Separate groups of people estimated the sizes of perceived or of remembered objects. In three independent experiments, both sets of data were well fit by power functions, and the exponent was reliably smaller by remembered than for perceived size.

Sampling experiments are reported that show that the uncorrected chi-square test of independence is exceptionally robust with respect to small expected frequencies in ^h R × C ^H contingency tables. In general, error rates that exceed .06 occurred only when both marginal probability distributions were extremely skewed and sample size was small. Nevertheless, the quality of the approximation of chi-square probabilities for exact multinomial probabilities was sometimes poor, although excessive errors in approximation by W. G. Cochran's (1952) criteria usually resulted from actual error rates being smaller, not larger, than the nominal level. A distinction is made between accuracy of approximation and control of the Type I error rate as considerations dictating the advisability of using an approximate test. (38 ref)

The apparent strength of subjective contours was investigated as a function of illumination level figure size. and viewing distance. Magnitude estimation. with a real contour standard as the modulus: was used to measure the perceived strength of the illusory contours. It was found that illuminatfon level and retinal size are both powerful determinants of the apparent strength of subjective contours, generating magnitude estimates varying from 20% to 96% of the strength of the real contour modulus. Particularly strong subjective contours were reported for figures of small retinal size (1.2 to 4.8 deg) under very dim illumination (.10 log lx).In addition to an experienced phenomenal reality. subjective contours and the objects delineated by them have a strong functional reality-in that they can cause and undergo visual masking, optical illusions, tilt aftereffects, binocular rivalry, and figural aftereffects in much the same way as real contours and objects (Fame.

With specially arranged inducing elements on a white surface of uniform luminosity, a phenomenally complete Necker cube can be seen in any array where only the 'corners' of the cube are physically represented. The subjectively seen bars of the cube disappear when the inducing 'discs' are seen as 'holes' in an interposing surface, through which the corners of a partially occluded cube are viewed. Illusory brightness effects are also observed in connection with the different organizations of this ambiguous figure.

A general purpose program is described for generating, analyzing, and graphing simulated data for experimental, multivariate, and contingency table designs. Applications include simulating actual or hypothetical research studies, conducting Monte Carlo investigations of the consequences of violating assumptions on statistical tests, generating randomized data sets for students to analyze and interpret as homework, and performing simulations demonstrating sampling theory. Data sets may be generated that have normal or nonnormal distributions, independent or correlated observations, and equal or unequal sample sizes. Data-analysis capabilities include oneway, two-way, and three-way ANOVA with any combination of independent-and correlated-groups factors, as well as simple-effects analyses, multiple pairwise comparisons, linear contrasts, and trend analyses.The use of computer simulation for teaching and research is a relatively well-established practice in psychology. I first employed simulation in the mid-1970s, to teach principles of sampling theory in courses on statistics, experimental design, and research methods (Bradley, Hotchkiss, Dumais, & Shea, 1976). The students in these courses would conduct simulations to assess the power of the F test as a function of effect size, sample size, variability, and so forth. Somewhat later, I developed software for sampling contingency tables from userspecified populations. These programs were used for CAl demonstrations (Bradley & Cutcomb, 1977), evaluation of the actual as opposed to the nominal Type I error rate of the chi-square statistic (Bradley, Bradley, McGrath, & Cutcomb, 1979), and empirical determination of the power of the chi-square test (Bradley & Seely, 1977). During this same period, I also developed software for generating independently randomized data sets for students to use in doing homework (Bradley, 1978;Bradley et al., 1976). While the data-generating algorithms were somewhat primitive, the program succeeded in providing students with simulated data sets based on actual studies, and in giving them feedback on the accuracy of their solutions.It soon became apparent that one program could incorporate all of the capabilities described above. Of course, the data-analysis and graphics capabilities of the software would have to be fairly extensive, because simulations involving a wide variety of research designs and dataanalysis methods should be supported. The software package written to fulfill these requirements is called DATASIM (Bradley, 1988), for data simulator. The original mainframe version of this software (Bradley, 1984a(Bradley, , 1984b has been supplanted by a more powerful and flexible version designed for the mM and mMCorrespondence maybe sentto Drake R. Bradley, Department of Psychology, Bates CoIlege, Lewiston, ME 04240. compatible PC (Bradley, 1988, in press). The latter version ofDATASIM is the focus of this paper; versions for Macintosh and Amiga personal computers are in development.DATASIM can generate simulated data for one-group, ...