[olivia schuter]

Hey: thanks for the answer.

This is my regression:

logit life_satisfaction i.income_scale i.sex i.employment_status i.age_recoded i.kids i.marital_status i.health_status log_pcBIP i.job_satisfaction , or vce(robust)

versus

logit life_satisfaction i.income_scale i.sex i.employment_status i.age_recoded i.kids i.marital_status i.health_status log_pcBIP , or vce(robust)

When leaving out job satisfaction, the results turn statisitically significant and the coefficients get bigger (before 0,98 and then 1,23 for example; in this case from negative to positive as it is an odds ratio).
I read in literature that job and life satisfaction are reciprocally related.... As there is no good instrument variable I could use for solving this endogeneity I will just leave out job satisfaction although this also would mean inefficient results or biased results due to omitted variable bias. BUt I think it is still better than leaving it inside.

What do you think?

[Carlo Lazzaro]

Olivia:
again, -ability- may play a role in your results.
That said, if you're confident that you can defend your model against the possible "charges" of endogeneity/omitted variable bias coming from supervisors/teachers/classmates/reviewers, removing -job_satisfaction- might be a cautionary approach (the issue about a valid instrument still holds, though).
As an aside, from the (very) little I know about that, yes/no scale for happiness might have both ceiling and floor effect (there're so many grey areas in everyone's feeling of happiness...).