Dice Question Streamline Icon: https://streamlinehq.com

Direction of selection bias from conditioning on fully attached US earners

Ascertain the direction and magnitude of selection bias introduced by restricting analyses to personnel who remain fully attached to the US labor market (i.e., have positive US earnings in every post‑interruption year) when estimating the effect of NIH R01 funding interruptions on personnel earnings.

Information Square Streamline Icon: https://streamlinehq.com

Background

To estimate earnings effects while lacking data on earnings outside the US, the paper analyzes a subsample of personnel who are continuously employed in the US after interruption. The authors note that such conditioning may introduce selection bias, but the direction of that bias is not clear.

Quantifying and addressing this bias is important for accurately measuring how funding interruptions affect earnings trajectories, especially given differential mobility across borders and sectors among treated and comparison groups.

References

Conditioning on positive earnings after treatment raises concerns about whether changes in sample composition bias our estimates. Though we cannot definitively determine the direction of the bias, we suspect that omitting personnel with true nonemployment spells is likely to dominate other sources of selection into and out of the fully-attached sample, in which case our estimate is a lower bound on the true effect of interruptions on earnings.

— Scientific Talent Leaks Out of Funding Gaps (2402.07235 - Tham et al., 11 Feb 2024) in Earnings and Job Mobility, subsubsection “Earnings” (Wage outcomes)