Selective attention and individual preferences in judgmental responses to multifeature patterns

Selective attention and individual preferences injudgmental responses to multifeature patterns* CYNTHIA M. MAVRIDES Laurentian University, Sudbury, Ont., Canada Three experiments were conducted to show that the physical features of star-shaped patterns were preferred and used consistently as a basis of judgment across all pairs of patterns and throughout various tasks. Superior prediction of responding was achieved when it was hypothesized that within any pattern pair, the feature having the relatively larger difference in level would be emphasized in responding. In other words, both eonsistent preferences and selective attention to relative discriminability of features effected judgments for pattern pairs. Responses to visual patterns are often assumed to be of a multi dimensional nature, and the rationale for this assumption, as weil as several concrete applications arising from it, can be found throughout the current psychophysical literature (Brown & Andrew, 1968; Brown & Owen, 1967; Stenson, 1968; Fenker & Brown, 1969; Mavrides & Brown, 1969, 1970). In most approaches it is assumed that psychological dimensions (generated and utilized by the 0) be ar some consistent and predictable relationship to physical features of the pattern which can be measured objectively (Mavrides & Brown, .1969, 1970). Given that a pattern is composed of several features, research in seleetive attention has conclusively demonstrated that Os are not always consistent in the utilization of one single feature, or a fixed set of features (Egeth, 1967). Experiments have further demonstrated the importance of individual preferences and variability within features in the predietion of changes in selective attention (Imai & Garner, 1965; Mavrides, 1970). It has been suggested previously (Mavrides, 1970) that selective attention might be ineorporated into linear predictive models with thc creation of an additional variable (dmax), which is based on the assumption that relative differences in feature levels for pairs of patterns would affect any responses based on pattern similarity. The predietive value of this assumption was demonstrated for two-feature patterns in an experiment where the effeets of individual preferences in selection were minimized. The following experiments were designed to evaluate the effectiveness of the dmax variable with multifeature patterns where individual preference was an additional determinant of responding and to discover if a strategy of feature *This research was supported by Operating Grant AP-A 747 from the National Research Council 01' Canada. Psychon. Sei., 1970, Vol. 21 (2) seleetion (dependant on relative differences in feature levels within pattern pairs) generalized across Os and across tasks with different required responses. PATTERNS Patterns consisted of symmetrical star-shaped outlines (black on a white background), with points of equal angles at equal distances along the circumference of the smallest enclosing circle. The following feature measures could be calculated for each pattern: (l) The radius of the smal1est enclosing eircle, (2) the number of points, (3) the ratio of the radius of the largest enclosed circle to the radius of the smal1est enclosing circle (indicating the depth of the stars' points), (4) the exterior interpoint angle, and (5) the interior point angle. EXPERIMENT 1 Os for the experiment were seven undergraduate university students. Twelve patterns were constructed for the experiment, having insignificant correlations between all the five feature measures. The number of points ranged from 2 to 18, the interpoint and point angles were all over 20 deg, the radius of the enclosing circle fell between 20 and 35 cm, and the ratio for the enc10sed circle was between .2 and .6. Os were presented with all possible paire d comparisons, excluding those involving the same pattern twice (N =66). Each of the Os was tested individually and was al10wed a maximum of 5 sec to make an oral difference judgment with respect to each pair of patterns on a 7-point seale (1 was "very !ittle differenee" and 7 was "very much difference"). Pairs of patterns were presented on an 8'h x 11 in. sheet of white paper, and Os were instructed to look at al1 pairs onee before making any judgments. Pairs were presented randomly to each O. Since it was assumed that prediction of responses would involve measures relating the patterns in each paired comparison (rather than direct feature measures of any single patterns), an absolute difference vector was computed on the basis of all five feature measures for each of the 66 pairs. The sign of the differences was not considered important, since the order of subtraction was arbitrary and the position during presentation would not affect the magnitude of the difference. A sixth variable [dmax (5)] was added to each vector. Differences for each measure were standardized to the unit normal (Z) across the 66 pairs, and dmax (5) was equal to the maximum Z score in each vector. The dmax measure was designed to reflect the variability in feature level for the maximum discriminating dimension, sinee it was the aim of the experiment to test whether or not difference judgments would be related to a measure of this type. Data and information provided by Os (discussed below) suggested the importance of a dmax measure based on only two of the total five measures; the same method was used to generate dmax (2). Results The dependent variable for each pair was themeanresponseofthesevenOs.Tablel presents intercorrelations of the five difference measures and their correlations with judged difference. All difference measures (except for the radius of the smallest enclosing circ1e) correlated significantly with judgments, as did dmax (5). When questioned after the experiment, Os claimed to have based their judgments on the number of points and the "sharpness" of points, while ignoring differences in circular extent and size. Data Iabte 1 Intercorretations of the Five Difference Measures, dmax(5), and Mean Difference Judgments for Experiment 1 2 t 2 3 4 5 6 .04 -.02 -.01 4 5 6 7 SI S2 S3 .01 .17 .62** .14 .65** .25** .26** .42** .43** .50** .59** .61 ** .03 .68** .30** .45** .69** .44** .04 .61** .28** .34** .50** .35** .08 .65** .26** .36** .69** .49** -.02 .57** .26** .42*' .58** .42** (1) Radius of the enclosing circle. (2) Nzllnber of points. (3) Ratio of the radii of the enclosed and encloring eircles. (4) Interior point angle. (5) Exterior interpoint angle. (6) dmax(5). (7) Differenee judgments. (S) Subject (indil'idual difference judgments). "*p <01 67 this Table 2 Interconelations Between the Two Preferred Difference Measures, dmax(2), and Difference Judgments for Experiments 1,2, and 3 Experiment 1 1 2 3 Experiment 2 2 3 4 .65** .88** .87** .68** .69** .76** 1 2 Experiment 3 2 3 4 .38 .79** .70** .76** .67** .77** 3 1 2 2 3 4 .38 .79** .70** .85** .72** .85** 3 (1) Number of poi (...truncated)