Loading...
Essays on Developing Treatment Effects Set-ups in Interactive Fixed Effects Models
karami, sonia
karami, sonia
Citations
Altmetric:
Genre
Thesis/Dissertation
Date
2021
Group
Department
Economics
Permanent link to this record
Collections
Research Projects
Organizational Units
Journal Issue
DOI
http://dx.doi.org/10.34944/dspace/6820
Abstract
This dissertation includes three chapters. The first two chapters mainly focuson developing treatment effect set-ups in interactive fixed effects (IFE) models. The
first chapter identifies and estimates distributional treatment effects on the treated
in the IFE models using a small number of panel data. The set-up introduced
in the first chapter is utilized to examine the effect of two tax-based investment
incentives in the United States, namely bonus depreciation and section 179, on the
local labor markets unemployment rate. The second chapter provides identification
and estimation of average treatment effects on the treated in the IFE models using a
small number of either repeated cross-sections or panel data. The approach proposed
in the second chapter is employed to assess the effect of job displacement on the
earnings of displaced workers in the United States. The third chapter investigates
the average impact of the Affordable Care Act Medicaid expansion on employment
in the local labor markets using the interactive fixed effects models.
Chapter 1, tilted DISTRIBUTIONAL TREATMENT EFFECTS ON THE
TREATED SET-UPS IN INTERACTIVE FIXED EFFECTS MODELS, extends
the previous literature on the causal effect of a binary treatment by developing a
distributional treatment effects on the treated (DTT) set-up that i) it controls for the
unobservable confounding factors with either time-invariant or time-varying effects on
the outcomes of interest ii) it allows for the heteroskedasticity in the error terms iii) it
requires a short number of panel data and iv) it captures any potential heterogeneity
in the response of outcomes to a binary treatment. In this chapter, I focus on the
specific cases when some unit-specific unobserved characteristics have a time-varying
impact on the outcomes of interest. These types of unobservable factors cannot be
“differenced out” using panel data. Here, untreated potential outcomes are assumed
to follow the IFE models; that is, the models that enter the unobservable factors with
a time-varying “effect” into the regression models in an analogous way to observed
covariates. Although this chapter utilizes a special case of the IFE models with only
one time-varying unobservable, the set-up can be easily expanded to the case with
more than one unobservable time-varying factors. Here, the distributional effects
of a binary treatment are estimated through a non-parametric method built on the
relationship between the probability density function of a random variable and its
corresponding characteristic function. The DTT estimation set-up in this chapter is
shown to be consistent when the number of time periods is finite, making this set-up
applicable to a wide range of empirical questions with limited access to data over
time. The set-up introduced in this chapter is utilized to investigate the effect of
two tax-based investment incentives in the United States, namely bonus depreciation
and section 179, on the local labor markets’ unemployment rate. The results of
this chapter suggest that these two investment incentives decrease the unemployment
rate of local labor markets. Also, the heterogeneity in the causal effect of these two
incentives on the unemployment rate is small and almost negligible.
Chapter 2, titled AVERAGE TREATMENT EFFECTS ON THE TREATED IN
INTERACTIVE FIXED EFFECTS MODELS (with Brantly Callaway), extends the
previous literature on the causal effect of a binary treatment by proposing an average
treatment effect set-up using the IFE models that requires only a few numbers of
either repeated cross-sections or panel data. In particular, we focus on the cases where
untreated potential outcomes follow a special case of the IFE models with a single
unobserved confounding factor whose effect is allowed to change over time, though we
also allow for the time fixed effects and unobserved unit-specific heterogeneity. The
models that we consider in this chapter generalize many commonly used models in
treatment effects literature, including difference-in-differences (DID) and unit-specific
linear trends models. Unlike most literature on the IFE models, we do not require
the number of time periods approaches infinity to reach consistency in estimating the
ATT. Using our approach, we show that ATT can be identified with as few as three
to four time periods and with panel or repeated cross-sections data. Here, our main
identification result relies on having the effect of some time-invariant covariates (e.g.,
race or sex) not vary over time. The developed set-up in this chapter is further used
to investigate the effect of job displacement on the earnings of displaced workers in
the United States. Our results suggest that the effect of job displacement on the
earnings appears to be massive “on impact” but decays zero in four years.
Chapter 3, titled THE EFFECT OF AFFORDABLE CARE ACT MEDICAID
EXPANSION ON THE EMPLOYMENT: AN INTERACTIVE FIXED EFFECTS
MODEL APPROACH, provides new empirical evidence on the effect of Affordable
Care Act (ACA) Medicaid expansion on the United States local labor markets
employment using the IFE models. Some evidence provided in this chapter suggests
the states expanding the ACA Medicaid expansion may differ systematically from
the non-expanding ones. Some observed differences between the expanding and nonexpanding
states include the percentage of Medicaid coverage before implementing
the policy, percentage of households below the Federal Poverty Line (FPL), and
demographic characteristics. Hence, the two groups of states may differ substantially
in some unobserved characteristics, such as jobs’ benefits in different industries within
each local labor market (including the employer-based insurance coverage). The
causal effect estimation under the DID method is biased in the presence of timevarying
unobservable confounding factors. Utilizing the IFE models in treatment
effects studies has an advantage over the DID in controlling both time-invariant
and time-varying unobservable characteristics. Using the treatment effect set-up
introduced by Callaway and Karami (2020); that is, a set-up that uses the IFE
models in the identification of average treatment effects, and 2010-2018 Quarterly
Census of Employment and Wages dataset, this chapter’s results suggest that ACA
Medicaid expansion decreases the employment for almost 1% during 2014-2018 in the
states that adopt this policy. The results of this chapter demonstrate a higher effect
(by almost 2%) in some sub-populations of the original sample (e.g., industries with
the highest percentage of low-paid employees).
Description
Citation
Citation to related work
Has part
ADA compliance
For Americans with Disabilities Act (ADA) accommodation, including help with reading this content, please contact scholarshare@temple.edu