Interpretation of my logit model. Help please

MSC · September 20th, 2005, 4:44 pm

I have estimated the following model using Logit Maximum Likelihood estimation:RG = -10.3445 -0.16151LPR + 1.3268LTR + utI'm not too sure on how to interpret this. Would it simply be that a one unit increase in LPR is likely to lead to a 0.16151 decrease (assign is negative) in RG and a one unit increase in LTR is likely to lead to a 1.3268 increase in RG

Antonio · September 20th, 2005, 5:04 pm

Hi,If you used a logit model, then you have to interpret, not the coefficients themselves, but the odds ratio corresponding to these coefficients. In a logistic model, with a binary reponse variable Y, you have P(Y=1)=1/(1+exp(-(a+bX))).When you estimate the coefficient b for instance, you interpret it as the change in the logit for a 1-unit move of X. If you want to consider the variable Y, and not the logit, you have to interpret exp(b). The exact interpretation depends on the characteristics of your variables : are they dichotomous, polytomous, continuous ?

MSC · September 20th, 2005, 5:27 pm

Hi Antonio, Thanks for your reply, I think I should be able to get the probabilities. I think I got confused when looking at the print out of my results from a computer package as I had learnt how to do this manually.Once again, many thanks

MSC · September 20th, 2005, 5:42 pm

Just one last question. Would a negative odds ratio mean that a unit increase will lead to a decrease in the dependent variable my the amount of the odds ratio?

Antonio · September 20th, 2005, 6:22 pm

Yes indeed. More formally if f is your logit function : f(x)=ln(p(x)/(1-p(x))). You have f(x)=a+bx and hence f(x+1)-f(x)=b. This is the logit slope coefficient, whatever its sign may be.

MSC · September 20th, 2005, 6:33 pm

Thank you very much. I can finally finsih of the last of my results.