EVALUATIVE CONDITIONING IS INSENSITIVE TO BLOCKING

————— Tom Beckers is affiliated to the Department of Psychology, University of Amsterdam, the Netherlands; Pascale de Vicq and Frank Baeyens are affiliated to the Department of Psychology, Katholieke Universiteit Leuven. This research was supported by a grant (GOA/2007/03) from the Research Council of the Katholieke Universiteit Leuven (FB). Correspondence concerning this article should be addressed to Tom Beckers, Department of Psychology, University of Amsterdam, Roetersstraat 15, 1018WB Amsterdam, the Netherlands. E-mail: tom.beckers@uva.nl EVALUATIVE CONDITIONING IS INSENSITIVE TO BLOCKING

One way in which those preferences may be acquired is through evaluative conditioning (EC; for a review, see De Houwer, Thomas, & Baeyens, 2001).Evaluative conditioning refers to the transfer of positive or negative valence to an initially neutral stimulus through associative learning.At a procedural level, evaluative conditioning is a form of Pavlovian conditioning: A neutral stimulus (the conditioned stimulus or CS) is repeatedly paired with a stimulus that somehow has biological or emotional significance (the unconditioned stimulus or US), and changes in the response to the CS are measured (e.g., through valence ratings or affective priming).However, at a functional level, a considerable amount of research suggests that EC has a number of characteristics that set it apart from other forms of Pavlovian conditioning (Baeyens, Vansteenwegen, Hermans, & Eelen, 2001a;Dwyer, Jarratt, & Dick, 2007; see Lipp & Purkis, 2005, for a review of work that suggests that EC does not have those distinct functional characteristics in a shock-conditioning procedure).One of those characteristics is the remarkable insensitivity of EC to extinction: once a like or a dislike for a CS has been established by repeatedly pairing that CS with a liked or disliked US, unreinforced presentations of the CS (that is, presentations of the CS without the US) have limited effect if any on the evaluative value of the CS (e.g., Baeyens, Eelen, & Crombez, 1995;De Houwer, Baeyens, Vansteenwegen, & Eelen, 2000).More generally, EC appears to be influenced by the number of CS-US pairings (i.e., contiguity), rather than the degree of contingency between the CS and the US, unlike other forms of Pavlovian conditioning (Baeyens, Hermans, & Eelen, 1993).EC also appears to be independent of awareness of the CS-US contingency (e.g., Baeyens, Eelen, Van den Bergh, & Crombez, 1990;De Houwer, Baeyens, & Eelen, 1994;Dickinson & Brown, 2007;Hammerl & Grabitz, 2000; but see Pleyers, Corneille, Luminet, & Yzerbyt, 2007), whereas contingency awareness seems to be crucial for other forms of Pavlovian conditioning to occur in humans (Lovibond & Shanks, 2002).Finally, EC seems not susceptible to modulation: if a CS is repeatedly paired with a liked or disliked US when it is accompanied by another stimulus (a feature or occasion setter), but not when presented in the absence of that stimulus, the CS will afterwards elicit a positive or negative evaluation irrespective of the presence or absence of the feature (Baeyens, Crombez, De Houwer, & Eelen, 1996;Baeyens, Hendrickx, Crombez, & Hermans, 1998).
A possible way to theoretically integrate these functional characteristics of EC is to consider EC as the behavioural expression of a very simple learning mechanism (Baeyens, Vansteenwegen, Hermans, & Eelen, 2001b;De Houwer et al., 2001).According to this hypothesis, during acquisition an association between CS and US representations is built up following a rudimentary Hebbian learning algorithm, the parameters of which include only temporal contiguity and stimulus salience.In a Hebbian learning device, the synchronous activation of CS and US representations results in an increase of associative strength, whereas the activation of the representation of the CS alone (or of the US alone) is ineffective in changing associative strength.In other words, according to a Hebbian learning rule the strength of an association between two stimuli increases whenever the stimuli co-occur, but remains unchanged when one of the two stimuli is presented in isolation.By consequence, EC does not extinguish, is not sensitive to CS-US contingency manipulations, and is not sensitive to conditionality of CS-US contingencies: Each of these phenomena precisely requires what is absent in a Hebbian rule, namely the causal efficacy of nonreinforced CS-only presentations.
As noted by De Houwer et al. (2001) in their comprehensive review of research on EC, despite the considerable amount of research on the functional properties and underlying processes of EC, there are currently hardly any data on the susceptibility of EC to cue competition.Cue competition in Pavlovian conditioning refers to the observation that the acquisition of a conditioned response to a CS is affected by the presence, during the CS-US pairings, of other stimuli that have an association with the US.The most prominent example of cue competition is the blocking effect (Kamin, 1969): blocking is said to occur when the conditioned response to a target CS X that is paired with the US in compound with a competing CS A is reduced because of separate pairings of CS A and the US (i.e., A→US training followed or preceded by AX→US training).In the context of EC, blocking would imply that a neutral stimulus X that is repeatedly paired with a liked (/disliked) outcome in compound with another stimulus A, would elicit less of an evaluative response or even remain neutral if A is also repeatedly paired with the liked (/disliked) outcome by itself.The lack of data is all the more remarkable, given that findings about cue competition (and most notably blocking) have had a major impact on theories of Pavlovian conditioning (e.g., Rescorla & Wagner, 1972), and that the 'Hebbian learning' account of EC implies that it should not be subject to blocking.Indeed, if mere cooccurrence of CS and US is the driving force behind EC, then the presence of another stimulus, whatever associative history that stimulus has with the US, should not have any impact on the degree of evaluative learning about the CS.Therefore, a lack of blocking would support the Hebbian account, whereas the observation of blocking would call for its qualification.Accordingly, De Houwer et al. (2001, p. 866) argue that "it is likely that data on cue competition in EC will have a major impact on theories of EC".
As noted, we know of no studies that have assessed the occurrence of blocking in EC.However, two studies have addressed the occurrence of other forms of cue competition, with mixed results.In two shock-conditioning experiments, Lipp, Neumann, and Mason (2001) compared evaluative ratings for a blocked cue (i.e., ratings for X after A→US training followed by AX→US training) to ratings for a protection-from-overshadowing cue (i.e., ratings for X after presentations of A without shock followed by AX→ US training).In one of both experiments, they obtained some evidence for a difference in evaluation between both cues, suggesting sensitivity of EC to cue competition.However, it is unclear whether EC results obtained in shock-conditioning procedures are applicable to EC in general (see above).In a more typical picture-picture evaluative conditioning procedure, Dwyer et al. (2007) assessed the sensitivity of EC to overshadowing, by comparing evaluative ratings for cues that had been paired with the US in isolation (i.e., ratings for X after X→US training) to ratings for cues that had been paired with the US in compound with another cue (i.e., ratings for X after AX→US training).Simultaneously pairing two CSs with the same US did not reduce the evaluative conditioning effect.Finally, a recent study by Dickinson and Brown (2007) is relevant here as well.The authors had participants drink solutions that contained either sugar or Tween (polysorbate20, an unpalatable but harmless substance with a soap-like taste).Each drink had a particular flavour and colour.They observed that flavours that were paired with sugar were afterwards rated more positively than flavours that were paired with Tween, even though participants could not reliably identify which flavours had been presented in a sugar solution and which ones had been paired with Tween during training.Moreover, the EC of the flavours was not influenced by whether or not the colours alone had been pretrained to predict the presence of sugar or Tween in the solutions.The authors do not interpret this as evidence that EC would be insensitive to blocking though, because the pretraining did not result in EC of the colours.Accordingly, their results do not illuminate whether the acquisition of an affective reaction to one stimulus A, through pairing of A with an affective outcome, would reduce the subsequent acquisition of an affective reaction to stimulus X if the AX compound is paired with that outcome.Rather, their results indicate that EC (which develops more readily for flavours than for colours when using a gustatory US) is independent of contingency knowledge (which develops more readily for colours than for flavours in this kind of paradigm).
In the present study, we aimed to investigate to what degree EC would be susceptible to blocking.A first crucial step was to identify an EC task that would yield a robust EC effect on the one hand, and allow the implementation of a blocking procedure on the other hand.The most robust EC effects have generally been obtained with the flavour-taste paradigm.However, the presentation of two flavours in an AX compound (as is necessary in a blocking procedure) would likely be perceived as one single, new flavour, instead of as a combination of a previously presented flavour A and a new flavour X.Another widely used EC paradigm, the picture-picture paradigm, in which neutral pictures are paired with positive or negative pictures, would be more easily adapted to a blocking procedure.However, the robustness of that procedure in yielding EC effects has been variable (De Houwer et al., 2001), as is the case for a number of other EC procedures that have been introduced over the years.Instead, we used a candy task that was recently developed in our lab and appeared to produce strong and consistent EC effects.In this task, which is administered in small groups, children are presented with a fortune wheel, on which various strings of symbols are depicted.The chil-dren turn the wheel in turn, and depending on the symbol string that the needle of the wheel ends at, they gain or lose a specified number of candies.Earlier studies have pointed out that after the game, children exhibit a preference for symbol strings that were associated with a gain of candies over strings that were associated with a loss of candies (Crabbé & Baeyens, 2009).In this candy game, we implemented a blocking contingency.In one group of children (blocking group; n = 36), after a first phase in which some strings were followed by gain (denoted as A+ trials) and others by losses (denoted as B-trials), in a second phase of the game additional symbol strings were presented along with the original strings on the fortune wheel, maintaining the same contingencies between symbol strings and gains or losses as before (i.e., AX+ and BY-trials; see Table 1).In a second group of children (control group; n = 39), the same second training phase was preceded by a first phase in which different strings were used (i.e., C+ and D-trials in the first phase, followed by AX+ and BY-trials in the second phase; see Table 1).Afterwards, we collected evaluative ratings for the various symbol strings and also administered a forced choice test.Blocking would be evident if X and Y were not evaluated differently in the blocking group, whereas they would be in the control group, or if the difference in evaluation and preference for X and Y would at least be markedly smaller in the former than in the latter group.

Participants
Seventy-five children (13 girls, 62 boys) between 9 and 11 years of age participated in the experiment.They were recruited from four classes of a local school in the vicinity of Antwerp.The class teachers accompanied the children and acted as supervisors during the experiment.Written informed consent was obtained from the children's parents before the start of the study; the children gave oral consent right before participating.

Stimuli and materials
The to-be-conditioned symbols strings were depicted on wooden fortune wheels (see Figure 1).The children were also supplied with 'translation sheets' on which the consequence that would follow each of the symbol strings (degree of loss or gain of candy; for examples see Figure 2) was indicated.Before the start of the experiment, each participant received a sheet with instructions ('rules of the game'; see below).After the fortune game, the children received two booklets with instructions and rating questions (see below).

Procedure
The experiment was run in two groups, in the classroom, on a regular school day.Children were randomly divided in groups of four to five.The children were told that they participated in a pilot study on the development of a new game that might be brought on the market in the future.They then received a sheet with instructions, along with the translation sheets.
Figure 1 Example of a game wheel used in the compound phase The instructions read (translated from Dutch): "You have been divided in groups.The goal of the game is to collect as many candies for your group as possible.The number of candies collected by the end of the game will be divided among the group.Each group receives a number of candies to start with.Each of you will turn the wheel in turn.By turning the wheel, you will gain or lose candies.You have received a sheet on which is indicated how many candies you gain or lose for particular symbol strings.When you have turned the wheel, you have to tell the supervisor how many candies you have won or lost.She will then give or take that number of candies.Pay close attention!You all have to be attentive to transmit the correct number of candies won or lost to the supervisor.If you make a mistake, you will have to return four candies.Therefore, it is important that you not only pay attention when it is your turn, but also when it is someone else's turn.After a number of turns, you will proceed to the next phase, which will be indicated by the supervisor.This phase is identical to the first phase, but now there will be more symbols on the wheel, and a new sheet.When the supervisor indicates that the game is over, you will have to fill out a few question booklets.If you have any questions, you can ask them now or in the course of the game".
For the blocking group (n = 36), in the first phase (elemental phase) of the game, one set of strings of three symbols on the wheel was associated with a gain of either one, two or three candies, depending on the order of the symbols (A+ strings).Another set of strings led to the loss of one, two, or three candies (B-strings).The assignment of symbol strings to gains or losses was counterbalanced across participant groups.In the second phase (compound phase), each of the strings of symbols from the first phase was supplemented by three additional symbols, and accompanied by the same outcome as during the first phase (AX+ and BY-strings).Each phase comprised 40 acquisition trials, resulting in a total of 80 acquisition trials.The control group received the same training as the blocking group, apart from the fact that different symbol strings were used for the elemental phase (C+ and D-strings; see Table 1 and Figure 2).For both the blocking and the control group, X and Y strings were presented to the right of the A and B strings on the game wheel during the compound phase.
At the end of the game, the translation sheets were collected.The children were then given the question booklets, with the instruction to answer the questions individually and in silence.They were told that the questions would not be graded and that their teacher would not get to see their answers.
The first booklet contained pairs of symbol strings, with the instruction to circle the string in each pair that the child liked best (forced choice task).For the blocking group, all the strings that during acquisition had resulted in a gain or loss of two or three candies were included in the forced choice task, as well as four new, supposedly neutral strings (making for 12 strings in total; to limit the number of items in the forced choice task, the strings resulting in a gain or loss of one candy were not included).For the control group, all the strings that had resulted in a gain or loss of two or three candies were included (making for 12 strings as well).Each symbol string was paired with every other symbol string, with the exclusion of pairs consisting of strings of the same intended valence.Each pair was presented once in the booklet, in the same randomized order for all participants, resulting in 48 choice pairs in the blocking group and 32 pairs in the control group.Pairs were presented on separate pages; the position of the strings in a pair was determined by chance.
The second booklet contained all 18 strings (i.e., also including the strings that had resulted in a gain or loss of one candy), with the instruction to rate each string on a scale from 0 (very ugly) to 10 (very pretty) (evaluation task).Each string was presented on a separate page, in the same randomized order for all participants.
For both tasks, participants were instructed to respond according to their first impression, without deliberating.For exploratory reasons, participants were asked to indicate why they thought they liked some symbol strings better than others at the end of each booklet.

Results
For the forced choice task, we counted how often participants selected the string that had yielded a gain in the candy game over the string that had yielded a loss, among the four pairs consisting of an A and a B string on the one hand and among the four pairs consisting of an X and a Y string on the other hand (resulting in a score between 0 and 4 for both A and X).We then subtracted 2 from these scores to correct for chance performance, such that a positive score reflects a preference for A strings over B strings or for X strings over Y strings, respectively, whereas a negative score reflects a preference for B strings over A strings or Y strings over X strings.Figure 3 presents the resulting preference scores for each group.Participants apparently preferred the strings that had resulted in a gain over strings that had resulted in a loss, and more so for A strings versus B strings than for X strings versus Y strings.Importantly, this pattern was roughly the same for both conditions.A 2 x 2 split-plot factorial ANOVA, with Group (Blocking, Control) as a between-subjects factor and String (A, X) as a within-subjects factor, confirmed that impression.The ANOVA yielded a main effect of string, F(1, 73) = 7.60, MSE = 1.05, p = .007.Both effects involving group were non-significant, F(1, 73) < 1. One-sample t-tests confirmed that the preference scores for A strings and X strings were both greater than 0, t(75) = 19.15,and t(75) = 14.68, respectively.Positive preference scores were obtained both in the blocking group, t(36) = 12.22 for A strings and t(36) = 10.20 for X strings, and in the control group, t(39) = 14.82 for A strings and t(39) = 10.43 for X strings (all ps < .001).
Figure 4 presents the ratings from the evaluation task, for each type of string and group.We first analysed the ratings for the strings presented in the elemental phase (i.e., the ratings for the A and B strings from the blocking group and the ratings for the C and D strings from the control group), to confirm that our procedure was successful at inducing evaluative likes and dislikes for strings of patterns paired with gains and losses of candy.Strings that had consistently resulted in a gain of candy appeared to be evaluated more positively than strings that had resulted in losses, irrespective of the degree of gain or loss (one, two, or three candies).A 2 x 2 x 3 split-plot factorial ANOVA with Group (Blocking, Control) as a between-subjects factor and Outcome (Gain, Loss) and Degree (1, 2, 3) as within-subjects factors confirmed this observation.The ANOVA yielded a main effect of Outcome, F(1, 73) = 99.20,MSE = 10.66,p < .001.None of the other main effects or interactions came close to a conventional level of significance, largest F(1, 73) = 1.56, p = .22. Figure 5 depicts the ratings for the X and Y strings from the compound phase.It appears that only the type of outcome that followed a string during the candy game affected evaluative ratings: strings that were associated with a gain were rated more positively than strings that were associated with a loss of candy.Remarkably, the ratings seemed unaffected by the training history of the strings that X and Y strings were paired with during the candy game; if anything, the evaluative conditioning effect seemed slightly larger in the blocking group than in the control group.A 2 x 2 x 3 ANOVA on the ratings, with Group (Blocking, Control) as between-subjects factor and Outcome (Gain, Loss) and Degree (1, 2, 3) as within-subjects factors confirmed these observations.The ANOVA resulted in a main effect of outcome, F(1, 73) = 46.62,MSE = 10.89,p < .001.All other effects were non-significant, F(2, 146) = 2.26, p = .11for the main effect of degree, largest F(1, 73) = 1.38, p = .26for the remaining effects.Planned comparisons revealed that the effect of outcome was highly reliable in both the blocking group, F(1, 73) = 30.77,MSE = 10.89,p < .001,and the control group, F(1, 73) = 16.66,MSE = 10.89,p < .001.Additionally, ratings for either X or Y strings did not differ between the blocking and the control group, F(1, 73) < 1 and F(1, 73) = 1.99,MSE = 10.41,p = .16,respectively.
One additional observation concerning the evaluative ratings deserves further attention.From the comparison between Figures 4 and 5 it appears that the evaluative conditioning effects are more outspoken for the elementally trained strings (A and B in the blocking group; C and D in the control group) than for the X and Y strings from the compound phase.A 2 x 2 x 2 x 3 ANOVA with Group (Blocking, Control) as a between-subjects factor and Phase (Elemental, Compound), Outcome (Gain, Loss) and Degree (1, 2, 3) as within-subjects factors on the ratings for strings A and B (for the blocking group), C and D (for the control group), and X and Y (for both groups) indeed yielded a significant outcome by phase interaction, F(1, 73) = 9.65, MSE = 5.17, p = .003.Phase was however not involved in any interaction with group, all Fs < 1.In other words, the somewhat weaker evaluative conditioning effect for X versus Y strings compared to the conditioning effect observed for the elementally trained strings was not linked in any way to the difference in learning history of the compounded strings between the blocking group and the control group.

Discussion
The candy game was clearly successful in producing an EC effect.Overall, the participants had a clear preference for symbol strings associated with a gain of candy over symbol strings associated with a loss, and the former strings were rated more positively than the latter.That finding corroborates the present procedure as a robust and reliable procedure for studying EC.
More importantly, the data indicate that EC is not susceptible to blocking.The degree of preference for target strings presented in a compound that resulted in gains over target strings presented in a compound that resulted in losses was not reliably different when the competitor string in the compound had previously acquired affective value through pairings of that competitor on its own with the same outcome from when the competitor string had not been trained before.Likewise, the difference in evaluative ratings between target strings followed by gains and target strings followed by losses during compound training was not affected by whether or not their competitors had been paired with the same gains or losses before.Notice that this result is unlikely to be due to a lack of sensitivity.On an overall measure of EC for the target strings, calculated by subtracting the mean evaluation of Y strings from the mean evaluation of X strings, the mean score in the blocking group is 2.49, with a 95% confidence interval ranging from 1.57 to 3.41.Given a mean of 1.76 for the control group, the true EC score can be assumed to be at most .19smaller in the blocking group than in the control group, which would arguably still amount to a negligible degree of blocking.In fact, if anything the EC effect for the target cues seems slightly larger, rather than smaller, in the blocking group.It is also unlikely that the lack of blocking in our EC procedure is related to the age of the participants.Previous research has demonstrated that cue competition readily occurs in associative learn-ing tasks presented to children as young as 3 years of age, and that blocking occurs in such tasks under similar circumstances in children as in adults (Beckers, Van den Broeck, Renne, Vandorpe, De Houwer, & Eelen, 2005;Beckers, Vandorpe, Debeys, & De Houwer, 2009).
Before we discuss the finding that EC is insensitive to blocking in more detail, another observation deserves some attention.Overall, EC effects for the X and Y strings, although equal in size for the blocking and for the control group (hence, no blocking effect), were smaller than the EC effects observed for the elementally trained strings from the first phase.There are at least two possible explanations for that observation.One reason may be superior learning of initial over later information (i.e., some kind of proactive interference, be it because of increasing fatigue, distraction, processing limitations, or some other factor).Alternatively, or additionally, participants may have been paying somewhat less attention to the X and Y symbol strings than to the A and B strings during the compound phase, due to the fact that the A and B strings were always presented as the left part of a compound and the X and Y strings as the right part.As such, X and Y strings were presented at the end of a string compound in terms of a left-to-right reading direction.That arrangement may have limited the amount of attention paid to those strings by some of the participants; evidence suggests that although a lack of contingency awareness does not disrupt EC, a lack of attention to the stimuli that are presented does (Corneille, Yzerbyt, Pleyers, & Mussweiler, 2009;Field & Moore, 2005).Whatever the explanation, it does not diminish the fact that the difference in training history of the A and B competitor strings between the blocking and the control group had no effect whatsoever on the evaluations and preferences acquired for the X and Y strings, amounting to a complete lack of blocking.
The observation that EC is not susceptible to blocking fits well with the idea that EC reflects the operation of a very simple, Hebbian learning system.According to a Hebbian logic, co-occurrence of CS and US is the sole factor that determines the strength of association between the CS and US (and therefore the strength of the evaluative response elicited by the CS).As such, the question whether or not the CS is redundant in terms of its signal value for the US (a crucial determinant of learning in traditional models of Pavlovian conditioning, e.g., Rescorla & Wagner, 1972) is of no impact on the degree of learning.The observation that EC is not sensitive to blocking therefore supports the contention that the Hebbian account of EC provides a comprehensive and unifying framework for the various functional characteristics of EC.
Note that the lack of blocking in EC is also consistent with recent propositional accounts of EC (De Houwer, 2007) and associative learning in humans in general (De Houwer, 2009).According to a propositional account, human performance in associative learning tasks reflects their truth-evaluation of propositions about the relation between CS and US.This approach assumes that knowledge is represented in a structured, content-rich manner that does not only specify which representations are related, but also how they are related.As such, it does away with the traditional concept of associations as unqualified, content-free links between representations that carry merely statistical information altogether.According to this approach, for different types of reactions to a CS (e.g., preparatory reactions versus evaluations), different propositions are relevant which sample the knowledge available in memory in different ways.As such, the same information structures present in memory may yield different functional characteristics for different kinds of conditioned responses.Recent research suggests that particularly blocking and other forms of cue competition are not the solid phenomena that their central, hard-wired status in contemporary associative learning theories would suggest, neither in causal learning (e.g., Beckers, De Houwer, Pineño, & Miller, 2005), nor in human electrodermal conditioning (Mitchell & Lovibond, 2002); rather, the available evidence suggests that their occurrence crucially depends on the appropriateness and possibility of truth-evaluating propositions that will yield cue competition (De Houwer, 2009;De Houwer, Beckers, & Vandorpe, 2005).Remarkably, even in Pavlovian conditioning in animals, the occurrence of blocking seems sensitive to manipulations that affect the propositions that the animals should truth-evaluate (Beckers, Miller, De Houwer, & Urushihara, 2006).Therefore, if one assumes that, unlike other conditioned responses, evaluations and preferences rely on the truth-evaluation of propositions of which the truth does not depend on predictive redundancy, the lack of blocking that we observed is compatible with this general framework.
As such, the observation that EC is not susceptible to blocking may be compatible with both a Hebbian account of EC, as with a propositional account of human associative learning in general.Most importantly, however, the present findings help to fill a long-standing empirical gap in our knowledge concerning the functional properties of EC.

Figure 2
Figure 2 Examples of translation sheets used in the elemental phase (panel A: translation sheet for the blocking group; panel B: translation sheet for the control group) and in the compound phase (panel C), corresponding to the game wheel in Figure 1 (adapted from Dutch) Figure 3 Preference scores for rewarded strings, by group.Error bars represent standard errors of the means

Figure 4
Figure 4Evaluative ratings for strings presented in the elemental phase, by group and outcome.Error bars represent standard errors of the means

Figure 5 Figure 5
Figure 5Evaluative ratings for the X and Y strings from the compound phase, by group and outcome.Error bars represent standard errors of the means

Table 1
Design of the experiment Note.A, B, C, D, X, Y each represent three different symbol strings; + and -represent a gain or loss of one, two, or three candies, respectively; / indicates interspersed trials.