The assumptions of a repeated measures ANOVA are that the continuous dependent variable is approximately normally distributed, the categorical independent variable (e.g., experimental group) has three or more levels, no outliers in any of the repeated measurements, and sphericity (constant variance across time points). All assumptions are required to be met for a repeated measures ANOVA to be an appropriate statistical analysis approach. In a review of 58 preclinical animal studies (7), it was found that the checking of assumptions of repeated measures analyses were not accurately reported. Specifically, assumptions related to variance were not described. The sphericity assumption is a strong assumption that may not be accurate when there are repeated measurements. Mauchly’s test can determine if the sphericity assumption is reasonable (8), and there are adjustments such as the Huynh-Feldt and Greenhouse-Geisser corrections that can account for the violations of the sphericity assumption (9–11). These adjustments are easy to apply with statistical software. Results from analyses that utilize these approaches should be stated in descriptions of statistical methods. Transforming the continuous dependent data so that they are approximately normally distributed or using Friedman’s test (nonparametric version of the repeated measures ANOVA) can be done when the normality assumption is invalid.

A repeated measures ANOVA requires a balanced number of repeated measurements for each experimental unit. Due to this requirement, experimental units with missing measurements are completely excluded from the analysis (i.e., complete case analysis), which results in the sample size decreasing and the type II error increasing (12, 13). By excluding experimental units with missing data, the statistical power also decreases. Sample sizes are typically smaller in basic science research, and any reduction in the sample size due to missing data can greatly effect the results (6).

Missing data can occur in basic science for various reasons. The reporting of missing data in basic science research is not standard practice as it is in clinical trials (14). Therefore, the description of statistical analysis methods in peer-reviewed publications could include the sample size of the complete cases only (7) or may not describe if an imputation technique was used for the missingness. Simple imputation approaches such as mean imputation and last observation carried forward are considered as remedies for missing data and alternatives to the complete case approach (2). Another missing data technique is multiple imputation. Instead of replacing the missing observation with a single value, like in simple imputation, multiple imputation replaces the missing observation many times, utilizing distributional properties and information from the observed data. However, these imputation approaches have their limitations, given small sample sizes that are commonly observed in basic science research (12, 15). The type of missingness should be determined when selecting an imputation technique. The three types of missingness are missing at random, missing completely at random, and missing not at random. Missing at random describes the scenario when the missing data are independent of the unobserved measurement. Missing completely at random is missing data that are independent of the observed and unobserved measurements. Missing not at random is data that are missing due to the unobserved measurement. Multiple imputation is applicable when missing at random occurs.