Interpretation of Results in clogit model

[Mina Sami]

Hello,
My papers' question is : Killed people in region on the firm location decision by opening a new store.


After Clogit model I found the coefficient of killed is -.002301


How can I interpret this in useful way?

As , I think the interpretation is:
When number of killed increases by 1 person, probability that firm locates decreases by 0.002301.

Alternatively can I use these words to be more meaningful:

When number of killed increases by 1000 persons, probability that firm locates decreases by 2.301


Does this work?
Any other suggestions?

Thank you in Advance

[Maarten Buis]

Your paper's question is not a question. I suspect you mean: Does the number of people killed in a region influences whether or not a firm opens a store/factory/office/... in that region? However I am not sure.

Coefficients of clogit cannot be interpreted the way you did. If you want to interpret the right hand side as a linear function, then the left hand side of your model is not the probability but the log odds. The easiest way to get interpretable coefficients is to take the exponential of your coefficients, which will give you the odds ratios. Stata makes that easy, all you have to do is add the or option to clogit. In your case, a extra person killed decreases the odds of a firm locating there by (exp(-0.002301) - 1)*100% = -0.2%.

The easiest way to get the effect of 1000 kills increase is to create a new variable gen kills1000 = kills/1000 and add kills1000 instead of kills to your model. This will give you a coefficient of about (exp(-0.002301*1000)-1)*100 = -90%.

Remember that the odds is not a probability: the odds is the expected number of successes per failure, while a probability is the expected number of success per trial.

[Mina Sami]

Thank you for your useful help. I wrote the question briefly. I am what you wrote exactly:

Does the number of people killed in a region influences whether or not a firm opens a factory in that region?

But, in fact after running the command
clogit, or

I got a coefficient of killed ! (+ 0.999) not (-0.2%!)

[Maarten Buis]

That sounds about right. Based on the coefficients you reported I computed the odds ratio as exp(-0.002301)= 0.998, so an extra kills changes the odds by a factor 0.998. If you multiply something by 0.998, it will degrease by (0.998 - 1)*100% = -0.2%. The difference between 0.999 and 0.998 is probably due to rounding; what Stata reports witht he or option is more precise that the computation I did based on "only" 6 decimal digits.