Unequal sample sizes

Steve Simon


For some reasons, it seems to unnerve people when the sample size in the treatment and control group are not the same. They worry about whether the tests that they would run on the data would be valid or not. As a general rule, there is no reason that you cannot analyze data with unequal sample sizes. There are a few procedures, such as the Tukey follow-up test in Analysis of Variance, that require some adaptations. You also may not have as much precision as you’d like to have. The rule of thumb is that precision is determined predominantly by the group with the smaller sample size. Finally, some violations of assumptions (such as unequal variances) may be more serious when the samples sizes are unequal.

There are some situations where you might deliberately recruit unequal sample sizes. This occurs most commonly when it costs much less to recruit patients from one group than from another. Suppose the treatment group requires ten thousand dollars worth of therapy and the control group only requires a hundred dollars worth of therapy. For a fixed budget, you would be better off recruiting more control patients. With a bit of calculus you can show that the optimal degree of unbalancedness is related to the square root of the ratio of costs.

You can find an earlier version of this page on my original website.