Spencer Bennett

An Effective Paradigm for Conditioning Visual Perception in Human Subjects

Repeated pairing of an auditory conditioned stimulus with a weak visual unconditioned stimulus produced extended image sequences and visual responses conditioned to the tone alone. The experiment is set into the context of Pavlovs view of Helmholtzs “unconscious inference” thus providing experimental evidence linking the higher mental process of perception with classical conditioning.

Semantic generalization investigated using second-order conditioned visual afterimages.

A series of experiments is reported during which three subjects were initially classically conditioned to produce visual percepts to a sound only, second-order conditioned to transfer that response to an animal name, and finally taught a “foreign language”. The results suggest that semantic generalization occurs whether the semantic relationships are formed either before or after the conditioned response is established.

Multiple Groups Analysis

The previous chapters have dealt with methods from which orthogonal factors emerge. Clusters of related variables are then obtained subsequently by orthogonal or oblique rotations to simple structure. This chapter is concerned with a method called multiple groups analysis from which an oblique factor matrix is obtained directly and all factors are extracted simultaneously. The number of such groups (and therefore factors) is assumed at the outset either on the basis of a priori knowledge of the field of study or by a systematic procedure which the interested reader can find in more advanced texts, such as Harman [1967]. It is, however, not serious if an incorrect assumption about the number of groups is made initially.

Principal Factor Analysis

In Chapter 2 we examined the method of factor analysis using the centroid technique. It was stated then that the aim of the analysis was to explain the correlations between the original observed variables in terms of their correlations with a smaller set of factors. In this chapter we will examine two other methods of analysis, principal factor analysis and principal component analysis. The approach of the two methods is similar and their aim, to aid interpretation of the underlying structure of the interrelationships between variables, is the same. But there is in fact, as we shall see later, a fundamental difference between the two methods.

Factor Analysis: the Centroid Method

We begin with a matrix or table of correlations between a set of variables. Since we are employing product-moment correlations as our starting-point, it is important that the assumptions underlying their use are met. To re-iterate briefly: ideally, the distributions of the variables should be continuous and reasonably normal (at least they should not be bimodal or markedly skewed); discontinuous variables will be discussed further in Chapter 7. Regression should be linear, and samples should be large (at least several hundred) to ensure reliability of the resulting correlations.

Rotation of Factors

The results of a factor analysis may be represented graphically as points in n-dimensional space where the factors represent the axes and the factor loadings the co-ordinates of each point. This is only conveniently realisable graphically when an analysis yields a two-factor solution. To illustrate this idea let us consider an analysis having five variables and giving the two-factor solution shown in Table 3.1

Concluding Remarks and Overview

The main aims of factor analysis can be re-iterated here: (a) To produce a more parsimonious description within a domain of study. For example, in the body-size problem (Chapter 2), nine variables measuring various aspects of body-size were reduced to three, giving approximately the same amount of information. (b) To test theories about the interrelationships between variables. For example: (i) in the analysis of intellectual ability (Chapter 4), the theory that out of eight tests of mental ability three factors (convergent, non-verbal divergent and verbal divergent thinking) would emerge could be tested; (ii) in the body-size problem, the theory that there are distinct body-types (short and stocky versus tall and lean) could be tested. (c) To establish functional relations between variables (one of the goals of science). Factor analysis can be employed to isolate variables which it may not be possible to measure directly, but which can be computed from a set of observable and directly measureable but otherwise unsatisfactory measures. This involves factor measurement which will be described in Section 9.2. (d) To analyse people or objects into types, which will be described in Section 9.3. (e) Finally, as a preliminary to regression analysis, to analyse the factorial structure of criterion variables, and hence point the way to those variables which are most likely to be usefully included in a regression equation. This will be discussed below in Section 9.4.

The Analysis of Qualitatitive Data

In previous chapters we have dealt principally with two problems, that of factor-analysing relationships between correlated variables and that of discriminating between groups and classifying individuals or objects into categories. Throughout, it was assumed that the measurement of the variables under consideration was in quantitative terms. Furthermore, it was also assumed that the variables in question were normally distributed (at least approximately) and that the regression relationship was linear.