Inferring Economic Impacts from a Program's Physical Outcomes: An Application to Forest Protection in Thailand

Abstract

Economists typically estimate the average treatment effect on the treated (ATT) when evaluating government programs. The economic interpretation of the ATT can be ambiguous when program outcomes are measured in purely physical terms, as they often are in evaluations of environmental programs (e.g., avoided deforestation). This paper presents an approach for inferring economic impacts from physical outcomes when the ATT is estimated using propensity-score matching. For the case of forest protection, we show that a protection program's ex post economic impact, as perceived by the government agency responsible for protection decisions, can be proxied by a weighted ATT, with the weights derived from the propensity of being treated (i.e., protected). We apply this new metric to mangrove protection in Thailand during 1987-2000. We find that the government's protection program avoided the loss of 12.8% of the economic value associated with the protected mangrove area. This estimate is about a quarter smaller than the conventional ATT for avoided deforestation, 17.3 percentage points. The difference between the two measures indicates that the program tended to be less effective at reducing deforestation in locations where the government perceived the net benefits of protection as being greater, which is the opposite of the relationship that would characterize a maximally effective program.

Keywords: Forest; Impact evaluation; Mangrove; Propensity score matching; Protected area; Random utility model; Roy model; Thailand; Treatment effect.

Fig. 1

Map of mangrove sampling frame. Note: Sampling frame (i.e., areas with mangrove cover in either 1987 or 2000) is shown in black. Gray boundaries demarcate Thai provinces

Fig. 2

Local illustration of the mangrove zoning system (portion of Krabi Province on Thailand’s Andaman Sea coast)

Fig. 3

Distribution of the average treatment effect for avoided deforestation (${\tau }_{AT{T}^Q}$) by quantiles of the predicted propensity score (20 quantiles)