Reverse Causality Test with Panel Data

[Carys Wright]

Hi there,

I was wondering if it was possible for some assistance on how to carry out a reverse causality test with panel data from 1997-2016. I am aware the Granger model uses VAR - but how does this work with panel data as when I try and log my dependent variable I get syntax error ' repeated time values in sample'.

My original regression is examining how the consumption of gin affects alcohol-related deaths, controlling for total expenditure on healthcare (TEH), gross domestic household income (GDHI) and drinking behaviour (DB).

ARD_it = a_o +a₁Consumption_it +a₃TEH_it +a₂GDHI_it +a₁DB_it + dyear +e_1it

The coefficients between ARD and TEH suggest a strong positive correlation - so I want to examine the reverse causality between this.

Any help would be greatly appreciated.

Carys

[Phil Bromiley]

One way is to estimate the models with xtivreg or xtivreg2. You set up both causal models. You use appropriate controls and xtivreg should give you consistent estimates. Alternatively, xtivreg2 has a built in test for endogeneity.

[Carys Wright]

Hi,

Thanks for your response. Yes - I've used instrumental variable test within my analysis. But how do i run a reverse causality test?

Would i use TEH as a instrumental variable for what other dependent variable?

If you could go in to more detail that would be great.


Carys