Issue of Multicollinarity in areg

[Shifa Fatimah]

Hello I am facing a problem with my regression and any help would be nice.
I follow the following commands:

1. Xtset panel year id1

(Id1 are the unique observation and year is the time invariant factor)
Than I try running the regression with the following command:

1. areg mig_dummy lagSPEI i.age_cat_5 i.wealth_index i.Gender, absorb(id1)

where i.age_cat_5 is age in 5 year categories, wealth index is in 3 categories and Gender is in three categories. However when I run the regression I get the problem:

. areg mig_dummy lagSPEI i.age_cat_5 i.wealth_index i.Gender, absorb(id1)
note: 20.age_cat_5 omitted because of collinearity
note: 25.age_cat_5 omitted because of collinearity
note: 30.age_cat_5 omitted because of collinearity
note: 35.age_cat_5 omitted because of collinearity
note: 40.age_cat_5 omitted because of collinearity
note: 45.age_cat_5 omitted because of collinearity
note: 50.age_cat_5 omitted because of collinearity
note: 55.age_cat_5 omitted because of collinearity
note: 60.age_cat_5 omitted because of collinearity
note: 65.age_cat_5 omitted because of collinearity
note: 3.wealth_index omitted because of collinearity
note: 4.wealth_index omitted because of collinearity
note: 2.Gender omitted because of collinearity

Could anyone tell me what am I doing wrong? Is my id corelated to the panel specifying variable? What are the remedies which you suggest at this stage.

[Carlo Lazzaro]

Shifa:
--xtset-ting your data before -areg- is useless;
- the omitted variables are probably collinear with -id1-.

That said, if you have an unique observation per -id- and a categorical regressand (-mig_dummy-) why using a linear regression model instead of, saying, -logit- or -probit-?

[Shifa Fatimah]

Thank you for your response.

I am trying to see the fixed effect variability in the model which is better captured through linear regression model as compared to a binary one. Moreover, my data is divided into groups i.e. same observations for the same person across a time where the ID1 is the unique identifier, therefore, I used the absorb command so that the fixed effect variation is accounted for.

Could you tell me a remedy for this?

I arrived at the areg model after doing:
xtmixed mig_dummy i.lagSPEI_24 i.age_cat_5 i.wealth_index i.Gender ||id1

But since my data is very big i.e. 28979 observations across 1099 groups it was taking very long for the regression to compute the results.

I must mention that my data is longitudinal data.

[Carlo Lazzaro]

Shifa:
thanks for clarifying.
If you actually have longitudinal data (es, the same sample is measured repeatedly along theoretically equally spaced times) and you're interested in fixed effect, and assuming that your approach makes sense (admittedly, I've never heard about that), you may want to consider -xtreg,fe-. The drawback of -fe- is that any time-invariant predictor will be wiped out (ie, you will not have coefficient for them in your regerssion output).