Instrumental Variables (as Two Stage Least Squares 2SLS)

[Seed Al]

Hello guys,

Please I need help, I have questions about Instrumental Variables (Two Stage Least Squares 2SLS)

The dependent variable is log pre-death.
Independent variables are Poor or fair health (%), Adult smoking (%), Adult obesity (%), Excessive drinking (%), Uninsured underage 65 (%), college degrees (%), and Unemployment (%).
The research question is "Does unemployment lead to death? Evidence from Panel Data".

How can I define (endogenous variables, exogenous variables, and instrumental variables)?



Is this correct?
ivregress 2sls log_death poorhealth Smokers Obese ExcessiveDrinking College (unemployed = Uninsured), first

where
log_death is the dependent variable
poorhealth Smokers Obese ExcessiveDrinking and College are the exogenous variables
unemployed is the endogenous variable
uninsured is the instrumental variable

[Joro Kolev]

Yes, your command is doing what your have described that you want to do.

[Maxence Morlet]

Whether or not you're getting causal results is another issue. You are instrumenting unemployment with insurance status, which means that insurance status must have a (strong) effect on unemployment (I would have though it to be collinear but I'm really not an expert in the field), but it also means that insurance status is exogenous. That is much harder to believe in your context.

To make it more plausible you may want to check out Sergio Correia's

ivreghdfe

.

You mentioned you have panel data, right?

[Seed Al]

Thank you for your response.
Yes, I have panel data.
Do you mean the uninsured cannot be the instrumental variable?

What variables can I add to answer my question, if so?

[Maxence Morlet]

Well you're going to want to add unit and time fixed-effects (check out the community-contributed command ivreghdfe), but especially find an instrument which affects unemployment, and especially affects your outcome through unemployment only. It might the case for insurance status, although it just seems difficultly plausible...