Calculate Correlation or Covariance for non-equal Sample Sizes

Computing the covariance and correlation (the covariance is embedded in the correlation formula) depends on the fact, that the measurements are paired. This means, that each value from sample A can be matched with a value from sample B.

If the samples do not have the same size, this is not possible. So, you cannot calculate either the covariance nor the correlation in a reasonable way.

Anyway, the question is, what you really want to calculate in your case and whether it is meaningful is to use the correlation for this. Perhaps, another statistical technique or test would be a better choice.

Of course, the problem can also be that there are missing data in your second measurement because the data was not captured correctly or it was not possible to capture it. In this case, you have the classic "Missing Data Problem", for which there are several solutions. One ansatz would be to remove the whole data row. So, if there are missing values in sample B, using this approach, you have to delete the corresponding values from sample A to make the sample sizes equal.
Last update on 2023-05-02 | Created on 2014-02-09