Contingency Learning 1 Running head: CONTINGENCY LEARNING AND RESPONSE PREDICTION The Stroop Effect: Why Proportion Congruent has Nothing to do with Congruency and Everything to do with Contingency

The item-specific proportion congruent (ISPC) effect is the observation that the Stroop effect is larger for words that are presented mostly in congruent colours (e.g., BLUE presented 75% of the time in blue), and smaller for words that are presented mostly in a given incongruent colour (e.g., YELLOW presented 75% of the time in orange). One account of the ISPC effect, the modulation hypothesis, is that participants modulate attention based on the identity of the word (i.e., participants allow the word to influence responding when it is presented mostly in its congruent colour). Another account, the contingency hypothesis, is that participants use the word to predict the response that they will need to make (e.g., if the word is YELLOW, then the response is probably orange). Reanalyses of data from Jacoby, Lindsay, and Hessels (2003) along with results from new experiments are inconsistent with the modulation hypothesis, but entirely consistent with the contingency hypothesis. A response threshold mechanism for using contingency information is proposed and tested. Contingency Learning 3 The Stroop Effect: Why Proportion Congruent has Nothing to do with Congruency and Everything to do with Contingency Contingency learning is a lively area of research (e.g., Jacoby, Lindsay, & Hessels, 2003; Musen & Squire, 1993; Schmidt, Crump, Cheesman, & Besner, 2007). One popular domain for this research is the Stroop paradigm. The standard Stroop effect is the finding that participants take longer to identify the print colour of an incongruent colour word (e.g., the word GREEN printed in red; GREENred) relative to a congruent colour word (REDred; Stroop, 1935; see MacLeod, 1991, for a review). The magnitude of the Stroop effect changes when the proportion of congruent items is manipulated (Lowe & Mitterer, 1982). In particular, the Stroop effect increases as the proportion on congruent trials increases. The standard account of the influence of proportion congruent (Cheesman & Merikle, 1986; Lindsay & Jacoby, 1994; Lowe & Mitterer, 1982) is that the detection of these proportions allows participants to modulate attention to the word, thereby changing the size of the Stroop effect. However, Schmidt et al. (2007) suggest that simple contingency learning provides a sufficient account of the proportion congruency effect. Here, we provide a reanalysis of the Jacoby et al. (2003) data and report new results from our own laboratory that are consistent with the Schmidt et al. (2007) suggestion and at the same time are problematic for the modulation hypothesis. Possible mechanisms by which participants use contingency information to control responding will also be considered. Proportion Congruency and the Modulation Hypothesis A number of experiments (Cheesman & Merikle, 1986; Lindsay & Jacoby, 1994; Lowe & Mitterer, 1982) have shown that the magnitude of the Stroop effect can be modulated by varying the proportion of congruent trials. Specifically, the Stroop effect is larger when most of the items in the experiment are congruent (high proportion congruent) than when most of the Contingency Learning 4 items are incongruent (low proportion congruent). The standard explanation of this effect (Cheesman & Merikle, 1986; Lindsay & Jacoby, 1994; Lowe & Mitterer, 1982), here termed the modulation hypothesis, is that participants strategically modulate attention to the word depending on the proportion of congruent items. For instance, when the word and colour match most of the time (high proportion congruent), participants attend to the word more than usual. This will speed up responses on congruent trials (increased facilitation) and slow down responses on incongruent trials (increased interference), making for a larger Stroop effect. In contrast, when the word and the colour mismatch most of the time (low proportion congruent), participants make a greater effort to ignore the word. This decreases both facilitation from congruent words and interference from incongruent words, making for a smaller Stroop effect. Thus, the modulation account holds that participants use information about proportion congruency to decide the degree to which they will attend to the word and thus to allow the word to impact performance in colour identification. The modulation hypothesis seems to be the most widely accepted explanation for the proportion congruent effect. For instance, Lowe and Mitterer (1982) conclude that their findings “demonstrate the strategic modulation of selective attention” (p. 698) and Lindsay and Jacoby (1994) state that the effect “suggests that when most items are incongruent, subjects somehow inhibit the influence of word-reading processes, relative to when most items are congruent” (p. 225). Although intuitively appealing, the modulation hypothesis has difficulty explaining some findings. Given the assumption that participants modulate attention to the word, it would be expected that this modulation would be task-wide. That is, words should be ignored throughout a low proportion congruent block and attended to throughout a high proportion congruent block. Contingency Learning 5 However, Jacoby et al. (2003) manipulated proportion congruency for each item (i.e., each colour word) such that some words were presented most often in their congruent colour (e.g., BLUEblue) and other words were presented most often in a particular incongruent colour (e.g., ORANGEyellow). A proportion congruent effect was still observed, even though high and low proportion congruent stimuli were intermixed. Jacoby et al. called this finding the item-specific proportion congruent (ISPC) effect. As Jacoby et al. point out, this finding is difficult to accommodate within the modulation hypothesis framework, because it would have to be assumed that participants are modulating attention to the word on a trial-by-trial basis depending on the identity of the word (e.g., if the word is BLUE, then the word is attended, but if the word is ORANGE, then the word is ignored). In essence, to defend the modulation account it would have to be maintained that participants decide whether to attend to the word after they have already read it. Response Prediction and the Contingency Hypothesis A different account of the ISPC effect, here termed the contingency hypothesis, is that participants implicitly learn contingencies (i.e., correlations) between words and responses and then use said contingencies to predict the specific response associated with each distracting word (Schmidt et al., 2007). For instance, if the word ORANGE is presented most often in yellow, then upon processing the word ORANGE participants will (unconsciously) predict that the correct response will be yellow. This response prediction allows participants to shortcut some processing (thus speeding responding) if the predicted response is indeed the correct one. When a word accurately predicts the correct response (e.g., BLUEblue, where BLUE is presented most often in blue), we call this a high contingency trial. When the word predicts the wrong response (e.g., BLUEgreen), we call this a low contingency trial. When the word does not predict a specific Contingency Learning 6 response (e.g., PINKbrown, where PINK is presented equally often in all colours), we call this a medium contingency trial. It is important to note that we are not the first to point out a role for contingency in the Stroop task. Dishon-Berkovits and Algom (2000; see also, Malara & Algom, 2003; Sabri, Melara, & Algom, 2001) have argued that when words and colours are correlated, participants will pick up on this cue, attend to the content of the word, and use that content to aid responding. For instance, in three experiments using a word-word version of the Stroop task they demonstrated that a Stroop effect is only observed when the correlation between words and colours is non-chance. When the correlation is chance, they claim, participants do not attend to the content of the distracting word. While their account is not as specified as the current account (indeed, their argument could even be construed as a variant of the modulation hypothesis), their work has highlighted the importance of correlational cues on responding. Although the modulation hypothesis has difficulty explaining the ISPC effect, the contingency hypothesis does not. According to the contingency hypothesis, in the high proportion congruent condition responses to congruent trials will be faster than usual, because participants can use the word to successfully predict the response (e.g., BLUEblue; high contingency). The same advantage does not occur for incongruent trials in this condition because, for instance, BLUE does not accurately predict an incongruent green response (low contingency). Thus, by speeding congruent but not incongruent trials, the difference between incongruent and congruent trials (the Stroop effect) will be larger (i.e., relative to a condition where words are not predictive of responses). Similarly, in the low proportion congruent condition responses will be faster for incongruent items, because participants can use the word to successfully predict the response (e.g., ORANGEyellow; high contingency). The same advantage Contingency Learning 7 does not occur for congruent trials in this condition because, for instance, ORANGE predicts a yellow, not an orange response (low contingency). Thus, by speeding incongruent but not congruent trials, the overall Stroop effect will be smaller. According to the contingency hypothesis, then, proportion congruency manipulations are confounded with contingency. Specifically, as demonstrated in Table I, in the high proportion congruent condition the magnitude of the Stroop effect is inflated due to confounding higher word-response contingencies for the congruent (.75) relative to incongruent (.25) items. Similarly, in the low proportion congruent condition the Stroop effect is reduced due to confounding higher word-response contingencies for the incongruent (.75) relative to congruent (.25) items. Fixing this confound is as simple as rearranging the cells in the design. Rather than comparing high contingency congruent trials with low contingency incongruent trials (the high proportion congruent condition), we can compare high contingency congruent trials (from the high proportion congruent condition) with high contingency incongruent trials (from the low proportion congruent condition). Similarly, rather than comparing low contingency congruent trials with high contingency incongruent trials (the low proportion congruent condition), we can compare low contingency congruent trials (from the low proportion congruent condition) with low contingency incongruent trials (from the high proportion congruent condition). After making this adjustment, the contingency hypothesis predicts a main effect of Stroop trial type (congruent, incongruent), a main effect of contingency (high, medium, low), and, more critically, no interaction between the two. In contrast, the modulation hypothesis predicts an interaction, because incongruent trials should be more affected by attention given that the majority of the Stroop effect is interference, with little or no facilitation from congruent trials (see MacLeod, 1991, for a review). Contingency Learning 8