Baseline Balance - T-Test - Missing Values

[Aakash Gupta]

I want to create a balance table on STATA. It is a baseline dataset conducted as a part of an RCT. However, some of the variables have missing values. The number of missing values the variables have varies from 18 to 97. N is 2000, and the data is at the household level. How do I account for these missing values while running the t-test? What is the command I use on STATA?

Thanks!

[Leonardo Guizzetti]

There are two issues here. The first is that the two-sample t-test can only accommodate known values from two groups. Missing values will always results in data being omitted. The best you can do is given an indication of how many missing, if any, for each variables in your descriptive table.

The second, is the more important issue. That is, does it make sense to show a balance table *with statistical tests* when describing arms from an RCT. There are very good reasons not to perform such tests. The reason is that due to randomization, the expectation is that the groups will be balanced at baseline. This is even more likely with larger trials, such as yours. Unless you have specific knowledge that the randomization mechanism has been significantly compromised in some way, then any statistically significant results from your inferential testing are due to type 1 error since these must be due to chance imbalance. Logically, this practice makes no sense and does not add any value to the reader. (note: important imbalances can be adjusted for in analysis, but this is altogether a different issue.)

Senn, S. (1994), Testing for baseline balance in clinical trials. Statist. Med., 13: 1715-1726. https://doi.org/10.1002/sim.4780131703