Cynthia G. Parshall

Innovative Items for Computerized Testing

As computer-based testing (CBT) becomes a dominant, if not the dominant,medium for delivering assessments, interest in the potential of innovative items has grown. Innovative items are those that make use of features and functions of the computer to deliver assessments that do things not easily done in traditional paper-and-pencil assessments.

Tests of Independence in Contingency Tables with Small Samples: A Comparison of Statistical Power.

The chi-square test of independence is a frequently used nonparametric test. Although the Type I error estimates for small-sample chi-square statistics applied to 2 x 2 tables have been relatively thoroughly studied, the Type II error rate (or power) has not been as extensively investigated. Additionally, most of the previous research has been limited to 2 x 2 tables and has not been extended to tables of larger dimensions. The objectives of the present research were (a) to investigate the small-sample power of several statistical tests in 2 x 2 and 3 x 3 contingency tables, and (b) to examine the Type I error rates for these tests under studied conditions. Various marginal distributions, sample sizes, and effect sizes were examined.

Comparative performance of three statistical tests of homogeneity for sparse i x j contingency tables

The Type I error rates and statistical power of three tests of homogeneity of variance for sparse I × J contingency tables were investigated in a Monte Carlo study. The Pearson chi-square, likelihood ratio chi-square, and the Read and Cressie power-divergence statistic with λ = 2/3 were compared under a variety of table dimensions, sample sizes, marginal distributions, and effect sizes. The results suggested that, for small samples, the Pearson chi-square evidenced power advantages in large tables, while the power-divergence statistic was more powerful in small tables. The power estimates for these tests converged with large samples. The likelihood ratio chi-square showed excessively liberal Type I error rates and was not recommended for the analysis of sparse tables.