Evidence for distinct serial processes in animals: The multiplicative-factors method

Abstract

The assumption that responses are controlled by distinct serially arranged processes was used by Sternberg (1969a) to explain the result, observed in many human experiments, that two factors have additive effects on reaction time (RT). With slight changes, Sternberg's explanation of additive factors with RT also explains the result, observed in animal experiments, that two factors have multiplicative effects on response rate. This article describes and interprets 17 examples of multiplicative factors from response-rate experiments with rats, pigeons, and goldfish, as well as some other animal evidence for distinct serial processes. The examples suggest and/or support new and old ideas about generalization, attention, timing, learning, motivation, and response production. Most important, the animal evidence makes the case for distinct serial processes considerably stronger. Since the procedures used in the two sets of experiments (human and animal) have little in common, distinct serial processes may control behavior in a very wide range of situations.Sternberg (l969a) introduced the additive-factor method (AFM), a guide to designing and interpreting reaction time (RT) experiments, and used the method to show that there was evidence for distinct serially arranged processes (which Sternberg called processing stages) in a number of human RT experiments. The data that he used to illustrate the method are unusual. They come from factorial RT experiments, and the key result, found in a number of experiments, was that pairs of factors (treatments) had close-to-additive effects on mean RT. One unusual feature of the data is the precision of the additivity. In most cases, an additive model fit so well that in graphs of the data, all of the points (representing the data) touched the fitted lines (representing the model). In a few cases, it was not obvious that the lines had not been drawn through the points. Ten of Sternberg's examples provided enough data to calculate the percentage of variance explained by an additive model; on the average (median over the 10 cases), an additive model describes 99.89% of the variance. In one set of experiments, not done with the method in mind, an additive model described RTs with median absolute error of 1.8 msec (median taken over five pairs of factors). In an experiment designed for the method, the errors were .9 msec (one pair offactors) and.4 msec (same pair at a different level of a third factor), with RTs on the order of 500 msec. Sternberg (1971) described 9

Additive and interactive effects on response time distributions in visual word recognition.

“…For example, Becker (1979) investigated the effects of word frequency (high vs. low frequency) and semantic context (related vs. unrelated context) on word recognition and reported that word frequency interacts with semantic context. This suggests that word frequency and semantic context influence a common stage.Following this early classic work, demonstrations of clear additivity have been observed across diverse studies (see Sternberg, 1998, for an extensive review; see also Roberts, 1987;Sanders, 1990), supporting the claim that additive effects reflect discrete stages of processing and that different factors can selectively influence these stages. Given that the notion of discrete stages seems simplistic and architecturally implausible, it is perhaps unsurprising that additive-factor logic has encountered some resistance and skepticism (e.g., Broadbent, 1984, pp.…”

Separate modifiability, mental modules, and the use of pure and composite measures to reveal them

“…17 For any level G k of G, the p.eect on M AB of a factor F with two levels F 1 Y F 2 is the value of the measure when F F 2 divided by its value when F F 1 . 18 As shown by Anderson (1996) and Roberts (1987), analyses of such data bene®t from numeric scaling of the factor levels; see Sections 13,15,and Appendix A.15. and Appendix A.9.1. However, factorial experiments are essential with composite measures, because the predictions take the form of statements about how factor eects combine.…”

“…In Appendix A.15 I discuss a similar numerical scaling of factor levels (but one that linearizes their main eects on the untransformed RR data) and explain how Roberts (1987) used this representation to help discriminate among the simple multiplicative model of Eq. (4) and the three hybrid models expressed by Eq.…”

“…Among the relatively few factorial experiments by students of animal behavior, Roberts (1987) discovered a large number of diverse examples of factors having multiplicative eects on response rate, , one from a study by Clark (1958). Such a pattern can arise if the combination rule for the contributions to of dierent processes (in¯uenced selectively by those factors) is multiplication; Roberts has considered what sort of process model would create such a combination rule.…”

“…In both panels the levels of HD plotted on the x-axis have been assigned numerical values so as to linearize the main eect of that factor; the goal is to make any systematic deviations of data from model easier to discern. Roberts (1987) considered several process models in relation to the multiplicative combination rule, and favored one that is equivalent to the following: Processes A and B are arranged sequentially. Each operates by receiving a series of discrete signals (``pulses'') and transmitting each pulse with some probability.…”