topic Re: Number of observations for logistic regression in Statistical Procedures

Number of observations for logistic regression

freedom93 — Tue, 06 Sep 2016 21:35:46 GMT

Hello,

I'm new SAS user; I performed a logistic regression of 68 observations, but this is not significant. A friend advised me to take two or three times the number of observations that I have is to say the same thing but in two or three times. When I do this, then the regression is significant. But have I the right to do so?
If not, what can I do? I think this is my échatillon of studies was small.

Re: Number of observations for logistic regression

PGStats — Tue, 06 Sep 2016 22:21:17 GMT

Don't do that. One of the key assumptions for inference in logistic regression is the independence of observations. Duplicated obs are definitely not independent.

What you can do:

Plot the data. Make sure there are no outliers.
Consider using another statistical model.
Get more independent data.

Good luck.

Re: Number of observations for logistic regression

ballardw — Tue, 06 Sep 2016 22:47:55 GMT

How many independent variables are you using? How many levels are there for each of those? Too many varaibles/levels will often leave nothing signficant as there are not enough observations in any group.

Also, models are NOT required to have a signficant difference. The fact that there isn't may well mean that there is not sufficient influence on the population from your variables.

Re: Number of observations for logistic regression

Reeza — Tue, 06 Sep 2016 23:49:06 GMT

This implies that your study was under powered, you didn't have enough samples to find a significant difference - based on the observed difference.

Re: Number of observations for logistic regression

Ksharp — Wed, 07 Sep 2016 01:55:04 GMT

You have small sample data. Try EXACT logistic regression.