---

On December 2, 2025, the Global Network Seminar was held, and the seminar welcomed Professor Insik Min from the Department of Economics at Kyung Hee University.

Under the theme “Understanding Impact Evaluation in Vulnerable Labor Markets: Evidence from Vietnam,” Professor Min discussed selection bias and approaches to addressing it in policy impact evaluation using observational data, drawing on evidence from Vietnam’s labor market and social insurance system. The lecture was organized around three key questions: “Who works?”, “Who participates?”, and “Who benefits more?”

In the first part of the lecture, Professor Min explained sample selection bias using the returns to education in Vietnam as an example. He highlighted that labor market participation is not random and that analyzing wages only among those who are employed may lead to biased estimates because the decision of whether to work is not taken into account. As an approach to addressing this issue, he introduced Heckman’s two-step estimation method. Using data from the 2020 Vietnam Household Living Standards Survey (VHLSS), the analysis identified statistically significant selection bias and demonstrated that conventional OLS estimates that do not account for this selection may overestimate the true returns to education in the formal sector.

The lecture then addressed self-selection and the estimation of policy effects when participation in a policy or program is determined by individuals’ voluntary choices. Using Vietnam’s Voluntary Social Insurance (VSI) as an example, Professor Min explained that simply comparing participants and non-participants does not allow researchers to distinguish the effect of the program itself from pre-existing differences in individual and household characteristics. He therefore introduced methods such as Nearest Neighbor Matching and Propensity Score Matching (PSM), which compare participants and non-participants with similar observable characteristics, including age, income, urban or rural residence, and household size. He also discussed the limitations of matching approaches, particularly their inability to fully account for unobservable factors such as risk preferences and preferences related to health.

In the latter part of the lecture, Professor Min moved beyond average policy effects to discuss treatment effect heterogeneity, focusing on the question of “who benefits more.” He introduced the Conditional Average Treatment Effect (CATE) as an approach for examining how the effects of VSI participation vary across individuals and groups with different characteristics. Using VHLSS 2020 data and healthcare utilization as an outcome, he demonstrated how individual-level treatment effects can be estimated while considering characteristics such as age, gender, education, income, residential location, occupation, and health status. The results revealed differences in treatment effects across individuals that could not be captured by the average treatment effect alone. While positive effects were observed for many individuals, the magnitude of these effects varied considerably.

The seminar provided participants with a valuable opportunity to understand the importance of appropriately accounting for sample selection and self-selection when evaluating labor market and social policy interventions using observational data. Through empirical applications of Heckman’s two-step estimation method, matching approaches, and CATE using actual data from Vietnam, participants also deepened their understanding of the importance of examining not only the average effects of policies, but also “for whom” and “to what extent” those effects occur.