Aggregation Biases in Estimates of the Value per Statistical Life: Evidence from Longitudinal Matched Worker-Firm Data in Taiwan

This study uses a unique longitudinally-linked employer–employee dataset to estimate the magnitude of bias in estimating the value per statistical life (VSL) that arises from the conventional use of industry-average occupational risk. This unique dataset, covering workers in Taiwan over the period 1998–2002, allows us to distinguish among potential sources of bias including omitted variables and to control for the potential endogeneity of firm-specific job risk with respect to unobserved worker and firm characteristics. We find that VSL estimates based on risk data aggregated by three-digit manufacturing SIC codes are biased downward by an order of magnitude compared with estimates using firm-specific risk that control for endogeneity.