The Effect of Cancer on the Employment of Older Males in the U.S.: Attenuating Selection Bias using a High Risk Sample

Background: Obtaining unbiased estimators of the effect of health shocks on employment is an important topic in both health and labour economics. This is particularly relevant to cancer, where improvements in screening and treatments have led to increases in survival for nearly all types of cancer. In order to address the issue of selection bias, I estimate the effect of cancer on employment for a sample who are very likely to get cancer (male workers over the age of 65) thus attenuating the impact of many cancer risk factors and leading to a natural balance in observable characteristics.

Data: I use data from the Health and Retirement Study, which is a large, longitudinal, biennial, nationally representative data set. It contains information on the respondents’ health, wealth, employment, and other demographic information. For this analysis, I use the first 11 waves of data (spanning the years 1992 to 2012), and stack the waves into two simple before and after time periods.

Methods: The analysis is conducted in three stages. First, I examine the plausibility of the identification strategy with numerous tests of covariate balance. The effect of cancer on labour supply is then estimated with multiple regression and propensity score matching. Finally, I examine the correlation of cancer with potentially unobserved variables.

Results: This identification strategy leads to a natural balance in the pre-diagnosis covariates between the cancer and the non-cancer groups in numerous tests. The estimated effect of cancer on labour supply appears insensitive to different specifications and control groups. The estimates also appear to be uncorrelated with omitted confounders. It is arguable that, under these conditions, this particular estimator suffers from negligible selection bias.

Conclusion: With the number of workers working past the traditional retirement age of 65, and with the retirement age set to increase in the future, these results could provide important information on how future workers (who normally would have retired) will behave when diagnosed with cancer while in employment.