Response Variables in the Estimation of Relative Frequency

Abstract

6 groups of 20 Ss each observed a sequential series of digits, 4s and 8s, and judged what proportion (P) of the total was made up by one of the digits. Three kinds of response were used, estimates of percentage, the ratio of one frequency to the other, and the two frequencies themselves. P was varied from .1 to .9, either by holding total frequency constant, or by holding one frequency constant and varying the second. All responses were converted to log ratios, and the relationship of log estimated ratio to log ratio was found to be generally linear. There were significant differences in scales of proportion for the three response modes, and for the two methods of P variation. Percentages, ratios and frequencies gave different slopes, which were possibly a function of an assimilation-contrast effect. Differences due to methods of P variation were thought to result primarily from differences in total frequency at certain levels of P.