Statistical Strategies for Small Sample Research

Newer statistical models, such as structural equation modeling and hierarchical linear modeling, require large sample sizes inappropriate for many research questions or unrealistic for many research arenas. How can researchers get the sophistication and flexibility of large sample studies without the requirement of prohibitively large samples? This book describes and illustrates statistical strategies that meet the sophistication/flexibility criteria for analyzing data from small samples of fewer than 150 cases. Contributions from some of the leading researchers in the field cover the use of multiple imputation software and how it can be used profitably with small data sets and missing data; ways to increase statistical power when sample size cannot be increased; and strategies for computing effect sizes and combining effect sizes across studies. Other contributions describe how to hypothesis test using the bootstrap; methods for pooling effect size indicators from single-case studies; frameworks for drawing inferences from cross-tabulated data; how to determine whether a correlation or covariance matrix warrants structure analysis; and what conditions indicate latent variable modeling is a viable approach to correct for unreliability in the mediator. Other topics include the use of dynamic factor analysis to model temporal processes by analyzing multivariate; time-series data from small numbers of individuals; techniques for coping with estimation problems in confirmatory factor analysis in small samples; how the state space model can be used with surprising accuracy with small data samples; and the use of partial least squares as a viable alternative to covariance-based SEM when the N is small and/or the number of variables in a model is large.