Tell me more ×
Cross Validated is a question and answer site for statisticians, data analysts, data miners and data visualization experts. It's 100% free, no registration required.
up vote 1 down vote favorite

I am doing multinomial logistic regression.

Dependent variable is categorical with 6 categories. 6 Independent variables are measured on an interval scale (included as covariates in SPSS)

Now I am watching the LR test. None of the independent variables have a $\chi^2$ value and no degrees of freedom are shown, so significance is not specified. (except for the constant, which is significant and has 5 df's). One variable is redundant in the parameter estimates.

When I omit the redundant IV in the model, there are $\chi^2$values, df's and significance values for the independent variables. Was the first model over-parametrised or something?

what is the explanation for this situation? And is it valid to omit one independent variable in the model?

share|improve this question
In a multinomial logistic regression, there is a relation between parameters, i.e. the summation of the parameters is equal to 1. Then, you have to consider a side constraint to make the parameters identifiable. – hbaghishani May 17 '12 at 8:05
One variable is redundant in the parameter estimates How did you know that? Were there any warnings about it printed out? Also, note that multinomial logistic regression works on the basis of subpopulation of cases defined by unique combination of values of IVs, not on the basis of individual cases. If your IVs are continuous with many values, that can pose a problem. You might better categorize them. – ttnphns May 17 '12 at 8:36
@ttnphns the output of spss stated that my last parameter is set to zero because it is redundant. The independent variables are measured on a 0-1 scale. Each independent variable contains approximately 80 unique values. Each independent variables has 3610 observations. I think it is not possible to categorize the independent variables for practical purposes. Advice? – jochem damen May 17 '12 at 9:56
1  
One way to see what is going on is to look at collinearity diagnostics, particularly condition indexes and proportions of variance explained by them. – Peter Flom May 17 '12 at 10:29
Post your SPSS output. I am suspecting you are not interpreting this right. This is the first time I hear of "redundant" explanatory variables (nothing is truly independent in social science research, so the term "independent variable" is best avoided). There is an identification issue in multinomial logistic regression, namely that only the differences of the coefficients of the same variable $k$ between different outcomes $j, j'$ are identified, $\beta_{jk}-\beta_{j'k}$, so usually the coefficients in one outcome are arbitrarily set to zero. Is this what you are talking about? – StasK May 17 '12 at 17:02
show 3 more comments

Know someone who can answer? Share a link to this question via email, Google+, Twitter, or Facebook.

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Browse other questions tagged or ask your own question.