How do employees fare in leveraged buyouts? Evidencefrom workplace safety records∗
Jonathan CohnThe University of Texas-Austin
Nicole NestoriakBureau of Labor Statistics
Malcolm WardlawThe University of Texas-Dallas
April 17, 2016
Abstract
This paper presents evidence that workplace injury rates decline substantially afterLBOs of publicly-traded firms but not of already-private firms. The decline is greaterfor firms likely facing greater pressure from financial markets pre-LBO to focus on short-term performance. The results suggest that LBOs enable previously-public firms tobecome more far-sighted in spending on activities that improve workplace safety, withpositive consequences for employee well-being. The decrease in injury risk could partlyexplain the recently-documented fall in employee earnings after public firm LBOs.
∗Jonathan Cohn: [email protected], (512) 232-6827. Nicole Nestoriak: [email protected], (202) 691-7408. Malcolm Wardlaw: [email protected], (972) 883-5903.We would like to thank Andres Almazan, Aydoğan Altı, Christa Bouwman, Andres Dongangelo, CesareFracassi, John Griffin, Mark Jansen, Adam Kolasinski, Sam Krueger, Tim Landvoigt, Inessa Liskovich, Al-bert Sheen, Laura Starks, Sheridan Titman, Mindy Zhang, and seminar participants at the University ofAmsterdam, University of Texas at Austin, and Texas A&M for comments.
How do employees fare in leveraged buyouts?Evidence from workplace safety records
Abstract
This paper presents evidence that workplace injury rates decline substantially afterLBOs of publicly-traded firms but not of already-private firms. The decline is greaterfor firms likely facing greater pressure from financial markets pre-LBO to focus on short-term performance. The results suggest that LBOs enable previously-public firms tobecome more far-sighted in spending on activities that improve workplace safety, withpositive consequences for employee well-being. The decrease in injury risk could partlyexplain the recently-documented fall in employee earnings after public firm LBOs.
The contribution of private equity (PE) firms to the economy is the subject of a long-
standing debate between two narratives. PE executives argue that they creates value by
solving agency conflicts within corporations and improving operational efficiency (Gompers,
Kaplan, and Mukharlyamov, 2015), in sync with the arguments of prominent economists
such as Jensen (1989). However, others argue that PE firms merely expropriate rents from
employees, customers, and other non-financial stakeholders of the companies they acquire by
abrogating implicit contracts (Shleifer and Summers, 1988), without actually creating (and
possibly destroying) value. Politicians often weigh in on this debate as well. Former Danish
Prime Minister Poul Rasmussen argues that PE buyouts “leave the company saddled with
debt and interest payments, its workers are laid off, and its assets are sold, ... benefiting
neither workers nor the real economy.”
This paper adds to the discussion by studying the impact of PE-backed leveraged buyouts
(LBOs) on workplace injury rates at acquired firms, using establishment-level data from
the Bureau of Labor Statisics’ (BLS’) annual Survey of Occupational Injuries and Illnesses
(SOII). Recent estimates place the annual cost of workplace injuries at over $250 billion
in the U.S. alone (Leigh, 2011), making understanding the drivers of workplace injury risk
an important research objective. Studying injury rates also allows us to assess both of
the narratives in the debate over PE. Improvements in operational efficiency after LBOs
could lead to improved workplace safety and hence lower injury rates. On the other hand,
workplace safety is generally governed by implicit contracts with employees, and abrogation
of these contracts by PE buyers could result in higher injury rates.
Studying U.S. LBOs between 1997 and 2007, we present evidence that LBOs of publicly-
traded companies are accompanied by a large decline in injury rates. We also present ev-
idence that this decline is driven in part by alleviation of pressure from public markets to
cut spending, including spending on activities that may enhance workplace safety. Further
supporting this interpretation, we observe no clear changes in injury rates around LBOs
1
of already-private firms. While we cannot dismiss the possibility that some omitted factor
associated with being acquired in an LBO drives the results, we present evidence contrary to
several specific alternative explanations. Overall, our analysis suggests that, by taking com-
panies private and alleviating pressure to behave myopically, LBOs can lead to significant
improvements in workplace safety.
We begin by comparing trends in injury rates at establishments of firms acquired in
LBOs and similar-sized “control” establishments in the same 4-digit SIC code industry.1
For public-firm LBOs, pre-buyout injury rate levels and trends are similar across the two
groups. They remain flat in control establishments in the post-buyout period, but fall in LBO
establishments in the second year post-buyout and remain lower through at least the fifth year
post-buyout. These patterns hold for both injuries in general and specifically for relatively
serious injuries — those resulting in time away from work or temporary reassignment of
duties. Difference-in-differences estimates support a 0.86 percentage point fall in average
injury rate from the five years before to the five years after a public firm LBO, controlling
for establishment characteristics and establishment and industry-year fixed effects. This
represents a 17% fall relative to the mean injury rate for the sample period.
In contrast, injury rates are actually higher in the five years after LBOs of already-private
firms than in the five years before. However, the increase appears to reflect the continuation
of a pre-existing trend, with rates converging towards those of control establishments, rather
than a discrete shift around the time of the LBO. The lack of a decrease in injury rates after
private firm LBOs suggests that there may be something special about going private that
drives the decrease in injury rates after public firm LBOs.2
An LBO of a public firm might decrease injury rates in part because PE owners are freer1Industry is assigned at the establishment level in the BLS data. This allows us to control for differences
in industry practices at a finer level than a firm-level industry designation would allow.2Further exploring why injury rates are lower and upward-trending in the establishments of private firms
subsequently acquired in LBOs than in other establishments would be challenging because of a lack generaldata on private firms.
2
than public corporate managers to make long-run investments without concerns about the
impact on short-run profitability. Because of concerns about job loss, compensation loss,
reputation loss, or becoming a takeover target, managers may even forgo even positive NPV
investments that would decrease earnings in the short-run (Stein (1988), Graham, Harvey,
and Rajgopal (2005)). Curtailed spending on activities such as equipment maintenance,
employee training, and safety monitoring would be expected to adversely affect workplace
safety. So might decreased capital investment, as newer technologies tend to be safer than
older ones.3 Consistent with the argument that LBOs unlock long-run investment, Lichten-
berg and Siegel (1990) find an increase in research and development after LBOs.
If lengthening of investment horizon leads to decreased injury rates after public firm
LBOs, then we should observe a larger decrease when managers face more pressure to focus
on short-term performance pre-LBO. Consistent with this explanation, we find that injury
rates decline more after LBOs of firms with more short-horizon institutional shareholders and
stock analyst coverage (possible causes of such pressure), and more discretionary accounting
accruals (a possible response to such pressure). While we cannot rule out the possibility that
such characteristics are correlated with other factors that might lead to larger reductions in
injury rates after LBOs, it is difficult to conceive of what those factors might be.
Even though injury rates decline after establishments are acquired in public firm LBOs,
these LBOs might still lead to an overall increase in injury risk if they increase employment in
higher-injury rate establishments relative to lower-injury rate establishments. We therefore
also examine how employment at the establishments in our sample evolves after public firm
LBOs. Consistent with Davis, et al. (2014), we find that employment drops overall on average
after public firm LBOs. We find that this drop is larger in establishments with low pre-LBO
injury rates, consistently with a proportional increase in employment at high-injury rate3See Section 1 of Cohn and Wardlaw (2015) for a detailed discussion of how investment can impact
workplace safety.
3
establishments. However, the aggregate injury rate when we pool employees across all LBO
establishments falls substantially after LBO relative to control establishments, suggesting
that the within-establishment drop in injury rates is the dominant effect.
We consider several factors that could complicate interpretation of the drop in injury rates
after public firm LBOs. It is difficult to distinguish given jobs becoming safer from from
the elimination of dangerous jobs, as we lack employee-level data. However, the elimination
of more jobs in safe than dangerous establishments might be difficult to reconcile with a
systematic elimination of dangerous jobs. Underreporting due to pressure form PE owners
is difficult to reconcile with the lack of injury rate decreases after buyouts of private firms.
Injury rates decline less in establishments that decrease employment post-LBO, suggesting
that underreporting due to concerns about layoffs is unlikely to explain the results.4 Selection
by PE buyers of LBO targets that would have improved even absent the LBO does not appear
to account for the concentration of the decrease in injury rates in establishments of firms
facing more pressure to invest myopically. Indeed, one might expect injury rates to increase
in these firms over time in the absence of a change in investment policy.
Our results may partly explain the modest decrease in average employee compensation
after LBOs of public firms documented by Davis, et al. (2014). Existing evidence of a
compensating wage differential for injury risk (see the survey of Viscusi and Aldy (2003))
suggests that employees may be willing to accept a reduction in compensation in exchange
for improved workplace safety. A decrease in compensation demanded due to a decrease in
injury rates could explain up to 65% of the drop in average compensation, though we lack
wage data with which to test this channel explicitly.
Our paper contributes to a growing literature examining the impact of PE ownership
on a firm’s employees. In addition to Davis, et al. (2014), others such as Kaplan (1989),4Consistent with such reporting incentives, Boone, et al. (2011) show that workers are less likely to report
moderate workplace accidents when they believe that the likelihood of layoffs at their employer is high.
4
Muscarella and Vetsuypens (1990), Lichtenberg and Siegel (1990), Wright, Thompson, and
Robbie (1992), Amess and Wright (2007), Boucly, Sraer, and Thesmar (2011), and Davis,
et al. (2014) study changes in employment around LBOs, with mixed results. Agrawal and
Tambe (2014) find that employees of LBO-acquired firms are subsequently employed for a
longer amount of time over their careers than employees of non-acquired firms. Our paper
adds to this literature by providing evidence that LBOs are accompanied by improvements
in workplace safety, an important dimension of employee well-being.
Our paper also contributes to the literature examining the operational efficiency conse-
quences of PE buyouts. Lichtenberg and Siegel (1990) and Davis, et al. (2014) find substan-
tial improvements in total factor productivity after LBOs in the 1980s and in the 1990s-2000s,
respectively. Similarly, Kaplan (1989), Muscarella and Vetsuypens (1990), Smith (1990), and
Smart and Waldfogel (1994) find significant improvements in profitability measures after U.S.
public firm LBOs in the 1980s, as do studies of LBOs in Europe (Wright, Thompson, and
Robbie (1992), Amess and Wright (2007), Boucly, Sraer, and Thesmar (2011)). However,
Guo, Hotchkiss, and Song (2011) and Cohn, Mills, and Towery (2014) fail to find consistent
improvements in these metrics after U.S. public firm buyouts in the 1990s and 2000s. Bern-
stein, et al. (2015) find that industries in which PE firms invest tend to grow, suggesting
spillover effects within industry. Bernstein and Sheen (2015) find that restaurants commit
fewer health violations after being acquired by a PE firm. Our paper adds to the evidence
that LBOs of public firms are accompanied by operational improvements.
Finally, our paper contributes to a small literature examining how the allocation of a
firm’s financial claims impacts workplace safety. Cohn and Wardlaw (2015) present evidence
that financing constraints, including those induced by leverage, can contribute to higher
workplace injury rates.5 This may appear at odds with the decrease in injury rates after5Nie and Zhao (2015) find that on-the-job deaths among Chinese coal miners increase with employer
leverage.
5
LBOs, which are leverage-increasing. However, LBO debt may have limited impact on
investment because it is typically concentrated and therefore easier to renegotiate. There is
also evidence that PE owners inject capital into their portfolio companies both at the time
of the buyout and afterwards as needed (Hotchkiss, Stromberg, and Smith (2014), Cohn,
Mills, and Towery (2014)), which should loosen financing constraints. Our paper is the first
of which we are aware to examine the impact of ownership on workplace safety, and suggests
that ownership may be of first-order importance.
The remainder of the paper proceeds as follows. Section 1 describes the data and sample
we use in the study. Section 2 presents a series of figures showing the evolution of injury rates
and other variables in LBO and control establishments. We present difference-in-differences
analysis in Section 3. In Section 4, we conduct cross-sectional analysis to further explore
the causes of changes in injury rates around LBOs. Section 5 presents analysis of changes in
establishment employment characteristics after LBOs. Finally, Section 6 concludes.
1 Data and Sample
In this section, we describe the data that we use in the paper and the sample construction
process. We construct our sample by combining data on whole-firm LBOs with annual
establishment-level injury data from the BLS’ SOII. We obtain data on public firm LBOs
from Cohn, Mills, and Towery (2014) and on private firm LBOs from Cohn, Hotchkiss, and
Towery (2015). These papers build samples of LBOs using data from SDC Platinum and
Dealogic, supplemented with news articles to verify the classification of each transaction and
remove improperly classified transactions. Both samples consist of LBOs of non-bankrupt
U.S. corporations with at least $10 million in assets. The public LBO sample includes LBOs
between 1995 and 2007, while the private LBO sample includes LBOs between 1995 and
2011.
6
The BLS conducts the SOII each year by collecting injury and illness data based on
OSHA recordkeeping requirements. This involves gathering data for hundreds of thousands
of establishments each year in a stratified sampling process. Employers covered under the
Occupational Safety and Health Act and employers selected to be part of the BLS survey are
required to maintain a log recording any injuries “that result in death, loss of consciousness,
days away from work, restricted work activity or job transfer, or medical treatment beyond
first aid.” These employers must make their injury logs available to OSHA inspectors and
supply the data contained in the log to the BLS. The SOII is used primarily to produce
aggregate statistics on the state of occupational risk in various industries in the U.S. Annual
establishment-level SOII data is available for the period 1996 through 2012.
Table 1 shows the percentage of injuries in the U.S. in 2012 by different causes (Panel
A) and types (Panel B) as reported in the BLS’ annual news release on employer-related
workplace injuries and illnesses. The leading causes of workplace injuries are contact with
objects, falls, and physical over-exertion, while the most common injury types are sprains,
strains or tears, soreness and pain, bruises and contusions, cuts and lacerations, and fractures.
— Table 1 here —
Each establishment in the SOII data has a unique identifier. Each establishment-year
record includes establishment name, location, SIC code, number of injuries (Injuries),
number of injuries resulting in days away from work, restricted activity, or job transfer
(DARTInjuries), average number of employees (Employees), and total number of hours
worked (HoursWorked). We use this data to construct annual measures of the injury rate
at each establishment. Our primary injury rate measure is Injuries/Employee, which is
Injuries divided by Employees. We also construct the measureDARTInjuries/Employee,
which is DARTInjuries divided by Employees, and which captures the rate of relatively
serious injuries. Finally, we compute Log(Employees), which is the natural log an establish-
7
ment’s reported average employment over the year, and HoursWorked/Employee, which is
HoursWorked divided by Employees. The only firm-level identifier in the SOII data is the
parent firm’s employer identification number (EIN).
As the BLS data begins in 1996, we only consider LBOs taking place in 1997 and later.
As the public LBO sample ends in 2007, we only consider LBOs taking place in 2007 and
before in both the public and private firm LBO samples for consistency. Thus our LBO
sample period is 1997-2007. This period includes the large LBO wave of the mid-2000s. We
start with 317 public firm LBOs and 555 private firm LBOs.
Before merging the LBO data with the BLS data, we remove LBOs of firms in the finance
industry (12 public firm LBOs and four private firm LBOs) or that engage in franchising (20
public firm LBOs and four private firm LBOs). We make the latter determination by visiting
company websites and searching for other information on the Internet regarding franchising
opportunities. Removing franchisers is important, as a franchiser may have limited control
over the operational practices of its franchisees.6 This leaves us with 285 public firm LBOs
and 547 private firm LBOs that we then attempt to merge with the BLS data.
For public firm LBOs, we are able to use EINs from Compustat to match some estab-
lishments in the BLS data to LBO firms. However, Compustat provides only a single EIN,
while firms often have multiple EINs, and different establishments belonging to the same
firm often report different EINs. An added challenge is that EINs are only available in the
BLS data for the period 2002-2012. To address this limitation, we assign a parent firm to an
establishment-year in the 1996-2001 period if the establishment is matched to that parent
firm based on EIN for any year after 2001. We refer to matches based on EIN as “EIN
matches.”6Bernstein and Sheen (2015) compare company-owned and franchised restaurants within the same restau-
rant chain in their assessment of the impact of PE buyouts on restaurant health code violations in order tocontrol for chain-specific factors. We cannot employ this approach, as our data do not allow us to identifywhether a given location is company- or franchisee-owned.
8
After identifying EIN matches, we obtain additional matches by manually comparing
each LBO firm’s name to establishment names in the BLS data. In addition to looking for
obvious matches, we use information from corporate websites, Bloomberg Business, and news
articles to identify other names under which a firm might operate. If we cannot determine
with near certainty that an establishment belongs to a given LBO firm, we do not create the
match. We refer to matches based on name as “Name matches.”
Finally, for each establishment matched on the basis of name, we search for other
establishment-years in the BLS data with the same EIN. We assume that these have the
same parent company, as an EIN is unique to a company. We check these by hand to verify
that they are likely matches. However, we note that these matches may be slightly less
reliable than EIN and Name matches. We refer to these as “Name-EIN matches.” The result
of the matching is 13,682 establishments (26,419 establishment-years) matched to 242 public
LBO firms.
Matching private firms acquired in LBOs is more challenging. Noting that Compustat
generally covers only publicly-traded firms, we do not have an EIN with which to match
private firms to BLS establishments. We therefore can only obtain Name and Name-EIN
matches. We are able to match 5,718 establishments (11,292 establishment-years) to 316
private LBO firms.7
We focus on changes in injury rates over a window from five years before to five years
after a buyout. We therefore eliminate all LBO establishment-years outside of the five-
year window on either side of the LBO. We also eliminate LBO establishments not in the
BLS data at least once in the five years before and at least once in the five years after the
LBO. This reduces the sample to 4,668 establishment-years in the public LBO sample and
2,298 establishment-years in the private LBO sample. In our main analysis, we eliminate7The resulting link files for both public and private LBO establishments are stored at the BLS and can
be made available to researchers on-site.
9
LBO establishments with fewer than 100 employees in the most recent pre-buyout year in
the injury data and their associated control establishments. Meaningful injury rates are
difficult to calculate for small establishments, as the inability of an employee to suffer a
fractional injury results in a preponderance of both zero and very high injury rates for these
establishments.8 This leaves us with 2,182 establishment-years for the public LBO sample
and 1,429 establishment-years for the private LBO sample.
One challenge in assessing the impact of LBOs on injury rates is a powerful downward
trend in injury rates over time. To address this challenge, we construct control samples of
establishments not acquired in buyouts but in the same industries as the LBO establishments
and of similar size. We construct separate control samples for public firm and private firm
LBOs. We then compare changes in injury rates at LBO establishments around the time of
the LBO to changes in injury rates in control establishments.
Note that, because the establishments surveyed in the SOII are a sample, a given estab-
lishment is likely to appear in the data in some years but not in others. To construct the
control sample, for each LBO establishment, we first identify all non-LBO establishments
with the same 4-digit establishment-level SIC code that are in the BLS data for at least the
same years during the five-year window on each side of the buyout year. For example, if an
establishment in the same industry belongs to a firm taken private in an LBO in 2003 and is
in the BLS data in 1999, 2002, 2005, and 2007, we include as possible control establishments
only non-LBO establishments which are also observed in the BLS data in 1999, 2002, 2005,
and 2007. A non-LBO establishment fitting this criterion might also be in the BLS data in
additional years, say 2001 and 2008, but we ignore these non-overlapping years. Matching
on the exact years ensures that the distribution of observation years in the LBO and control
samples are the same.
Finally, we choose as matched control establishments for each LBO establishment the five8In a robustness check, we relax the minimum threshold to 50 employees.
10
potential controls that are closest in size (smallest absolute difference in Log(Employees))
in the last pre-buyout year in the data. We exclude cases where reported employment
is greater than twice or less than half that of the LBO establishment. If fewer than five
potential controls satisfy this restriction, we use as many as are available. Overall, this
process ensures that control establishments are of approximately the same size and exactly
the same industry as treated establishments. The resulting final public LBO sample consists
of 2,182 LBO establishment-years and 8,218 control establishment-years. The final private
LBO sample consists of 1,429 LBO establishment-years and 5,385 control establishment-
years.
It is worth noting that we do not require that a control establishment matched to a pub-
lic (private) firm LBO establishment also belong to a publicly-traded (private) firm itself.
We refrain from imposing this constraint primarily because of data limitations. Attempt-
ing to match all public firm years to establishment-years to enforce this contraint would
be prohibitively time-consuming. Moreover, beginning with a broader sample of possible
control establishments allows us to more closely match based on important establishments
characteristics such as establishment industry and size. We believe that this provides a more
comparable control group across the primary determinants of injury exposure.
Table 2 describes the sample of LBO firms. Panel A shows the number of public and
private LBO firms at each step in the sample construction process. Panel B shows the
number of LBOs in the final sample by year. Panel C summarizes financial statement
statistics calculated from Compustat for the 137 public LBO firms represented in the final
sample.9
— Table 2 here —
Table 3 describes the final sample of establishments. Panel A reports the number of9There is no corresponding summary for private LBO firms because Compustat covers only publicly-
traded firms.
11
LBO establishments in the final sample by the approach used to match it to an LBO firm.
For the public LBO sample, most of the LBO establishments are matched on either EIN or
Name. However, for the private LBO sample, a majority are matched on EIN-Name. As
noted previously, these matches may be slightly less reliable than those matched on EIN or
Name. Panel B tabulates the number of control establishments for each LBO establishment
in the sample. It shows that most LBO establishments in both the public and private LBO
samples are matched to five firms (the maximum number allowed). Panel C reports means
of different variables in LBO and control establishments, as well as the results of t-tests for
differences in means. Panel D reports the number of establishment-year observations in each
year relative to the year of the LBO.
Despite the fact that we match only on establishment industry and size, Panel D shows
that LBO and control establishments in the public LBO sample are similar on all observed
dimensions, including injury rates in the pre-buyout period. While we cannot rule out the
possibility that the two groups in these experiments differ on unobservable dimensions, we
take comfort in the fact that they are similar on observable dimensions.
— Table 3 here —
Establishments in the private LBO sample are also similar to control establishments in
terms of size and work intensity. However, private firm LBO establishments have consid-
erably lower injury rates than control establishments in the pre-buyout period. This raises
doubts about whether we can reasonably treat establishments as being randomly assigned
to the treatment and control groups in the case of the private LBO sample.
2 The Evolution of Injury Rates Around LBOs
We begin our analysis by presenting a series of plots of injury rates at LBO and con-
trol establishments in each year around the buyout year. Figure 1 shows these plots
12
for the public LBO sample. Figures 1a and 1b plot mean Injuries/Employee and
DARTInjuries/Employee, respectively. Figures 1c and 1d plot industry-adjusted rates,
where we first substract the mean rate for establishments in the same year and 4-digit SIC
code industry. The points in these latter two plots are equivalent to the mean residuals from
a regression of injury rates on industry-year fixed effects.
— Figure 1 here —
The plots show similar patterns. While there is independent variation in injury rates
across LBO and control establishments pre-buyout, there are no obvious differences in
trends in the pre-buyout period, suggesting that the parallel trends assumptions required
for difference-in-differences estimation to be valid is likely to be satisfied.10 The figures also
show that injury rates for LBO establishments fall below those of control establishments in
the second year post-buyout, and continue to remain below through the fifth year after the
buyout. All show patterns consistent with injury rates declining in acquired establishments
after LBOs relative to those of non-LBO establishments.
Figure 2 shows the analogous plots for the private LBO sample. As already noted, in-
jury rates at private firm LBO establishments are considerably lower than those of control
establishments before the buyout. They exhibit a strong upward trend throughout both
the pre- and post-buyout period, effectively converging towards the injury rates of control
establishments several years after the buyout. These differential trends, combined with dif-
ferences in average pre-buyout injury rates, complicate interpretation. We do not, however,
observe any obvious sharp changes in injury rates at LBO establishments, relative to control
establishments, right around the time of or shortly after private firm LBOs.
— Figure 2 here —10The large difference in injury rates five years before the buyout is difficult to explain and appears to be
idiosyncratic. The results we present in the remainder of the paper are essentially unchanged if we omit yeart − 5 (see Appendix Table AIV).
13
Finally, Figure 3 plots trends in employment and worker utilization at LBO and control
establishments in the years around the LBO. Figure 3a plots Log(Employees) for the public
LBO sample. This figure shows a decrease in employment around the LBO, consistent
with the conclusions of Davis, et al. (2014). Figure 3b plots HoursWorked/Employee for
the public LBO sample. There are no consistent patterns here. Figures 3c and 3d show
the analogous figures for private firm LBOs. There is no indication of a change in either
Log(Employees) or HoursWorked/Employee after these buyouts.
— Figure 3 here —
While the patterns presented in Figure 1 suggest a significant drop in injury rates starting
the second year after public firm LBOs, they do not account for other factors that may
be important drivers of injury risk. In the next section, we implement a regression-based
difference-in-differences approach to account for these drivers as best we can, and to perform
formal statistical tests of changes in injury rates after LBOs.
3 Difference-in-differences analysis
In this section, we describe our difference-in-differences approach and present the results
from implementing it. We only estimate differences-in-differences for the public LBO sample,
as the combination of differences in observable characteristics and differential trends pre-
buyout suggests that difference-in-differences estimation for the private LBO sample would
be invalid, making the estimates difficult to interpret. The first difference is between the
pre- and post-buyout periods. The second difference is between LBO and non-LBO control
establishments. Denoting establishment by i, year by t, and 4-digit SIC code industry by j,
our primary regression specification is the following:
14
Injuries/Employeeit = αi + φjt + βPostLBOt + γLBOFirmi ∗ PostLBOit
+ δLog(Employees)it + ηHoursWorked/Employeeit + εit. (1)
We also estimate specifications where DARTInjuries/Employee is the dependent variable.
The indicator LBOFirm equals one for LBO establishments and zero for control estab-
lishments. The indicator PostLBO equals one for observations in the five year post-LBO
period and zero for those in the five year pre-LBO period. We exclude establishment-year
observations from the LBO year itself, as the establishment is under PE ownership only
part of that year. Each control establishment is assigned the same LBO year as the LBO
establishment to which it is matched. We include both establishment fixed effects (αi) and
industry-year fixed effects (φjt) to account for any unobserved time-invariant establishment
factors and time-varying industry factors that might impact injury rates. Because the buyout
year varies across establishments, we can separately identify industry-year fixed effects from
the treatment effect itself. Note that the main effect of LBOFirm is not included because it
does not vary within establishment, and would therefore be absorbed by the establishment
fixed effects. The coefficient γ captures the estimated change in injury rate from before to
after a buyout for LBO firms relative to non-LBO firms and is the object of interest in the
regressions.
Table 4 presents difference-in-difference estimates based on regression equation (1). Stan-
dard errors clustered at the firm level are reported below each point estimate. Columns (1)
and (2) present estimates where we include only year and industry-year fixed effects, re-
spectively. The exclusion of establishment fixed effects in these two regressions allows us to
include the main effect of LBOFirm. Column (3) presents actual estimates of (1), where we
include both industry-year and establishment fixed effects. Column (4) shows results where
15
we reduce the minimum number of employees at the time of the buyout for inclusion in the
sample from 100 to 50. Column (5) shows results from estimation of an establishment fixed
effects Poisson model with industry-year dummies, where we set the exposure variable to
Employees.11
— Table 4 here —
The statistically insignificant coefficients on LBOFirm in the first two columns suggest
no differences in injury rates in public firm LBO and control establishments pre-LBO. The
statistically insignificant coefficients on PostLBO in all five columns suggest that control
firms do not experience changes in injury rates from before to after the buyout year. The
negative coefficients on the interaction of LBOFirm and PostLBO - the coefficient of pri-
mary interest - suggest that injury rates decline at public firm LBO establishments relative to
non-LBO establishments around LBOs. The coefficient is statistically significant at the 5%
level in all of the OLS regressions, but is statistically insignificant in the Poisson regression
in column (5).
The coefficient on LBOFirm ∗PostLBO in column (3), where we include establishment
and industry-year fixed effects, suggests that injury rates fall by 0.86 percentage points in
LBO establishments after LBOs relative to control establishments. This decline represents a
substantial 17% of the 5.12 percentage point mean annual injury rate reported by the BLS
over the sample period. The decline implied by the coefficients in columns (1) and (2) is
larger (1.18 percentage points in each) and in (4) is slightly smaller (0.78 percentage points).
Appendix A shows results where we construct matched samples using four alternative
approaches to verify that the results are robust to the sample formation process. The first11The advantage of the Poisson model relative to an OLS model is that it explicitly accounts for the
non-negative, non-fractional nature of the injury data. The drawbacks are that the estimates it producesare more difficult to interpret and it imposes the assumption that the conditional mean and variance of theinjury rate are the same. While violation of this assumption does not bias estimates, it does reduce theirefficiency (Wooldridge, 2002, ch. 19).
16
involves matching on multiple characteristics using propensity score matching to further
ensure the similarity of the LBO and control groups. The second involves using only the
single closest match (rather than up to five matches) in terms of establishment size to
each LBO establishment to construct the control sample. The third involves excluding
observations in year t−1 to ensure that the large number of establishment-year observations
in that year relative to other pre-LBO years is not skewing the results. The fourth involves
excluding observations in year t − 5 to ensure that the big deviation between LBO and
control establishments in injury rates in that year is not driving the results. The results do
not appear sensitive to the use of reasonable alternative sample formation processes.
It is worth considering the size of the decline in injury rates implied by the estimates in
Table 4 in comparison to the 7% reduction in wages per employee after public firm LBOs
that Davis, et al. (2014) document. This 7% decline translates into a wage drop of $3,013 for
a worker earning the 2013 mean U.S. compensation of $43,041. In a survey article, Viscusi
and Aldy (2003) conclude that existing evidence supports an average compensating wage
differential per additional expected injury that translates into $60,000 to $210,000 in 2013
dollars. Based on these estimates, an 0.86 percentage point decline in injury rate should lower
wages demanded by employees by between $558 and $1,953. Thus our estimates can account
for 19% to 65% of the estimated $3,013 fall in compensation for the average worker.12 It is
important to note, however, that our data do not include wages, so we cannot test whether
the drop in earnings and drop in injury risk are actually linked.
We next estimate the same set of regressions using DARTInjuries/Employee (the rate
of relatively serious injuries) rather than Injuries/Employee (the rate of all injuries) as the
dependent variable. Table 5 shows the results. The negative coefficients on the interaction of
LBOFirm and PostLBO suggest that relatively serious injuries also decrease after LBOs.12Note that neither this interpretation, nor the results of Davis, et al. (2014), imply that any individual em-
ployee experiences a reduction in compensation. For example, even if wages are sticky, average compensationper employee will fall if newly-hired employees receive lower wages.
17
The estimate in column (3) suggests that DART injury rates decrease (increase) by 0.39
percentage points after public firm LBOs, or 15% of the mean annual DART injury rate
reported by the BLS over the sample period.
— Table 5 here —
The results in Table 4 suggest that injury rates decrease (increase) from the last five
years before to the first fives years after public firm LBOs. We further explore the timing of
these changes by estimating the following regression:
Injuries/Employeeit = αi + φjt +∑
K∈(0,5)βKPostLBOY rKit
+∑
K∈(0,5)γKLBOFirmi ∗ PostLBOY rKit
+ δLog(Employees)it + ηHoursWorked/Employeeit + εit. (2)
This regression equation allows the difference in injury rates across LBO and control
establishments relative to the pre-LBO period to vary across each of the post-buyout years.
Here, K represents the number of years an observation occurs after the buyout year, with
K = 0 representing the buyout year itself. Unlike in estimating (1), we include observations
from the buyout year itself here since the regression specification allows the injury rate in
that year to differ from the post-buyout period. The γK coefficients capture the difference
in injury rates in post-buyout year K relative to the difference in the pre-buyout period
and are the objects of interest. Table 6 presents the results from these regressions, with the
specifications mirroring those of Table 4.
— Table 6 here —
18
The patterns here are consistent with those shown in Figure 1. Injury rates in public firm
LBO establishments do not change relative to those of non-LBO establishments in the year
of the buyout (K = 0) itself and are only slightly lower than those at control establishments
in year 1 post-buyout, before becoming significantly lower in year 2 post-buyout. Beyond
that point, they remain substantially lower than in the pre-buyout period through year 5
post-buyout, the last year we include in the sample window. The slight delay in the decline
in injury rates after the buyout is consistent with changes in workplace safety and investment
practices taking time to impact observed injury rates.
4 Cross-sectional analysis
The previous section presents evidence that injury rates drop after LBOs of publicly-
traded firms, consistent with these LBOs improving workplace safety. In this section, we
explore the possible cause of this drop. We do so by analyzing how the decrease in injury
rates varies cross-sectionally with pre-buyout firm characteristics and characteristics of the
buyout itself. We implement these cross-sectional tests by estimating regressions of the
general form:
Injuries/Employeeit = αi + φjt + βPostLBOt + γLBOFirmi ∗ PostLBOit
+ θPostLBOt ∗ Characteristici
+ λLBOFirmi ∗ PostLBOit ∗ Characteristici
+ δLog(Employees)it + ηHoursWorked/Employeeit + εit. (3)
Charactersitic is a characteristic of the buyout firm or of the buyout, and therefore
does not vary within establishment. The main effects of LBOFirm and LBOFirm ∗
Characteristic are both fully absorbed by the establishment fixed effects αi and are there-
19
fore omitted from the regression equation. The coefficient λ on the triple interaction term
LBOFirm ∗ PostLBO ∗ Characteristic is the object of interest in these regressions.
We explore three different possible causes of the decrease in injury rates after public
firm LBOs. The first of these relates is motivated by a large body of existing evidence that
managers of publicly-traded firms make myopic investment decisions because they are con-
cerned about short-term profitability. Faced with such concerns, managers may turn down
even positive NPV long-run investments. This may result in the curtailment of spending
on activities such as equipment maintenance, employee training, and safety monitoring that
directly contribute to workplace safety, especially since many of these activities are directly
expensed for accounting purposes. Cuts to investment in capital assets could also adversely
impact workplace safety, as newer technologies tend to be safer than the ones they replace.
If the alleviation of pressure to focus on short-term profitability drives the decline in injury
rates after public firm LBOs, then we should observe larger declines for firms where this
pressure is greater. We consider three proxies for the extent to which managers face pressure
to focus on short-term profitabiltiy. The first is the the fraction of the firm’s shares held
by “transient” institutional investors as identified by Bushee (1998). These are institutions
that either have high portfolio turnover or engage in momentum trading strategies. Because
their holding periods are short, these investors may pressure management to focus on short-
run performance. Bushee (1998) shows that firms with more transitory institutional investor
shareholdings are more likely to reduce research and development spending in order to reverse
an earnings decline, consistent with a response to such pressure. We use Thomson 13(f)
holdings data to identify the percentage of a firm’s shares held at year-end by instutitions
(AllInstHoldings). We then use Bushee’s classification of transient investors, available from
his website, in conjunction with the 13(f) holdings data to measure the percentage of shares
held by transient investors (TransientInstHoldings). We also include the interaction of
AllInstHoldings with LBOFirm ∗ PostLBO in the regression when we estimate equation
20
(3) for TransientInstHoldings to be sure that we are picking up only differential sensitivity
with transient investor holdings and not institutional investor holdings as a whole.
Our second proxy for short-term pressure is the number of stock analysts covering a firm.
Greater analyst coverage is likely to focus more attention on the firm’s short-run financial
performance. We calculate the number of analysts issuing at least one earnings forecast
for a firm in a given year from the I/B/E/S earnings forecast database. We then create
an indicator variable HighAnalystCoverage, which equals one if at least six analysts (the
median number in the I/B/E/S universe) cover the firm and zero otherwise.
Our third proxy is based on the firm’s earnings management practices. A firm facing
strong pressure to focus on short-run performance may engage in aggressive accounting
practices in order to boost reported earnings. We employ a commonly-used measure of
accounting aggressiveness — abnormal discretionary accruals as computed using the mod-
ified Jones model. We create an indicator variable PosAbnormalAccruals, which equals
one if abnormal accruals are positive and zero otherwise. While the first two proxies are
intended to capture characteristics influencing managers to focus on the short run, the third
measure reflects a response to such influence. If injury rates decline after LBOs because
managers of publicly-traded firms face pressure to focus on short run profitability, then λ
should be negative when Characteristic is TransientInstHoldings, HighAnalystCoverage,
or PosAbnormalAccruals.
The second channel we consider is the possibility that an LBO reduces injury risk by
alleviating financing constraints and therefore allowing more long-term investment. Cohn
and Wardlaw (2015) present evidence that financing constraints are associated with higher
injury rates. An LBO could relax financing constraints via capital infusion at the time of
the LBO and subsequently. While we are aware of no evidence that LBOs of public firms
relax financing constraints, there is evidence that they relax financing constraints in private
firms (Boucly, Sraer, and Thesmar (2011), Cohn, Hotchkiss, and Towery (2015)), and that
21
acquisitions by other operating companies relax financing constraints in public firms (Erel,
Jang, and Weisbach (2015)).
We consider three indices intended to measure the degree to which a firm is financially
constrained as proxies for financing constraints. These are the Kaplan and Zingales (1997)
index, the Whited and Wu (2006) index, and the Hadlock and Pierce (2010) age-size index.
Higher values of these indices imply tighter financing constraints. We define indicator vari-
ables HighKZIndex, HighWWIndex, and HighHPIndex that are equal to one for values
of the relevant index greater than the sample median, and zero otherwise. If LBOs of public
firms lead to a decline injury rates by alleviating financing constraints, then λ should be
negative when these three proxies are used for Characteristic in estimating equation (3).
The third channel we consider is simple improvements in management. Holding invest-
ment policy fixed, poor management could lead to higher injury rates if it weakens oper-
ational efficiency. While manager quality is difficult to measure, we consider two proxies
relating to the structure of the LBO itself. The first is whether the CEO is replaced at the
time of the LBO. Such a replacement may indicate that the LBO is motivated in part by
the opportunity to replace low-quality management. We define the indicator CEOTurover,
which equals one if the CEO is replaced at the time of the LBO, and zero otherwise.
The second proxy is MBO, and indicator equal to one if the LBO is structured as a
management buyout, and zero otherwise. An MBO could indicate managerial strength, with
managers signaling confidence in their skill by taking a large equity stake. This proxy could
also capture information about public market pressure, as an MBO may be motivated largely
by a desire to escape this pressure. Information on CEO turnover and MBOs is obtained
from Cohn, Mills, and Towery (2014). If the decline in injury rates after public firm LBOs is
driven by improvements in management, then λ should be negative when Characteristic is
set to CEOTurnover and positive when it is set to MBO. It is worth noting, however, that
we identify CEO turnover at the time of the LBO is only 13 cases, and only 17 of the LBOs
22
are structured as MBOs. Thus the statistical power of these tests is likely to be limited.
Table 7 reports estimates of regression equation (3). Only the coefficients on LBOFirm∗
PostLBO and LBOFirm ∗ PostLBO ∗ Characteristic are shown for the sake of brevity.
— Table 7 here —
As the results in columns (1) through (3) show, injury rates decline more after LBOs
in acquired establishments relative to non-LBO establishments when the firm has more
transient investors, more analyst coverage, and higher abnormal accruals pre-LBO. In fact,
the insignificant coefficients on LBOFirm∗PostLBO suggest that injury rates do not decline
at a detectable level for firms with very low levels of transient investor holdings, little analyst
coverage, or low levels of abnormal accruals. These results support the hypothesis that the
drop in injury rates after public firm LBOs is at least partly attributable to LBOs allowing
firms to escape public market pressure and hence to focus more on long-run investment. It is
worth noting that the correlations among TransientInstHoldings, HighAnalystCoverage,
or PosAbnormalAccruals are fairly low, suggesting that they are not all simply capturing
the same information.13
The only other statistically significant triple interaction coefficients in Table 7 are those
involving HighWW and HighHP , two of the financing constraints indices. However, the
signs of these two coefficients disagree. The negative coefficient on the HighHP triple inter-
action suggests that injury rates decline more for firms that are more financially-constrained
pre-buyout, while the positive coefficient on the HighWW triple interaction suggests the
opposite. Thus we can draw no conclusions about the role of financing constraints in driving
the drop in injury rates after public firm LBOs. There is also no evidence that improvements
in management quality play a role.13The pairwise correlation between TransientInstHoldings and HighAnalystCoverage is 0.3379, be-
tween TransientInstHoldings and PosAbnormalAccruals is 0.0873, and between HighAnalystCoverageand PosAbnormalAccruals is 0.1224.
23
5 Employment dynamics after public firm LBOs
The analysis in this study focuses primarily on examining within-establishment changes
in injury rates after LBOs. To add further color to this analysis, this section investigates
employment dynamics around LBOs with two questions in mind. First, how does the dis-
tribution of employment change across establishments with differing levels of injury risk?
Second, how do changes in injury risk after LBOs vary with changes in employment?
The results in Sections 3 support a decrease in injury risk within establishment after
LBOs. However, post-LBO changes in the distribution of employment across establishments
with differing levels of injury risk could also impact overall levels of injury risk. We therefore
examine changes in establishment-level employment after LBOs using the same difference-
in-differences approach described in Section 3. Table 8 presents this analysis.
We begin by simply estimating a variant of equation (1), substituting log(Employees)
for Injuries/Employee. Column (1) reports the results from this regression. The coefficient
on LBOFirm ∗ PostLBO, which is statistically significant at the one percent level, implies
a 12.8% average reduction in employment relative to pre-buyout levels. This is consistent
with, though slightly larger than, the estimates of Davis, et al. (2014). This provides some
comfort that the sample of LBO establishment surveyed by the BLS is not unusual in some
way.
— Table 8 here —
Next, we examine how the fall in employment after buyouts varies across establishments
with establishment-level injury risk. We use two measure of this risk. The first is the injury
rate of an establishment in the final year reported before the buyout year. The second is the
average 4-digit SIC code injury rate over the sample period computed using all of the SOII
data. We then estimate regressions of the form of equation (3), setting Chararteristic to
each of these two measures.
24
Columns (2) and (3) present the results. The variable of interest is the interaction of
LBOFirm∗PostLBO with each of the two injury risk measures. The positive coefficient on
this triple interaction term in column (2), which is statistically significant at the one percent
level, indicates that employment falls by less in relatively dangerous establishments than in
relatively safe establishments pre-buyout. The positive coefficient on this triple interaction
in column (3) provides confirmatory evidence, though this coefficient is not statistically
significant at the ten percent level.
We also examine how employee utilization changes around LBOs. LBO owners might
compensate for reduced employment by requiring remaining employees to work longer hours.
Column (4) presents estimates of equation (1) where HoursWorked/Employee is the depen-
dent variable. The coefficient on LBOFirm ∗ PostLBO is statistically insignificant. Thus
it appears that the reduction in employment corresponds to a real reduction in actual hours
worked.14
The results in column (2) suggest that a shift in employment towards higher-injury rate
establishments after LBOs could at least partly offset the negative impact of the within-
establishment decrease in injury rates on overall injury risk. We therefore examine how
overall injury rates — pooling employees across all LBO and control establishments sep-
arately — evolve after LBOs. To do so, we sum Injuries and Employment for all LBO
establishments in our final sample and also for all control establishments, and then divide the
the pooled injuries by the pooled employment. This is equivalent to calculating mean injury
rates for LBO and control establishments, weighting each establishment-year observation by
the establishment’s reported employment in that year. Thus these pooled injury rates will
reflect the impact of relative changes in employment across higher- and lower-injury rate
establishments after LBOs. Figure 4 presents a plot of these pooled injury rates in each year
relative to the LBO year.15
14Estimates of equation (1) where HoursWorked is the dependent variable confirm this.15Note that, unlike the plots in Figures 1 and 2, this plot does not show error bands around the means.
25
— Figure 4 here —
The figure shows that overall injury rates fall considerably for LBO establishments but
not control establishments starting in the second year post-LBO, even after accounting for
changes in employment across establishments. Thus the impact of the fall in injury rates
within establishment after LBOs on overall injury rates appears to dwarf the impact of the
relative decrease in employment in lower-injury rate establishments.
The results in columns (2) and (3) of Table 8 potentially aid with interpretation of the
cause of the fall in injury rates within establishment. These rates could fall because LBO
acquirers systematically outsource relatively dangerous jobs, rather than because remaining
jobs become safer. Without employee-level data, we cannot directly test this explanation,
which would be interesting in its own right. However, the fact that employment, if any-
thing, falls more in relatively safe establishments, implies that LBO acquirers would have to
eliminate relatively dangerous jobs within establishment, while at the same time eliminating
more jobs at relative safe establishments, for outsourcing to explain the fall in injury rates.
The results in column (1) of Table 8 raises concerns that employees of firms acquired
in LBOs may underreport injuries rather than risk “rocking the boat” at a time of high
termination risk. Suggesting that such a concern is valid, Boone, et al. (2011) find that
employees are less likely to report moderate injuries when they face a higher risk of being
laid off. While we cannot observe layoff risk ex ante, we do observe actual changes in
establishment-level employment after buyouts. Assuming that employees anticipate layoffs
correctly to at least some degree, the decrease in injury rates should be greater among
establishments experiencing a drop in employment ex post if it is driven by reporting changes
due to increased layoff risk. We explore this possibility by estimating equation (3), setting
Characteristic to EmpDecrease, an indicator equal to one if the number of employees at
Because we are pooling all LBO and control establishments in this plot, we observe only a single aggregateinjury rate for each group.
26
the establishment decreased from the last observation pre-buyout to the first observation
post-buyout, and zero if its employment remained the same or increased. Table 9 presents
the results.
— Table 9 here —
We present results from five regressions mirroring those in Table 4. The coefficient on the
triple interaction of LBOFirm, PostLBO, and EmpDecrease is positive and statistically
insignificant. This suggests that the drop in injury rates after LBOs is not concentrated
among firms reducing employment, and in fact decreases slightly less for these establishments.
This appears inconsistent with underreporting due to layoff concerns driving the drop in
reported injury rates in our sample.
6 Conclusion
Overall, the results presented in this paper suggest a positive effect of LBOs on workplace
safety, at least in LBOs of publicly-traded firms. This effect appears to be driven in part
by the alleviation of pressure from public markets to cut costs in order to meet earnings
expectations. Of course, LBOs are not random events, and one must be careful in reaching
conclusions about causality. Nevertheless, the results lend support to the argument that
LBOs of public firms improve operational performance by allowing firms to focus more on
long-run investment. Future work considering how injury rates and wages evolve together
around LBOs would be useful for further understanding the impact of these transactions on
employees.
27
ReferencesAgrawal, Ashwini, and Prasana Tambe, 2014, Technological Investment and Labor Out-comes: Evidence from Private Equity, Working paper.
Amess, Kevin, and Mike Wright, 2007, The wage and employment effects of leveraged buy-outs in the UK, International Journal of the Economics of Business 14, 179–195.
Bernstein, Shai, et al., 2015, Private equity and industry performance, Management ScienceForthcoming.
Bernstein, Shai, and Albert Sheen, 2015, The Operational Consequences of Private EquityBuyouts: Evidence from the Restaurant Industry, Working paper.
Boone, Jan, et al., 2011, Recessions are bad for workplace safety, Journal of health economics30, 764–773.
Boucly, Quentin, David Sraer, and David Thesmar, 2011, Growth lbos, Journal of FinancialEconomics 102, 432–453.
Bushee, Brian J, 1998, The influence of institutional investors on myopic R&D investmentbehavior, Accounting review pp. 305–333.
Cohn, Jonathan B., Edith S. Hotchkiss, and Erin M. Towery, 2015, The motives for privateequity buyouts of private firms: Evidence from U.S. corporate tax returns, Working paper.
Cohn, Jonathan B, Lillian F Mills, and Erin M Towery, 2014, The evolution of capitalstructure and operating performance after leveraged buyouts: Evidence from US corporatetax returns, Journal of Financial Economics 111, 469–494.
Cohn, Jonathan B., and Malcolm I. Wardlaw, 2015, Financing and Workplace Safety, Work-ing paper.
Davis, Steven J, et al., 2014, Private Equity, Jobs, and Productivity, American EconomicReview 104, 3956–90.
Erel, Isil, Yeejin Jang, and Michael S Weisbach, 2015, Do acquisitions relieve target firms’financial constraints?, The Journal of Finance 70, 289–328.
Gompers, Paul A, Steven N Kaplan, and Vladimir Mukharlyamov, 2015, What do privateequity firms say they do?, Journal of Financial Economics Forthcoming.
Graham, John R, Campbell R Harvey, and Shiva Rajgopal, 2005, The economic implicationsof corporate financial reporting, Journal of accounting and economics 40, 3–73.
Guo, Shourun, Edith S Hotchkiss, and Weihong Song, 2011, Do buyouts (still) create value?,The Journal of Finance 66, 479–517.
28
Hadlock, Charles J, and Joshua R Pierce, 2010, New evidence on measuring financial con-straints: Moving beyond the KZ index, Review of Financial studies 23, 1909–1940.
Hotchkiss, Edith S., Per Stromberg, and David C. Smith, 2014, Private equity and theresolution of financial distress, Working paper.
Jensen, Michael C, 1989, Eclipse of the public corporation, Harvard Business Review (Sept.-Oct. 1989).
Kaplan, Steven, 1989, The effects of management buyouts on operating performance andvalue, Journal of financial economics 24, 217–254.
Kaplan, Steven N, and Luigi Zingales, 1997, Do investment-cash flow sensitivities provideuseful measures of financing constraints?, The Quarterly Journal of Economics pp. 169–215.
Leigh, J. Paul, 2011, Economic Burden of Occupational Injury and Illness in the UnitedStates, Milbank Quarterly 89, 728–772.
Lichtenberg, Frank R, and Donald Siegel, 1990, The effects of leveraged buyouts on produc-tivity and related aspects of firm behavior, Journal of Financial Economics 27, 165–194.
Muscarella, Chris J, and Michael R Vetsuypens, 1990, Efficiency and organizational struc-ture: a study of reverse LBOs, The Journal of Finance 45, 1389–1413.
Nie, Huihua, and Huainan Zhao, 2015, Financial Leverage and Employee Death: Evidencefrom China’s Coalmining Industry, Working paper.
Shleifer, Andrei, and Lawrence H Summers, 1988, Breach of trust in hostile takeovers, inCorporate takeovers: Causes and consequences (University of Chicago Press, ).
Smart, Scott B, and Joel Waldfogel, 1994, Measuring the effect of restructuring on corporateperformance: the case of management buyouts, The Review of Economics and Statisticspp. 503–511.
Smith, Abbie J, 1990, Corporate ownership structure and performance: The case of man-agement buyouts, Journal of financial Economics 27, 143–164.
Stein, Jeremy C, 1988, Takeover threats and managerial myopia, The Journal of PoliticalEconomy pp. 61–80.
Viscusi, W Kip, and Joseph E Aldy, 2003, The value of a statistical life: a critical review ofmarket estimates throughout the world, Journal of risk and uncertainty 27, 5–76.
Whited, Toni M, and Guojun Wu, 2006, Financial constraints risk, Review of FinancialStudies 19, 531–559.
Wright, Mike, Steve Thompson, and Ken Robbie, 1992, Venture capital and management-led, leveraged buy-outs: a European perspective, Journal of Business venturing 7, 47–71.
29
Figure 1: Injuries over event time - public firms
This figure presents mean injury rates and DART injury rates for public firm LBO andcontrol establishments around the LBO year. Error bands represent one standard de-viation above and below the mean. Figure 1a presents Injuries/Employee. Figure1b presents DARTInjuries/Employee. Figure 1c presents 4-digit SIC code industry-adjusted Injuries/Employee. Figure 1d presents 4-digit SIC code industry-adjustedDARTInjuries/Employee.
(a) Injuries/Employee
.03
.05
.07
.09
Case
s/Em
ploy
ee
-5 -4 -3 -2 -1 0 1 2 3 4 5Years to LBO
LBO Non-LBO
(b) DARTInjuries/Employee
.02
.03
.04
.05
.06
Case
s/Em
ploy
ee
-5 -4 -3 -2 -1 0 1 2 3 4 5Years to LBO
LBO Non-LBO
(c) Industry-Adjusted Injuries/Employee
-.01
0.0
1.0
2.0
3Re
sidua
l - C
ases
/Em
ploy
ee
-5 -4 -3 -2 -1 0 1 2 3 4 5Years to LBO
LBO Non-LBO
(d) Ind-Adjusted DARTInjuries/Employee
-.005
0.0
05.0
1.0
15Re
sidua
l - C
ases
/Em
ploy
ee
-5 -4 -3 -2 -1 0 1 2 3 4 5Years to LBO
LBO Non-LBO
30
Figure 2: Injuries over event time - private firms
This figure presents mean injury rates and DART injury rates for private firm LBO andcontrol establishments around the LBO year. Error bands represent one standard de-viation above and below the mean. Figure 1a presents Injuries/Employee. Figure1b presents DARTInjuries/Employee. Figure 1c presents 4-digit SIC code industry-adjusted Injuries/Employee. Figure 1d presents 4-digit SIC code industry-adjustedDARTInjuries/Employee.
(a) Injuries/Employee
.03
.04
.05
.06
.07
Case
s/Em
ploy
ee
-5 -4 -3 -2 -1 0 1 2 3 4 5Years to LBO
LBO Non-LBO
(b) DARTInjuries/Employee
.015
.02
.025
.03
.035
.04
Case
s/Em
ploy
ee
-5 -4 -3 -2 -1 0 1 2 3 4 5Years to LBO
LBO Non-LBO
(c) Industry-Adjusted Injuries/Employee
-.02
-.01
0.0
1Re
sidua
l - C
ases
/Em
ploy
ee
-5 -4 -3 -2 -1 0 1 2 3 4 5Years to LBO
LBO Non-LBO
(d) Ind-Adjusted DARTInjuries/Employee
-.015
-.01
-.005
0.0
05Re
sidua
l - C
ases
/Em
ploy
ee
-5 -4 -3 -2 -1 0 1 2 3 4 5Years to LBO
LBO Non-LBO
31
Figure 3: Employment dynamics
This figure presents mean Log(Employees) and HoursWorked/Employee for LBO andcontrol establishments around the LBO year. Error bands represent one standard de-viation above and below the mean. Figures 3a and 3b present Log(Employees) andHours/Employee, respectively, for public firm LBOs. Figures 3c and 3d presentLog(Employees) and Hours/Employee, respectively, for private firm LBOs.
(a) Log(Employees) - public firms
5.2
5.4
5.6
5.8
6lo
g(Em
ploy
ees)
-5 -4 -3 -2 -1 0 1 2 3 4 5Years to LBO
LBO Non-LBO
(b) HoursWorked/Employee - public firms
1600
1700
1800
1900
Hour
s pe
r Em
ploy
ee-5 -4 -3 -2 -1 0 1 2 3 4 5
Years to LBO
LBO Non-LBO
(c) Log(Employees) - private firms
5.2
5.4
5.6
5.8
6lo
g(Em
ploy
ees)
-5 -4 -3 -2 -1 0 1 2 3 4 5Years to LBO
LBO Non-LBO
(d) HoursWorked/Employee - private firms
1500
1600
1700
1800
1900
Hour
s pe
r Em
ploy
ee
-5 -4 -3 -2 -1 0 1 2 3 4 5Years to LBO
LBO Non-LBO
32
Figure 4: Pooled injuries over event time - public firms
This figure presents pooled injury rates across for LBO and control establishments around theLBO year. These pooled injury rates are calculated by summing Injuries and Employeesseparately for all LBO and control establishments in each year relative to the LBO year, andthen dividing the summed injuries by the summed employees.
.035
.04
.045
.05
.055
.06
Inju
ries/
Em
ploy
ee
-5 -4 -3 -2 -1 0 1 2 3 4 5Years to LBO
LBO Non-LBO
33
Table 1: Injuries by Cause and Type
This table shows the percentage of private sector U.S. workplace injuries in 2012 by nature(Panel A) and cause (Panel B), as reported by the BLS. These percentages were computedfrom incident rates available at http://www.bls.gov/news.release/pdf/osh2.pdf.
Panel A: Percent injuries by natureNature of injury PercentSprains, strains, tears 38.16Soreness, pain, including back 14.67Bruises, contusions 8.33Fractures 8.03Cuts, lacerations 8.03Multiple traumatic injuries and disorders 3.07Heat (thermal) burns 1.49Carpal tunnel syndrome 0.89Amuptations 0.59Chemical burns 0.40Tendonitis (other or unspecified) 0.30All other natures 16.06
Panel B: Percent injuries by causeCause of injury PercentContact with objects 29.69Fall on same level 19.56Overexertion in lifting/lowering 14.44Violence and other injuries by persons or animal 8.38Transportation incidents 6.64Fall to lower level 6.29Exposure to harmful substances or environments 5.82Slips or trips without fall 5.47Repetitive motion 3.49Fires and explosions 0.23
34
Table 2: Sample ConstructionThis table presents information about the LBO firms in the sample. Panel A describes how the sample was constructed. PanelB reports the sources of matches with the BLS injury data. Panel C reports summary statistics for LBO firms in the samplefrom the year prior to the LBO. Assets equals total reported assets. Sales equals total reported sales. Debt/Assets equalsbook debt divided by book assets. Tobin′sQ equals the ratio of the firm’s market value to its book value. CashF low/Assetsequals the sum of income before extraordinary items and depreciation, divided by lagged assets. Capex/Assets equals capitalexpenditures divided by lagged assets.
Panel A: LBO sample constructionPublic firm LBOs Private firm LBOs
Starting 1997-2007 LBO sample 317 555
RemoveFinancial firms 12 4Franchisers 20 4
LBO firms to match to injury data 285 547
Matched to BLS data 242 316
At least one establishment in data in (-5,-1) & (+1,+5) 150 156
At least one establishment with at least one valid control match 143 147
Panel B: LBOs in final sample by yearYear Public firm LBOs Private firm LBOs
1997 4 71998 10 41999 12 42000 14 132001 7 42002 3 12003 15 52004 6 162005 15 262006 23 312007 34 36
Total 143 147
Panel C: Public LBO firm pre-buyout characteristicsMean Std. Dev 10th pctile Median 90th pctile
Assets $1,370M $3,562M $73M $387M $2,929MSales $1,220M $2,037M $83M $391M $3,197MDebt/Assets 0.251 0.223 0.000 0.217 0.587Tobin’sQ 1.168 0.728 0.543 0.910 2.158CashFlow/Assets 0.090 0.084 0.016 0.083 0.184Capex/Assets 0.071 0.096 0.013 0.049 0.141
35
Table 3: Summary StatisticsThis table presents information about the establishments in the sample. Panel A reports the number of LBO establishments inthe final sample by the approach used to match it to an LBO firm. Panel B tabulates the number of control establishments foreach LBO establishment in the sample. Panel C reports means of various establishment variables the last year in the sampleprior to the LBO for LBO and control establishments, with t-statistics for the difference in means shown to the right. Panel Dreports the number of establishment-year observations in the final sample in each year relative to the LBO year. ***, **, and* indicate statistical significance of the differences at the 1%, 5%, and 10% level, respectively, based on a two-tailed t-test.
Panel A: Types of LBO establishment matchesType of match Public firm LBOs Private firm LBOs
EIN 132 0Name 215 121Name-EIN 159 196
Total 506 317
Panel B: Control establishments per LBO establishmentLBO establishments
# control establishments Public firm LBOs Private firm LBOs
1 66 422 45 273 48 154 19 155 328 218
Total 506 317
Panel D: Means of LBO and control establishment characteristics pre-buyoutPublic firm LBOs Private firm LBOs
LBO Control LBO Controlestab estab t-stat estab estab t-stat
Number 506 2,016 317 1,291
Employees 435.43 438.24 -0.07 349.45 339.85 0.39Log(Employees) 5.5735 5.5314 1.00 5.5026 5.4532 1.02HoursWorked/Employee 1,741.59 1,742.26 -0.03 1,628.28 1,643.52 -0.51Injuries/Employee 0.0651 0.0655 -0.13 0.0484 0.0595 3.34***DARTInjuries/Employee 0.0362 0.0383 -1.05 0.0263 0.0342 3.49***
Panel C: Establishment-year observations by year relative to LBO yearObservations Observations
Year Public firm LBOs Private firm LBOs Year Public firm LBOs Private firm LBOst− 5 591 288 t+ 1 1,081 693t− 4 612 348 t+ 2 998 746t− 3 801 385 t+ 3 973 710t− 2 877 460 t+ 4 949 663t− 1 1,422 1,041 t+ 5 904 667t 1,192 813 Total 10,400 6,814
36
Table 4: Injury Rates Around Public Firm LBOs: Difference-in-Difference EstimatesThis table presents estimates of the change in LBO establishment injury rates relative to control establishments after LBOsof publicly-traded firms. The sample consists of establishment-years belonging to LBO and control establishments in the fiveyears before and five years after the LBO. Columns (1) through (4) of each show results from OLS regressions, where thedependent variable is Injuries/Employee. Column (5) shows results from an establishment fixed effects Poisson regression,where the dependent variable is Injuries and the exposure variable is Employees. LBOFirm is an indicator equal to one if theestablishment belongs to a firm acquired in an LBO and zero otherwise. PostLBO is an indicator equal to one in the year afterthe LBO year and zero otherwise. Log(Employees) equals the log of the establishment’s average reported employment for theyear. HoursWorked/Employee equals reported hours worked divided by reported average employment. “Minimum employees”refers to the minimum number of establishment employees pre-LBO required for an LBO establishment to be included in thesample. Standard errors clustered at the firm level are shown below each point estimate. ***, **, and * indicate statisticalsignificance at the 1%, 5%, and 10% level, respectively, based on a two-tailed t-test.
OLS Poisson(1) (2) (3) (4) (5)
LBOFirm 0.0056 0.0064(0.0070) (0.0041)
PostLBO 0.0023 -0.0007 0.0045 -0.0002 0.0239(0.0041) (0.0064) (0.0043) (0.0046) (0.0470)
LBOFirm * PostLBO -0.0118*** -0.0118*** -0.0086*** -0.0078** -0.1236(0.0042) (0.0032) (0.0033) (0.0034) (0.0820)
Log(Employees) -0.0058*** -0.0012 -0.0022 -0.0019 -0.1075**(0.0013) (0.0014) (0.0024) (0.0018) (0.0426)
HoursWorked/Employee 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0003***(0.0000) (0.0000) (0.0000) (0.0000) (0.0001)
Establishment FE No No Yes Yes YesYear FE Yes No No No YesYear x Industry FE No Yes Yes Yes NoMinimum employees 100 100 100 50 100
Observations 9,208 9,208 9,208 13,427 8,931Adjusted R2 0.0780 0.1217 0.1694 0.0489
37
Table 5: DART Injury Rates Around Public Firm LBOs: Difference-in-Difference EstimatesThis table presents estimates of the change in LBO establishment DART injury rates relative to control establishments afterLBOs of publicly-traded firms. DART injuries are those serious enough to require days away from work, reduced activities, ortransfer. The sample consists of establishment-years belonging to LBO and control establishments in the five years before andfive years after the LBO. Columns (1) through (4) of each show results from OLS regressions, where the dependent variableis Injuries/Employee. Column (5) shows results from an establishment fixed effects Poisson regression, where the dependentvariable is Injuries and the exposure variable is Employees. LBOFirm is an indicator equal to one if the establishmentbelongs to a firm acquired in an LBO and zero otherwise. PostLBO is an indicator equal to one in the year after the LBOyear and zero otherwise. Log(Employees) equals the log of the establishment’s average reported employment for the year.HoursWorked/Employee equals reported hours worked divided by reported average employment. “Minimum employees”refers to the minimum number of establishment employees pre-LBO required for an LBO establishment to be included in thesample. Standard errors clustered at the firm level are shown below each point estimate. ***, **, and * indicate statisticalsignificance at the 1%, 5%, and 10% level, respectively, based on a two-tailed t-test.
OLS Poisson(1) (2) (3) (4) (5)
LBOFirm 0.0017 0.0023(0.0043) (0.0022)
PostLBO -0.0010 -0.0000 0.0005 0.0005 0.0179(0.0025) (0.0051) (0.0036) (0.0035) (0.0560)
LBOFirm * PostLBO -0.0065** -0.0063*** -0.0039** -0.0029* -0.1058(0.0026) (0.0021) (0.0017) (0.0017) (0.0841)
Log(Employees) -0.0040*** 0.0004 -0.0009 -0.0000 -0.0759(0.0008) (0.0008) (0.0015) (0.0011) (0.0518)
HoursWorked/Employee 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0003***(0.0000) (0.0000) (0.0000) (0.0000) (0.0001)
Establishment FE No No Yes Yes YesYear FE Yes No No No YesYear x Industry FE No Yes Yes Yes NoMinimum employees 100 100 100 50 100
Observations 9,208 9,208 9,208 13,427 8,709Adjusted R2 0.0596 0.1005 0.1350 0.0336
38
Table 6: Evolution of Injury Rates After Public Firm LBOsThis table presents estimates of the injury rate each year post-LBO relative to pre-LBO level for LBO establishments relativeto control establishments. The sample consists of establishment-years belonging to LBO and control establishments in the fiveyears before, year of, and five years after the LBO. Columns (1) through (4) of each show results from OLS regressions, wherethe dependent variable is Injuries/Employee. Column (5) shows results from an establishment fixed effects Poisson regression,where the dependent variable is Injuries and the exposure variable is Employees. LBOFirm is an indicator equal to one ifthe establishment belongs to a firm acquired in an LBO and zero otherwise. PostLBOY rK is an indicator equal to one forobservations in the Kth year after the LBO year and zero otherwise. Log(Employees) equals the log of the establishment’saverage reported employment for the year. HoursWorked/Employee equals reported hours worked divided by reported averageemployment. “Minimum employees” refers to the minimum number of establishment employees pre-LBO required for an LBOestablishment to be included in the sample. Standard errors clustered at the firm level are shown below each point estimate.***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively, based on a two-tailed t-test.
OLS Poisson(1) (2) (3) (4) (5)
LBOFirm 0.0055 0.0064(0.0070) (0.0041)
PostLBOYr0 -0.0023 0.0023 0.0027 0.0047 -0.0001(0.0026) (0.0035) (0.0033) (0.0020) (0.0036)
PostLBOYr1 -0.0017 -0.0008 0.0050 0.0020 0.0170(0.0035) (0.0048) (0.0043) (0.0045) (0.0554)
PostLBOYr2 0.0005 0.0032 0.0040 -0.0033 -0.0547(0.0043) (0.0068) (0.0067) (0.0067) (0.0703)
PostLBOYr3 0.0002 0.0063 0.0041 -0.0068 0.0008(0.0052) (0.0074) (0.0080) (0.0083) (0.0860)
PostLBOYr4 -0.0022 0.0011 0.0054 -0.0052 0.0296(0.0055) (0.0083) (0.0101) (0.0096) (0.1095)
PostLBOYr5 -0.0113* -0.0028 0.0025 -0.0074 -0.0354(0.0059) (0.0089) (0.0115) (0.0113) (0.1379)
LBOFirm * PostLBOYr0 -0.0047 -0.0048 -0.0009 -0.0027 0.0132(0.0038) (0.0031) (0.0026) (0.0027) (0.0563)
LBOFirm * PostLBOYr1 -0.0043 -0.0047 -0.0054 -0.0031 -0.0593(0.0036) (0.0038) (0.0036) (0.0040) (0.0895)
LBOFirm * PostLBOYr2 -0.0134*** -0.0141*** -0.0112** -0.0091 -0.1733*(0.0051) (0.0048) (0.0054) (0.0056) (0.0937)
LBOFirm * PostLBOYr3 -0.0163*** -0.0145*** -0.0103** -0.0090** -0.1908**(0.0060) (0.0036) (0.0041) (0.0036) (0.0924)
LBOFirm * PostLBOYr4 -0.0125** -0.0134*** -0.0099*** -0.0106*** -0.2116***(0.0058) (0.0038) (0.0038) (0.0039) (0.0908)
LBOFirm * PostLBOYr5 -0.0122* -0.0126*** -0.0066 -0.0087* -0.0469(0.0060) (0.0046) (0.0043) (0.0049) (0.1107)
Log(Employees) -0.0059*** -0.0011 -0.0020 -0.0016 -0.1154***(0.0013) (0.0013) (0.0021) (0.0016) (0.0390)
HoursWorked/Employee 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0003***(0.0000) (0.0000) (0.0000) (0.0000) (0.0000)
Establishment FE No No Yes Yes YesYear FE Yes No No No YesYear x Industry FE No Yes Yes Yes NoMinimum employees 100 100 100 50 100
Observations 10,400 10,400 10,400 15,258 10,279Adjusted R2 0.0745 0.1195 0.1570 0.1231
39
Table 7: Injury Rates Around LBOs: Variation with Pre-Buyout Firm CharacteristicsThis table presents estimates of cross-sectional variation in injury rate changes around LBOs. The sample consists ofestablishment-years belonging to LBO and control establishments in the five years before and five years after the LBO. OnlyLBO establishments with at least 100 employees pre-LBO are included. All column show estimates from OLS regressions ofthe following form:
Injuries/Employeeit = αi + φjt + βPostLBOt + γLBOFirmi ∗ PostLBOit
+ θPostLBOt ∗ Characteristici
+ λLBOFirmi ∗ PostLBOit ∗ Characteristici
+ δLog(Employees)it + ηHoursWorked/Employeeit + εit.
LBOFirm is an indicator equal to one if the establishment belongs to a firm acquired in an LBO and zero otherwise.PostLBO is an indicator equal to one in the year after the LBO year and zero otherwise. Characteristic equals a char-acteristic of the LBO establishment’s parent firm measured the year before the LBO, where the characteristic varies by column.TransitoryInstHoldings equals the fraction of a firm’s shares held by transitory investors as identified by Bushee (1998) andcomputed from Thomson 13(f) holdings data. AllInstHoldings equals the fraction of shares held by all institutional investorsas computed from Thomson 13(f) holdings data. HighAnalystCoverage is an indicator equal to one if the firm is covered bysix or more stock analysts and zero otherwise. PosAbnormalAccurals is an indicator equal to one if the firm’s abnormal dis-cretionary accruals are positive and zero otherwise. HighKZIndex, HighWWINdex, and HighHPIndex are indicators equalto one if the firm’s Kaplan and Zingales (1997), Whited and Wu (2006), or Hadlock and Pierce (2010) SA index, respectively,is above the sample median and zero otherwise. MBO is an indicator equal to one if the LBO took the form of a manage-ment buyout. CEOTurnover is an indicator equal to one if the CEO was replaced at the time of the LBO. Log(Employees)equals the log of the establishment’s average reported employment for the year. HoursWorked/Employee equals reportedhours worked divided by reported average employment. The variables PostLBO, PostLBO ∗Characteristic, Log(Employees),and HoursWorked/Employee are included in the regressions, but the coefficients are not shown for the sake of brevity. Allregressions include establishment and industry-year fixed effects. Standard errors clustered at the firm level are shown beloweach point estimate. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively, based on atwo-tailed t-test.
Short-term pressure Financing constraints Management issues(1) (2) (3) (4) (5) (6) (7) (8)
LBOFirm * PostLBO 0.0056 -0.0021 0.0073 -0.0052 -0.0165*** -0.0022 -0.0088** -0.0095**(0.0213) (0.0056) (0.0070) (0.0059) (0.0063) (0.0082) (0.0043) (0.0037)
LBOFirm * PostLBO * -0.0326**TransitoryInstHoldings (0.0156)
LBOFirm * PostLBO * -0.0156AllInstHoldings (0.0248)
LBOFirm * PostLBO * -0.0204**HighAnalystCoverage (0.0098)
LBOFirm * PostLBO * -0.0400**PosAbnormalAccruals (0.0133)
LBOFirm * PostLBO * -0.0150HighKZIndex (0.0117)
LBOFirm * PostLBO * 0.0138*HighWWIndex (0.0072)
LBOFirm * PostLBO * -0.0241**HighHPIndex (0.0123)
LBOFirm * PostLBO * 0.0043MBO (0.0107)
LBOFirm * PostLBO * 0.0071CEOTurnover (0.0047)
Observations 9,104 9,208 8,182 7,093 8,970 9,208 9,208 9,208Adjusted R2 0.1705 0.1710 0.1756 0.2038 0.1723 0.1700 0.1695 0.1695
40
Table 8: Employment, Worker Utilization, and Injury Rates Around Public Firm LBOsThis table presents estimates from regressions examining the relationship among employment, worker utilization, and injuryrate around LBOs of publicly-traded firms. The sample consists of establishment-years belonging to LBO and control es-tablishments in the five years before and five years after the LBO. The dependent variable in the first, third, and fourthcolumns is Log(Employees). The dependent variable in the second column is HoursWorked/Employee. The dependentvariable in the fifth column is Injuries/Employee. LBOFirm is an indicator equal to one if the establishment belongs toa firm acquired in an LBO and zero otherwise. PostLBO is an indicator equal to one in the year after the LBO year andzero otherwise. EstabInjuryRate equals the establishment’s Injuries/Employee the last year observed prior to the LBO.IndustryInjuryRate equals the mean 4-digit SIC code Injuries/Employee for the full BLS sample. EmpDecrease is anindicator equal to one if an establishment’s employment declines from the last year observed pre-LBO to the first year observedpost-LBO and zero otherwise. Log(Employees) equals the log of the establishment’s average reported employment for the year.HoursWorked/Employee ewquals reported hours worked divided by reported average employment. “Minimum employees”refers to the minimum number of establishment employees pre-LBO required for an LBO establishment to be included in thesample. Standard errors clustered at the firm level are shown below each point estimate. ***, **, and * indicate statisticalsignificance at the 1%, 5%, and 10% level, respectively, based on a two-tailed t-test.
HoursWorked/Dep var Log(Empl) Log(Empl) Log(Empl) Employee
PostLBO -0.0493 -0.0208 0.0257 -52.7516(0.1045) (0.0987) (0.1369) (90.8095)
LBOFirm * PostLBO -0.1200*** -0.2420*** -0.1801*** 18.3074(0.0391) (0.0686) (0.0666) (17.9480)
PostLBO * EstabInjuryRate -0.1343(0.3350)
LBOFirm * PostLBO * 1.7693***EstabInjuryRate (0.5477)
PostLBO * IndustryInjuryRate -1.5033(2.0709)
LBOFirm * PostLBO * 0.9546IndustryInjuryRate (0.6972)
Establishment FE Yes Yes Yes YesYear x Industry FE Yes Yes Yes YesMinimum employees 100 100 100 100
Observations 9,208 9,208 9,208 9,208Adjusted R2 0.1593 0.1595 0.1624 0.0854
41
Table 9: Injury Rates Around Public Firm LBOs: Difference-in-Difference EstimatesThis table presents estimates of how the change in LBO establishment injury rates relative to control establishments afterLBOs of publicly-traded firms varies with reductions in employment post-buyout. The sample consists of establishment-yearsbelonging to LBO and control establishments in the five years before and five years after the LBO. Columns (1) through (4)of each show results from OLS regressions, where the dependent variable is Injuries/Employee. Column (5) shows resultsfrom an establishment fixed effects Poisson regression, where the dependent variable is Injuries and the exposure variableis Employees. LBOFirm is an indicator equal to one if the establishment belongs to a firm acquired in an LBO and zerootherwise. PostLBO is an indicator equal to one in the year after the LBO year and zero otherwise. EmpDecrease is anindicator equal to one if the establishment’s employment decreases from the last reported year before to the first reported yearafter the LBO and zero otherwise. Log(Employees) equals the log of the establishment’s average reported employment for theyear. HoursWorked/Employee equals reported hours worked divided by reported average employment. “Minimum employees”refers to the minimum number of establishment employees pre-LBO required for an LBO establishment to be included in thesample. Standard errors clustered at the firm level are shown below each point estimate. ***, **, and * indicate statisticalsignificance at the 1%, 5%, and 10% level, respectively, based on a two-tailed t-test.
OLS Poisson(1) (2) (3) (4) (5)
LBOFirm 0.0106 0.0110(0.0086) (0.0075)
PostLBO 0.0004 -0.0006 0.0036 0.0002 0.0208(0.0042) (0.0067) (0.0047) (0.0047) (0.0559)
LBOFirm * PostLBO -0.0144** -0.0151** -0.0144** -0.0139** -0.1510(0.0062) (0.0062) (0.0069) (0.0064) (0.0923)
EmpDecrease -0.0105*** -0.0014(0.0025) (0.0024)
LBOFirm * EmpDecrease -0.0078 -0.0073(0.0100) (0.0079)
PostLBO * EmpDecrease 0.0017 -0.0003 0.0012 -0.0006 0.0114(0.0024) (0.0026) (0.0027) (0.0021) (0.0557)
LBOFirm * PostLBO * 0.0036 0.0052 0.0096 0.0104* 0.0625EmpDecrease (0.0066) (0.0065) (0.0073) (0.0061) (0.1688)
Log(Employees) -0.0062*** -0.0012 -0.0016 -0.0014 -0.1004**(0.0013) (0.0013) (0.0023) (0.0017) (0.0450)
HoursWorked/Employee 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0003***(0.0000) (0.0000) (0.0000) (0.0000) (0.0001)
Establishment FE No No Yes Yes YesYear FE Yes No No No YesYear x Industry FE No Yes Yes Yes NoMinimum employees 100 100 100 50 100
Observations 9,208 9,208 9,208 13,427 8,931Adjusted R2 0.0906 0.1220 0.1699 0.0500
42
A Additional Tables
Table AI: Injury Rates Around LBOs: Difference-in-Difference Estimates - Matching onMultiple CharacteristicsThis table presents estimates from a series of regressions where the dependent variable is Injuries/Employee. The sampleconsists of establishment-years belonging to public firms acquired in LBOs during the sample period and matched controlestablishments. Unlike in Table 4, LBO establishments are matched to control establishments using propensity score matching,where log(Employees), HoursWorked/Employee, and Injuries/Employee are used to estimate an establishment’s propensityto be acquired as part of an LBO. Columns (1) through (4) of each show results from OLS regressions. Column (5) showsresults from an establishment fixed effects Poisson regression. Here, the dependent variable is Injuries and the exposurevariable is Employees. LBOFirm is an indicator equal to one if the establishment belongs to a firm acquired in an LBO andzero otherwise. PostLBO is an indicator equal to one in the year after the LBO year and zero otherwise. Log(Employees)equals the log of the establishment’s average reported employment for the year. HoursWorked/Employee equals reportedhours worked divided by reported average employment. “Minimum employees” refers to the minimum number of establishmentemployees pre-LBO required for an LBO establishment to be included in the sample. Standard errors clustered at the firm levelare shown below each point estimate. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively,based on a two-tailed t-test.
OLS Poisson(1) (2) (3) (4) (5)
LBOFirm 0.0062 0.0069*(0.0070) (0.0039)
PostLBO 0.0031 -0.0010 0.0085* 0.0065 0.0149(0.0042) (0.0065) (0.0048) (0.0047) (0.0570)
LBOFirm * PostLBO -0.0125*** -0.0123*** -0.0095*** -0.0084** -0.0843(0.0043) (0.0031) (0.0033) (0.0034) (0.0821)
Log(Employees) -0.0057*** 0.0000 -0.0034 -0.0026 -0.0982**(0.0013) (0.0013) (0.0024) (0.0019) (0.0402)
HoursWorked/Employee 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0003***(0.0000) (0.0000) (0.0000) (0.0000) (0.0001)
Establishment FE No No Yes Yes YesYear FE Yes No No No YesYear x Industry FE No Yes Yes Yes NoMinimum employees 100 100 100 50 100
Observations 9,208 9,208 9,208 13,427 8,931Adjusted R2 0.0830 0.1241 0.1725 0.0522
43
Table AII: Injury Rates Around LBOs: Difference-in-Difference Estimates - Single ControlEstablishmentThis table presents estimates from a series of regressions where the dependent variable is Injuries/Employee. The sampleconsists of establishment-years belonging to public firms acquired in LBOs during the sample period and matched control estab-lishments. Compared to Table 4, we include only the closest (as opposed to five closest) matches for each LBO establishmentsas control establishments. Columns (1) through (4) of each show results from OLS regressions. Column (5) shows resultsfrom an establishment fixed effects Poisson regression. Here, the dependent variable is Injuries and the exposure variableis Employees. LBOFirm is an indicator equal to one if the establishment belongs to a firm acquired in an LBO and zerootherwise. PostLBO is an indicator equal to one in the year after the LBO year and zero otherwise. Log(Employees) equals thelog of the establishment’s average reported employment for the year. HoursWorked/Employee equals reported hours workeddivided by reported average employment. “Minimum employees” refers to the minimum number of establishment employeespre-LBO required for an LBO establishment to be included in the sample. Standard errors clustered at the firm level are shownbelow each point estimate. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively, based ona two-tailed t-test.
OLS Poisson(1) (2) (3) (4) (5)
LBOFirm 0.0058 0.0061(0.0075) (0.0041)
PostLBO 0.0055 0.0005 0.0075 0.0025 -0.0243(0.0066) (0.0094) (0.0064) (0.0073) (0.0704)
LBOFirm * PostLBO -0.0122** -0.0115*** -0.0099*** -0.0086** -0.1167(0.0050) (0.0037) (0.0036) (0.0031) (0.0892)
Log(Employees) -0.0077*** -0.0017 -0.0074*** -0.0061*** -0.1515***(0.0023) (0.0020) (0.0026) (0.0022) (0.0575)
HoursWorked/Employee 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0004***(0.0000) (0.0000) (0.0000) (0.0000) (0.0001)
Establishment FE No No Yes Yes YesYear FE Yes No No No YesYear x Industry FE No Yes Yes Yes NoMinimum employees 100 100 100 50 100
Observations 3,868 3,668 9,208 5,568 3,767Adjusted R2 0.0753 0.1232 0.1740 0.0568
44
Table AIII: Injury Rates Around LBOs: Difference-in-Difference Estimates - Excluding Yeart− 1This table presents estimates from a series of regressions where the dependent variable is Injuries/Employee. The sampleconsists of establishment-years belonging to public firms acquired in LBOs during the sample period and matched controlestablishments. Compared to Table 4, the sample used here excludes observations in the last year pre-LBO. Columns (1)through (4) of each show results from OLS regressions. Column (5) shows results from an establishment fixed effects Poissonregression. Here, the dependent variable is Injuries and the exposure variable is Employees. LBOFirm is an indicator equalto one if the establishment belongs to a firm acquired in an LBO and zero otherwise. PostLBO is an indicator equal to onein the year after the LBO year and zero otherwise. Log(Employees) equals the log of the establishment’s average reportedemployment for the year. HoursWorked/Employee equals reported hours worked divided by reported average employment.“Minimum employees” refers to the minimum number of establishment employees pre-LBO required for an LBO establishmentto be included in the sample. Standard errors clustered at the firm level are shown below each point estimate. ***, **, and *indicate statistical significance at the 1%, 5%, and 10% level, respectively, based on a two-tailed t-test.
OLS Poisson(1) (2) (3) (4) (5)
LBOFirm 0.0083 0.0095*(0.0089) (0.0050)
PostLBO 0.089* -0.0051 0.0015 0.0005 0.0158(0.0053) (0.0081) (0.0073) (0.0071) (0.0745)
LBOFirm * PostLBO -0.0146*** -0.0148*** -0.0099*** -0.0085** -0.0978(0.0053) (0.0043) (0.0037) (0.0043) (0.1037)
Log(Employees) -0.0055*** -0.0013 -0.0017 -0.0007 -0.1029**(0.0013) (0.0014) (0.0021) (0.0018) (0.0509)
HoursWorked/Employee 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0003***(0.0000) (0.0000) (0.0000) (0.0000) (0.0001)
Establishment FE No No Yes Yes YesYear FE Yes No No No YesYear x Industry FE No Yes Yes Yes NoMinimum employees 100 100 100 50 100
Observations 7,786 7,786 7,786 11,223 7,329Adjusted R2 0.0744 0.1205 0.1632 0.0537
45
Table AIV: Injury Rates Around LBOs: Difference-in-Difference Estimates - Excluding Yeart− 5This table presents estimates from a series of regressions where the dependent variable is Injuries/Employee. The sampleconsists of establishment-years belonging to public firms acquired in LBOs during the sample period and matched controlestablishments. Compared to Table 4, the sample used here excludes observations in the fifth year pre-LBO. Columns (1)through (4) of each show results from OLS regressions. Column (5) shows results from an establishment fixed effects Poissonregression. Here, the dependent variable is Injuries and the exposure variable is Employees. LBOFirm is an indicator equalto one if the establishment belongs to a firm acquired in an LBO and zero otherwise. PostLBO is an indicator equal to onein the year after the LBO year and zero otherwise. Log(Employees) equals the log of the establishment’s average reportedemployment for the year. HoursWorked/Employee equals reported hours worked divided by reported average employment.“Minimum employees” refers to the minimum number of establishment employees pre-LBO required for an LBO establishmentto be included in the sample. Standard errors clustered at the firm level are shown below each point estimate. ***, **, and *indicate statistical significance at the 1%, 5%, and 10% level, respectively, based on a two-tailed t-test.
OLS Poisson(1) (2) (3) (4) (5)
LBOFirm 0.0029 0.0034(0.0065) (0.0038)
PostLBO 0.0022 -0.0029 0.0018 0.0024 0.0159(0.0039) (0.0066) (0.0048) (0.0046) (0.0645)
LBOFirm * PostLBO -0.0090** -0.0087*** -0.0060* -0.0065** -0.0991(0.0039) (0.0029) (0.0032) (0.0032) (0.0733)
Log(Employees) -0.0058*** -0.0013 -0.0024 -0.0020 -0.1100**(0.0013) (0.0014) (0.0028) (0.0021) (0.0432)
HoursWorked/Employee 0.0000*** 0.0000*** 0.0000*** 0.0000*** 0.0003***(0.0000) (0.0000) (0.0000) (0.0000) (0.0000)
Establishment FE No No Yes Yes YesYear FE Yes No No No YesYear x Industry FE No Yes Yes Yes NoMinimum employees 100 100 100 50 100
Observations 8,617 8,617 8,617 12,491 8,258Adjusted R2 0.0847 0.1235 0.1710 0.0520
46