fixed effect or random effect, comparatively small sample(N= 404 / T=2), short panel

[Chungmann Kim]

Hi all, I am trying to do the analysis on studying diversity index influencing farmer's productivity.

I have two wave(2 years gap) and 202 balanced samples(total 404) for each wave.

Since I thought diversity index has endogeneity problem, I used IV for the analysis.

Now, i have problem whether to choose random or fixed effect in my situation..

If I run with FE which consumes a lot of obs(202) in my case, coefficients are against our common sense and few of them are significant...

but random effect? results are acceptable and significant....(also same when estimated with pooled OLS)


However, if I run Hausman Test, the results are saying that fe is more suitable...


xtset hhid year, yearly
xtivreg ln_yield_val_ ///
/*socio-economi*/ hhhead* education minority ///
/*agricultural input as value*/ ln_labor4total ln_irri4total ln_cheftz4total ln_orgftz4total ///
/*agricultural input as quantity(kg)*/ culareaT ///
/*risk*/ creditcard ///
/*aid*/ t1_aid_mach ///
/*weather variability*/ m_dro3_10_ky pr_CoV_ky ///
/*diversity*/ (diversity_s = diversity6 SRF LRCoVmean_5), fe (or re)

estimate store fe
estimate store re
hausman fe re, sigmamore


My question is,

1) Is my approach reasonable? running xtivreg and hausman test to check which is more suitable?

2) Do I need to stick to the fe, even though results are against the theory?

3) Any other suggestions or insights on this analysis?


Kindly give me some advice.

Many Thanks,

Manny

[Phil Bromiley]

You didn't get a quick response. You'll increase your chances of a useful response by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

This is a bit of a tricky issue. With fe, you're only seeing the effect of changes in diversity within a farm on productivity. However, it may be that much of the action in diversity is associated with stable farm level differences. The Hausman essentially tests whether the within effects equal the between. But there may be good reasons to think that the effects are different. Within reflects weather, prices (if you use value of production), and perhaps fine tuning of the amount of different crops or other outputs produced, but between may reflect basic choices in farm plan and business model.

You might consider xthybrid which estimates both at once. I'm not sure if it handles endogeneity, but if it doesn't, you can manually do an equivalent estimate by including the panel means as regressors using 2SLS.