Using random effects or not?

[Joel Eliasson]

Hi,

I have data of like 10k observation where each respondents have answered how many days a week ("never", "less than once a week", "1-2", "3-4", "5-7") they would consume meat if the tax increases in steps by 1eur/kg (0-10eur/kg). So 11 observations per individual. How do I best examine this in stata? If I just do a pooled ols, my data will be biased I suppose? Is the best approach to set the panel variable to the id of each respondents, or to just cluster the standard errors?

I.e., what is preferred?

reg amount tax age education income etc., r
reg amount tax age education income etc., vce(cluster id)

xtset id
xtreg amount tax age education income etc., r
xtreg amount tax age education income etc., re

Or are there other alternatives that is better?

Thanks!


(And yes I should use e.g., an ordered probit model in the end, but just want to solve the this problem first)

[Carlo Lazzaro]

Joel:
welcome to this forum.
Provided that, as you stated, you regressand is not continuous, I'd discard -regress- and give -xtreg,fe- a shot.
With such a sample size, the -vce(cluster clusterid)- standard errors are called for.

[Joel Eliasson]

Thanks Carlo.

When I try the fe approach all of my control variables are omitted due to collinearity. Is this because I don't have a time dimension?

To maybe better understand my data, I have attached a picture of the first 3 individuals (id:s).

[Carlo Lazzaro]

Joel:
the issue rest on the well-known -fe- estimator's hunger for within-panel time-invariant variables !
That said, two asides:
1) one sentence in my previous reply was misleading and shoud be rephrased as follows:

Provided that, as you stated, you regressand is not continuous, but pretending that it is so, I'd discard -regress- and give -xtreg,fe- a shot.

2) please do not post screenshots but use CODE delimiters to post what you typed and what Stata gave you back and/or use -dataex- to share an excerpt/example of your dataset. The reason undepinning these recommendations are well reported in the FAQ. Thanks.