Blaming the victims of your own mistakes: How visual search accuracy influences evaluation of stimuli

Abstract

Even without explicit positive or negative reinforcement, experiences may influence preferences. According to the affective feedback in hypotheses testing account preferences are determined by the accuracy of hypotheses: correct hypotheses evoke positive affect, while incorrect ones evoke negative affect facilitating changes of hypotheses. Applying this to visual search, we suggest that accurate search should lead to more positive ratings of targets than distractors, while for errors targets should be rated more negatively. We test this in two experiments using time-limited search for a conjunction of gender and tint of faces. Accurate search led to more positive ratings for targets as compared to distractors or targets following errors. Errors led to more negative ratings for targets than for distractors. Critically, eye tracking revealed that the longer the fixation dwell times in target regions, the higher the target ratings for correct responses, and the lower the ratings for errors. The longer observers look at targets, the more positive their ratings if they answer correctly, and less positive, following errors. The findings support the affective feedback account and provide the first demonstration of negative effects on liking ratings following errors in visual search.

Keywords:

• Conjunctive visual search
• Error monitoring
• Preferences
• Conflict
• Affective feedback
• Fixation dwell times

This study has been supported in parts by Russian Foundation for Humanities (grant number 12-36-01298a2, AC) and by Saint-Petersburg State University (research grant number 8.38.287.2014, AC), and by grants from the Research fund of the University of Iceland (ÓIJ and ÁK) and the Icelandic Research Fund (ÁK).

This study has been supported in parts by Russian Foundation for Humanities (grant number 12-36-01298a2, AC) and by Saint-Petersburg State University (research grant number 8.38.287.2014, AC), and by grants from the Research fund of the University of Iceland (ÓIJ and ÁK) and the Icelandic Research Fund (ÁK).

Notes

¹ In fact, the opposite post-decisional effect should be present for correct decisions as well. Thus, positive ratings for targets after correct answers probably confound the positivity stemming from the match between target template and object and the positivity stemming from the correct decision. The present study does not distinguish between the two, however.

² In Experiment 1 there was also a group of participants for whom target and distractors were presented for 1200 ms instead of 600 ms. In this group the low number of errors (14% of all answers, 2 errors per observer) did not allow a proper analysis of target liking following errors. Thus, this group was excluded from analyses and is not described here.

³ Our study followed a quasi-experimental design. Thus, liking ratings can be confounded both by between-stimulus differences (one face can be more attractive than another) and by between-observer differences (one observer can be in a better mood than another and provide higher ratings). In addition, observers may differ in their interpretation of the scale: some may try to use the full range of the provided scale, while others may consider most stimuli neutral and use only a small range of the possible values (liking rating SDs measured for each observer in Experiment 1 varied from 0.28 to 3.10). Usually, a by-subject aggregation is used to control for between-subject differences, but it does not allow for controlling the within-trial variables, such as target sex or trial position nor for between-stimulus differences. To avoid this problem, liking ratings were first Z-standardised individually for each observer. The standardisation allows comparing data from different distributions, thereby diminishing the effects of between-subject differences in strategies used to rate the stimuli. Second, we conducted linear mixed-effects regression with random effects of stimuli and random within-subject effects of answer accuracy and object type (Judd, Westfall, & Kenny, 2012). Random effects for stimuli provided control for the differences in mean ratings of stimuli. Random within-subject effects of answer accuracy and object type controlled for the possible difference in individual reactions to correct and incorrect answers and to targets and distractors. Finally, we also added stimulus gender (male vs. female), subject gender (male vs. female), their interaction, and response time during the search as covariates to control for their possible confounding influence (the effect of these variables is not reported for the sake of brevity). This analysis also allows the inclusion of dwell times and is therefore preferable for consistency of analytic approaches in Experiment 1 and Experiment 2. For the ease of understanding, results are presented as Type III F-tests using a Satterthwaite approximation of the degrees of freedom.

⁴ Portions of the research in this paper use the FERET database of facial images collected under the FERET programme, sponsored by the DOD Counterdrug Technology Development Program Office.

⁵ During pilot testing no effect of premasks was found, so they were removed to shorten experiment duration.