent topic within AC as well as related EC research (see Stahl et al., 2009; Sweldens et al., 2014). The current study’s design and procedures as well as the hypotheses were highly complex. It would have been challenging for participants to guess the correct hypotheses about AC effects due to initial pairings and US-attribute-reversal affecting CS assessment. The funneled demand awareness and Experiment 2 showed interactions with the relevant effects based of the most funneled question (Q3.1); however, in line with Bar-Anan and colleagues (2010), this might simply be due to the fact that participants who show the predicted effect also have introspective access to the reported change, and not that the introspective access causes the effect due to participants compliance with experimental demands. Experiment 3 then found no systematic influences and when asking for demand compliance and very few participants reported strategically controlling their ratings at all. Given the combined data of Experiments 2 and 3, the current study’s effects seem to be independent of demand awareness and demand compliance.

On a theoretical level, the present data shows the existence of a CS-US association (cf. Figure 1). This sets AC apart from STI (Uleman, 1987) and Transference (STT; Skowronski et al., 1998). STT effects occur when perceivers transfer an attribute to a communicator who ascribes an attribute to another person (Brown & Bassili, 2002; Carlston & Skowronski, 2005; Crawford et al., 2007). The general notion for STTs and STIs is that targets are associated with traits, or more generally, attributes. Conversely, the current study shows associations between targets (here, CSs) and the communicators (here, USs).

However, the theoretical conclusions have two caveats. First, the present data only show the existence of CS-US associations, but they do not preclude the existence of direct CS-attribute associations. One might propose a third model in addition to the two models in Figure 1 in which the CS is associated with both the US and the attribute. If one assumes additive effects for CS assessments, such a model would be congruent with the data that shows no full CS assessment reversals, but only a negation of the original effect. That is, the CS-attribute (athletic) association and the CS-US (now with a reversed attribute: non-athletic) association might cancel each other.

Second, the proposed indirect CS-attribute structure might rest on the procedural specifics of the paradigm. All three experiments presented the CS followed by the US, akin to trace-conditioning in classic learning paradigms. For EC, Sweldens, Van Osselaer, and Janiszewski (2010) showed that changing US valence post-conditioning (i.e., US-revaluation) only influenced CS evaluation if one specific CS was paired with one specific US, but not if a CS was paired with many USs of the same valence (see also Stahl & Unkelbach, 2009). They assumed that one-to-one pairings lead to CS-US associations. In the present case, this would imply indirect CS-attribute associations. One-to-many pairings might lead to valence misattribution and thereby, in the present case, to direct CS-attribute associations. Furthermore, Hütter and Sweldens (2013) showed that misattribution in EC only occurs when CS and US are presented simultaneously, but not when they are presented sequentially. Such procedural differentiations would also fit with the suggestion by Förderer and Unkelbach (2014) that CSs are associated with what is accessible or salient. In the sequential one-to-one case, the US might be the salient stimulus. One-to-many conditioning situation might make the common attribute of the USs salient, thereby leading to direct CS-attribute associations.

Finally, we have argued with associations, a term that implies associative models of learning in terms of memory links between mental representations (see Gawronski & Bodenhausen, 2006). However, the current conception is largely compatible with propositional models of learning (De Houwer, 2009; Shanks, 2007, 2010). Whether one conceptualizes Figure 1’s links as mere associations or propositional cognitions does not fundamentally change the predictions or the interpretations of the present data. Different predictions emerge only if one manipulates the semantic implications of these links (e.g., “is similar to” vs. “is different from”; see Fiedler & Unkelbach, 2011; Förderer & Unkelbach, 2012), which we did not do in the present study.

These caveats may indicate that AC effects are due to multiple processes; it is up to further research to show whether propositional or “mere” associative effects underlie the observed CS change post-conditioning, whether there are direct CS-attribute associations, or whether even non-associative processes contribute to AC effects. Nevertheless, the present data clearly suggest indirect CS-attribute associations, setting AC effects apart from STT effects and suggesting a novel mechanism underlying impression formation and how people come to ascribe attributes to stimuli in general.

Conclusion

The current AC study shows that if US attributes change post-conditioning (i.e., US-attribute-reversal), participants’ assessments of CSs change as well. This provides evidence for indirect CS-attribute associations, setting AC apart from effects of spontaneous trait transference. In addition, the data support the specificity of AC effects (i.e., musicality and athleticism) beyond mere evaluative effects. The possible nature of these CS-US attribute links as well as the possibility of other processes underlying AC provides a fertile ground for further research.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The present research was funded by an Emmy Noether grant from the Deutsche Forschungsgemeinschaft (UN 273/1-1) awarded to the second author.

Supplemental Material

The online supplemental material is available at

Notes

References

Baeyens

Eelen

Van den Bergh

Crombez

(1992). The content of learning in human evaluative conditioning: Acquired valence is sensitive to US-revaluation. Learning and Motivation, 23, 200-224.

Bar-Anan

De Houwer

Nosek

B. A.

(2010). Evaluative conditioning and conscious knowledge of contingencies: A correlational investigation with large samples. The Quarterly Journal of Experimental Psychology, 63, 2313-2335.

Bar-Anan

Nosek

B. A.

(2012). Reporting intentional rating of the primes predicts priming effects in the affective misattribution procedure. Personality and Social Psychology Bulletin, 38, 1194-1208.

Bazińska

Wojciszke

(1996). Drawing inferences on moral and competence-related traits from the same behavioural information. Polish Psychological Bulletin, 27, 293-299.

Brown

R. D.

Bassili

J. N.

(2002). Spontaneous trait associations and the case of the superstitious banana. Journal of Experimental Social Psychology, 38, 87-92.

Carlston

D. E.

Skowronski

J. J.

(2005). Linking versus thinking: Evidence for the different associative and attributional bases of spontaneous trait transference and spontaneous trait inference. Journal of Personality and Social Psychology, 89, 884-898.

Crawford

M. T.

Skowronski

J. J.

Stiff

Scherer

C. R.

(2007). Interfering with inferential, but not associative, processes underlying spontaneous trait inference. Personality and Social Psychology Bulletin, 33, 677-690.

De Houwer

. (2009). The propositional approach to associative learning as an alternative for association formation models. Learning & Behavior, 37, 1-20.

Festinger

(1957). A theory of cognitive dissonance. Stanford, CA: Stanford University Press.

10.

Fiedler

Unkelbach

(2011). Evaluative conditioning depends on higher order encoding processes. Cognition and Emotion, 25, 639-656.

11.

Fiedler

Unkelbach

(2014). Regressive judgment: Implications of a universal property of the empirical world. Current Directions in Psychological Science, 23, 361-367.

12.

Förderer

Unkelbach

(2011). Beyond evaluative conditioning! Evidence for transfer of non-evaluative attributes. Social Psychology and Personality Science, 2, 479-486.

13.

Förderer

Unkelbach

(2012). Hating the cute kitten or loving the aggressive pit-bull: EC effects depend on CS-US relations. Cognition and Emotion, 26, 534-540.

14.

Förderer

Unkelbach

(2013). On the stability of evaluative conditioning effects: The role of identity memory, valence memory, and evaluative consolidation. Social Psychology, 44, 380-389.

15.

Förderer

Unkelbach

(2014). The moderating role of attribute accessibility in conditioning multiple specific attributes. European Journal of Social Psychology, 44, 69-81. doi:10.1002/ejsp.1994

16.

Förderer

Unkelbach

(2015). Attribute conditioning: Changing attribute-assessments through mere pairings. Quarterly Journal of Experimental Psychology, 68, 144-164.

17.

Galli

Gorn

(2011). Unconscious transfer of meaning to brands. Journal of Consumer Psychology, 21, 215-225.

18.

Gawronski

Bodenhausen

(2006). Associative and propositional processes in evaluation: An integrative review of implicit and explicit attitude change. Psychological Bulletin, 132, 692-731.

19.

Hastie

Park

(1986). The relationship between memory and judgment depends on whether the judgment task is memory-based or on-line. Psychological Review, 93, 258-268.

20.

Heider

F. F.

(1946). Attitudes and cognitive organization. Journal of Psychology: Interdisciplinary and Applied, 21, 107-112.

21.

Hütter

Sweldens

(2013). Implicit misattribution of evaluative responses: Contingency-unaware evaluative conditioning requires simultaneous stimulus presentations. Journal of Experimental Psychology: General, 142, 638-643.

22.

Hütter

Sweldens

Stahl

Unkelbach

Klauer

K. C.

(2012). Dissociating contingency awareness and conditioned attitudes: Evidence of contingency-unaware evaluative conditioning. Journal of Experimental Psychology: General, 141, 539-557.

23.

Imhoff

Schmidt

A. F.

Bernhardt

Dierksmeier

Banse

(2011). An inkblot for sexual preference: A semantic variant of the affect misattribution procedure. Cognition and Emotion, 25, 676-690.

24.

Kim

Allen

C. T.

Kardes

F. R.

(1996). An investigation of the mediational mechanisms underlying attitudinal conditioning. Journal of Marketing Research, 33, 318-328.

25.

Mackintosh

N. J.

(1983). Conditioning and associative learning. New York, NY: Oxford University Press.

26.

Meersmans

De Houwer

Baeyens

Randell

Eelen

(2005). Beyond evaluative conditioning? Searching for associative transfer of nonevaluative stimulus properties. Cognition and Emotion, 19, 283-306.

27.

Olson

Kendrick

Fazio

(2009). Implicit learning of evaluative vs. Non-evaluative covariations: The role of dimension accessibility. Journal of Experimental Social Psychology, 45, 398-403.

28.

Payne

Brown-Iannuzzi

Burkley

Arbuckle

N. L.

Cooley

Cameron

Lundberg

K. B.

(2013). Intention invention and the affect misattribution procedure: Reply to Bar-Anan and Nosek (2012). Personality and Social Psychology Bulletin, 39, 375-386.

29.

Shanks

D. R.

(2007). Associationism and cognition: Human contingency learning at 25. Quarterly Journal of Experimental Psychology, 60, 291-309.

30.

Shanks

D. R.

(2010). Learning: From association to cognition. Annual Review of Psychology, 61, 273-301.

31.

Skowronski

J. J.

Carlston

D. E.

Mae

Crawford

M. T.

(1998). Spontaneous trait transference: Communicators take on the qualities they describe in others. Journal of Personality and Social Psychology, 74, 837-848.

32.

Stahl

Unkelbach

(2009). Evaluative learning with single versus multiple unconditioned stimuli: The role of contingency awareness. Journal of Experimental Psychology: Animal Behavior Processes, 35, 286-291.

33.

Stahl

Unkelbach

Corneille

(2009). On the respective contributions of awareness of unconditioned stimulus valence and unconditioned stimulus identity in attitude formation through evaluative conditioning. Journal of Personality and Social Psychology, 97, 404-420.

34.

Sweldens

Corneille

Yzerbyt

(2014). The role of awareness in attitude formation through evaluative conditioning. Personality and Social Psychology Review, 18, 187-209.

35.

Sweldens

Van Osselaer

S. J.

Janiszewski

(2010). Evaluative conditioning procedures and the resilience of conditioned brand attitudes. Journal of Consumer Research, 37, 473-489.

36.

Todorov

Said

C. P.

Engel

A. D.

Oosterhof

N. N.

(2008). Understanding evaluation of faces on social dimensions. Trends in Cognitive Sciences, 12, 455-460.

37.

Uleman

J. S.

(1987). Consciousness and control: The case of spontaneous trait inferences. Personality and Social Psychology Bulletin, 13, 337-354.

38.

Walther

Gawronski

Blank

Langer

(2009). Changing likes and dislikes through the backdoor: The US-revaluation effect. Cognition and Emotion, 23, 889-917.

39.

Walther

Nagengast

Trasselli

(2005). Evaluative conditioning in social psychology: Facts and speculations. Cognition and Emotion, 19, 175-196.

40.

Willis

Todorov

(2006). First impressions: Making up your mind after 100 ms exposure to a face. Psychological Science, 17, 592-598.

41.

Winter

Uleman

J. S.

(1984). When are social judgments made? Evidence for the spontaneousness of trait inferences. Journal of Personality and Social Psychology, 47, 237-252.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.18 MB