The effect of interstimulus interval on sequential effects in absolute identification

Abstract

In absolute identification experiments, the participant is asked to identify stimuli drawn from a small set of items which differ on a single physical dimension (e.g., 10 tones which vary in frequency). Responses in these tasks show a striking pattern of sequential dependencies: The current response assimilates towards the immediately preceding stimulus but contrasts with the stimuli further back in the sequence. This pattern has been variously interpreted as resulting from confusion of items in memory, shifts in response criteria, or the action of selective attention, and these interpretations have been incorporated into competing formal models of absolute identification performance. In two experiments, we demonstrate that lengthening the time between trials increases contrast to both the previous stimulus and the stimulus two trials back. This surprising pattern of results is difficult to reconcile with the idea that sequential dependencies result from memory confusion or from criterion shifts, but is consistent with an account that emphasizes selective attention.

Keywords:

• Absolute identification
• Interstimulus interval
• Sequential effects
• Assimilation
• Contrast

Acknowledgments

This work was supported Economic and Social Research Council Grant RES-000-23-1372. We thank Donald Laming and an anonymous reviewer for their incisive comments on an earlier version of this article.

Notes

¹ When separate regression analyses are conducted for a number of different participants, it is common to use inferential statistics, such as t tests and ANOVA, to test whether the mean coefficients differ from zero or differ between conditions (Lorch & Myers, 1990). However, an alternative, arguably superior, approach is provided by multilevel analysis (Raudenbush & Bryk, 2002). For all of the analyses reported here we conducted corresponding multilevel analyses and found the same pattern of significant results.

² A reviewer asked whether the results for the subject-averaged data reflected the findings from individual participants. In the stimulus-only regression, the S _{n − 1} coefficient decreased when the ISI was lengthened for 27 of the 37 participants (p = .008, two-tailed binomial test), and the S _{n − 2} coefficient decreased for 25 of the 37 participants (p = .047). The coefficients for the S _n , S _{n − 3}, S _{n − 4}, and S _{n − 5} terms decreased for 17, 15, 22, and 18 participants, respectively (all ps > .3). For the stimulus–response regression, increasing the ISI led to a decrease in the S _{n − 1} coefficient for 26 of the 37 participants (p = .02). Similarly, 26 participants showed a decrease in the coefficient for S _{n − 2}. The coefficients for S _n , R _{n − 1}, and R _{n − 2} decreased for 15, 17, and 15 participants, respectively (all ps > .3). These results match those of the averaged data.

³ In addition to Equations 1 and 2, several alternative regression equations for the assessment of sequential effects have been proposed (DeCarlo & Cross, 1990; Lockhead, 1984; Lockhead & King, 1983; in addition, a reviewer suggested regressing the current response only on the preceding sequence of responses). These approaches produced essentially the same results as those reported here.