If, as you say, land and wealth distribution have not varied much in the time dimension then you are not missing much information by only having one observation date for this regressor. You could treat this as the average value for this variable. The "between" estimator averages all variables over the time dimension and applies OLS to this cross-section of cross-section of averages. There is also a "hybrid" estimator - convert all time-varying variables to deviations around their country-specific means and apply Random Effects estimation to a model that includes both these transformed variables and also, as additional regressors, the country-specific means for both the time-varying and also the time-invariant regressors. (Google "panel hybrid estimator" for more detailed explanations.)

Dear Vince,

thank you very much. My argumentation was in line with yours (i.e., the literature suggests that no much variation in wealth and land distribution took place). However I was wondering if panel data is still the best approach. I was working with xtabond2, since agricultural expansion can be better understood as a dynamic process. I will also try the hybrid regression. Thank you so much for your help.

No harm in using panel data (the more data the better it is) regardless of how small the over-time variations are. So yes panel data is still a better option and i agree with your argument there that expansion can be better understood as a dynamic process (and again panel data can allow this). At least you can account for the unobserved fixed effects with the panel data to avoid potential bias. Good luck

Thank you Hayatullah.

I'm also would apply a panel data analysis, controlling for unobserved fixed effects. If both land and wealth concentration have not varied significantly over your time interval, results should be OK

Thank you Moreno. I am actually using a dynamic panel now (arellano-bond system GMM). I think it addresses the data limitations problems.