Is it possible to run Somers' D (and its accompanying significance tests) on a very small sample? In my case, I have a 2x3 (one variable is ordinal, the other 0/1 coded) table but only 36 cases. If it is (not) okay, does anyone know the exact assumption?

PS: A similar question is asked here: Measure of Association between two Ordinal Variables


1 Answer 1


It's OK to do any analysis on a small sample if you completely understand the data quality and possible sample selection biases. But it is very important to include self-contained statistical quantifications that capture the limitations of the information base used in the analysis. In your case this can be done in the frequentist statistics setting using a compatibility interval (aka confidence interval). For example the R Hmisc package rcorr.cens function will give you the standard error of $D_{xy}$ which can be used to get an approximate compatibility interval.

Note: compatibility is related to the fact that the confidence interval includes the set of $D_{xy}$ values that if null hypothesized would not be rejected at the $1 - \alpha$ level.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.