The project I'm working on attempts to explain why an NGO is present (1=present,0=not present) in some countries (15) but not others. As I'm working with country-level data, my sample size is naturally limited to the number of existing countries (193). Moreover, many countries criminalize the existence of the NGO in question (meaning the non-existence of the NGO in those countries requires no explanation), which further reduces the sample size to around 60-70 countries. Will this be enough for conducting a logistic regression? There are around 10 predictor variables I'd like to test, but I'm not sure whether any of the predictions will be meaningful. If not logistic regression, what methodology, if any, can I use to demonstrate significant relationships (given my small N)? Thanks in advance!

  • 1
    $\begingroup$ Look at this recent answer and links from that page. With 15 cases in your smallest outcome, even a single predictor is on the edge. $\endgroup$
    – EdM
    Oct 16, 2017 at 16:26

1 Answer 1


As I understand it, you have 10 predictor variables and 15 countries with the NGO in question. You simply do not have enough to fit such a model (cf., here). Nor is there some other magical model that will let you assess this. The problem isn't the model, but that you don't have enough information to work with. Some other model also wouldn't have enough information to work with. The options I can think of are:

  1. Find some other dataset that allows you to expand the amount of information available (e.g., state-level data).
  2. Work out informative prior probability distributions from other sources of information in your field and fit a Bayesian model. Note that because you don't have much information, the result of the Bayesian model will be based largely on what you believed already.
  • $\begingroup$ Would a simple correlational analysis work (if I merely want to demonstrate a relationship) or is my N even too small for that? $\endgroup$ Oct 16, 2017 at 17:10
  • $\begingroup$ @ZachGoldberg a correlation is a (type of) model. "The problem isn't the model, but that you don't have enough information to work with." $\endgroup$ Oct 16, 2017 at 17:13
  • $\begingroup$ What if I limit the number of predictors to 4-5? $\endgroup$ Oct 16, 2017 at 17:23
  • $\begingroup$ If you read the linked thread, you will see that whether you have enough data for 1 variable is debatable. You definitely don't have enough for 2. $\endgroup$ Oct 16, 2017 at 17:26
  • $\begingroup$ Is the situation different for experimental studies? I've seen some where N < 100 $\endgroup$ Oct 16, 2017 at 17:46

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.