Improving Data Infrastructure Through Survey Data Integration

Sam Hwang, University of British Columbia
Seunghee Lee, Korea Development Institute

Survey datasets are important for economic research because they provide information that is often missing in administrative datasets. For example, they are indispensable for investigating the mechanisms through which treatment effects materialize. However, survey datasets typically have smaller sample sizes than administrative datasets, leading to imprecise estimates. This imprecision may contribute to publication bias, where researchers refrain from publishing “null results” due to the difficulty of getting them accepted. In this paper, we show that when multiple survey datasets are available, they can be combined by reweighting observations in each dataset, substantially increasing the sample size and improving the precision of estimates. We apply our method to two survey datasets from South Korea and demonstrate that the effect of a nationwide education reform on intergenerational mobility can be precisely estimated when both datasets are used, but not when only one is used. Our method can help mitigate publication bias and improve the precision of estimates for various policies whose treatment effects are nonzero but small.

No extended abstract or paper available

 Presented in Session 185. Building Data Infrastructure