Hello
I am conducting a series of ordinal regression analyses using SPSS Version 25 as I have done many times in the past. I am coming up with an error every time with the same variable and I can't work out what the problem is. If anyone could offer some advice that would be very much appreciated!
I have ran several ordinal logistic regressions with proportional odds with the same independent variables and different dependent variables. All assumptions have been met and there has been no issues when running these... except... in the results there is one variable that doesn't show up properly. This is the case for all regressions I've run with this same problem variable (which is a five point likert scale from very supportive to very unsupportive indicating how supportive people perceive their partners to be). For every regression results table this variable shows up with nothing but a dot (.) in the significance and 95% CI columns and the value 1.000 in the OR column (Exp_B).
This variable is coded as ordinal and is no different to the other variables in the regression models. I have created a new variable in case there is some error and inputted the same values and it still happens. I have used the variable in other analyses (chi square) and it worked fine. I can't work out what the problem is and why this variable won't produce results.
Any guidance would be fantastic!
Thanks so much
Perhaps you are running out of degrees of freedom in the model? I'd you have df in the output look there or try dropping some other variables from the model and see what happens. Or perhaps this variable is entirely confounded with another variable? Again you might remove some other variables from the model and see if you can isolate the confounding one.
Thank you Salvatore!
I have gone back and checked the degrees of freedom as you suggested and noticed that the problematic variable's estimate says 0a . Below the parameter estimates table is stated 'a. this parameter is set to zero because it is redundant.' What does this mean and what solution might there be?
I am going to suggest something different. Suppose that you have multicollinearity in your IVs. Then lasso variable selection will tell you that and automatically make the correction. Examples and software are described here:
https://stats.stackexchange.com/questions/84185/lasso-to-identify-important-variables-in-ordered-logistic-regression.
For a general discussion and references see the attached. Best wishes, David Booth
@JUlia the variable that has the 0 estimate is redundant in the sense that it is strongly related to other information in the model and thus can be discarded. This is one thing that lasso also does. Best, D. Booth
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS