Shepard et al. (1961) stated that where there are two, equal-sized categories constructed from the eight stimuli it is possible to produce from varying three binary stimulus dimensions, there are only six logically distinct category structures. Shepard et al. (1961) labeled these structures as Types I through VI (see e.g. Nosofsky et al., 1994, Figure 1, for details). The CIRP concerns the relative difficulty of learning these category structures, as indexed by classification accuracy. The result, expressed in terms of accuracy, is:
I > II > [III, IV, V] > VI
The experiment reported by Nosofsky et al. (1994) provides the data for this CIRP.
A data frame with the following columns:
Type of category structure, as defined by Shepard et al. (1961). Takes values : 1-6
Training block. Takes values: 1-16
Mean error probability, averaged across participants
Wills & O'Connell (n.d.) discuss the derivation of this CIRP. In brief, the effect has been demonstrated independently on three occasions (Nosofsky et al., 1994; Shepard et al., 1961; Smith et al., 2004). Kurtz et al. (2013) discuss and report a number of experiments that indicate that the advantage of Type II over Type IV may depend on specific details of the experimental design, including the stimulus dimensions being easily verbalizable, and the participants being explicitly instructed to look for rules.
Nosofsky et al. (1994) was selected as the CIRP because it had
acceptable sample size (N=40 per Type), and included simulations of
the results with a number of different formal models. Inclusion of
this dataset in
catlearn thus permits a validation of
catlearn model implementations against published simulations.
In Nosofsky et al. (1994) the stimuli varied in shape (squares or triangles), type of interior line (solid or dotted), and size (large or small). Each participant learned two problems. Each problem was trained with feedback, to a criterion of four consecutive sub-blocks of eight trials with no errors, or for a maximum of 400 trials.
The data are as shown in the first 16 rows of Table 1 of Nosofsky et al. (1994). Only the first 16 blocks are reported, for comparability with the model fitting reported in that paper. Where a participant reached criterion before 16 blocks, Nosofsky et al. assumed they would have made no further errors if they had continued.
Andy J. Wills firstname.lastname@example.org
Nosofsky, R.M., Gluck, M.A., Plameri, T.J., McKinley, S.C. and Glauthier, P. (1994). Comparing models of rule-based classification learning: A replication and extension of Shepaard, Hovland, and Jenkins (1961). Memory and Cognition, 22, 352-369.
Kurtz, K.J., Levering, K.R., Stanton, R.D., Romero, J. & Norris, S.N. (2013). Human learning of elemental category structures: Revising the classic result of Shepard, Hovland, and Jenkins (1961). Journal of Experimental Psychology: Learning, Memory & Cognition, 39, 552-572.
Shepard, R.N., Hocland, C.I., & Jenkins, H.M. (1961). learning and memorization of classifications. Psychological Monographs, 75, Whole No. 517.
Smith, J.D., Minda, J.P. & Washburn, D.A. (2004). Category learning in rhesus monkeys: A study of the Shepard, Hovland, and Jenkins (1961) tasks. Journal of Experimental Psychology: General, 133, 398-414.
Wills, A.J. & O'Connell (n.d.). Averaging abstractions. Manuscript in preparation.