The first question is, are you using panel or cross-section? If you are using panel data, it might be a good idea to estimate one variant of a fixed effect model (with household fixed effects), if not as the main estimation model, maybe just for robustness to complement and compare your results.

So endogeneity comes mainly from three sources: measurement errors, simultaneity, and omitted variables. Not sure if the first two would be a problem, unless you think "home ownership" is measured with errors or that happiness sore causes people to buy homes (i.e. x causes y but i don't see how y is likely to cause x). However, the omitted variable bias is maybe your potential problem. There should be uncontrolled confounding variables that are correlated with both "home ownership", and with the error term. Obviously, you can't control for everything that drives the happiness score. So, can you think of any "unbservables" that are omitted from your model or not included in your model but they are there in the error term, that could affect both home ownership and happiness score? for example, heterogeneous life styles, ability...etc

I hope this helps.

Good luck.

Hi Hayatullah, I have unbalanced panel data over 2 waves and have tried pooled OLS.

It is accurately addressed research issue. Ownership of immovable property creates responsibility and set benchmark for performance comparison from generation to generation. Rest need to assess

Thanks all. However, my question still remains to be answered.