Mobile Menu
Back to Blog

On Balance: Publication Selection Biases in Stated Preference Estimates of the Value of a Statistical Life
02/18/2021

By W. Kip Viscusi, Vanderbilt Law School

International studies valuing mortality risk changes often rely on stated preference estimates of the value of a statistical life (VSL). Because labor market data in most countries are not as reliable as the fatality rate statistics in the United States, stated preference evidence for the VSL provides a popular research strategy for obtaining country-specific estimates. Unfortunately, this article finds that this literature is subject to rampant publication selection effects, leading to huge biases in the estimated VSL levels.
 
Publication selection biases may arise at different stages of the research process. If, for example, researchers and editors use U.S. estimates of the VSL as the reference point for reasonable estimates of the VSL, the U.S. value will influence which estimates researchers choose to submit to journals and which estimates are accepted for publication and ultimately published. Previous research has found that publication selection effects plague the VSL literature overall (Stanley and Doucouliagos, 2012), but do not significantly affect labor market estimates of the VSL based on the Census of Fatal Occupational Injuries data (Viscusi, 2015). 

This article examines the potential presence of publication selection effects using a sample of 1,148 stated preference estimates of the VSL from throughout the world. If there are no biases, estimates of the VSL should be symmetrically clustered around the true VSL level. In the case of the stated preference data, the distribution is highly skewed, with many precisely estimated values being near zero, and with a long tail of imprecisely estimated high values of VSL. Using standard statistical tests to adjust for publication selection effects, the article finds that the mean VSL level is reduced from $8.5 million (in 2015 USD) to a bias-adjusted value of $3.2 million. The extent of the publication selection biases is not uniform. High international VSL estimates are most seriously affected. At the 90th percentile of the distribution of VSL estimates, the mean estimate is $15.8 million, but the bias-corrected value is $4.2 million.

Similar biases also affected the labor market estimates in the U.S. literature. However, in that case, it was feasible to identify a subset of studies not significantly affected by such biases. To explore whether international stated preference studies similarly have a set of studies that is not distorted by publication selection effects, the authors also considered ten different subsamples of studies, such as stated preference studies published in peer reviewed journals, estimates pertaining to health risks or environmental risks, and estimates from lower and medium income countries. All these groups were subject to substantial publication selection biases, which sometimes reduced the estimated VSL by up to 80%.

What then are researchers and policy analysts to do when valuing mortality reductions outside the U.S.? The approach advocated by the authors is to use the U.S. estimates based on the CFOI data as the baseline, and then to extrapolate these values to other countries using their estimated international income elasticity of 1.0. Tables of such estimates appear in Masterman and Viscusi (2017) and Viscusi (2019). Notwithstanding the downward adjustments for income relative to the U.S., the projected VSL levels generally exceed the VSL amounts currently used outside the U.S. (Viscusi, 2018).

Using an average international income elasticity value has the advantage that the procedure is readily transferable and is based on income differences, which are a principal driver of variations in the VSL. The international elasticity value of 1.0 is also robust, as it is reflected in both stated preference and revealed preference estimates. However, the procedure ignores elasticity differences that may arise because of international differences in attitudes toward risk, such as religious and cultural factors (Hammitt and Robinson 2011).

The evidence of biases in the stated preference literature on the VSL does not imply that stated preference studies have no valid policy role. Other health risks that have been the focus of stated preference studies may not be subject to biases such as those that arise from anchoring on the revealed preference VSL estimates. There are many health risks such as risks of cancer for which reliable stated preference evidence is not available. Stated preference studies can continue to serve a complementary role (Alberini, 2019; Viscusi and Dalafave, 2020).
Post a new comment
CAPTCHA Image
Refresh Image