Measuring habituation in infants: an approach using regression analysis

Abstract

The effectiveness of different habituation criteria was examined by means of computer simulations. A criterion based on fitting a second-order polynomial regression function to the looking time data was described. This criterion produced more accurate estimation of looking times as well as higher experimental power for detecting novelty effects, compared to the traditional windowed running average criterion or to a criterion based on linear regression. The polynomial regression approach probably has this advantage because it utilizes all of the available looking time data, rather than just the data in the current windowed average, and because it is sensitive to nonlinear trends in looking time. This new habituation criterion is easy to implement on a laboratory computer, and it should increase session lengths by no more than one trial or so, compared to windowed average criteria. With regard to test-retest reliability, all of the habituation criteria that were evaluated appear to have low reliability on average, with high sample-to-sample variability. These undesirable reliability characteristics are attributable to the high variability of infants' attentional behavior.