Three Essays on Worker-Firm Dynamics

This dissertation uses three different sources of matched employer-employee (MEE) data
to study how firm-level heterogeneity influences workers’ labor market outcomes. Chapters I
and II deal with implicit contracts or long-term unenforceable agreements between firms and
workers that govern outcomes such as the split between wage and non-wage compensation,
the duration of employment relationships, worker productivity, and risk sharing. Chapter
III considers the statistical challenge of creating a MEE dataset by linking a household-level
survey with establishment-level administrative data in the absence of unique identifiers.
In chapter I, I study the determinants of retirement behavior by exploiting the recent and
widespread elimination of traditional pensions and subsequent adoption of 401(k) plans by
U.S. employers. Using thousands of firm-level natural experiments, I show that unexpected
losses in future compensation engendered by pension plan transitions induce a 1 percentage
point increase in retirement on impact. Affected workers who do not retire immediately
lengthen their careers and exhibit a 2 percentage point reduction in retirement 10 years after
the pension plan transition. Observed heterogeneity in retirement behavior is indicative of
differences in wealth and in preferences for leisure. Using these treatment effects as estimation
targets, I fit a structural model of retirement and saving in which workers with different levels
of wealth and different preferences for leisure exhibit heterogenous behavior when faced with
a pension plan transition. I use simulations from the model to show that a counterfactual
policy that eliminates Social Security payroll taxes for workers over age 60 increases the
average retirement age by one year and provides substantial welfare gains to older workers.
In chapter II, which is co-authored with Parag Mahajan, we use MEE data from Germany
to show that cohorts entering the labor market during a recession experience a 4.9 percent
reduction in wages cumulated over the first decade of labor market experience. While 40
percent of the recession-induced wage loss is explained by workers matching with lower
paying firms, we use a revealed preference-based algorithm to show that fully three-fourths
of the losses in employer-specific pay are compensated for by non-pay amenities. The higher
non-pay amenities that we associate with recessionary labor market entrants are consistent
with the view that employers that hire during business cycle downturns exhibit less cyclically
sensitive labor demand and provide greater long-term job security. Our findings indicate
that the welfare cost of labor market entry during recessions is less severe than pecuniary
estimates would imply.
In chapter III, which is co-authored with John Abowd, Joelle Abramowitz, Margaret
Levenstein, Kristin McCue, Trivellore Raghunathan, Ann Rodgers, Matthew Shapiro, and
Nada Wasi, we illustrate an application of record linkage between a household-level survey
and an establishment-level administrative dataset in the absence of unique identifiers. Record
linkage in this setting is challenging because the distribution of employment across firms is
highly asymmetric. To address these difficulties, we use a supervised machine learning model
to probabilistically link survey respondents in the Health and Retirement Study (HRS) with
employers and establishments in the Census Business Register (BR) to create a new data
source which we call the CenHRS. We use multiple imputation to propagate uncertainty
from the linkage step into subsequent analyses of the linked data. The linked data reveal new
evidence that survey respondents’ misreporting and selective nonresponse about employer
characteristics are systematically correlated with wages.