Is it correct to find that an explanatory variable was found to be statistically significant with the chi-square test but insignificant with the logistic regression analysis model?

- 63,378
- 26
- 142
- 467

- 31
- 1
-
3Does your logistic regression model only have one explanatory variable in it? – Nov 19 '15 at 19:04
-
1No it has several explanatory variables,about 10 of them. – Etta Nov 19 '15 at 19:19
-
1It sounds as if you are running the risk of double dipping. You can't use associations with $Y$ to decide which variables to test for associations with $Y$. Besides that, adjusted and unadjusted (for covariates) analyses naturally differ. – Frank Harrell Nov 19 '15 at 19:27
-
Does this mean that the results might be faulty or the approach wrong? – Etta Nov 19 '15 at 19:45
-
2Correct. Look at some of the rules of thumb for the required sample size as a function of the number of _candidate_ predictors, and pre-specify a full model and use it as the basis for your conclusions. – Frank Harrell Nov 19 '15 at 21:23
1 Answers
Cleared up and partially answered in comments:
This logistic regression model has several explanatory variables, about 10 of them. – Etta
It sounds as if you are running the risk of double dipping. You can't use associations with Y to decide which variables to test for associations with Y. Besides that, adjusted and unadjusted (for covariates) analyses naturally differ. – Frank Harrell
Does this mean that the results might be faulty or the approach wrong? – Etta
Correct. Look at some of the rules of thumb for the required sample size as a function of the number of candidate predictors, and pre-specify a full model and use it as the basis for your conclusions. – Frank Harrell
See also Variable selection for predictive modeling really needed in 2016?

- 63,378
- 26
- 142
- 467