I suppose regression analysis with few observations has some peculiarities. Here are results for my logistic regression.

$R^2_{\rm LR}$ = 0.7, Number of 1 = 6 (4 predicted), Number of 0 = 78 (all predicted)

Is it good-fitting? If the number of observation is 10 times higher I would have on doubt but here I'd like to ask for an expert's opinion.


Whether it is a "good" fitting is context dependent.

Is the fact that 2 of 6 '1's were missed OK or not? That depends on what a '1' is. If this is about, say, airplanes crashing on takeoff, then that's pretty bad. In other circumstances it could be pretty good.

What is your context?

| cite | improve this answer | |
  • $\begingroup$ Actually, my objective is not a prediction but testing the impact of independent variables. The dependent variable is election win/lose. $\endgroup$ – 8k14 Dec 15 '13 at 15:30
  • $\begingroup$ For looking at impact of independent variables, you should look at odds ratios or other effect size measures. Unfortunately, with a 0 in one cell, OR doesn't work. But you can look at difference in proportions, for example. $\endgroup$ – Peter Flom Dec 15 '13 at 15:32
  • $\begingroup$ Thanks. I use z-test for each independent variable. I'm now interested in overall fitting of my model. $\endgroup$ – 8k14 Dec 15 '13 at 15:41
  • $\begingroup$ If you are interested in the impact of independent variables, use effect size (that's what you said in one comment). If you are interested in overall fit, then think of the context. $\endgroup$ – Peter Flom Dec 15 '13 at 15:44
  • $\begingroup$ I need to know overall fit in order to justify my results concerning the impact of independent variables. I'm not going to predict anything with the aid of my model so why does the context matter? $\endgroup$ – 8k14 Dec 15 '13 at 16:23

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.