Multiple Logistic Regression: Interpreting Insignificant Variables

[xjuhneevuhx]

I ran a multiple regression, using the following categorical dependent variables which ranked current job satisfaction:

0=not satisfied at all
1=somewhat not satisfied
2=neutral (base outcome)
3=somewhat satisfied
4=definitely satisfied

one of my independent variables is job_number (# of jobs held in the past). there is a significant correlation (p<.05) in category 4, and category 3, but NOT significant correlation (p>.05) in category 1, and category 0.

how to i interpret this? Is the IV job_number relevant to the model at all? please help!


notes: used mlogit command on STATA 12

[Mean Joe]

I ran a multiple regression...<snip>...there is a significant correlation

I think you're using the term correlation in a layman's sense; maybe more correct to say significant relationship.

Assuming you did the multiple regression correctly, job_number is a sig. factor for a person being satisfied (category 3, and 4), but is not significant in a person unsatisfied (category 0-2).

So job_number could be relevant to the model, as it leads to significant differences between the groups of people who are satisfied/unsatisfied. Make sure the effect on each of the categories is in the same direction. eg the significant result in categories 3 and 4 should go with both positive, or both negative, effects in those categories too. If the effect in category 3 is positive, but negative in category 4, then the result looks dubious and needs a real good explanation.