# Cross Tabulation vs Logistic Regression

I ran a logistic regression with 20 variables. This model is to figure out what factors contribute to a greater likelihood of being a donor. Dependent Variable= donor (y/n) (1/0) Independent Variable in question = BEC member (y/n) (1/0) I ran the logistic regression and MLE for BEC was -.4265 and the odds ratio was .426. The odds ratio seems to say that for every person who is a BEC member there is a less likelihood that they donate. This doesn't seem to be consistent with the cross tab because if someone is a BEC member then they seem to have a much greater chance of being a donor. Am I reading the cross tab incorrectly? Or was there something wrong with my logistic regression model? • It looks to me that the categories (donor/non-donor) have been flipped in your logistic regression. It could be that your software reversed the labels , e.g., it took the first category as the 'event' and it happened to be 0. Might be worth checking it out May 9, 2017 at 18:29
• I thought the same thing but my code had model donor(event='1')
– Mike
May 9, 2017 at 18:45
• Would the Odds Ratio for this be 18.67? 150/22462 = .0066/.9933 = .006723 14/38895 = .00036/.9996 = .00036 .006723 / .00036 = 18.67?
– Mike
May 9, 2017 at 21:35
• When I ran the same logistic regression model with ONLY the BEC variable the odds ratio came out to be 18.67. If I run it with all 20 variables then the odds ratio is .426. Why is there such a difference?
– Mike
May 10, 2017 at 12:42
• It is good that you mention this. I thought you ran the logistic regression only with BEC and Donor before as well. One reason for this large difference could be a strong correlation between BEC and another variable. Check whether BEC highly correlates with any of the other 19 variables. Then, run the model once with all 20 variables and once with those correlating variables removed. If BEC has a coefficient close to the 18.67 in the latter case, you have found the problem. May 10, 2017 at 22:18