I am performing an econometric analysis on the impact of certification participation on the income of farmers growing a specific commodity in Latin America.

A socio-economic survey has been taken on N farmers, these include: individual farmers, cooperative non-certified farmers, and cooperative certified farmers. If I want to asses the impact of the certification scheme on total income, agricultural income and livelihoods of the farmers. My question is, how do I disentangle the effect of cooperative impact and certification impact. For my specific certification scheme a farmer must be a member of a recognized farmer cooperative. So how can I asses the impact of the certification scheme, accounting for the benefits of the cooperative membership?

I have started with an OLS regression to asses any significant impact of a chosen set of variables on total income. My next step would either be:

Propensity score matching (PSM)? : does this technique allow me to disentangle the effect of certification impact from cooperative impact? My sample contains individual, cooperative farmers (no certification scheme) and cooperative farmers with certification scheme.

Instrumental variables (IV)?: from FGD’s and interviews it has become clear that management from the cooperatives has a certain selection bias to which farmers are granted this specific certification. What is more, farmers do not always know that they are listed as certified or what benefits they can receive. Is it thus possible to find a good instrument, given this selection bias and uninformed farmers?

Thank you for your input.

It seems clear that your goal is to compare the potential outcomes of cooperative farmers were they to receive certification and were they not to. The individual farmers are unrelated to your causal question (but they may provide information in the analysis). So, any analysis that would allow you to compare the potential outcomes in those circumstances would be adequate.

Whether you should use regression, propensity score methods, or IV depends on the assumptions you are willing to make. If you have a good instrument for certification, then IV may be a good option. Knowledge of being certified would not be a good instrument because it is an effect of certification, and an instrument must be a cause of certification. A plausible instrument might be whether farmers received a pamphlet encouraging them to certify.

If you have enough variables in your data set to account for confounding (i.e., all common causes of certification and total income), then you can use regression or propensity score methods to estimate the effect of certification.