Many randomized clinical trials capture normally distributed outcomes at a limited number of fixed time points. It is common to analyze treatment effects in this setting using the "mixed model repeated measures" approach, particularly when the treatment effect at the final time point is of primary interest. Small sample corrections to the denominator degrees of freedom for Wald tests help control Type I error probability, but inflation can occur when there are missing data. Assuming an MAR mechanism, we propose a degrees of freedom adjustment that replaces the total sample size by an effective sample size, obtained by summing each subject's contribution to the information at the final time point. This contribution is computed as the R2 value from a regression (using subjects with the requisite data) of the final value on the values at time points for which the subject has available data. Subjects with complete data thus contribute fully to the effective sample size and those with missing data contribute a fraction depending on how well their observed data can inform their missing final visit data. Simulation studies are presented illustrating the benefits of the proposed approach.