4
$\begingroup$

I need to compare $16$ quantitative variables, measured for two groups, A and B. I thought of applying MANOVA. However, there are only $4$ and $9$ cases for groups A and B respectively. I looked for solutions, namely when number of DVs $>n$ (for example Many dependent variables, few samples: is this an example of "large $p$, small $n$" problem?), but I'm afraid I'm a bit confused.

What can be done? Remove variables? Reduce dimensionality? Duplicate cases? Use an alternative to MANOVA not sensitive to the small number of cases (is there are any)?

$\endgroup$
6
$\begingroup$

Any method will be "sensitive to a small number of cases", meaning that a small study has lower power to detect features of interest than a larger one. With multivariate data, you are not only comparing means -- you are typically estimating a covariance matrix as well, and this requires more data to do with any accuracy.

It's worth considering what happens in MANOVA. Basically, the algorithm looks for linear combinations of the variables that do a good job of distinguishing the groups. Effectively, it has dimension reduction built in. If your variables are actually independent, MANOVA will not do better than a bunch of univariate comparisons (t-tests).

Let's look at your suggestions:

  1. Remove variables. Certainly, if you think some variables won't distinguish the groups, then remove them.

  2. Reduce dimensionality. Best option, if you have subject-matter reasons for combining the variables in some way. Perhaps the sum of your variables would be a good measure of something. Instead of letting MANOVA find the best linear combination, you supply one, based on theoretical considerations, and do a t-test of the scores.

  3. Duplicate cases. This won't work. You can't make a singular matrix become non-singular by duplicating rows.

  4. Use an alternative to MANOVA not sensitive to a small number of cases. The two-sample t-test can actually detect group differences on very small samples. You could do t-tests on all your variables, but to be honest, you would need to adjust the significance level using a Bonferroni approach.

$\endgroup$
  • $\begingroup$ Thank you for you answer! I've never hear of Bonferroni correction. I'll take a look! $\endgroup$ – Hugo Oct 31 '14 at 12:28
  • $\begingroup$ +1. I took the liberty to improve formatting of your answer, hope you will approve; if not, roll back with apologies. $\endgroup$ – amoeba says Reinstate Monica Oct 31 '14 at 13:20
  • $\begingroup$ @amoeba - the formatting looks good. No roll backs. $\endgroup$ – Placidia Oct 31 '14 at 18:10
  • $\begingroup$ @hagc. The bonferroni is a simple adjustment of the p-value based on the highest type 1 error you can stand (say 0.05) and the number of tests. It's in most elementary stats texts - or Wikipedia. $\endgroup$ – Placidia Oct 31 '14 at 18:11
  • $\begingroup$ @hagc: Note that if you think that the answer satisfactorily resolves your question, you should consider "accepting" it by clicking on a green tick nearby (apart from upvoting, which you should do with any useful answers anyway!). Apologies if you know that. $\endgroup$ – amoeba says Reinstate Monica Oct 31 '14 at 19:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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