catlearn (version 0.8)

homa76: Category breadth CIRP

Description

In some category-learning experiments, category members are distortions of an underlying base pattern. Where this is the case, 'category breadth' refers to the magnitude of such distortions. Broad categories take longer to learn than narrow categories. Once trained to an errorless criterion, the effect of category breadth on performance on novel items depends on category size. For small categories, narrow categories are better than broad ones. For larger categries, the reverse is true. Homa & Vosburgh (1976) provide the data for this CIRP.

Usage

data(homa76)

Arguments

Format

A data frame with the following columns:

phase

Experimental phase (within-subjects). Takes values : 'train','imm'. The training phase is 'train', 'imm' is the immediate test phase.

cond

Category breadth (between-subjects). Takes values : 'mixed', 'uni-low'

stim

Stimulus type (within-subjects). Takes values : 'proto', 'low', 'med', 'high', 'old-low', 'old-med', 'old-high', 'rand'. All refer to novel stimuli in the test phase, except those beginning 'old-', which are stimuli from the training phase presented during the test phase. 'low', 'med', 'high' refer to distortion level. 'proto' are prototypes. 'rand' are a set of 10 random stimuli, generated from prototypes unrelated to those used in training. These random stimuli are not mentioned in the Method of the paper, but are mentioned in the Results section - they are presented at the end of the test session. Empty cell for training phase.

catsize

Category size (within-subjects). Takes values : 3, 6, 9. NA for training phase, where category size is not a meaningful variable given that the DV is blocks to criterion. Also NA for old stimuli; Homa & Vosburgh's (1976) Results section collapses across category size for old stimuli

val

For test phases: probability of a correct response, except for random stimuli, where 'val' is the probability with which the random stimuli were placed into the specified category. For training phase: number of blocks to criterion

Details

Wills et al. (n.d.) discuss the derivation of this CIRP. In brief, the effects have been independently replicated. Homa & Vosburgh (1976) was selected as the only experiment to contain all three independently replicated effects.

Homa & Vosburgh's experiment involved the supervised classification of nine-dot random dot patterns. Stimuli had three different levels of distortion from the prototype - low (3.5 bit), medium (5.6 bit), and high (7.7 bit). There were three categories in training, one with 3 members, one with 6 members, and one with 9 members. Participants were either trained on stimuli that were all low distortion (narrow categories), or on an equal mix of low, medium, and high distortion stimuli (broad categories). Training was to an errorless criterion. The test phase involved the presentation of the prototypes, old stimuli, and novel stimuli of low, medium, and high distortion.

The data for the prototype, and other novel test stimuli, were estimated from Figure 1 of Homa & Vosburgh (1976), using plot digitizer (Huwaldt, 2015). The data for old stimuli were estimated from Figure 3, using the same procedure. The data for the training phase, and for random stimuli, were reported in the text of Homa & Vosburgh (1976) and are reproduced here. All data are averages across participants.

Homa & Vosburgh's (1976) experiment also includes results for further test phases, delayed by either 1 week, or 10 weeks, from the day of training. These data are not the focus of this category breadth CIRP and have not been included.