THE EFFECT OF EVACUATION FROM HURRICANE
KATRINA ON HOUSTON-AREA EARNINGS
DAKSHINA G. DE SILVA, ROBERT P. MCCOMB, YOUNG-KYU MOH, ANITA SCHILLER, ANDRES J. VARGAS
TEXAS TECH UNIVERSITY
March 23, 2009
ABSTRACT
Hurricane Katrina forced an estimated 240,000 persons to be evacuated to Houston, TX, causing its
population to increase by 3% virtually overnight. Most of these evacuees were younger and less-educated than
existing residents and remained in the Houston area for at least a year. The objective of this paper is to estimate the
effect of this massive in-migration on workers’ earnings in non-tradable goods industries in the Houston
Metropolitan Statistical Area (MSA). An important consideration in analyzing this effect is that, given the
characteristics of the evacuees, their influx would have caused the supply of applicants for lower skilled jobs to
increase proportionately more than for higher skilled jobs. We utilize a difference-in-difference methodology in
which we compare relative earnings per worker within the low- skill non-traded goods industries in the Houston and
Dallas-Fort Worth MSAs before and after the Katrina-induced migration. To control for other localized shocks to
the Houston labor market, we perform the same comparison for high-skill industries. Unlike previous studies in this
realm, we control in addition for the influence of an increase in demand for local goods and services on demand for
labor in normally non-tradable goods and services activities. Failure to account for this induced right-ward shift in
demand for labor will result in a failure to correctly identify the impact on wage rates of the rightward shift in labor
supply due to the sudden in-migration. To do this, we use establishment-level data from the Quarterly Census of
Employment and Wages (QCEW) and gross sales and use tax receipts from the Texas Comptroller of Public
Accounts. We find evidence that the average payroll per employee in the low-skill non-tradable industries decreased
by 3.0% in the Houston MSA relative to the Dallas-Fort Worth MSA as a result of the Katrina-induced shift in labor
supply. We find no evidence of an effect in the set of high-skill non-tradable industries. Our findings also suggest
that failure to control for demand-side influences confounds this effect and results in the mistaken conclusion that
wages increased by 1% in the Houston MSA relative to the Dallas-Fort Worth MSA for both the low-skill and the
high-skill non-tradable industries. Keywords: Migration, Wages.
JEL Codes: J21, J31, J61
____________________________
We thank Jushan Bai and Pierre Perron for the use of their publicly available Gauss code. We also like to thank
Texas Tech University Wind Science and Engineering Research Center for their ongoing support.
I.
INTRODUCTION
Hurricane Katrina struck the Gulf Coast of the United States on August 29, 2005. According to
BLS estimates, an estimated 1.504 million individuals, aged 16 or older, evacuated from their
homes because of Katrina. . By October 2006, 27% or some 280,000 of the evacuees had still
not returned to the area in which they lived prior to the storm. The BLS estimated that about
one-third or about 94,000 of these evacuees were living in Texas in October 2006. Houston was
the most common destination in Texas for evacuees leaving New Orleans (Frey and Singer
2006). Indeed, based on CPS data for November 2005 through August 2006, McIntosh (2008)
reports estimates that the evacuation resulted in a 3.0% increase in the population of Houston.
Most of these evacuees were younger and less-educated than existing residents and remained in
the Houston area for at least a year.
This paper analyzes the impact of this large scale in-migration on average earnings across
a subset of industries within the Houston Metropolitan Statistical Area (MSA). An important
consideration in analyzing this effect is that, given the characteristics of the evacuees, their
influx would have caused the supply of applicants for lower skilled jobs to increase
proportionately more than for higher skilled jobs. We utilize a difference-in-difference
methodology in which we compare relative earnings per worker within the low- skill non-traded
goods industries in the Houston and Dallas-Fort Worth MSAs before and after the Katrinainduced migration. To control for other localized shocks to the Houston labor market, we
perform the same comparison for high-skill industries.
Unlike previous studies in this realm, we control for the influence of an increase in
demand for local goods and services on demand for labor in normally non-tradable goods and
services activities. Failure to account for this induced right-ward shift in demand for labor will
result in a failure to correctly identify the impact on wage rates of the rightward shift in labor
supply due to the sudden in-migration. To do this, we use fully disclosed establishment-level
data from the Quarterly Census of Employment and Wages (QCEW) and gross sales and use tax
receipts from the Texas Comptroller of Public Accounts. We find evidence that the average
payroll per employee in the low-skill non-tradable industries decreased by 3.0% in the Houston
1
MSA relative to the Dallas-Fort Worth MSA as a result of the Katrina-induced shift in labor
supply. We find no evidence of an effect in the set of high-skill non-tradable industries.
It should not be surprising that demand-side influences are important. The evacuees were
provided housing assistance and income support. In view of their numbers, they would have
caused a shift in the local demand for goods and services. Moreover, given the proximity of
Houston to the affected areas, it seems likely that both short-term and long-term recovery and
rebuilding efforts in the context of severely disrupted regional distribution channels would have
turned to the Houston market for material and expertise. The post-storm scarcity of local goods
in the affected region may have effectively resulted in the export of goods from Houston that
normally would only be demanded by local consumers. The increase in demand for goods and
services as a result of either or both of these circumstances would induce a non-negligible shift
in demand for labor. Our findings also suggest that failure to control for demand-side
influences confounds the effect of the rightward shift in labor supply and results in the mistaken
conclusion that wages increased by 1% in the Houston MSA relative to the Dallas-Fort Worth
MSA for both the low-skill and the high-skill non-tradable industries.
The paper is organized in three sections. In the first section we discuss briefly the
theoretical basis and the existing literature on immigration’s impact on labor market outcomes.
In the second section, we analyze the time-series of average payroll data by industry to compare
the apparent structural breaks that occurred in the Dallas and Houston MSAs over the period
1996-2006. We seek to determine if the two MSAs exhibited similar behavior before Katrina
and if there was divergence at the time corresponding to the Katrina evacuation. In the third
section, we conduct a differences-in-differences analysis of changes in payroll per employee
(average payroll) at both the firm and establishment levels before and after the storm in the two
MSAs.
II.
THEORETICAL BASIS AND PREVIOUS RESEARCH
Basic economic theory predicts that an outward shift in the labor supply curve will result
ceteris paribus in a decrease in the equilibrium wage rate and an increase in the level of
2
employment. Whether the total average payroll increases or decreases is a matter determined by
the elasticity of demand for labor services. On the other hand, an increase in the demand for
labor will lead ceteris paribus to an increase in both the equilibrium wage rate and employment
level, in which case an unambiguous increase in total average payroll will be observed. Demand
for labor is derived partly from conditions in firms’ output markets. As is normally assumed,
when demand for a firm’s or industry’s output increases and market prices rise, firms respond by
increasing output by means of engaging additional factors of production.
Most analyses of the impact of immigration on native or existing workers recognize that
local labor markets are segmented along occupational lines. For example, Card (2001) makes
the assumption that local labor markets are “stratified” along occupational lines. In his model,
he supposes that the market effect of a net increase in the local supply of workers within a given
occupation will be largely restricted to that occupation. As such, downward pressure on relative
wage rates will be greatest in the occupations in which immigrants substitute most readily for
existing workers. This view is predicated on the assumption that the occupational composition
of the immigrant inflow differs from the occupational composition of the existing labor force.
Otherwise, as Card (2001) notes, an immigrant population that has the same occupational shares
as the native population will leave relative wage rates unchanged.
If immigrants tend to be younger and less-educated than existing residents, immigration
will cause the supply of applicants for lower skilled jobs to increase proportionately more than it
will for higher skilled jobs. Since occupational employment varies across industries, not all
industries will be similarly affected by the shift in labor supply. A brief review of the Bureau of
Labor Statistics’ staffing patterns by industries leads to the conclusion that local, nonprofessional services, retail, and other non-tradables tend to employ lower-skilled occupations.
If this is the case, the effect of the Katrina-induced shift in labor supply should be most readily
observed in the set of non-tradable goods industries and, even more specifically, within the
subset of those industries that tend to have relatively more low-skill occupational employment.
Altonji and Card (1991), noting that many of the goods produced within a city are nontradables or enjoy some degree of imperfect substitutability due to transportation costs, suggest
“it is more reasonable to posit the existence of downward sloping labor demand functions at the
3
local level.” If this is the case, local and regional labor markets for the occupations represented
in the non-tradable activities would a fortiori be expected to reflect the standard model’s
forecast. Wage rates in these industries can thus rise, fall or remain unchanged depending on
the relative magnitudes of the shifts in the labor supply and demand curves. More importantly, if
demand is properly accounted for, we would expect relative wages to fall in those industries with
occupational employment that most closely matches the occupational characteristics of the
immigrant population.
There is an extensive literature on the effects of immigration on native populations’ labor
market outcomes. The use of inter-city comparisons has been a common methodology to
estimate relative labor market impacts in the U.S. where immigration rates have varied across
cities. However, this approach has been criticized due to the potential endogeneity of
immigration patterns and employment opportunities (see Carrington and de Lima, 1996 for an
overview). That is, migrants seeking employment opportunities move to areas with higher real
wage and/or employment rates. Thus, immigration itself leads to a leveling of relative real wage
and employment rates and masks the evidence of its effect. Moreover, an in-migration may
induce an out-migration of natives that blunts the immigration-induced shift in labor supply.
Lastly, if the net immigration substantially augments the labor force, it may serve to attract
labor-intensive industries that effectively absorb the increased labor supply to the point that wage
rates adjust back toward the pre-immigration equilibrium.
As a result of these criticisms, researchers have sought examples where the immigration
is clearly non-economic. For example, it may be a result of a natural hazard, political event, or
conflict. Card (1990) investigated the effect of the Mariel Boat-lift of 1980 on the Miami labor
market where the “Mariel immigrants” increased the Miami labor force by 7%. He points to the
exogenous nature of the immigration as enabling a “natural” experiment in the sense that
immigrants were not responding to economic incentives, rather to political circumstances.
Similar to the Katrina evacuees, the Mariel immigrants tended to be relatively less-skilled than
the native labor force in Miami. However, Card notes that the Mariel influx appears to have had
virtually no effect on the wages or unemployment rates of less-skilled workers. He raises the
point that if the Mariel immigrants displaced other immigrants and residents, the net effect on
4
wages would be a wash. He also considers the possibility that rapid growth in industries that
employ low-skilled persons may have enabled the rapid absorption of low-skilled immigrants.
He concludes, however, there was little change in the relative distribution of immigrant-intensive
industry in the Miami area.
Kugler and Yuksel (2006) showed that a large influx of lesser-skilled Central American
immigrants after Hurricane Mitch, who were quickly legalized and who moved towards Southern
ports of entry in the U.S., generated positive wage effects for college- and high school-educated
native men and women and earlier Latin American immigrant men, but not for less-skilled
workers. Hence, they suggested that low-skilled immigrants complement high-skilled natives.
But they also found some evidence of negative employment effects on less-educated natives
suggesting that immigrants may substitute for less-skilled natives.
McIntosh (2008) considered how wages and employment among Houston natives
changed with the addition of Katrina evacuees. Using CPS data for Houston and other
metropolitan areas not affected by the storm, she reported a 1.8% decline in wages and a 0.5%
decline in the probability of being employed among Houston natives after the storm. McIntosh
assumed that the shift in labor demand was much smaller in magnitude than the shift in labor
supply and, therefore, her analysis does not account for demand-side effects that the increased
regional purchasing power the migrants brought and hurricane recovery would have engendered.
Among the studies of immigrations outside the U.S. are Hunt’s (1992) examination of the
impact on the French labor market of the 900,000 repatriates from Algeria in 1962 and
Carrington and de Lima’s (1996) study of the 600,000 retornados who migrated to the much
smaller Portugal from Angola and Mozambique in the mid-1970s. Both studies are chosen to
take advantage of the economically exogenous nature of these population movements. Neither
study finds persuasive evidence of an impact on wage rates from immigration. In the French
case, the pieds noirs tended to be relatively high-skilled. Although her data do not permit her to
control for the effects of unionization in France, Hunt concludes that, despite the repatriates’
unemployment rates remaining well above those of non-repatriates for the lengthy period of her
study, there may have been only a weak effect on French wages at the national level. She does,
however, find evidence of regional variation in the effect on wages.
5
Carrington and de Lima (1996) in their study of the retornados (repatriates) from Angola
and Mozambique in the mid-1970s attempt to control for the unstable political environment in
Portugal during the period by using Spain as a control. The Spanish political experience and
institutional landscape was similar to Portugal’s during the decade. They find prima facie
evidence that Portuguese unemployment rates were pushed upward by the immigration, but note
that both Spain and France quickly followed suit, despite not having had any significant
immigration wave. They also fail to find much evidence of a significant impact on wage rates.
They conclude that comparisons of Portugal with Spain and France were quantitatively swamped
by the Europe-wide downturn in labor market conditions in the 1970s.
They also undertake a comparison between districts within Portugal although data
availability limits their analysis to the construction industry. It is worth noting that immigrants
were in fact under-represented in construction. Their intra-national comparison indicates that the
retornados may have had a strong adverse effect on Portuguese wages. However, the authors
express skepticism over these results and state their preference for the findings of the
international comparison.
As Card (2001) notes, “Overall, the evidence that low-skill workers are hurt by
immigration has not been particularly compelling.” In his study of the effect of immigration into
the U.S. during the decade of the 1980s on occupation-specific labor market outcomes, he finds
that wage and employment rates within occupations are lower in cities in which the relative
supply of the given occupation is higher. Where immigration has increased the relative supply of
specific occupations, it has affected wage and employment rates, although only modestly. He
finds that the immigration of the 1980s reduced the relative wages of laborers and lower-skilled
natives by at most 3% in high-immigration cities, and less elsewhere.
It remains the case that most studies on the effects of immigration on local labor market
conditions have found little evidence that wage rates fall significantly. Yet none of these studies
explicitly accounts for shifts or exogenous shocks in demand.
A proximate catastrophic hurricane and resultant large scale in-migration, such as was
observed in Houston toward the end of 2005, clearly imply a proportionate increase in demand
for certain imperfectly localized goods and services such as personal and retail services,
6
transportation, housing, etc. In contrast, demand for production that is largely exported from the
local economy and which trades in national or international markets will not be affected by local
or even regional demand shocks. There is thus a change in demand, attributable to the event,
only for localized goods and services. This then also implies an asymmetric response across the
local industrial landscape in terms of the induced demand for labor services.
An additional and important theoretical consideration is that real sales of localized goods
and services are endogenous. As opposed to a demand-side shift in output markets that causes
non-tradable industry output and demand for labor to increase, a drop in real wages due to a shift
in labor supply will result in lower marginal production costs, a decrease in output prices, and an
increase in equilibrium quantities. If the demand for output is elastic, the final effect will be an
increase in real sales.
How important might this supply-side effect in output markets be? Cortes (2006)
examined the impact of immigration in the U.S. in the 1990s on output prices of non-traded
goods. Excepting gardening and housekeeping, two industries that have very high
concentrations of non-native low-skill labor, she finds that a 10 percent increase in the share of
low-skilled immigrants decreases the prices of non-traded goods and services by 0.7 percent.1
She concludes that lower wages due to immigration are likely behind these price decreases. It is
not obvious how her results would apply in the post-Katrina Houston context, however. While
she finds that wage effects are much larger (and negative) for low-skilled immigrants than for
low-skilled native workers, she notes that this result appears to be related to English proficiency
and, perhaps, legal status, both of which contribute to imperfect substitution between native and
non-native low-skilled workers. Neither of these characteristics is generally operative in the
influx of Katrina evacuees into Houston. It is more likely that substitutability between lowskilled Katrina refugees and native Houstonians would be substantially greater, tending to be
near perfect substitutes.
1
She finds that a 10% increase in the share of immigrants in the labor force decreases prices in strictly local services
such as gardening and housekeeping by 2.1%, and other non-tradables by 0.9%. This analysis does not include
employment in the gardening and housekeeping industries except as those activities are subsumed in the building
services industries, and the employed individuals are covered by unemployment insurance.
7
A demand shock due to the hurricane itself is clearly exogenous. Thus the destruction of
infrastructure, productive capacity, and distribution channels in the region that was affected
directly by the hurricane would have an effect on demand in neighboring regions. Houston is the
largest and one of the closest metropolitan areas to New Orleans. Response and recovery inputs
would have been gotten from Houston. Certain goods that under normal circumstances would be
non-tradable goods due to transportation costs, such as retailing and other services, would
become exportable to the affected region. One example might be automobile sales and service in
the Houston area following the widespread damage and destruction of the fleet in the affected
Gulf region. Lastly, a significant portion of evacuees to Houston would have viewed themselves
as temporary residents but, having income and other resources available, would have presented
effective local demand for housing and services.
Separating the endogenous from the exogenous component in the observed shift in
demand is beyond the scope of this paper. We recognize the difficulty in precise interpretation
of the results that are based on controlling for demand for localized goods and services. We take
the position that, at least in Houston in the aftermath of Katrina, there is a clear rationale for
controlling for output market demand shocks due to the hurricane.
Our argument is, then, that if immigrants’ occupational characteristics and skill levels are
best suited to the occupational employment in the low-skill non-tradable goods industries, the
effect of the shift in labor supply would be greatest within those industries. If, by the same
token, demand shocks associated with the event are also concentrated in the low-skill nontradable goods industries, the demand-side influences will be largest for those occupations in
which the migrants seek employment. Therefore, controlling for demand is essential when
searching for the immigration’s supply-side effects on wage rates and employment within this
subset of industries.
Post-Katrina Houston
For the Houston MSA, net domestic migration measured between mid-year points was slightly
negative for 2003-2005. However, in 2005-2006, the Katrina time-frame, it was a positive
8
81,783.2 This represents an increase of about 1.8% in the region’s population due solely to net
migration in the ten months following the storm. Although the demographic profile of the
evacuees reflected the demographic profile of the affected regions, it differed in significant ways
from pre-Katrina Houston. For example, nearly one-third of the evacuees who did not return to
their previous area of residence did not have a high school diploma. This compares to 23% of
the residents of the Houston MSA who did not have a high school diploma at the time of the
2000 Census. Slightly more than 40% of the non-returnees were African-American, compared to
about 17% of the population of the Houston MSA in the 2000 Census.
It also appears that the evacuees that did not return to their previous areas of residence
had some difficulty assimilating into their new locality’s labor force. Groen and Polivka (2008)
provide information on the employment outcomes of evacuees who returned to their pre-storm
area of residence and of those that did not return. They report that the employment-population
ratio for non-returnees was 26 percentage points lower than the same ratio for residents of
unaffected areas and about 20 percentage points lower than the ratio of evacuees that returned to
their pre-storm areas. This of course may reflect the fact that individuals with known
employment opportunities in their pre-storm area of residence were more likely to return than
those evacuees whose employment opportunities were more limited regardless of whether they
returned or not.
III. DATA
In this study, we use two data sources to empirically examine the effect of in-migration
on the Houston economy due to Hurricane Katrina. We obtained fully disclosed payroll and
employment data for Texas from the Quarterly Census of Employment and Wages (QCEW)
from the Texas Workforce Commission. These data provide establishment-level monthly
employment3 and quarterly total payroll for workers covered by unemployment insurance (UI).
2
Texas A&M University Real Estate Center tabulation of U.S. Census data for the Houston and Dallas MSAs. Can
be seen at http://recenter.tamu.edu/Data/popm/pm3360.htm and http://recenter.tamu.edu/data/popm/pm1920.htm.
3
Monthly employment data under the QCEW program represent the number of covered workers who worked
during, or received pay for, the pay period including the 12th of the month. Excluded are members of the armed
forces, the self-employed, proprietors, domestic workers, unpaid family workers, and railroad workers covered by
the railroad unemployment insurance system. Wages represent total compensation paid during the calendar quarter,
regardless of when services were performed. Included in wages are pay for vacation and other paid leave, bonuses,
stock options, tips, the cash value of meals and lodging, and in some states, contributions to deferred compensation
9
Reporting is required under the Texas unemployment insurance (UI) program for all firms with
UI liability. Each observation includes the specific location (address) of the establishment,
business start-up date (the date on which UI liability begins), and the relevant six-digit North
American Industry Classification System (NAICS) code. Note that separate establishments
(branches or franchises) of the same firm are identified separately by unit number and address
under the parent firm's Employer Identification Number (EIN) and UI account number. This
panel data set runs from July 1999 through December 2007 and the full data set has more than 40
million observations for the State of Texas.
An important limitation to the QCEW data is that only aggregate payroll information is
provided. Since this variable is the product of total hours worked and wage rates, no specific
information on hours worked or wage rates can be directly identified. Inasmuch as there is an
inverse relationship between wage rates and quantity of labor demanded, short-run flexibility in
wage rates will result in offsetting changes in quantity of labor demanded, all else equal, and the
actual effect on wage rates of a shift in labor supply will be obscured by the use of average
payroll data. However, if labor demand elasticities are less than one in absolute value, wages
and average payrolls will move in the same direction. According to Hamermesh (1993), a
reasonable confidence interval for the absolute value of the elasticity of labor demand in the
manufacturing sector is [0.15, 0.75] with a point estimate of 0.30. Although other sectors of the
economy might have different point elasticities, Hamermesh states that a random firm should
have an elasticity of demand within this confidence interval. Thus the demand for labor is
reasonably assumed to be inelastic. Cotterill (1975) supports this view. He provides evidence
that elasticities of demand for low-skill labor tend to be consistently less than one in absolute
value. If so, the sign of changes in average wage rates can be inferred from the sign of changes
in average payrolls. Moreover, since wages and hours worked would be expected to move in
opposite directions, if the elasticity is in the interval (-1, 0), the percentage change in wage rates
(in absolute value) would be greater than the percentage change (in absolute value) in payrolls.
Therefore, a finding of a decrease in average payroll would imply a more than proportionate
decrease in average wage rates.
plans (such as 401(k) plans). The QCEW program does provide partial information on agricultural industries and
employees in private households.
10
The importance of controlling for labor demand is clear when investigating the impact of
migration on wages.4 Indeed, the analysis below suggests that significant demand-side effects
are present. Hence, controlling for the demand side effects becomes critical in examining the
impact of the Hurricane Katrina evacuees on Houston MSA industry wage rates compared to
Dallas MSA wage rates. We use gross sales receipts, collected from the Texas Comptroller of
Public Accounts, to control for demand-side effects. The unit of observation in the sales data is
quarterly gross sales by industry at the four-digit NAICS for a given county. This data set
extends from Q1:2002 through Q4:2007. It should be pointed out that the authors obtained both
of these data sets under an agreement of confidentiality and disclosure of the actual data is
subject to certain restrictions.
Since our analysis is confined to the differences between the Houston and Dallas-Fort
Worth MSAs, we use only Houston and Dallas MSA-level data. This reduces our sample to
slightly more than 5 million quarterly observations. We then exclude all tradable goods and
service industries, industries that are not common to Houston and Dallas, self-employed workers,
and firms that do not exist both before and after Hurricane Katrina. Our selection of nontradable industries is based on the work of Jensen and Kletzer (2005). They use the geographic
clustering of activities in the United States to identify industries that are traded domestically. In
particular, using employment information from the 2000 Decennial Census of Population Public
Use Micro Sample (PUMS), they estimate Gini coefficients of geographic concentration for each
industry at the Metropolitan Statistical Area level.5 They categorize industries with a Gini
coefficient below .1 as non-tradable and industries with a Gini coefficient greater than or equal to
.1 as tradable. This measure of tradability relies on the assumption that returns to scale, access to
transportation nodes, and proximity to natural resources lead tradable firms to congregate near
each other while non-traded activities will not exhibit geographic concentration in production.
4
Altonji and Card (1991), Borjas (2003), Ottaviano and Perri (2006).
The Gini coefficient in this context considers the distributional share of industrial activity relative to the
distributional share of population across MSAs in the United States. A Gini coefficient of zero indicates a perfectly
even distribution of a given industry’s total employment relative to the distribution of population, and thus suggests
strictly localized production and consumption of the given industrial output.
5
11
Tables 1.a and 1.b. provide a complete list of the 82 non-tradable industries we selected by 4digit NAICS sector.6
Given that most Katrina evacuees were younger and less-educated than existing Houston
residents, the effect should be most readily observed in the set of non-tradable industries that
tend to have relatively more low-skill occupational employment. Therefore, to focus our
empirical approach, we classify the set of non-tradable activities into low-skill and high-skill
industries. To do this, we use wages as a measure of skill relying on the assumption that the
determinants of productivity also determine wages. Industries in which the average wages of the
majority of establishments are on the lower end of the wage distribution are classified as lowskill, while industries where the average wages of a majority of establishments are at the upper
end of the wage distribution are classified as high-skill. Tables 1a and 1b show the number of
workers and establishments for these two groups of industries. The low-skill sector has 27
industries comprising 14,662 establishments and 487,894 workers in Houston. In Dallas, there
are 17, 028 establishments and 539,182 workers in the set of low skill industries. In the 55
industries that constitute the high-skill sector, Houston has 16,057 establishments with 721,255
workers as compared to 18,188 establishments with 774,740 workers in Dallas.
Table 1.a shows that industries in the entertainment and recreation sectors and in the
accommodation and food services sectors are primarily classified as low-skill. Among these,
full-service restaurants, limited-service eating places, grocery stores, and services to buildings
and dwellings comprise close to 50% of the low-skill totals in the two regions in terms of
employment shares. The data in Table 1.b suggest that high-skill industrial employment is
distributed across a larger number of industries, among which the healthcare and education
sectors, business and accounting services, non-residential construction, and automobile related
activities are relatively important. In retail trade activities, however, industries are balanced
between the low-skill and the high-skill sectors. Finally it is worth underscoring that all of the
included industries are equally represented in Houston and in Dallas in terms of both
employment and establishments.
6
Industries not equally represented in Houston and in Dallas were excluded from the sample. Offices of physicians
were also excluded from the sample due to problems in the measurement of their wages.
12
Figures 1 and 2 present the density function of the log of real average quarterly wages
for low and high-skill industries in Houston MSA and Dallas MSA before and after August
2005. Figure 1shows that quarterly wages for Houston-area low-skill industries shifted left
(declined) while high-skill wages remained unchanged after Hurricane Katrina. It also suggests
that the distributional shift for low-skill industries can be observed for all ranges of wages. In
contrast, Figure 2 shows that Dallas MSA’s wages for both low and high-skill industries show a
slight rightward shift indicating that wages have increased since August 2005. While Figures 1
and 2 suggests that low-skill industry wages declined in the Houston MSA after Hurricane
Katrina, we need to be cautious in interpreting these data. In the next section, we look for the
presence of structural breaks in industry-level wages after Hurricane Katrina.
III.
TIME SERIES MODEL ESTIMATION
Timing of the Katrina Impact
An important question to resolve in an analysis of the impact of the Katrina evacuation on the
Houston labor market is when that impact actually occurred. That is, recognizing that most
evacuees anticipate returning to their previous place of residence, most of the evacuees would
not likely enter the labor force until such time that they concluded that return was not possible or
preferable. Based on anecdotal reporting, many evacuees from New Orleans may have decided
fairly early after the event that a new start in a new place was preferred to going back. Indeed,
elements of the profile of non-returnees would seem consistent with a person who would tend to
be less risk-averse when considering such a change, i.e., younger, single and fewer children.
Katrina evacuees who remained outside of the directly affected area were also provided
with both financial and in-kind assistance as the circumstances in New Orleans, in particular,
limited the possibilities for their early return. For example, qualifying households were given
$2,358 as an initial payment for three month’s rental assistance, with the possibility of extension
for up to 18 months. Disaster Unemployment Assistance provided unemployment benefits for a
period up to 26 weeks to evacuees not covered by other unemployment insurance. Groen and
Polivka (2006, p. 47) report that 42.4% of non-returnees received some form of governmental
13
assistance. Those that did not qualify or apply for benefits must have had other resources at their
disposal. Given this financial support, there would have been some degree of discretion in
entering the labor market in the months following the evacuation.
In subsequent sections of this paper, we estimate the effect of the Katrina evacuation on
the Houston-area labor market by comparing the pre and post-storm differences in the difference
between average payrolls per employee by establishment and industry in the Houston and
Dallas-Fort Worth MSAs. The choice of Dallas enables us to control for state-specific (state
taxes and budget policies, for example) and broader regional influences on the markets in the two
sub-regions. Both, for example, were buffeted by the downturn in the technology sectors that
occurred early in the decade. We must however first address two questions. Prior to the storm,
did the relevant variables in these two regions move together? Were both regions similarly
affected by Katrina? While the Houston MSA realized a net migration of nearly 82,000 between
mid-2005 and mid-2006, the Dallas MSA itself had a net migration of some 43,000. For Dallas,
however, that was a difference of 32,000 (or 0.8%) over the previous year, whereas for Houston
the difference was 82,000 (or 1.7%) over the previous year or roughly twice as large for Houston
in relative terms.
We consider the possibility that the effect of Katrina on the Houston MSA labor market
may not have been immediate, but rather a delayed impact. Since evacuees tended to be
younger, single and less educated than the Houston population, employment opportunities would
be greater in industries which use relatively more low-skilled individuals. Thus, it is reasonable
to allow for the possibility that the delayed effect on the Houston MSA labor market was
asymmetric across industries rather than symmetric. For these reasons, we consider average
payroll per employee by industry and treat the break points as unknown a priori. This represents
a different approach to explore the labor market impact of the in-migration than those taken by
the existing studies on this issue. Moreover, this study accounts for the possibility of delayed
effects with unknown break points.
Structural Break Model
Bai and Perron's (1998, 2003) method (BP method) for estimating unknown multiple structural
breaks in dynamic linear regression models largely consists of two stages. The first stage is
14
testing and estimating the number of unknown structural breaks. For this, Bai and Perron
suggest several testing procedures. The first one is double maximum tests constructed under the
null hypothesis of no structural break against the alternative of an unknown number of breaks
given some upper bound. Another one is the sup FT (l+1|l) test that tests the null of l breaks
against the alternative of l+1 breaks. Bai and Perron recommend applying the double maximum
tests first to see whether at least one break exists. If the tests suggest the presence of at least one
break, then the number of breaks is determined sequentially following the sup FT (l+1|l)
statistics. According to Bai and Perron, this sequential approach leads to the most reliable results
and hence is recommended for empirical applications.
After identifying the number of breaks, the second stage of the BP method is to estimate
the break points along with coefficients of interest using the least squares principle. Consider a
linear regression model with m-1 breaks (m+1 regimes) which are identified in the
first stage),
y t   j   t , t  T j 1  1,..., T j ,
for j = 1, …, m+1 where δj is the mean level of the series in the jth regime.
For each m-partition of T, (T1, ..., Tm), least-squares estimates of coefficients (δjs) are obtained by
minimizing the sum of squared residuals
m 1
Tj
 [ y
j 1 t T j 1 1
j
  j ]2 .
Let ST (T1, ..., Tm) be the sum of squared residuals such that
m 1
S T (T1 ,..., Tm )  
Tj
[ y
j 1 t T j 1 1
j
 ˆ j ] 2 .
Then the break points estimates (Tˆ1 ,..., Tˆm ) are obtained as
(Tˆ1 ,..., Tˆm ) = argmin T1, ..., Tm ST (T1, ..., Tm).
With these estimates of break points, the associated regression parameters are then estimated.
Bai and Perron develop an efficient algorithm for the minimization problem based on the
principle of dynamic programming.
Structural Break Model Estimation Results
15
We applied the Bai and Perron (1998, 2003) structural break test methodology to quarterly real
average payrolls by industry in Houston and Dallas-Fort Worth from Q3:1999 – Q4:2007.
Average real payroll by industry is computed by summing total real quarterly payroll by 4 digit
NAICS code and dividing by total employment during the quarter in the given industry. The
quarterly CPI for the relevant quarter, available from the BLS for both the Houston and DallasFort Worth metropolitan areas, was used to deflate nominal payroll data. We consider 82 nontradable goods industries.
Table 3.a presents estimates of the post-Katrina structural break points and their 95%
confidence intervals for average real payroll by industry. The table shows that structural breaks
in real wages are more pronounced in the Houston MSA compared to the Dallas MSA. In
Houston, 12 out of 82 industries experienced abrupt changes in average real payrolls following
Hurricane Katrina while only one industry in Dallas had a break. The industries that appear to
have experienced a structural break are in line with one’s intuition. That is, localized goods and
services activities were affected. The specific industries that experienced structural breaks were
general rental centers, retailers, personal and household goods repair and maintenance, and
services to buildings and dwellings. Structural breaks occur with different delays after the
Hurricane between Q3:2005 and Q3:2006.
However, when we estimate structural breaks without considering establishment size, it is
possible that the establishment specific effects might be washed out by aggregation. Tables 3.b
and 3.c contain the structural break points estimation by industry and by establishment size.7 The
impact of Hurricane Katrina can be seen more clearly when we explicitly consider establishment
size. 51 out of 82 industries in Houston MSA experienced breaks compared to 19 out of 82
industries in Dallas MSA. The tables also indicate that the impact of the hurricane differs by
establishment size as well as by industry since not all the firms in the same industries were
equally affected. While structural breaks occur with different delays after the hurricane, there is
no commonality regarding establishment sizes. For example, both community care facilities for
the elderly and shoe stores show breaks in Q3:2006. However, the impact of the hurricane was
7
Establishment size 1 = average employment is less than 11; Establishment size 2 = average employment ≥11 and <
26; Establishment size 3 = average employment ≥26 and < 51 ; Establishment size 4 = average employment ≥51 and
< 101; and Establishment size 5 = average employment ≥101 .
16
mainly on smaller firms in community care facilities for the elderly while it was on the larger
firms among shoe stores.
These findings support our conjecture that Katrina had an effect on average payrolls in
Houston, but not Dallas. Of interest, however, is that this analysis suggests that the effect,
where present, was predominantly positive. This would be more consistent with a labor market
scenario in which wage rates were rising (given inelastic demand for low-skilled labor). This
suggests a demand-side effect that appeared within a very short time following the hurricane and,
contrary to a priori expectations, had a positive effect on wages and/or hours worked. These
findings underscore the importance of controlling for demand-side effects in an analysis of the
impact of the Katrina event on Houston labor markets.
IV.
Difference in Difference Estimation
Choice of period for Differencing
The structural break tests indicate that twelve industries in the Houston MSA had payroll breaks
just after Hurricane Katrina in 2005 in the Houston MSA, while only one industry in the Dallas
MSA experienced such breaks in that period. This evidence supports our conjecture that the
evacuation from Hurricane Katrina had an effect on Houston-area wages. There were however
other structural breaks in the Houston economy prior to 2005.8 According to Sanchez (2002),
the effect of the Enron Corporation collapse, Continental Airlines job cuts, and employment
reductions at Compaq Computer Corporation following the merger between Compaq and
Hewlett-Packard in 2002 all contributed to slow economic growth in the Houston MSA. Losses
from flooding from tropical storm Allison in June 2001 topped $6 billion in the Houston while
insured losses were only $2.5 billion. Arthur Andersen, LLP also laid off 7,000 of its employees
following its dissolution in the wake of the Enron failure.9 Sanchez (2002) states that the
economic effects of these incidents and 9/11 lingered for a long time.10 We believe these effects
8
These results can be provided upon request.
http://www.businessweek.com/magazine/content/02_12/b3775006.htm
10
Also see: "Enron collapse latest shock to Houston's economy," Forbes.com, January 29, 2002 (Reuters);
"Continental to furlough 3,000 in Houston, cut back flights," Houston Chronicle.com, September 15, 2001; "The
Impact of September 11 on U.S. Metropolitan Economies," The Milken Institute, January 2002;
9
17
quite likely continued through 2003 and, hence, use data beginning in December 2004 in the
differences-in-differences analysis.
In the difference-in-difference estimations, we use establishment-level real quarterly
payroll per covered employee based on average monthly employment during the quarter. Thus
the use of the term average payroll refers to values computed using this definitions. Real gross
sales data in the difference-in-difference analysis are aggregated to the industry and county
levels. The final data set contains 501,285 low-skill and 541,235 high-skill establishment-level
quarterly observations.
Table 2 provides summary statistics for low and high skill establishments in Houston and
Dallas before and after Hurricane Katrina. In both the low and high-skill industries real average
payrolls per employee are higher in Dallas than in Houston.11 For the low skill industries,
average wages in the Houston MSA is about $200 less than Dallas MSA for both before and
after Hurricane Katrina. This difference is about $300 for high skill non-tradable industries. The
difference in average wages for low and high skill industries are about $3,500 for both MSAs.
Table 2 also shows that the number of employees per establishment goes down after Hurricane
Katrina, especially for low-skill firms in the Houston MSA. However, employment ratios12 for
both MSAs are similar and show virtually no change after Hurricane Katrina. When considering
gross real sales after Katrina, we see that Houston had higher sales volume than Dallas.13
Model and Estimation Results
Given the labor skill characteristics of the Katrina evacuees, the effect of the induced
migration should be most readily observed in the industries that tend to use relatively more lowskill occupational employment. The differences-in-differences estimation approach that we use
http://pages.stern.nyu.edu/~tphilipp/papers/NewYorkTimes.pdf; and
http://news.bbc.co.uk/1/low/business/2007255.stm
11
Real payroll per employee is constructed by deflating the establishment level quarterly payroll per employee by
the CPI of the MSA where they are located.
12
The employment ratio is calculated by dividing the average number of employees for a given firm (or
establishment) in a given quarter by the total average number of employees in that industry (defined by the fourdigit NAICS) for a given county during the quarter. There may be a wage effect associated with firm size. Larger
firms may pay higher wages due to economies of scale associated with human resource functions and benefit
package costs.
13
Real gross sales are constructed by deflating the current gross sales by the CPI published by the BLS for the
relevant MSA.
18
compares the difference in average real payroll per person in low-skill industries at the
establishment level in the Houston and Dallas MSA counties before and after the population
influx generated by hurricane Katrina.
The estimation procedure can be summarized in the following reduced form wage equation:
y imt   0' H m  1' A   2' ( H m  A)   3' Eimt   4' S imt   5' ( S imt  H m )
  6' ( S imt  A)   7' ( S imt  H m  A)   imt
(1)
where yimt is the log of the average real payroll per employee of establishment i in metropolitan
area m at time t, Hm and A are fixed effects dummies for the Houston MSA and the post-Katrina
period, (Hm × A) is the interaction of the Houston MSA dummy with the post-Katrina period
dummy. This interaction term is intended to capture the magnitude of the supply side effect of
the evacuation on the Houston MSA average payroll per employee. Eimt is the average
employment of the establishment relative to the average employment of the industry in the
county in a given quarter, and Simt are the sales of the given industry relative to the average sales
of that industry in the county for the sample period. The interaction (Simt × Hm) controls for the
Houston MSA specific effects of the industry’s sales, the interaction (Simt × A) controls for a
differential effect of the industry’s sales during the post-Katrina period, and the interaction (Simt
× Hm × A) captures the effect of the industry’s sales on the Houston MSA during the post-Katrina
period. Lastly, imt is a random error term.
The estimator in equation (1) assumes that, in the absence of the Katrina induced
migration, changes in the average wages of establishments in low-skill industries would have
been similar for Houston and for Dallas. Given that we expect no effect of the in-migration
generated by Hurricane Katrina on the average payroll of firms in high-skill industries, we can
use the change in wages in high-skill industries to control for particular shocks that may have
affected the Houston labor market. For this purpose, we estimated the high-skill analog to
equation (1). In this regression, changes in the average wages of high-skill Houston firms,
relative to Dallas firms, are assumed to reflect Houston specific period effects. Therefore, if the
Katrina in-migration reduced the wages of lower-skilled Houstonians, one would expect to
observe a decline in the average wages of firms in the low-skill industries, at least relative to
firms in the high-skill industries.
19
Table 4 reports results of the establishment-level fixed-effects estimation of equation (1)
for low-skill and high-skill non-tradable industries. To observe the effect of including controls
for the induced increase in demand for labor services, we estimate the fixed effects regressions
with and without industry sales and the corresponding interactions. Columns 1 and 3 include
controls for demand-side effects, whereas columns 2 and 4 do not. The first row shows the
change in the average payroll per person that is common to both the Houston and Dallas MSAs
after Hurricane Katrina. The second row compares the difference in average payrolls between
the Houston and Dallas MSA counties before and after hurricane Katrina and measures the effect
of the population influx on the relative average payrolls of firms in the Houston MSA. The
effects of the employment ratio on average payrolls are in the third row. Finally, rows 4 through
7 show the effects of the increase in demand for labor services on firms’ average payrolls.
The entry in the first row of column 1 indicates that in the period following the inmigration generated by Hurricane Katrina, there is a significant 1.5 percent increase in the
average wages of low-skill establishments in the Dallas MSAs. In the Houston MSA, however,
there is a 1.5 percent drop in the average wages of low-skill establishments during the same time
period. Therefore, in the post-Katrina environment, there is a 3 percent statistically significant
reduction in the average payrolls per employee for low-skill firms in the Houston MSA relative
to the payrolls of those in the Dallas MSA as shown by the differences-in-differences coefficient
in the second row of column 1.
Regarding the effect of the increase in demand for labor services on firms’ average
payrolls, the coefficient in row 4 indicates that a one percentage point increase in an industry’s
relative sales generates a 4.4 percent increase in the average payroll of firms in that industry.
Based on the results reported in row 5, this effect appears to be only 1.2 percent for low-skill
firms in the Houston MSA before Hurricane Katrina. Furthermore, row 6 suggests that the effect
of sales on average payrolls decreases to 1.6 percent ( a 2.8 percentage points reduction) for
firms in the Dallas MSA after Hurricane Katrina, while the corresponding effect in the Houston
MSA increases by 2.7 percent (an increase of 1.5 percentage points) in the post Katrina period.
Thus, the difference in difference coefficient in row 7 suggests that there is a 4.2 percentage
point increase in the effect of sales on the average payroll of low-skill firms in the Houston
20
MSA after Hurricane Katrina, relative to the effect of sales on the average wages of comparable
firms in the Dallas MSA. In fact, all these effects are statistically significant at conventional
levels. It is important to note that the negative impact of the increase in labor supply generated
by Katrina on Houston’s average payrolls, as seen in row 2, is dominated by the positive effect
of the increase in sales measured in row 7. The fact that the difference-in-difference coefficient
becomes 1percent if we fail to control for demand-side effects, as shown in row 2 of column 2, is
further evidence of this point and highlights the importance of controlling for changes in demand
for location-specific goods and services to be able to identify the impact on wages from the sharp
increase in labor supply.
To control for idiosyncratic shocks affecting the Houston labor market, we perform the
same exercise for the control group of firms in high-skill industries. Column 3 in Table 4
presents the estimation results for high-skill industries when controlling for establishment-level
fixed effects and demand-side effects. The coefficient in the first row of that column indicates
that average payrolls in the Dallas MSA increased by 1.9 percent after Hurricane Katrina. In the
Houston MSA, however, there is also a 1.6 percent increase in the average wages of high-skill
establishments during the same time period. Therefore, there does not appear to be a significant
reduction in the average payrolls of high-skill firms in the Houston MSA relative to the average
payrolls of firms in the Dallas MSA following the hurricane, as reported in the differences-indifferences coefficient in the second row of column 3.
Regarding the effect of the increase in demand for labor services on firms’ average
payrolls, the coefficient in row 4 indicates that a one percentage point increase in a high-skill
industry’s relative sales generates a 4.8 percent increase in the average payrolls of firms in that
industry. In addition, row 5 shows that this effect is only 2.9 percent for high-skill firms in the
Houston MSA before Hurricane Katrina. Furthermore, row 6 suggests that the effect of sales on
average payrolls decreases to 2.9 percent ( a 1.9 percentage point reduction) for firms in the
Dallas MSA after Hurricane Katrina, while the corresponding effect in the Houston MSA stays
constant in the post Katrina period. Thus, the difference in difference coefficient in row 7 reveals
that there is a 1.7 percentage point increase in the effect of sales on the average payroll of lowskill firms in the Houston MSA after Hurricane Katrina, relative to the effect of sales on the
21
average wages of comparable firms in the Dallas MSA. All these effects are statistically
significant at conventional levels. In this case, due to the absence of a supply-side effect, most of
the results are driven by the positive effect of the increase in relative sales as presented in row 7.
This explains why the difference-in-difference coefficient becomes positive 1 percent and
statistically significant if we fail to control for demand side effects, as shown in row 2 of column
4.
Comparing the differences-in-differences coefficients between low- and high-skill
industries, we obtain a triple differences estimator that controls for idiosyncratic shocks affecting
the Houston labor market. The coefficients and standard errors for these triple difference
estimations are presented in the eighth row of Table 4. The one in column 1 controls for demandside effects, whereas the one in column 2 does not. The coefficient in column 1 shows a
statistically significant 2.7 percent fall in the average wages of low-skill establishments in
Houston, relative to the change for the same groups of industries in Dallas, and relative to the
corresponding change in high-skill industries. Furthermore, the same triple difference becomes
statistically insignificant if we fail to control for demand-side effects, as shown in the eighth row
of column 2, highlighting the importance of controlling for changes in demand for locationspecific goods and services se to be able to identify the impact on wages from the sharp increase
in labor supply.
In Figures 1 and 2, we show that average wages for low-skill industries in Houston has
shifted left and can be observed for all ranges of wages. This basic observation can be analyzed
using a quantile regression model with firm effects (see Koenker and Bassett [1982] and
Koenker [2004]). The quantile regression model allows us to estimate wage differences in the
distribution of low- and high-skill industries while controlling for other factors that contribute to
the variability of wages. We restrict the estimation to three quantiles: .25, .50, and 75. The results
are presented in Table 5. The analysis employs the specification described in equation (1). The
coefficient on the dummy variable Houston after August 2005 for the low-skill industries
estimation shows that Houston wages decreased by 2.6 percent at the .25 quantile , 4.3 percent
in the median regression, and 3.6 percent in the .75th quantile holding everything else constant.
We do not observe any significant declines for high-skill industries in Houston MSA after
22
Hurricane Katrina. These results are in agreement with our hypothesis that Katrina induced inmigration affected low-skill industry wages in Houston MSA. We also estimated the mean and
quantile regressions comparing only Harris County, the county where the city of Houston is
located, to the Dallas MSA. The results are not statistically different to those obtained when all
the Houston MSA counties are considered. We can provide these runs on request.
Robustness Analysis
In order to check for the robustness of our results, we estimate a number of alternative
specifications. First, we address the problem of within-group correlation raised by Moulton
(1990). If this is the case, the standard errors in our model may be underestimated. Therefore,
we employ clustered standard errors throughout the paper to overcome this potential problem.
Bertrand, Duflo, and Mullainathan (2004) show that clustered standard errors can be biased
downward in panel data if serial correlation is present. One approach that they recommend is to
collapse the time dimension of the data down to two periods – pre- and post-Hurricane Katrina.
In our application, we aggregate the pre- and post-quarterly real average payroll data by
establishment and skill group. Aggregation by establishment types will still allow for differences
in response to Hurricane Katrina while reducing the time dimension to two periods. We also
require each firm to be present in both periods in order to estimate the fixed effects models. In
this estimation we use only firms that existed at least 12 quarters (six quarters before and
minimum of six after August 2005.) The estimated coefficient for the ‘Houston after August
2005’ have the correct sign and similar point estimate (-.032) indicating the real wages declined
after hurricane Katrina. However, it is statistically not significant.
Next question is whether the Houston MSA around the period of the event was already
experiencing a downward trend in real average payrolls either in absolute terms or relative to the
Dallas MSA. To investigate this issue, we estimate the real payroll model using only firm and
establishment-level data from the pre-Hurricane Katrina period and include time variables to
measure the trends in real average payrolls in the Houston and Dallas MSAs over the relevant
period. The models include an overall trend term and the interaction of the trend term with a
dummy variable for Houston MSA, to test for differences in the trend across MSAs. We report
these results in Table 6 for both High and Low-skill industries. The estimated trend terms show
that there is a general decline in real wages for low skill industries and increase in real wages for
23
high skill industries in both Houston and Dallas MSAs at the establishment level. However, the
results indicate that the trend is not statistically different between the Houston and Dallas MSAs
for both high and low skill industries. Hence, we can conclude that Houston’s real average
payrolls were not trending downward prior to Hurricane Katrina, either in an absolute sense or
relative to the Dallas MSA.
V.
Conclusions
Given that a large proportion of the 240,000 Katrina evacuees who went to Houston, TX,
remained for at least one year, the region experienced a nearly 3% increase in population. Most
of these evacuees were younger and less-educated than existing residents and remained in the
Houston area for at least a year. The objective of this paper is to estimate the effect of this
massive in-migration on workers’ earnings in non-tradable goods industries in the Houston
MSA. The underlying assumption has been that, the influx of lower-skilled workers would have
caused the supply of applicants for lower skilled jobs to increase proportionately more than for
higher skilled jobs. The effect of this situation should be most readily observed in the industries
that use relatively higher proportions of lower-skilled labor.
We utilize a differences-in-differences methodology in which we compare relative
earnings per worker within the relevant industries in the Houston and Dallas-Fort Worth MSAs
before and after the Katrina-induced migration. Unlike previous studies in this realm, we control
for the influence of an increase in demand for local goods and services on labor demand in
normally non-tradable goods and services activities. We find evidence that average payrolls in
the Houston-area non-tradable industries that utilize relatively higher levels of low-skilled
workers decreased by 3.0% in the Houston MSA relative to the Dallas-Ft. Worth MSA due to the
sharp jump in population in the aftermath of Katrina. We find no evidence of a significant effect
in the set of high-skilled non-tradable industries.
Comparing the differences-in-differences coefficients between the same sets of low and
high-skill industries in the Houston and Dallas MSAs, we obtain a triple differences estimator
that indicates a statistically significant 2.7 percent fall in the average wages of low-skill
establishments relative to high-skill establishments in the Houston MSA when compared to the
24
Dallas MSA given the assumption of inelastic demand for labor, this implies a greater than 1.3%
decrease in average wage rates in the non-tradable goods industries.
Furthermore, the same triple difference becomes statistically insignificant if we fail to control for
demand-side effects, highlighting the importance of controlling for changes in demand for
location-specific goods and services for identification of the impact on wages from the sharp
increase in labor supply.
Unlike most previous studies on the effect of immigration on wages of native workers,
this study finds a significant and negative impact on Houston-area wages due to the mass influx
of people generated by Hurricane Katrina. We posit two possible reasons for this result. First,
there would likely have been greater substitution possibilities between low-skill Katrina refugees
and low-skill native workers than in other contexts in which migrants generally were not citizens
and probably lacked English proficiency. Second, by controlling for demand-side effects, we are
able to identify the shift in the labor supply curve. In fact, our findings also suggest that failure
to control for demand-side influences obscures the effect of in-migration on Houston wages and
renders it statistically insignificant.
There are clearly some difficulties in interpreting results when controlling for demand.
All immigration naturally brings its own demand for localized goods and services. But, unlike
other types of immigrants, domestic in-migrants due to hurricane evacuation do not necessarily
arrive with the intention to relocate permanently. Moreover, given the financial resources that
were provided the Katrina evacuees, it is possible the initial demand-side effects encouraged
their permanent relocation and participation in the Houston-area labor market by offering greater
apparent employment opportunities than were available in their previous place of residence. As
their employment participation increases, the depressing effect on low-skill industry wages is
observed.
25
REFERENCES
Altonji, Joseph and Card, David. “The Effects of Immigration on the Labor Market Outcomes
of Less-Skilled Natives.” In Immigration, Trade and the Labor Market, edited By John M.
Abowd and Richard Barry Freeman. Chicago: University of Chicago Press, 1991.
Bai, Jushan and Pierre Perron (1997). “Estimating and testing linear models with multiple
structural changes,” Econometrica 66: 47-78.
Bai, Jushan and Pierre Perron (2003). “Computation and analysis of multiple structural change
models,” Journal of Applied Econometrics 18: 1-22.
Bertrand, Marianne, Duflo, Esther and Sendhil Mullainathan (2004) “How Much Should We
Trust Differences-in-Differences Estimates?” Quarterly Journal of Economics 119(1): 24975
Borjas, George J. (2003). “The Labor Demand Curve Is Downward Sloping: Reexamining the
Impact of Immigration on the Labor Market.” Quarterly Journal of Economics, Vol. 118,
No. 4 (November), pp. 1335-1374.
Card, David. (1990). “The Impact of the Mariel Boatlift on the Miami Labor Market.” Industrial
and Labor Relations Review, Vol. 43, No. 2 (January), pp. 245-257.
____. (2001). “Immigrant Inflows, Native Outflows, and the Local Market Impacts of Higher
Immigration.” Journal of Labor Economics, Vol. 19, No. 1 (January), pp. 22-64.
Cotterill, Philip (1975). “The Elasticity of Demand for Low-wage Labor,” Southern Economic
Journal, vol. 41, no. 3: 520-525.
Carrington, William, and Pedro deLima. (1996). “The Impact of 1970s Repatriates from Africa
on the Portuguese Labor Market.” Industrial and Labor Relations Review, Vol. 49, No. 2
(January), pp. 330-347.
Cortes, Patricia (2006). “The Effect of Low-skilled Immigration on U.S. Prices: Evidence from
CPI Data”, Unpublished Manuscript, University of Chicago, Graduate School of Business
Frey and Singer ( 2006). Katrina and Rita Impacts on Gulf Coast Populations: First Census
Findings, The Brookings Institution, Metropolitan Policy Program.
http://www.brookings.edu/metro/pubs/20060607_hurricanes.pdf
Groen and Polivka (2008). The Effect of Hurricane Katrina on the Labor Market Outcomes of
Evacuees, U.S. Bureau of Labor Statistics, Working Paper 415
26
____. (2008). “Hurricane Katrina Evacuees: Who They Are, Where They Are, and How They
Are Faring.” Monthly Labor Review, Vol. 131, No. 3, March 2008, pp. 32-51.
Hamermesh, Daniel (1993). Labor Demand, Princeton, N.J.: Princeton University Press.
Hunt, Jennifer. (1992). “The Impact of the 1962 Repatriates from Algeria on the French Labor
Market.” Industrial Labor Relations Review, Vol. 45, No. 3 (April), pp. 556-572.
Jensen, J. Bradford and Lori G. Kletzer (2005) “Tradable Services: Understanding the Scope and
Impact of Services Offshoring.” Brookings Trade Forum, Offshoring White-Collar Work,
Brookings Institution Press, pp. 75-133
Koenker, R. and Bassett, G., 1978. Regression quantiles. Econometrica 46 (1), 33–50.
Koenker, R., 2004. Quantile regression for longitudinal data. Journal of Multivariate Analysis
91, 74–89.
Kugler, Adriana and Mutlu Yuksel. (2006). “Effects of Low-Skilled Immigration on U.S.
Natives: Evidence from Hurricane Mitch.” Working paper, University of Houston.
McIntosh (2008). ”Measuring the Labor Market Impacts of Hurricane Katrina Migration:
Evidence From Houston, TX.” Working paper, Princeton University, Department of
Economics.
Moulton, Brent R. (1990) “An Illustration of a Pitfall in Estimating the Effects of Aggregate
Variables on Micro Units,” Review of Economics and Statistics, 72(2): 334-38.
Ottaviano, Gianmarco I.P. and Giovanni Perri. (2006). “Rethinking the Effects of Immigration
on Wages.” NBER Working Paper No. 12497.
Sanchez, Adrian (2002), “Houston's Downturn Comes Later but Will Linger Longer”, Bank
Trends, Number 02-03, March. Federal Deposit Insurance Corporation.
27
0
.2
Density
.4
.6
.8
Figures 1: Wage Distributions for Houston MSA
7
8
9
10
Log average quarterly wage
11
12
Houston Low Skill Before Aug 2005
Houston Low Skill After Aug 2005
Houston High Skill Before Aug 2005
Houston High Skill After Aug 2005
0
.2
Density
.4
.6
.8
Figures 2: Wage Distributions for Dallas MSA
7
8
9
10
Log average quarterly wage
11
12
Dallas Low Skill Before Aug 2005
Dallas Low Skill After Aug 2005
Dallas High Skill Before Aug 2005
Dallas High Skill After Aug 2005
28
Table 1.a Number of Workers and Establishments in Low-skill Non-Tradable Industries Industry
Amusement Parks and Arcades
Book, Periodical, and Music Stores
Child Day Care Services
Civic and Social Organizations
Clothing Stores
Community Care Facilities for the Elderly
Consumer Goods Rental
Department Stores
Drinking Places (Alcoholic Beverages)
Drycleaning and Laundry Services
Full-Service Restaurants
Gasoline Stations
Grocery Stores
Lawn and Garden Equipment and Supplies
Limited-Service Eating Places
Office Supplies, Stationery, and Gift Stores
Other Amusement and Recreation Industries
Other General Merchandise Stores
Other Schools and Instruction
Personal Care Services
Services to Buildings and Dwellings
Shoe Stores
Special Food Services
Specialty Food Stores
Sporting Goods, Hobby, and Musical Instruments
Traveler Accommodation
Used Merchandise Stores
Total low-skill non-tradable
Employment
Average Number
Share of Total
Houston
Dallas
Houston Dallas
1,089
2,461
.002
.005
3,210
3,684
.007
.007
11,577
15,361
.024
.028
5,001
5,755
.010
.011
27,199
25,144
.056
.047
4,025
5,417
.008
.010
4,500
7,788
.009
.014
28,703
12,635
.059
.023
5,269
6,161
.011
.011
9,344
8,269
.019
.015
84,745
116,802
.174
.217
10,198
18,711
.021
.035
58,131
37,114
.119
.069
1,878
2,279
.004
.004
78,688
96,078
.161
.178
9,976
9,656
.020
.018
12,876
20,160
.026
.037
29,186
36,099
.060
.067
2,974
5,026
.006
.009
7,811
8,367
.016
.016
41,849
33,523
.086
.062
6,473
5,030
.013
.009
10,774
10,240
.022
.019
3,384
2,742
.007
.005
8,049
5,294
.016
.010
18,584
35,915
.038
.067
2,400
3,471
.005
.006
487,894
539,182
1.000
1.000
29
Establishments
Average Number
Share of Total
Houston
Dallas
Houston Dallas
28
34
.002
.002
167
182
.011
.011
527
706
.036
.041
114
164
.008
.010
693
861
.047
.051
66
61
.005
.004
194
213
.013
.013
49
59
.003
.003
370
312
.025
.018
528
603
.036
.035
1,950
2,300
.133
.135
1,280
1,370
.087
.080
1,019
836
.070
.049
159
169
.011
.010
2,297
2,872
.157
.169
386
478
.026
.028
481
532
.033
.031
394
486
.027
.029
275
366
.019
.021
726
827
.050
.049
1,340
1,809
.091
.106
267
328
.018
.019
186
218
.013
.013
264
181
.018
.011
281
351
.019
.021
461
503
.031
.030
160
208
.011
.012
14,662
17,028
1.000 1.000
Table 1.b Number of Workers and Establishments in High-skill Non-Tradable Industries
Industry
Accounting, Tax Preparation, Bookkeeping
Activities Related to Credit Intermediation
Administration of Economic Programs
Administration of Environmental Quality
Automobile Dealers
Automotive Parts, Accessories, and Tire
Automotive Repair and Maintenance
Building Material and Supplies Dealers
Business Support Services
Colleges, Universities, and Professional
Commercial and Industrial Machinery
Death Care Services
Depository Credit Intermediation
Direct Selling Establishments
Electric Power Generation, Transmission
Electronic and Precision Equipment Repair
Electronics and Appliance Stores
Elementary and Secondary Schools
Executive, Legislative, and Other General
Furniture Stores
General Medical and Surgical Hospitals
General Rental Centers
Grantmaking and Giving Services
Health and Personal Care Stores
Highway, Street, and Bridge Construction
Home Furnishings Stores
Home Health Care Services
Individual and Family Services
Jewelry, Luggage, and Leather Goods Stores
Justice, Public Order, and Safety Activities
Land Subdivision
Local Messengers and Local Delivery
Medical and Diagnostic Laboratories
Newspaper, Periodical, Book, and Directories
Nonresidential Building Construction
Nursing Care Facilities
Employment
Average Number
Houston
Dallas
16,818
3,453
2,919
1,408
22,718
13,974
14,334
26,963
28,030
32,610
6,824
2,308
36,864
1,117
5,548
2,290
6,850
158,898
7,368
5,701
65,571
4,157
626
26,022
12,997
7,411
28,874
9,886
2,398
20,175
935
868
5,253
4,619
39,511
7,906
24,467
7,577
1,759
323
27,205
11,052
18,157
36,441
29,538
27,635
3,192
1,610
38,768
1,005
8,218
4,370
32,391
151,395
12,576
8,821
79,039
2,031
843
16,721
13,165
9,584
32,041
7,692
6,380
11,232
2,497
715
6,001
12,246
16,237
16,429
30
Share of Total
Houston Dallas
.023
.005
.004
.002
.031
.019
.020
.037
.039
.045
.009
.003
.051
.002
.008
.003
.009
.220
.010
.008
.091
.006
.001
.036
.018
.010
.040
.014
.003
.028
.001
.001
.007
.006
.055
.011
.032
.010
.002
.000
.035
.014
.023
.047
.038
.036
.004
.002
.050
.001
.011
.006
.042
.195
.016
.011
.102
.003
.001
.022
.017
.012
.041
.010
.008
.014
.003
.001
.008
.016
.021
.021
Establishments
Average Number
Share of Total
Houston
Dallas
Houston Dallas
1,124
212
35
16
506
537
1,752
558
470
31
329
143
681
78
37
181
485
195
99
252
47
119
47
832
124
305
316
292
310
96
91
61
111
161
717
81
1,252
261
45
6
573
710
1,908
614
639
43
241
97
920
95
32
203
793
244
152
326
65
101
49
919
170
370
317
281
324
131
122
52
132
275
765
156
.070
.013
.002
.001
.032
.033
.109
.035
.029
.002
.020
.009
.042
.005
.002
.011
.030
.012
.006
.016
.003
.007
.003
.052
.008
.019
.020
.018
.019
.006
.006
.004
.007
.010
.045
.005
.069
.014
.002
.000
.032
.039
.105
.034
.035
.002
.013
.005
.051
.005
.002
.011
.044
.013
.008
.018
.004
.006
.003
.051
.009
.020
.017
.015
.018
.007
.007
.003
.007
.015
.042
.009
Table 1.b Number of Workers and Establishments in High-skill Non-Tradable Industries continued…
Industry
Offices of Dentists
Other Ambulatory Health Care Services
Other Heavy and Civil Engineering Construction
Other Miscellaneous Store Retailers
Other Motor Vehicle Dealers
Other Professional, Scientific, and Technical
Outpatient Care Centers
Personal and Household Goods Repair
Radio and Television Broadcasting
Religious Organizations
Remediation and Other Waste Management Services
Residential Building Construction
Social Advocacy Organizations
Utility System Construction
Vending Machine Operators
Vocational Rehabilitation Services
Waste Collection
Waste Treatment and Disposal
Water, Sewage and Other Systems
Total High-skill non-tradable
Employment
Number
Houston
Dallas
Establishments
Number
Share of Total
Houston
Dallas
Houston Dallas
8,839
3,135
4,285
3,249
2,615
11,363
3,935
1,322
2,880
306
2,150
11,163
524
20,015
441
1,654
1,484
3,272
4,414
10,243
4,422
1,627
9,370
2,959
10,632
7,220
1,392
4,070
594
1,142
13,073
711
10,188
900
2,547
2,912
7,366
4,018
.012
.004
.006
.005
.004
.016
.005
.002
.004
.000
.003
.015
.001
.028
.001
.002
.002
.005
.006
.013
.006
.002
.012
.004
.014
.009
.002
.005
.001
.001
.017
.001
.013
.001
.003
.004
.010
.005
1,242
96
81
340
142
687
146
197
43
34
93
812
46
336
45
38
62
67
118
1,309
71
66
464
157
805
157
201
56
37
55
878
36
242
50
32
50
56
87
.077
.006
.005
.021
.009
.043
.009
.012
.003
.002
.006
.051
.003
.021
.003
.002
.004
.004
.007
.072
.004
.004
.025
.009
.044
.009
.011
.003
.002
.003
.048
.002
.013
.003
.002
.003
.003
.005
721,255
774,740
1.000
1.000
16,057
18,188
1.000
1.000
31
Share of Total
Houston Dallas
Table 2: Summary Statistics for Houston and Dallas
Variable
Establishment Level Data
Quarterly average real payroll per employee
Quarterly average number of employees per establishment
Quarterly average employment ratio
Industry Level Data
Quarterly average real sales a
Low-Skill Non-Tradable Industries
Houston
Dallas
Before
After
Before
After
3,753
(8.265)
36.166
(1.145)
.0115
(.000)
312
(1.463)
Quarterly average relative real sales
.915
(.001)
a: in millions of dollars. Standard errors of the means are in parentheses.
3,800
(6.763)
30.637
(.727)
.0114
(.000)
3,983
(8.578)
33.467
(.676)
.012
(.000)
3,970
6.854)
29.927
(.463)
.012
(.000)
7,358
(14.201)
46.001
(1.461)
.014
(.000)
7,553
(11.636)
44.247
(1.093)
.015
(.000)
7,650
(13.730)
43.389
(1.081)
.017
(.000)
7,821
(11.177)
42.102
(.817)
.017
(.000)
318
(1.046)
1.045
(.001)
151
(.505)
.890
(.001)
170
(.432)
1.059
(.001)
199
(.989)
.923
(.002)
238
(.981)
1.034
(.002)
132
(.782)
.874
(.001)
143
(.638)
1.065
(.001)
32
High-Skill Non-Tradable Industries
Houston
Dallas
Before
After
Before
After
Table 3.a Multiple Structural Break Estimation by Industry
Industry
Break Point at Katrina (95% C.I.)
[Before: After Break Average Payroll]
Houston
Dallas
Low-skill Non-Tradable Industries
Book, Periodical, and Music Stores
Civic and Social Organizations
Drinking Places (Alcoholic Beverages)
Services to Buildings and Dwellings
2006:2** (2006:2-OC)
[3887:4735]
2006:3** (2006:2-2006:4)
[4543:5102]
2006:3** (2006:1-2007:1)
[2988:3368]
2006:1** (2004:4-2007:2)
[5081:5350]
―
―
―
―
High-skill Non-Tradable Industries
Administration of Economic Programs
2005:3** (2005:1-2006:1)
―
[9553:10265]
Administration of Environmental Quality Programs
2005:3** (2004:4-2006:2)
―
[9109:9776]
General Rental Centers
2006:2** (2005:3-2007:2)
―
[9460:10158]
Land Subdivision
2006:3** (2005:4-OC)
―
[11679:13128]
Nursing Care Facilities
2006:3** (2006:1-2007:4)
―
[5219:5421]
Other Miscellaneous Store Retailers
2006:1** (2005:3-2006:4)
―
[5609:5905]
Personal and Household Goods Repair and
2005:3** (2004:2006:1)
―
Maintenance
[5172:5607]
Utility System Construction
2006:3* (2004:4-2006:4)
2006:3** (2005:3-OC)
[9335:10948]
[9198:8932]
** And * denote breaks are significant at the 5% and 10% level, respectively. 95% confidence interval and average
real payroll before and after break are reported in the parenthesis. In the case of multiple breaks, only before and
after payroll at Katrina are reported. OC denotes out of the sample coverage.
33
Table 3.b Multiple Structural Break Estimation by Industry and Establishment Size:
Low-skill Non-Tradable Industries
Industry
Size
Break Point at Katrina (95% C.I.)
[Before: After Break Average Payroll]
Houston
Dallas
Book, Periodical, and Music Stores
1
Book, Periodical, and Music Stores
3
Book, Periodical, and Music Stores
4
Business Support Services
3
Child Day Care Services
1
Child Day Care Services
4
Civic and Social Organizations
1
Civic and Social Organizations
2
Civic and Social Organizations
3
Civic and Social Organizations
5
Clothing Stores
3
Clothing Stores
4
Clothing Stores
5
Community Care Facilities for the Elderly
2
Drinking Places (Alcoholic Beverages)
1
Drinking Places (Alcoholic Beverages)
3
Drinking Places (Alcoholic Beverages)
5
Drycleaning and Laundry Services
2
Drycleaning and Laundry Services
3
Drycleaning and Laundry Services
4
Drycleaning and Laundry Services
5
Full-Service Restaurants
4
Full-Service Restaurants
5
Office Supplies, Stationery, and Gift Stores
2
2006:2** (2006:2-OC)
[3845:4660]
2006:3** (2006:1-OC)
[4320:5430]
2006:3** (2005:4-2006:4)
[4559:7239]
2006:3** (2005:2-2007:1)
[8808:10185]
2006:1** (2005:3-2007:3)
[3164:3050]
2005:3** (2005:2-2007:4)
[5187:4905]
2006:3** (2006:1-2006:4)
[4351:4879]
2006:3** (2005:4-OC)
[5079:5714]
2006:3** (2006:2-2006:4)
[5749:7540]
2006:3** (2006:2-2006:4)
[3813:4641]
2005:3** (2004:3-2007:4)
[3861:3519]
2006:2** (2005:4-2006:4)
[3416:3020]
2006:3** (2005:3-2006:4)
[3913:5403]
2006:3** (2006:2-2007:2)
[3701:4632]
2006:3** (2006:1-2007:1)
[2942:3441]
2005:4** (2005:3-2006:4)
[2864:3132]
2006:2** (2006:1-2006:3)
[3384:2936]
2006:3** (2005:4-2007:1)
[3491:3683]
2006:3** (2006:2-2007:1)
[4229:4785]
2005:4** (2004:3-OC)
[5229:4387]
2006:3** (2005:4-2006:4)
[5796:6692]
2006:3** (2006:1-OC)
[3405:3487]
2006:3* (2006:2-OC)
[3756:3835]
2006:3** (2005:2-OC)
[4742:5245]
34
2006:3** (2005:4-OC)
[4515:5213]
―
2005:3* (2005:1-OC)
[4250:3340]
―
―
―
―
―
―
2006:3** (2006:2-2006:4)
[5221:7163]
2005:4** (2004:3-2006:1)
[3812:3365]
―
2006:3** (2006:2-2006:4)
[3751:4419]
―
―
―
―
―
―
―
―
2006:3** (2005:4-OC)
[3503:3637]
―
―
Table 3.b Multiple Structural Break Estimation by Industry and Establishment Size:
Low-skill Non-Tradable Industries continued…
Industry
Size
Break Point at Katrina (95% C.I.)
[Before: After Break Average Payroll]
Houston
Dallas
Office Supplies, Stationery, and Gift Stores
5
2006:3** (2005:4-2006:3)
―
[4488:9570]
Personal Care Services
3
2006:2** (2005:3-2007:3)
―
[5668:5342]
Personal Care Services
4
2006:3** (2005:2-2007:3)
―
[6715:7491]
Services to Buildings and Dwellings
1
2006:1** (2005:2-2007:1)
―
[5107:5440]
Shoe Stores
5
2006:3** (2006:2-OC)
―
[3974:2832]
** And * denote breaks are significant at the 5% and 10% level, respectively. 95% confidence interval and average
real payroll before and after break are reported in the parenthesis. In the case of multiple breaks, only before and
after payroll at Katrina are reported. OC denotes out of the sample coverage.
35
Table 3.c Multiple Structural Break Estimation by Industry and Establishment Size:
High-skill Non-Tradable Industries
Industry
Size
Break Point at Katrina (95% C.I.)
[Before: After Break Average Payroll]
Houston
Dallas
Activities Related to Credit Intermediation
4
2005:4* (2004:3-2007:4)
[12494:9611]
2006:1** (2005:4-2006:3)
[9878:12246]
2005:3** (2004:4-2006:2)
[8620:9421]
2006:3** (2006:1-2006:4)
[8078:10749]
2006:3* (2005:3-2007:2)
[9911:10286]
2006:3** (2005:4-2006:3)
[7183:10545]
2006:3** (2006:2-2007:3)
[5426:6658]
2005:3** (2004:4-2005:4)
[10832:12655]
Administration of Economic Programs
2
Administration of Environmental Quality
Programs
Administration of Environmental Quality
Programs
Automobile Dealers
1
3
Automotive Parts, Accessories, and Tire Stores
5
Automotive Repair and Maintenance
4
Commercial and Industrial Machinery and
Equipment (except Automotive and Electronic)
Repair and Maintenance
Commercial and Industrial Machinery and
Equipment (except Automotive and Electronic)
Repair and Maintenance
Death Care Services
3
5
2006:3** (2004:4-2006:1)
[12658:14527]
―
3
Death Care Services
4
Department Stores
2
2006:3* (2005:2-2006:4)
[6451:6875]
2006:3** (2006:2-2006:4)
[6973:8304]
―
Depository Credit Intermediation
5
Direct Selling Establishments
5
Elementary and Secondary Schools
5
Executive, Legislative, and Other General
Government Support
Furniture Stores
1
2
Grantmaking and Giving Services
2
Home Furnishings Stores
5
Individual and Family Services
5
Land Subdivision
3
Land Subdivision
4
Lawn and Garden Equipment and Supplies
Stores
1
2006:1* (2005:3-2006:2)
[6455:7098]
2006:3** (2006:1-2007:2)
[5697:6669]
2006:1** (2005:1-2006:2)
[9314:13968]
2005:4** (2004:4-2006:1)
[9800:12960]
2006:3** (2005:1-2006:4)
[15825:41558]
2006:2** (2005:3-OC)
[6932:7291]
2005:3** (2005:2-OC)
[6968:7201]
2006:3* (2005:3-2007:3)
[7685:8162]
2006:3** (2005:4-2007:1)
[10218;11565]
2006:3** (2006:2-2006:4)
[3827:3590]
2006:3** (2006:2-2007:2)
[6342:6719]
2006:3** (2005:4-2007:1)
[20704:32490]
2005:4** (2005:3-2006:1)
[12824:7242]
2006:2** (2005:4-2007:2)
[4734:5325]
4
36
2005:4* (2005:2-OC)
[12120:13232]
―
―
2006:3* (2006:2-2007:1)
[10719:12593]
2006:4** (2005:3-2007:1)
[10752:9769]
―
―
―
―
―
―
2006:3** (2006:1-2007:1)
[8666:9168]
―
―
2006:3** (2005:3-2007:1)
[5443:6058]
2006:3** (2006:1-2007:1)
[6322:6833]
2005:3** (2004:3-2005:4)
[16343:14038]
―
2006:3** (2005:4-OC)
[5015:5301]
Table 3.c Multiple Structural Break Estimation by Industry and Establishment Size:
High-skill Non-Tradable Industries continued…
Industry
Size
Break Point at Katrina (95% C.I.)
[Before: After Break Average Payroll]
Houston
Dallas
Limited-Service Eating Places
3
Local Messengers and Local Delivery
2
Local Messengers and Local Delivery
3
Medical and Diagnostic Laboratories
1
Medical and Diagnostic Laboratories
3
Offices of Dentists
3
Offices of Dentists
5
Other Ambulatory Health Care Services
1
Other General Merchandise Stores
2
Other General Merchandise Stores
3
Other Heavy and Civil Engineering Construction
3
Other Miscellaneous Store Retailers
1
Other Miscellaneous Store Retailers
3
Other Motor Vehicle Dealers
5
Other Professional, Scientific, and Technical
Services
Other Professional, Scientific, and Technical
Services
Outpatient Care Centers
4
Personal and Household Goods Repair and
Maintenance
Personal and Household Goods Repair and
Maintenance
Religious Organizations
1
Remediation and Other Waste Management
Services
Residential Building Construction
2
2005:4** (2003:3-2006:1)
[2857:2696]
2005:4** (2005:1-2007:2)
[8020:6489]
2006:3** (2006:2-2007:3)
[7052:7856]
2005:4** (2004:3-2006:3)
[11890:10464]
2005:4** (2005:1-2006:1)
[9082:10072]
2005:3** (2004:1-2006:2)
[10297:12862]
2006:3** (2005:3-OC)
[15253:17315]
2005:4** (2005:2-2006:2)
[7966:7097]
2005:4** (2005:3-2006:4)
[4035:3713]
2006:3** (2005:4-2006:4)
[4387:5335]
2006:3** (2005:4-2006:4)
[10809:12417]
2006:3** (2005:4-2007:1)
[5708:6107]
2006:3** (2006:2-2007:4)
[5379-4728]
2005:3** (2005:2-2006:4)
[8392:10364]
2006:3** (2006:2-OC)
[8990:10260]
2006:3** (2005:2-2006:4)
[7824:9658]
2006:3** (2005:4-2006:4)
[12793:10584]
2005:3** (2004:4-2006:1)
[4965:5416]
2006:3** (2005:4-2007:1)
[6443:7623]
2006:2** (2006:1-OC)
[6873:6188]
2005:3** (2004:2-2007:1)
[8604:9643]
2006:3** (2005:3-2007:1)
[12558:14393]
5
5
2
2
5
37
―
―
―
―
―
2005:4** (2005:2-2007:2)
[12195:10104]
―
2006:2** (2005:4-2007:1)
[8485:7715]
2006:2** (2005:3-2006:4)
[4134:3564]
―
―
―
―
―
―
2005:4** (2005:-2006:1)
[9346:12735]
―
―
2005:4** (2005:3-OC)
[7743:6644]
―
―
―
Table 3.c Multiple Structural Break Estimation by Industry and Establishment Size:
High-skill Non-Tradable Industries continued…
Industry
Size
Break Point at Katrina (95% C.I.)
[Before: After Break Average Payroll]
Houston
Dallas
Utility System Construction
3
2005:3** (2005:1-2005:4)
―
[9257:10521]
Utility System Construction
5
2006:3** (2006:2-2006:4)
―
[10772:112708]
Vocational Rehabilitation Services
2
2005:4** (2005:1-2006:1) 2005:4** (2003:4-2006:1)
[7444:5987]
[7611:6622]
Waste Collection
3
2005:4** (2005:2-OC)
―
[7869:6971]
Water, Sewage and Other Systems
2
2006:3** (2006:1-2007:1)
―
[7661:8841]
Water, Sewage and Other Systems
5
2006:3** (2006:2-OC)
―
[9453:8414]
** And * denote breaks are significant at the 5% and 10% level, respectively. 95% confidence interval and average
real payroll before and after break are reported in the parenthesis. In the case of multiple breaks, only before and
after payroll at Katrina are reported. OC denotes out of the sample coverage.
38
Table 4: Regression Results for Log of Real Payroll per Employee
Variable
Wages after hurricane Katrina (1)
Houston wages after hurricane Katrina ( 2)
Employment ratio( 3)
Relative real sales( 4)
Houston relative real sales( 5)
Relative real sales after hurricane Katrina( 6)
Houston relative real sales after hurricane Katrina( 7)
Triple Difference 2 (low-Skill) = 2(high-Skill)
Number of Observations
Adjusted R2
Low-Skill Non-Tradable
With Sales
Without Sales
(1)
(2)
.015**
-.007***
(.007)
(.002)
-.030***
.009***
(.008)
(.003)
-.950***
-.955***
(.093)
(.093)
.044***
(.008)
-.032***
(.008)
-.028***
(.008)
.042***
(.009)
-.272***
-.001
(.001)
(.004)
501285
501285
.806
.806
39
High-Skill Non-Tradable
With Sales
Without Sales
(3)
(4)
.019***
.007***
(.004)
(.003)
-.003
.010***
(.005)
(.003)
-.523***
-.531***
.061
(.061)
.048***
.003
-.019***
.004
-.019***
.003
.017***
(.004)
541235
.812
541235
.811
Table 5: Quantile Regression with Sales
Variable
Wages after hurricane
Katrina (1)
Houston wages after
hurricane Katrina ( 2)
# Obs.
Low Skilled Industries
q.25
q.50
q.75
With Sales
With Sales
With Sales
.035***
.029***
-.000
(.005)
(.003)
(.004)
-.026***
-.043***
-.036**
(.006)
(.005)
(.006)
501285
501285
501285
High Skilled Industries
q.25
q.50
q.75
With Sales
With Sales
With Sales
.030***
.015***
.001
(.002)
(.002)
(.002)
-.001
-.004*
.001
(.003)
(.002)
(.003)
541235
541235
541235 Table 6: Robustness Results
Variable
Employment Ratio
Relative Sales
Relative Sales × Houston MSA
Time
Time × Houston MSA
# Obs.
Adj R2
With Time Trend
Low Skilled Industries
High Skilled Industries
With Sales
Without
With
Without
Sales
Sales
Sales
-1.101***
-1.100***
-.707***
-.707***
(.160)
(.161)
(.105)
(.104)
.075***
.054***
(.005)
(.002)
-.049***
-.018***
(.006)
(.003)
-.001**
.000
.000
.002**
(.000)
(.000)
(.000)
(.000)
-.000
-.001
.001
.001
(.000)
(.001)
(.000)
(.001)
191243
191243
206042
206042
.846
.846
.847
.846
40