newbie: logit/ologit model, significance of the independent variables

[Vitalii Korol]

Hi everybody,

I am using Stata 13 (mac student version) in logit/probit analysis for the first time. The task is to see which of the regressors have the most influence over the two dependent variables.
y_var1 is binary
y_var2 is ordinal
x_var(s) are mostly binary or ordinal

The following model were run:
logit y_var1 x_var1 x_var2 ... x_varN
ologit y_var2 x_var1 x_var2 ... x_varN

For logit, the following methods were considered for sorting significant variables:
1) variables with value P(over)|z| that is less than 0.05
2) calculation of marginal effect evaluated at the means, variables with the highest f(Z)b coefficient

For ologit:
1) variables with value P(over)|z| that is less than 0.05

Question:
I would appreciate it, if you tell me whether used methodologies for identifying the regressors' effect on dependent variables are correct?
If I am missing an important theory background, would you please point me at the material that I need to learn?

Best regards

[Ivica Rubil]

Vitalii,

check this Stata Press book: http://www.stata-press.com/books/reg...ent-variables/.
I guess you'll find there what you need.

Best
ir

[Joseph Luchman]

Hi Vitalii,

If you're interested in discerning which variables have the _most_ influence over the dependent variables - and you're reasonably confident about your models' specification - you could try using - domin - (SSC).

- domin - implements dominance analysis, which is used to evaluate independent variable importance in predicting some dependent variable. Relative importance _seems_ like what you may be looking for based on your description above. I'd recommend looking at the helpfile for -domin - to see if it would be helpful in answering your question.

- joe