topic Re: Logistic Regression Modeling and Sample Size in Statistical Procedures

Logistic Regression Modeling and Sample Size

BTAinRVA — Tue, 26 Mar 2019 15:20:57 GMT

Good morning all!

I'm doing logistic regression modeling with 7 categorical predictor variables and it seems everything is coming up significant and is being included in the final model. Is this simply due to having a rather large sample size (50k+)?

Thanks,

Brian

Re: Logistic Regression Modeling and Sample Size

Reeza — Tue, 26 Mar 2019 15:25:47 GMT

How many levels in each category? You're likely overpowered, which means you need to start looking at effect sizes to determine if it's practically significant.

Re: Logistic Regression Modeling and Sample Size

StatDave — Tue, 26 Mar 2019 15:27:52 GMT

If you are using a WEIGHT statement, this is quite common unless you also use the NORMALIZE option in that statement.

Re: Logistic Regression Modeling and Sample Size

BTAinRVA — Tue, 26 Mar 2019 16:48:51 GMT

Reeza,

Thanks for the reply! The eight variables have 5, 6, 6, 2, 15, 2, 9, and 4 levels.

Brian

Re: Logistic Regression Modeling and Sample Size

BTAinRVA — Tue, 26 Mar 2019 16:57:34 GMT

StatDave,

Thanks for the reply. I'm not using the weight statement as my data is not aggregated. I have one line per person and a response variable, comply, that is 0 or 1depending on if the person is in compliance.

Brian

Re: Logistic Regression Modeling and Sample Size

Reeza — Tue, 26 Mar 2019 17:07:06 GMT

That's actually 4+5+5+1+14+1+8+4 = 42 variables but still with 50,000 rows you're likely fine. Assuming no interaction. I would be adding confidence intervals to my estimates and seeing if the effects are large enough to matter.