Ologit and a very high Z Score

[sara jh]

Dear all,

I would like to kindly ask what is the reason for having a very high Z score for one variables. Actually this variable is sub variable of another variable. The main interest variable equals to one when that variable either positive or negative, while one of the control variable equals to 1 only when main interest variable is negative. The correlation between the control variable and the main interest independent variable is 89%.

Thanks

[Maarten Buis]

Large z-values are not a problem. If anything they indicate strong evidence against the null hypothesis, which is typically what you want.

However, beware of how you interpret your variables. I assume you have a variable, lets call it x, with three possible outcomes -1, 0, and 1. You have one variable of interest, lets call it i1, that is 1 when x is either -1 or 1 and 0 otherwise. Than a control variable, lets call that i2, which is 1 if x is -1 and 0 otherwise. This is summarized in the table below:

x	i1	i2
-1	1	1
0	0	0
1	1	0

If you include both i1 and i2, and you are interested in the interpretation of i1, then you interpret it as the effect of a unit change in i1 while keeping i2 constant. There are two ways in which i1 could increase by one unit, either x goes for 0 to -1 or x goes from 0 to 1. I assume if that given that you chose this unusual coding, that that is the comparison you want. However, there is only one way in which i1 could change by one unit and i2 remains constant, and that is if x goes from 0 to 1. If x goes from 0 to -1 then i2 changes from 0 to 1, i.e. does not remain constant. So if you add i1 and i2 to your model, then the interpretation of i1 is only the comparison of x=0 to x=1. For the same reason i2 is only the comparison of x=1 to x=-1.

[sara jh]

Thank you very much.

Which I mean by very large is more than 200 Z Score. Is that ok?

[Maarten Buis]

Sure, why would it not be ok?