Research Settings and Procedures

A common question or concern is about the proper sample size that should be drawn. The required sample size is based on several factors, including the type(s) of analysis, variable measurement, and general study objectives. For example, for multivariate analyses (e.g., multiple regression, factor analysis), the common rule of thumb is that a sample size should be ten times larger than the total number of variables analyzed. If the investigator is interested in subgroup analyses (e.g., investigating relationships among age groups), larger numbers of subgroups would require larger samples, especially in studies that do not contain sampling quotas for subgroups. In structural equation modeling (SEM), the sample size depends on the number of variables or indicators and model fit; it could be as low as two to three times the number of variables if there is a good fit. The important factor in hypothesis or model testing is to use samples that can produce sufficient variance in the distribution of variable values. If the distribution of the values on variables is highly skewed, larger samples might be needed to obtain adequate distribution of values (e.g., via transformation of variable values). As a rule of thumb, for hypothesis testing involving two variables and estimation of reliability Alpha, a few hundred elements are needed (about 300, according to Nunnally 1967). Three to five hundred would be adequate for all practical purposes. And using larger samples would be counterproductive and dangerous, unless subgroup analyses are required. Very large sample sizes tend to produce significant relationships between variables, when such relationships actually do not exist.