Public Company Audits and City-Specific Labor Characteristics

Public Company Audits and City-Specific
Labor Characteristics*
MATTHEW J. BECK, Michigan State University
JERE R. FRANCIS, University of Missouri–Columbia and University of

Technology–Sydney†
JOSHUA L. GUNN, University of Pittsburgh
ABSTRACT
Prior research emphasizes the centrality of audit offices in understanding auditing practices,
and documents significant interoffice variation in audit outcomes based on industry exper-
tise and office size. Our study examines how two city-specific labor characteristics also
affect audit offices and local audit markets: the city’s average educational attainment, and
the number of accountants in a city, which proxy for a city’s human capital. Our argument
draws on the urban economics literature and predicts that the level of human capital in a
city is positively associated with an audit office’s ability to conduct high-quality audits. As
expected, there is a positive association between audit quality (quality of audited earnings
and accuracy of going-concern reports) and average education level in the city in which the
lead engagement office is located. This association is generally significant for both Big 4
and non-Big 4 offices, but is relatively stronger for non-Big 4 firms that are more tied to
local labor markets. A company is also more likely to choose a non-Big 4 auditor in cities
with higher educational levels and relatively more accountants, and there is evidence of
higher non-Big 4 audit fees as a city’s education level increases. Collectively, these results
suggest that local labor characteristics affect audit offices, audit quality, and the ability of
non-Big 4 auditors to compete with Big 4 auditors in the audits of public companies.
Audits des societes cotees et caracteristiques de la main-d’œuvre

propres a la ville

RESUM
E
Les etudes anterieures sont axees sur la centralite des bureaux d’audit dans la comprehen-
sion des methodes d’audit et font etat, preuves a l’appui, de variations importantes des
resultats de l’audit entre les bureaux, selon les competences sectorielles et la taille des
bureaux. Les auteurs se penchent ici sur la facßon dont deux caracteristiques de la main-
d’œuvre propres a la ville influent egalement sur les bureaux d’audit et les marches d’audit
regionaux : le niveau de scolarite moyen et l’effectif comptable de la ville, indicateurs ser-
vant de variables de substitution au capital humain de la ville. Les auteurs fondent leur
argumentation sur la documentation en economie urbaine et posent l’hypothese selon
laquelle le niveau de capital humain d’une ville serait en relation positive avec la capacite
* Accepted by Jeffrey Pittman. We thank Jeff Pittman and two anonymous referees for their feedback and helpful
guidance. We also appreciate comments on earlier versions of the paper from workshops at Boston College,
Santa Clara University, University of Missouri–Columbia, University of Technology Sydney, and the Interna-
tional Symposium on Audit Research. Data availability: Data used are publicly available from sources identified
in the paper.
†
Corresponding author.
Contemporary Accounting Research Vol. 35 No. 1 (Spring 2018) pp. 394–433 © CAAA
doi:10.1111/1911-3846.12344
City Labor Characteristics and Public Audits 395
d’un bureau d’audit de mener des audits de qualite superieure. Leur analyse confirme l’exis-
tence d’un lien positif entre la qualite de l’audit (qualite des resultats audites et exactitude
des rapports sur la continuite de l’exploitation) et le niveau de scolarite moyen de la ville
ou se trouve le bureau responsable de la mission. Ce lien est generalement significatif tant
pour les bureaux des Quatre Grands cabinets que pour les bureaux des cabinets n’apparte-
nant pas a ce groupe, mais il est relativement plus marque dans le cas des cabinets ne fai-
sant pas partie des Quatre Grands qui sont lies de plus pres aux marches du travail
regionaux. Une societe est egalement davantage susceptible de choisir un auditeur n’appar-
tenant pas aux Quatre Grands dans les villes o u le niveau de scolarite est plus eleve et ou
l’effectif comptable est relativement plus grand, et certaines donnees revelent que les hono-
raires d’audit des cabinets n’appartenant pas aux Quatre Grands augmentent avec la hausse
du niveau de scolarite d’une ville. Dans leur ensemble, ces resultats semblent indiquer que
les caracteristiques de la main-d’œuvre regionale influent sur les bureaux d’audit, sur la
qualite de l’audit et sur la capacite des auditeurs n’appartenant pas aux Quatre Grands de
rivaliser avec les auditeurs des Quatre Grands dans les audits des societes cotees.
1. Introduction
We examine if audits of publicly listed companies in the United States are affected by local
labor characteristics of cities where lead audit engagement offices are located.1 Prior
research documents the importance of audit offices and interoffice variation in understand-
ing the quality of public company audits (e.g., Francis and Yu 2009; Reichelt and Wang
2010; Choi, Kim, Kim, and Zang 2010; Francis, Michas, and Yu 2013). Building on this
literature, we argue that the quality of audits performed by audit offices is also affected by
labor characteristics of the cities in which they are located. In particular, since audit firms
depend on labor as a primary input to the audit process, we focus on two city-specific
labor characteristics: the average education level in a city, and the number of accountants
in a city relative to the national average.
For expositional convenience, we use the term human capital to refer to these two
city-level labor characteristics. Human capital is a broad term used in the economics litera-
ture to conceptualize individuals’ knowledge, education, skills, training, and experience
that allows for productive labor (Becker 1962). Our study draws specifically on the urban
economics literature which focuses on a city’s aggregate human capital and its effect on
the productivity and economic growth of cities. This literature reports robust evidence that
the two city-specific labor characteristics which we focus on in this study, the size of the
labor force (number of accountants, in our case) and educational attainment (also called
schooling), are positively associated with a city’s overall productivity.2 Given the robust-
ness of these findings, we expect to observe a similar positive association between a city’s
human capital and the production of high-quality audits to public companies.
The audit industry is geographically decentralized because auditors are typically
located in the same cities as their clients.3 This characteristic of audit offices makes them
similar to traditional “bricks-and-mortar” retailers which are tied to particular geographic
locales for both labor and sales (Ettredge et al. 2014). However, unlike traditional bricks-
1. The lead engagement office is identified from the letterhead of the accounting firm’s audit report in the 10-
K. To the extent other offices participate in an audit for clients with multilocation operations, there is
potential for measurement error in attributing all of the audit effort to one city/engagement office which
would create noise and bias against the expected association between audit quality and the human capital
of a city (Ferguson, Francis, and Stokes 2003, footnote 9).
2. Although we use the term human capital to refer to both characteristics for expositional convenience, it is
important to note that they are conceptually and empirically distinct. We discuss this more in section 2.
3. For 72 percent of our sample, the company headquarters and audit engagement office are in the same city.
Ettredge, Sherwood, and Sun (2014) report a similar market-share statistic.
CAR Vol. 35 No. 1 (Spring 2018)

396 Contemporary Accounting Research
and-mortar retailers, auditors are service firms dependent upon having access to skilled labor
with specific knowledge and expertise. As a result, the human capital in a particular city
where an audit office is located may not necessarily be well-suited to an audit firm’s needs.
The location of an auditor’s office represents a trade-off in terms of reducing the
transaction costs that result from being far away from clients versus locating in places
where conditions, such as local labor characteristics, are most advantageous for the audi-
tor. Other industries which require labor with particular skills have well-known industry
clusters, such as the financial sector in New York, or the software industry in Silicon Val-
ley. The audit industry’s decentralization suggests that being close to clients is the more
important concern for auditors’ location decision (Choi et al. 2010). Nonetheless, there
may still be certain cities which provide benefits to audit offices similar to those that they
might otherwise obtain in some hypothetical auditor industry cluster. In particular, some
cities will have a relatively larger accounting labor pool, and some cities have a higher
level of educational attainment. In section 2, we review the literature on both of these
characteristics, and hypothesize an association with audit outcomes.
While we expect all auditors to benefit from being located in cities with greater human
capital, we do not expect human capital to affect all audit offices equally. Big 4 accounting
firms have national networks that allow skills and knowledge to be allocated within the
firm, making offices less captive to local labor conditions.4 As a result, the human capital
advantages from locating in a city with a larger or better educated labor force may affect
Big 4 firms less. Thus, we expect the effects of human capital to be relatively stronger for
non-Big 4 audit offices which are more tied to local labor markets.5
What do we find? The size of the labor pool is insignificant, but a city’s level of educa-
tional attainment is consistently associated with better audits (higher earnings quality and
more accurate going-concern reports), with a stronger effect for non-Big 4 offices, as pre-
dicted. To illustrate, when going from a city with relatively low human capital (Tampa,
FL) to a city with relatively high human capital (Chicago, IL), abnormal accruals decrease
by 0.8 percent of assets for Big 4 offices and by 2.5 percent of assets for non-Big 4 offices.6
Analyzing the 28 percent of the sample where audit offices and client headquarters are
located in different cities provides additional evidence that human capital affects earnings
through the auditor, as opposed to the client. The non-Big 4 results hold in this subsam-
ple, though there is no evidence of a Big 4 effect.
Non-Big 4 auditors also experience a reduction in going-concern error rates as a city’s
educational attainment becomes larger. The error reduction is approximately 20 percent
for non-Big 4 offices when going from Tampa, FL to Chicago, IL, whereas Big 4 auditors
experience an insignificant change, all else equal. Importantly, this test is less likely than
our accruals tests to be confounded by the effects of human capital on clients.
These results are robust to propensity score matching to control for client characteris-
tics with matching performed for auditor selection (Big 4/non-Big 4) and selection into
high human capital cities, multilevel mixed modeling as an additional method to account
for different levels of grouping in the data, and instrumental variables estimation with
2SLS equations that use lagged values of human capital and the presence of a land grant
college as instruments for educational attainment.
4. In section 5, we also examine the effect of human capital on the large national firms, BDO and Grant
Thornton.
5. Nonetheless, there is still likely an effect for Big 4 firms that is not fully mitigated by national networks.
For example, much of Big 4 entry-level recruiting on college campuses remains “local” in nature.
6. Throughout the paper, we use these two cities as examples to illustrate economic significance because they
represent approximately the interquartile shift for both human capital variables. More specifically, Chicago
is at the 80th (81st) percentile for educational attainment (number of accountants) and Tampa is at the
28th (34th).

We also find that in cities with greater human capital, companies are relatively more
likely to be audited by non-Big 4 auditors. Both the size of the local labor pool and the
city’s average education level are statistically and economically significant in explaining
whether a public company is audited by a non-Big 4 auditor. Going from Tampa, FL
(lower human capital) to Chicago, IL (higher human capital), the predicted odds of choos-
ing a non-Big 4 auditor increase by approximately 39 percent. Collectively, our evidence
suggests that greater human capital levels the playing field and reduces the difference
between Big 4 and non-Big 4 auditors.
Finally, we investigate the association between audit fees and human capital. There
are no Big 4 effects, but we do observe higher non-Big 4 fees as the average education
level in a city increases, which is consistent with relatively higher-quality non-Big 4 audits
in these cities and the ability to charge higher fees. Going from Tampa, FL (lower human
capital) to Chicago, IL (higher human capital) leads to higher fees of around 6 percent.
However, there is also a decrease in audit fees when the supply of accountants increases,
all else equal. These two effects largely offset each other in cities with similar percentile
values for education and labor pool size.
By linking research on city-level human capital with the recent literature on audit
office variation, we show that local labor characteristics can significantly affect the quality
of audited earnings, accuracy of going-concern reports, and a firm’s likelihood of choosing
a non-Big 4 auditor. We also contribute to research on office-level audit quality. Arguably,
one of the most important determinants of audit quality is the talent and knowledge of
the individual auditors performing engagements. This study provides insight into how
external forces like local labor characteristics can influence audit quality.
The paper proceeds as follows. The next section reviews relevant literature and more
fully develops the study’s hypotheses about city-specific human capital and auditing. Sec-
tion 3 discusses the empirical models. Section 4 presents the results and section 5 discusses
robustness checks. Section 6 summarizes and concludes the study, discusses limitations,
and suggests other areas of accounting research that might usefully investigate the effects
of human capital.
2. Background and hypothesis development

Recent auditing research emphasizes the centrality of audit offices in understanding auditing
practices and documents significant interoffice variation in audit outcomes based on an
office’s industry expertise and overall size (e.g., Francis, Stokes, and Anderson 1999; Cras-
well, Stokes, and Laughton 2002; Ferguson et al. 2003; Choi et al. 2010; Reichelt and Wang
2010; Francis et al. 2013). Large audit firms operate through a network of local practice
offices with significant local office autonomy with respect to contracting and administering
audit engagements (Reynolds and Francis 2000). Even though there is firm-level oversight
and quality controls, as former SEC Commissioner Stephen Wallman (1996) notes, the most
critical decisions on an audit are made by the lead engagement office. Given evidence of
intercity variation in audit offices, the goal of this study is to extend the analysis of offices
to include the potential effect of human capital in cities where practice offices are located.
Human capital and cities

The first city-level labor characteristic which we consider is the relative size of the account-
ing labor pool. The notion that labor productivity increases with the size of the local labor
pool is a long-standing proposition in economics, dating at least to Marshall (1920).
Rosenthal and Strange (2004) credit Marshall with identifying three channels through
which labor market concentration (“agglomeration”) can lead to increased productivity:
sharing of inputs, labor market pooling, and knowledge spillovers.

Empirical studies have found an association between city and/or industry size and
positive economic outcomes. For example, Glaeser and Mare (2001) find that workers in
cities (with larger labor markets) earn higher wages than nonurban workers (with smaller
labor markets). This wage difference accrues over time and stays with urban workers
when they move out of cities. Moomaw (1981) also concludes that large cities have sub-
stantial productivity advantages, and that the productivity advantages are more substan-
tial for service firms. Henderson (2003) finds that the number of local same-industry
plants has strong productivity spillovers for high-tech manufacturing firms, and this effect
is stronger for smaller firms because they are more reliant on local labor. In related
research, Rosenthal and Strange (2008) find evidence in support of an “urban rat race,”
where productivity is higher in larger cities because hard-working professionals self-select
into large cities.7
The second labor market characteristic that we consider is average educational attain-
ment (or human capital “depth”; Fu 2007), which has been the focus of more recent urban
economics research. The theory is that individuals’ investments in human capital generate
externalities that accrue to others that are within close geographic proximity (e.g., a city),
in addition to the private benefits that accrue to the individual (Rauch 1993; Glaeser and
Mare 2001; Glaeser 1999; Lucas 1988; Moretti 2004a,b). This occurs as individual eco-
nomic agents expand their personal human capital (knowledge) through meetings and
interactions with other agents randomly selected from a distribution of agents and ideas.
In contrast to the spillovers based on the size of labor pool, knowledge spillovers
based on educational attainment depend on the vertical differences (i.e., “depth”) in what
people know: “If all of us know the same thing, we cannot learn from each other” (Jova-
novic and Rob 1989, 569). In the same way, accountants might expand their learning and
knowledge through interactions with nonaccountants. For example, an individual auditor
may interact with a stock broker or financial analyst who can provide valuable insight
that might affect the fair value of a client’s investments, or who may have industry-specific
knowledge about product developments or emerging technologies affecting a client’s cur-
rent and future prospects.
While the exact mechanism through which these interactions take place remains some-
thing of a “black box” (Jovanovic and Rob 1989), the point is that these interactions, ran-
dom and otherwise, are more likely to be productive when the overall level of human
capital in a city is higher. The empirical evidence on the association between average edu-
cational attainment and labor productivity is robust, having been shown to be associated
with higher wages and higher rents (Rauch 1993), higher individual wages even when con-
trolling for individual ability (Moretti 2004a), manufacturing productivity (Moretti 2004c),
economic growth (Glaeser, Scheinkman, and Shleifer 1995), patent generation (Glaeser
and Saiz 2004), and service firm formation rates (Acs and Armington 2004).
Yet another reason the average educational attainment in a city is associated with pro-
ductivity gains is that workers in these cities might have unobserved characteristics that
make them more productive, for example, if high-ability workers self-sort into certain
cities (Rosenthal and Strange 2008). As Moretti (2004b, 2246) puts it, “a lawyer in New
York is likely to be different from a lawyer in El Paso, TX.” In a similar way, auditors in
high human capital cities might very well be different from auditors in low human capital
cities. In addition, when high-performing individuals are surrounded by other high per-
formers working hard and gaining success, there is social pressure and competition to be
successful and productive (Mas and Moretti 2009).
7. Separating the effects of these different mechanisms, especially the self-selection of high-ability workers into
high human capital cities, is an important econometric concern in the urban economics literature (Moretti
2004b). As noted previously, we discuss the implications of this as it relates to our study in section 6.

Human capital and auditing

Although previous urban economics research has not focused specifically on auditors,
there is little reason to suspect that the robust empirical relationship between human capi-
tal and productivity does not also hold in the audit setting. For example, audit offices are
a good setting in which to analyze the effect of a city’s labor characteristics because audit-
ing is a labor-intensive service industry (Francis 2011). While most, if not all, audit firms
will have established policies and procedures to try to ensure uniform audit quality across
offices, Lopez and Peters (2012) find evidence that individual auditor conditions transcend
the quality of a firm’s control mechanisms. This is further confirmed in a survey of audi-
tors and investors where there was 100 percent agreement that the education, training,
and experience of the individual auditors are a major factor in the production of audit
quality (Christensen, Glover, Omer, and Shelly 2016).8
The auditing and human capital literatures both focus on local city characteristics.
However, in our context, the Big 4 national networks of offices and professionals are also
important. When an audit office is in a low human capital city, the national networks of
the Big 4 auditors can, to some extent, offset this adverse effect on labor quality by relo-
cating staff from other offices, either temporarily or permanently. The Big 4 networks can
also provide a partial cushion against temporary shortages in local human capital that
could potentially affect an audit office. Current partner rotation requirements for public
company audits can strain local offices as partners are not permitted to serve for more
than five consecutive years. The Big 4’s national networks can locate a replacement part-
ner with suitable expertise if one is not present locally.9 In contrast, non-Big 4 offices do
not have the cushion of a national network and are exposed more directly to local labor
market conditions. Bills, Swanquist, and Whited (2016) also find that Big 4 audit firms
are better able to absorb short-term shocks in audit office capacity. Thus, the ability of
non-Big 4 auditors to provide high-quality quality audits will be relatively more depen-
dent upon the level of human capital available locally (Fiolleau, Hoang, Jamal, and
Sunder 2013).
Empirical predictions
The human capital literature leads us to expect cities with high human capital will have
improvements in the quality of accounting labor that results in better audits. Prior
research documents a positive relation between statistical properties of earnings and indi-
cators of auditor quality, such as industry expertise and auditor size (DeFond and Zhang
2014). Building on this literature, as a city’s human capital increases, the audited earnings
of an office are expected to be of higher quality (smaller abnormal accruals, smaller work-
ing capital accruals, and smaller accrual estimation errors), all else equal. Further, these
effects are expected to be stronger for non-Big 4 auditors, as they do not have the benefit
of national networks of professionals to cushion against the lack of local human capital.
Hypothesis 1 in alternative form is:
8. Aside from academic studies, the Public Company Auditing Oversight Board (PCAOB) also understands
this concept and notes that while firm policies and procedures are important “the quality of a firm’s work
ultimately depends on the integrity, objectivity, intelligence, competence, experience, and motivation of per-
sonnel who perform, supervise and review the work” (PCAOB QC Section 20.12). On the practitioner side,
Ian Powell, the UK Chairman of PwC notes that people are the firm’s biggest asset, and the majority of
the firm’s investment should go towards assuring that personnel have the appropriate skills and abilities,
rather than to infrastructure or audit methodology improvements (Starr 2011).
9. Several of the authors of this study had personal experience working for a Big 4 audit firm, and, anecdo-
tally, this has become a common and widespread practice. Many audit partners tell us they did not expect
to be traveling so much at this stage of their career.

Hypothesis 1. The quality of audited earnings improves (smaller abnormal accruals,

working capital accruals, and accrual estimation errors) as the level of human capital
in an audit office’s locale increases, with a relatively stronger effect for non-Big 4
offices.
Audits are jointly produced by clients and their auditors, and audited earnings are the
result of auditor-client negotiations (Antle and Nalebuff 1991; McCracken, Salterio, and
Gibbins 2008). Because auditors and clients are typically in the same city, there is a possi-
bility that the human capital effect is acting through the client to improve earnings quality,
rather than the auditor. An alternative testing strategy is to analyze the auditor’s going-
concern assessment, which is an audit outcome that is more directly under the control of
the auditor and unrelated to the financial statements, per se. We focus on the accuracy of
going-concern opinions because this is a better measure of audit quality in the sense of
higher levels of skill and knowledge (Lennox 1999), as opposed to the frequency of going-
concern reports, which is typically interpreted as a measure of auditor independence (e.g.,
DeFond, Raghunandan, and Subramanyam 2002). We expect that improvements in audit
quality in cities with greater human capital will result in more accurate going-concern opi-
nions, with a relatively stronger effect on non-Big 4 offices:
Hypothesis 2. The accuracy of auditor going-concern opinions increases as the level of

human capital in an audit office’s locale increases, with a relatively stronger effect
for non-Big 4 offices.
Acs and Armington (2004) find that city-level human capital is associated with service
firm formation rates. In our data for 120 cities, we find a similar result for accounting
firms. There are more unique accounting firms involved in the audits of public companies
in cities with a higher level of educational attainment, and in cities where the accounting
labor market is relative larger than the national average.10 If it is easier for service firms,
such as audit firms, to form in cities with high human capital, then non-Big 4 auditors
may also be better able to compete with the Big 4 firms in these markets. In addition, if
Hypothesis 1 is correct and the quality of non-Big 4 audits is affected to a greater extent
by the city’s human capital, this will increase the ability of non-Big 4 auditors to compete
for clients that demand higher-quality audits.11 As a result, we predict the likelihood of
choosing a non-Big 4 auditor is positively related to a city’s human capital:
Hypothesis 3. A company is more likely to choose a non-Big 4 audit firm versus a Big 4
audit firm when the level of human capital in the audit office’s locale is higher.
The final test analyzes the relation between audit fees and human capital. Following
Francis (2011), we view audit quality as a continuum whereby clients demand differential
audit quality and auditors supply differential audit quality to meet the demand. If non-Big
4 audit quality improves at a relatively faster rate in the presence of human capital, as
10. The correlation between the number of audit firms in a city and the two human capital variables, educa-
tional level and labor pool size, is 0.29 and 0.43, respectively.
11. As noted in the introduction, an alternative explanation is that human capital in a city works through the
client’s personnel rather than the auditor’s personnel, and leads to higher-quality pre-audited financial
reporting. Further, if high-quality auditing is viewed as a substitute to pre-audit earnings quality, then this
could predict a higher likelihood of non-Big 4 auditors. On the other hand, if they are complements (simi-
lar to the “confirmation” hypothesis discussed by Ball, Jayaraman, and Shivakumar 2012) then higher-
quality pre-audited earnings could lead to a lower likelihood of non-Big 4 auditors. Our tests investigating
whether human capital works through the auditor or the client are reported in section 4.

hypothesized in Hypothesis 1, and assuming there is a demand for quality that is priced,
then human capital would cause non-Big 4 audit fees to rise at a faster rate as well, all else
equal. However, if non-Big 4 audit quality increases at a relatively faster rate than Big 4
auditors, then non-Big 4 auditors will become more competitive with Big 4 auditors in the
public company audit market. With the increase in non-Big 4 audit quality, some clients
will now have more options to choose from in selecting an auditor with the capability and
expertise to perform their audit. This will increase the competition within the audit market
of those cities and could potentially decrease audit fees (Pearson and Trompeter 1994;
Numan and Willekens 2012). Alternatively, if auditors in high human capital cities are
more productive, fewer hours may be required per audit which would also lower fees.
Therefore, it is difficult to predict ex ante whether the increase in productivity lowers the
price of the audit, or if the increase in audit quality raises the price of the audit, and we state
Hypothesis 4 in null form as:
Hypothesis 4. Audit fees are not associated with the level of human capital in an audit
office’s locale, and the relationship is not significantly different for non-Big 4 audi-
tors compared to Big 4 auditors.
3. Research design
Measuring human capital
The study’s proxy for average educational attainment in a city is the fraction of the city’s
population over the age of 25 with at least a Bachelor’s degree (EDUCATION). This is
the most widely used variable in the urban economics literature to capture the average
skill level of a city’s workforce and the potential for externalities through knowledge/pro-
ductivity spillovers (e.g., Rauch 1993; Moretti 2004a,c; Acs and Armington 2004).12 The
point of this proxy is not to measure the educational attainment of individual auditors. In
fact, we would expect very few, if any, auditors working for public accounting firms dur-
ing our sample period not to have a college degree. Rather, the idea is to capture the gen-
eral effect of knowledge spillovers on audit offices located in cities with higher educational
attainment.
Data on educational attainment comes from the U.S. Census Bureau’s American
Community Survey (hereafter, ACS). The ACS provides data according to several geogra-
phical definitions, including core-based statistical areas (CBSA) which we adopt as our
definition of a city. As measured by the Office of Management and Budget, a CBSA is an
area surrounding an urban center of “at least 10,000 people and adjacent areas that are
socioeconomically tied to the urban center by commuting.”13
The ACS provides one-, three-, and five-year estimates of human capital. The surveys
that generate these estimates are conducted over time, and provide averages of the charac-
teristic during the relevant time period, as opposed to estimates at a particular point in
time as the decennial census has historically provided. We use the five-year estimates from
2005 to 2009 because the five-year estimates include larger sample sizes when surveying
the population, and thus, the estimates of educational attainment are more accurate than
the three- or one-year surveys. ACS began collecting data in 2003, but information is only
12. In section 5, we also investigate results using the percentage of the population with at least a Master’s
degree and find similar results. Other studies have used the high school dropout rate as alternative measure
of human capital (e.g., Acs and Armington 2004). We chose college attainment because it is more likely to
capture the kind of knowledge spillovers that are likely to affect auditors and because it is the variable that
receives the broadest support in the urban economics literature.
13. See http://www.census.gov/population/metro/about/ for a more detailed discussion of the construction of
CBSAs. CBSAs include both metropolitan and micropolitan statistical areas.

available for a smaller number of test pilot CBSAs before 2005, so the 2009 ACS results
are the first year in which the five-year estimates are available for all locations.14
Following the U.S. Department of Labor (DOL), the size of the city’s labor pool is
measured as the “labor quotient,” which is a measure commonly used in regional and
urban economics to measure the concentration of an economic activity (in this case,
accounting employment) in a particular area relative to a larger economic area. Following
the DOL’s approach, we define ACCTLQ as the city-level ratio of people employed as
accountants and auditors in a city to total employment, divided by the national-level ratio.
Thus, ACCTLQ implicitly controls for city population. For ease of exposition, the test
variable ACCTLQ is described as measuring the supply (number) of accountants in a city.
Untabulated results indicate that the study’s results are robust to the use of yearly raw
employment data in a city in lieu of ACCTLQ and a measure scaled by the number of
businesses instead of total employment. This data is provided on an annual basis, so we
measure it separately for each CBSA-year.15
Model specification
The research design for Hypotheses 1, 2, and 4 estimates empirical models separately for
clients of Big 4 and non-Big 4 auditors, and tests equality of the human capital coefficients
between the two samples using a Chow (1960) test. We adopt this approach, instead of a
pooled model with an interaction on our test variable for two reasons: (i) having fewer
interactions eases economic interpretation of the coefficients, and (ii) this approach
imposes fewer assumptions on the model, as all coefficients are allowed to vary across the
non-Big 4/Big 4 subsamples.
Earnings quality
The relation between human capital and earnings quality (Hypothesis 1) is tested by esti-
mating the model in equation (1):
X
EQ ¼ b0 þ b1 EDUCATION þ b2 ACCTLQ þ b3-6 CITYCONTROLS
X
þ b7-23 FIRMCONTROLS þ Industry and Year FEs þ e: ð1Þ
In our main analysis, we include both EDUCATION and ACCTLQ in the model
(rather than testing them separately) because prior literature in urban economics generally
treats Marshallian externalities and human capital depth externalities as two distinct con-
cepts (Fu 2007), with Marshallian externalities arising from the clustering of firms and
workers in the same industry or occupation, and human capital depth externalities arising
from vertical differences in knowledge. Since we view them as different concepts, including
them both in the model allows us to test each of them while controlling for the other.
However, because these two variables are correlated (see Table 2), section 5 discusses the
effects of testing them separately.
Three measures of earnings quality (EQ) are tested: abnormal accruals, working capi-
tal accruals, and accrual estimation errors, with variables constructed such that earnings
quality is assumed to be decreasing as each measure increases in value. Abnormal accruals
14. We considered using a year-specific measure of EDUCATION, but the data are nearly identical from year to
year. To demonstrate, in untabulated results, we obtained the ACS one-year estimates for all CBSAs in the final
sample from 2005 to 2012 and regress EDUCATION on a set of CBSA indicator variables. The r2 from this
regression is over 0.98, indicating less than 2 percent of the variation is attributable to changes across time.
15. OES defines accountants and auditors as those who “examine, analyze, and interpret accounting records
to prepare financial statements, give advice, or audit and evaluate statements prepared by others [and/or]
install or advise on systems of recording costs or other financial and budgetary data [excluding] tax exami-
ners and collectors, and revenue agents.” (http://www.bls.gov/oes/current/oes132011.htm).

are estimated with the modified Jones model (Jones 1991; Dechow, Sloan, and Sweeney
1995), controlling for firm performance (Kothari, Leone, and Wasley 2005). The second mea-
sure of earnings quality is the absolute value of working capital accruals (ABS_WCACCR)
(Dechow, Ge, Larson, and Sloan 2011). The third measure of earnings quality is accrual
estimation error, which is estimated following Dechow et al. (2011) as a cross-sectional
version of the model in Dechow and Dichev (2002). See the Appendix for detailed variable
descriptions.
The two human capital test variables are the percentage of a population with a bache-
lor’s degree or higher (EDUCATION), and the per capita number of accountants and
auditors employed in the CBSA relative to the national average (ACCTLQ). The measures
of human capital and all city-level controls are matched to firm-years based on the auditor
office location named on the audit opinion filed with the SEC and reported by Audit Ana-
lytics. Hypothesis 1 predicts that city-level human capital will be positively associated with
earnings quality. Thus, we expect b1 and b2 to be negative, indicating greater earnings
quality (smaller accruals and smaller estimation error) when human capital (EDUCATION
and ACCTLQ) is larger. We also predict that the coefficients will be relatively larger (more
negative) for non-Big 4 auditors.
City controls are important to assure the human capital variables in the study are not
confounded with other city characteristics. The set of city-level controls includes variables
from prior literature, although we make no directional prediction for these variables.
COMMUTE is the percentage of the city’s population that commutes more than 45 min-
utes per day. It is included to control for the cost in time of working in the city and could
affect audit quality by reducing the amount of total amount of time available to auditors
and making them less efficient as a result. Similar to EDUCATION, we use the ACS five-
year estimates for greater sample size and because there is essentially no variation across
time in this variable.16
Because many of the city-level variables are static across time in our sample, we follow
Acs and Armington (2004) and include two additional control variables intended to cap-
ture changing supply and demand effects in the city. Population growth (POPGROWTH)
and income growth (INCGROWTH) are measured as the percentage change in city popu-
lation and median income from 2005 to 2012, respectively, and are computed from data in
the ACS one-year surveys for 2005 and 2012. The ACS is not available before 2005, which
is why we measure income and population growth beginning in 2005. These variables
could influence the choice of audit firms to open or close offices in a particular city and/or
affect audit quality if it is related to the ability of audit offices to hire and retain qualified
staff. Finally, as Kedia and Rajgopal (2011) show that regulatory scrutiny increases when
firms are located close to SEC regional offices we include an indicator variable SECREG-
OFFICE equal to one if the auditor office is located in a city with a SEC regional head-
quarters or the SEC national office, and equal to zero otherwise.17
We also include auditor-related control variables to address the concern that human
capital is simply capturing omitted auditor characteristics. Audit office size and industry
expertise are associated with earnings quality in prior research (Francis and Yu 2009;
Reichelt and Wang 2010). These factors may also affect the local capacity of audit firms
and may be related to auditor choice. Audit office size (LNOFFICE) is the natural log of
total yearly audit fees of the office performing the audit. INDEXPERT is an indicator
variable equal to one if the auditor is the city-industry leader, as determined by yearly
16. Similar to EDUCATION, we obtained yearly estimates for COMMUTE from the one-year ACS surveys and
regressed this variable on a set of CBSA indicator variables. The R2 from this regression was above 0.98.
17. Following Kedia and Rajgopal (2011), we control for the SEC’s national office which is located in Wash-
ington, D.C., along with the five SEC regional headquarters: New York, NY; Miami, FL; Chicago, IL;
Denver, CO; and Los Angeles, CA.

market share of audit fees where industry is defined as 2-digit SIC codes, and zero other-
wise (Ferguson et al. 2003). Both LNOFFICE and INDEXPERT are measured using the
entire Audit Analytics population. We also include TENURE, an indicator variable equal
to one if the audit firm’s tenure is less than or equal to three years, and INFLUENCE, the
percentage of an audit office’s total fees attributable to the client, because prior research
has documented a relationship between these variables and audit quality (Johnson, Khu-
rana, and Reynolds 2002; Reynolds and Francis 2000).
We also include a set of firm-year controls motivated from prior research. LNAS-
SETS is the natural log of total assets to control for firm size. LOSS is an indicator vari-
able equal to one when net income is negative, ROA (income before extraordinary items
divided by total assets), and LEVERAGE (total debt scaled by total assets) control for
firm risk. TOBINQ measures the firm’s investment opportunities. SALE_GTH is the
annual percentage change in total revenues. OCF is operating cash flows, scaled by total
assets. STD_OCF is the rolling three-year standard deviation of OCF. We also include
variables for lagged absolute total accruals (scaled by assets) and debt/stock issues
(LAGABSACCRUALS, ISSUE) since Dechow et al. (2011) show these variables to be
correlated with earnings management.
Because entrepreneurs and start-up firms are attracted to cities with high levels of
human capital (e.g., Silicon Valley and technology firms), this creates the potential for
self-selection in our findings since start-up firms might be more likely to choose one type
of auditor, have lower-quality earnings since their cash flows are less predictable, and be
less willing to pay higher audit fees.18 We include three variables to control for this selec-
tion effect. LIFECYCLE is a measure of the firm’s stage in its financial lifecycle, and is
calculated as the difference between total common equity and retained earnings, scaled by
total assets (DeAngelo, DeAngelo, and Stulz 2006). The variable is increasing in value for
firms that are more dependent upon outside equity compared to financing from retained
earnings. ACCUMDEF is an indicator variable equal to one for firm-years with a negative
value of retained earnings in COMPUSTAT, and equal to zero otherwise. Lastly, IPO is
an indicator variable equal to one if the firm first appeared in COMPUSTAT within the
last three years. See the Appendix for detailed descriptions and data sources.
Going-concern accuracy
The following logistic regression model is estimated to test Hypothesis 2:
X
PrðGCERROR ¼ 1Þ ¼ a0 þ a1 EDUCATION þ a2 ACCTLQ þ a3-6 CITYCONTROLS
X
þ a7-23 FIRMCONTROLS þ Industry and Year FEs þ e: ð2Þ
GCERROR is an indicator variable equal to one when there was either a Type 1 or Type
2 error associated with the going-concern opinion. Type 1 errors (false positives) occur
when the auditor issued a going-concern opinion but the client did not declare bankruptcy
within 12 months of the date of the audit opinion date, and Type 2 errors (false negatives)
occur when the auditor did not issue a going-concern opinion and the client entered into
bankruptcy within 12 months of the audit opinion date. Bankruptcy data are obtained
from Audit Analytics. We expect a1 and a2 to be negative and significant for both Big 4
and non-Big 4 subsamples, and further expect both coefficients will be larger (more nega-
tive) for non-Big 4 auditors than Big 4 auditors. We estimate this equation on a
18. One might also argue that self-selection creates bias in the opposite direction; for example, “good” firms
choose to locate in cities with high levels of human capital, and this influences their choice of auditor. We
control for this possibility by including LAGABSACCRUALS, which controls for the level of accruals and
potential for earnings management.

subsample of firms with evidence of financial distress, as these are the firms for which the
going-concern decision is most salient for auditors.
Auditor choice
The following logistic regression model is estimated to test Hypothesis 3:
X
PrðNONBIG4tþ1 ¼ 1Þ ¼ c0 þ c1 EDUCATION þ c2 ACCTLQ þ c3-6 CITYCONTROLS
X
þ c7-23 FIRMCONTROLS þ Industry and Year FEs þ e: ð3Þ
NONBIG4 is an indicator variable equal to one when a company is audited by a non-Big

4 auditor and equal to zero otherwise. All right-hand side variables are lagged one year as
the decision to choose an auditor is typically made before the current-year financial state-
ments are available. City-level and firm-level controls are identical to those discussed
above for the model in equation (1). Hypothesis 3 predicts the coefficients on EDUCA-
TION and ACCTLQ will be positive, indicating non-Big 4 auditor choice is more likely as
city-level human capital increases.
Audit fees
Hypothesis 4 is tested with the following regression model:
X
LNAUDITFEE ¼ f0 þ f1 EDUCATION þ f2 ACCTLQ þ f3-6 CITYCONTROLS
X
þ f7-24 FIRMCONTROLS þ Industry and Year FEs þ e: ð4Þ
LNAUDITFEE is the natural log of audit fees reported in Audit Analytics for the firm-year.
The test variables EDUCATION and ACCTLQ are as defined above. We include the same
set of controls as the other models, and also include an indicator variable for whether or
not a going-concern opinion was issued in the prior year, as this has been shown to be posi-
tively associated with audit fees (Hay, Knechel, and Wong 2006). Hypothesis 4 does not
make a specific directional prediction for the coefficients on EDUCATION or ACCTLQ.
Sample selection
The sample begins with firm-year observations available on both COMPUSTAT and
Audit Analytics from 2001 to 2013. We delete financial firms, (6000 ≤ SIC ≤ 6999), utility
firms (4900 ≤ SIC ≤ 4949), and firm-year observations with missing data. We lose 2013
firm-year observations because the Dechow and Dichev (2002) model requires next-period
cash flows.19 We then add to this data set the city-level variables obtained from ACS and
the Department of Labor. The final sample has 24,944 firm-year observations (3,804
unique firms), spread across 120 unique CBSAs for the sample period 2001–2012. When
performing the tests of going-concern errors, firms must have evidence of being in finan-
cial distress, which we define as any of the following: negative earnings or operating cash
flows, current ratio less than one, or negative retained earnings.
Descriptive statistics
Table 1 lists all of the 120 CBSAs in the study and their city-level variables, sorted by
EDUCATION from high to low. Also presented are descriptive statistics for these vari-
ables and the number of firm-year observations in the final sample for each city (N). On
average, 29 percent of residents aged 25 and over have a bachelor’s degree or higher
(EDUCATION). The mean (median) value of ACCTLQ is 0.896 (0.882). Values less
(more) than one indicate that the CBSAs accountants per capita is smaller (larger) than
19. Results are comparable for all other empirical models when 2013 firm-year data is included in the sample.

TABLE 1
406
CBSA listing and descriptive statistics
Panel A: City-level variable summary statistics
EDUCATION ACCTLQ POPGROWTH INCGROWTH COMMUTE SECREGOFFICE
Mean 0.289 0.896 0.094 0.090 0.117 0.050

Median 0.277 0.882 0.093 0.088 0.102 0.000
Std. 0.063 0.224 0.057 0.064 0.054 0.219
25th percentile 0.249 0.749 0.057 0.047 0.077 0.000
75th percentile 0.319 1.026 0.133 0.124 0.153 0.000

Panel B: CBSA listing
N POP- INC- SECREG-

CBSA name (firm-years) EDUCATION ACCTLQ GROWTH GROWTH COMMUTE OFFICE
Contemporary Accounting Research
Boulder, CO 12 0.566 0.867 0.101 0.194 0.105 0

Washington-Arlington-Alexandria, DC-VA-MD-WV 342 0.467 1.311 0.114 0.160 0.293 1
Bridgeport-Stamford-Norwalk, CT 301 0.434 0.517 0.047 0.079 0.206 0
San Francisco-San Mateo-Redwood City, CA 527 0.432 1.225 0.078 0.101 0.216 0
San Jose-Sunnyvale-Santa Clara, CA 1,493 0.432 1.002 0.081 0.099 0.122 0
Boston-Cambridge-Quincy, MA 1,517 0.416 1.002 0.075 0.119 0.211 0
Raleigh-Cary, NC 120 0.413 0.841 0.259 0.112 0.128 0
Madison, WI 11 0.407 1.202 0.110 0.120 0.075 0
Austin-Round Rock, TX 138 0.387 1.393 0.268 0.125 0.145 0
Trenton-Ewing, NJ 114 0.387 1.331 0.064 0.143 0.186 0
Minneapolis-St. Paul-Bloomington, MN-WI 1,150 0.373 1.215 0.079 0.061 0.123 0
Denver-Aurora, CO 591 0.371 1.201 0.117 0.079 0.157 1
Seattle-Bellevue-Everett, WA 356 0.367 1.261 0.117 0.166 0.184 0
New York-White Plains-Wayne, NY-NJ 2,915 0.352 1.370 0.036 0.111 0.314 1
Lincoln, NE 12 0.346 0.973 0.146 0.049 0.053 0
Manchester, NH 5 0.341 0.360 0.022 0.094 0.164 0
Atlanta-Sandy Springs-Marietta, GA 666 0.340 0.827 0.111 0.026 0.238 0
(The table is continued on the next page.)
TABLE 1 (continued)
N POP- INC- SECREG-

San Diego-Carlsbad-San Marcos, CA 457 0.340 0.980 0.112 0.056 0.129 0

Baltimore-Towson, MD 199 0.339 0.909 0.056 0.140 0.208 0
Fargo, ND-MN 2 0.338 0.881 0.206 0.188 0.043 0
Hartford-West Hartford-East Hartford, CT 227 0.338 1.339 0.064 0.052 0.096 0
Lexington-Fayette, KY 11 0.334 0.773 0.166 0.106 0.071 0
Portland-Vancouver-Hillsboro, OR-WA 194 0.329 0.948 0.097 0.116 0.135 0
Chicago-Joliet-Naperville, IL 1,246 0.328 1.105 0.025 0.047 0.254 1
Logan, UT-ID 12 0.327 0.494 0.170 0.153 0.072 0
Des Moines-West Des Moines, IA 54 0.323 1.383 0.131 0.106 0.058 0
Worcester, MA-CT 36 0.322 0.864 0.055 0.090 0.193 0
Columbus, OH 190 0.322 1.127 0.116 0.079 0.091 0
Albany-Schenectady-Troy, NY 84 0.321 1.130 0.068 0.153 0.094 0
New Haven, CT 16 0.320 0.949 0.053 0.106 0.124 0
Philadelphia, PA 897 0.318 0.813 0.062 0.089 0.198 0
Charlotte-Gastonia-Concord, NC-SC 290 0.318 0.824 0.204 0.076 0.133 0
Kansas City, MO-KS 201 0.317 1.108 0.073 0.057 0.096 0
Bellingham, WA 5 0.317 0.913 0.141 0.164 0.085 0
Omaha-Council Bluffs, NE-IA 110 0.313 1.048 0.100 0.119 0.055 0
Rochester, NY 123 0.312 0.829 0.059 0.098 0.070 0
Santa Rosa-Petaluma, CA 11 0.311 0.700 0.076 0.046 0.161 0
Honolulu, HI 30 0.307 1.054 0.104 0.094 0.178 0
Richmond, VA 163 0.307 1.169 0.122 0.043 0.113 0
Milwaukee-Waukesha-West Allis, WI 393 0.305 0.980 0.055 0.068 0.082 0
Oxnard-Thousand Oaks-Ventura, CA 10 0.303 0.827 0.063 0.116 0.167 0
Indianapolis-Carmel, IN 173 0.302 1.127 0.105 0.019 0.111 0
Dallas-Plano-Irving, TX 928 0.301 0.972 0.140 0.117 0.170 0
City Labor Characteristics and Public Audits
Los Angeles-Long Beach-Glendale, CA 1,779 0.300 0.984 0.019 0.093 0.197 1

407

TABLE 1 (continued)
408
N POP- INC- SECREG-

Columbia, SC 17 0.299 1.272 0.190 0.051 0.108 0

Sacramento-Arden-Arcade–Roseville, CA 53 0.298 1.011 0.086 0.023 0.143 0
Salt Lake City, UT 325 0.298 0.856 0.126 0.164 0.084 0
Kalamazoo-Portage, MI 7 0.296 0.082 0.068 0.042 0.072 0
Anchorage, AK 15 0.295 0.800 0.136 0.191 0.101 0

Albuquerque, NM 14 0.294 0.985 0.147 0.045 0.114 0
Nashville-Davidson-Murfreesboro, TN 170 0.293 1.034 0.171 0.098 0.153 0
Charleston-North Charleston-Summerville, SC 3 0.290 0.868 0.186 0.109 0.124 0
Jackson, MS 9 0.286 1.143 0.097 0.107 0.094 0
St. Louis, MO-IL 396 0.285 0.984 0.032 0.050 0.135 0
Miami-Miami Beach-Kendall, FL 504 0.283 0.998 0.063 0.054 0.181 1
Sioux City, IA-NE-SD 2 0.282 0.970 0.156 0.196 0.041 0

Houston-Sugar Land-Baytown, TX 1,019 0.281 1.221 0.172 0.176 0.200 0
Knoxville, TN 9 0.279 0.740 0.108 0.076 0.078 0
Providence-Fall River-Warwick, RI-MA 69 0.278 1.004 0.022 0.044 0.139 0
Billings, MT 9 0.278 0.776 0.109 0.142 0.054 0
Akron, OH 24 0.277 0.945 0.025 0.052 0.114 0
Pittsburgh, PA 345 0.277 0.712 0.019 0.171 0.153 0
Ogden-Clearfield, UT 10 0.277 0.549 0.153 0.149 0.116 0
Boise City-Nampa, ID 38 0.277 0.994 0.183 0.040 0.082 0
Syracuse, NY 42 0.276 0.714 0.055 0.149 0.074 0
Cincinnati-Middletown, OH-KY-IN 216 0.276 0.953 0.055 0.088 0.109 0
Harrisburg-Lebanon-Carlisle, PA 17 0.276 1.276 0.105 0.085 0.094 0
Asheville, NC 3 0.274 0.582 0.129 0.097 0.086 0
Orlando-Kissimmee-Sanford, FL 154 0.273 0.922 0.141 0.036 0.159 0
Phoenix-Mesa-Glendale, AZ 326 0.273 1.030 0.120 0.040 0.160 0
Spokane, WA 65 0.272 0.769 0.113 0.178 0.073 0
TABLE 1 (continued)
N POP- INC- SECREG-

Virginia Beach-Norfolk-Newport News, VA-NC 9 0.271 0.934 0.058 0.123 0.110 0

Oklahoma City, OK 64 0.270 1.024 0.137 0.174 0.077 0
Little Rock-North Little Rock-Conway, AR 39 0.269 0.891 0.135 0.108 0.097 0
Buffalo-Niagara Falls, NY 111 0.267 1.028 0.020 0.113 0.069 0
Greenville-Mauldin-Easley, SC 78 0.264 0.747 0.135 0.075 0.089 0
Reno-Sparks, NV 17 0.263 0.767 0.108 0.037 0.081 0
Wichita, KS 5 0.263 0.717 0.086 0.151 0.044 0
Cleveland-Elyria-Mentor, OH 458 0.263 1.122 0.007 0.037 0.113 0
Detroit-Livonia-Dearborn, MI 279 0.262 1.146 0.032 0.036 0.155 0
Birmingham-Hoover, AL 95 0.261 0.764 0.059 0.042 0.151 0
Grand Rapids-Wyoming, MI 162 0.260 0.894 0.038 0.079 0.093 0
Jacksonville, FL 90 0.260 0.878 0.110 0.059 0.141 0
Allentown-Bethlehem-Easton, PA-NJ 2 0.256 0.697 0.076 0.048 0.181 0
Greensboro-High Point, NC 95 0.256 0.699 0.111 0.031 0.088 0
Tampa-St. Petersburg-Clearwater, FL 311 0.255 0.786 0.088 0.047 0.154 0
Fayetteville-Springdale-Rogers, AR-MO 32 0.252 0.844 0.201 0.055 0.086 0
New Orleans-Metairie-Kenner, LA 147 0.252 0.836 0.079 0.103 0.154 0
Appleton, WI 3 0.251 0.751 0.075 0.068 0.053 0
Tulsa, OK 126 0.249 1.053 0.092 0.138 0.077 0
Peoria, IL 14 0.248 0.691 0.066 0.047 0.070 0
Baton Rouge, LA 5 0.246 0.763 0.143 0.184 0.165 0
San Antonio, TX 95 0.245 0.793 0.191 0.126 0.126 0
Medford, OR 1 0.245 0.709 0.070 0.050 0.054 0
Memphis, TN-MS-AR 114 0.244 0.807 0.071 0.105 0.109 0
Lafayette, LA 3 0.244 0.681 0.146 0.083 0.103 0
Dayton, OH 63 0.243 0.914 0.034 0.003 0.077 0
Davenport-Moline-Rock Island, IA-IL 17 0.242 0.966 0.034 0.066 0.061 0

409

410
TABLE 1 (continued)
N POP- INC- SECREG-

Springfield, MO 2 0.242 0.643 0.145 0.104 0.086 0

South Bend-Mishawaka, IN-MI 12 0.241 0.973 0.049 0.063 0.073 0

Roanoke, VA 23 0.239 0.693 0.101 0.210 0.092 0
Louisville-Jefferson County, KY-IN 129 0.238 0.883 0.095 0.098 0.094 0
Duluth, MN-WI 10 0.238 0.830 0.068 0.135 0.077 0
Erie, PA 4 0.235 0.582 0.054 0.041 0.051 0
Fort Wayne, IN 21 0.233 0.698 0.057 0.055 0.065 0
Chattanooga, TN-GA 31 0.224 0.739 0.109 0.044 0.091 0
Toledo, OH 78 0.221 0.663 0.020 0.017 0.066 0

Reading, PA 3 0.216 0.780 0.078 0.028 0.138 0
Shreveport-Bossier City, LA 25 0.215 0.776 0.076 0.211 0.069 0
Winchester, VA-WV 3 0.214 0.620 0.136 0.021 0.232 0
Las Vegas-Paradise, NV 164 0.213 0.713 0.165 0.027 0.100 0
Scranton-Wilkes-Barre, PA 6 0.212 0.802 0.066 0.179 0.091 0
Charleston, WV 12 0.205 1.086 0.009 0.053 0.121 0
Evansville, IN-KY 15 0.199 0.897 0.057 0.076 0.078 0
Fresno, CA 6 0.194 0.604 0.098 0.022 0.085 0
Canton-Massillon, OH 5 0.191 0.738 0.012 0.002 0.097 0
Battle Creek, MI 7 0.188 0.693 0.006 0.039 0.074 0
El Paso, TX 6 0.188 0.748 0.159 0.278 0.074 0
Elkhart-Goshen, IN 17 0.176 0.776 0.033 0.035 0.046 0
Brownsville-Harlingen, TX 7 0.149 0.385 0.107 0.299 0.069 0
Note: All variables are as defined in the text and in the Appendix.
the national average.20 Some cities like Washington, DC and San Jose, CA have relatively
high values of both EDUCATION and ACCTLQ, but this is not always the case. For
example, Boulder, CO has the largest value of EDUCATION but its ACCTLQ value is
below the sample median, and Columbia, SC has a very large value of ACCTLQ, but
EDUCATION is only equal to the sample mean.
Turning to other city controls, the average CBSA grew in population by 9.4 percent
from 2005 to 2012. Median incomes rose by 9.0 percent over the same period, and, on aver-
age, 11.7 percent of the population commutes more than 45 minutes to work every day.
Table 2 provides descriptive statistics for firm-level variables, as well as the city-level
control and test variables, using the full sample of firm-years. Differences in descriptive
statistics between Tables 1 and 2 for the city-level variables are the result of calculating on
the sample of cities (n = 120) in Table 1 compared to calculating on the sample of firm-
years (n = 24,944) in Table 2. All firm-level continuous variables are winsorized at the 1st
and 99th percentiles of the distributions.
Table 3 presents Pearson correlations for all variables in the regressions, although cor-
relations do not control for other factors and should be interpreted with caution. The test
variable, EDUCATION, is negatively correlated with LNAUDITFEE, but positively corre-
lated with NONBIG4. EDUCATION is negatively associated with ABS_DD, as expected,
shows no relationship with ABS_WCACCR, and is positively associated with ABS_DA.
ACCTLQ is positively associated with LNAUDITFEE and NONBIG4. However, contrary to
Hypothesis 1, ACCTLQ is positively correlated with two earnings-quality proxies.21 An analy-
sis of Table 3 also reveals that our two human capital variables, EDUATION and ACCTLQ,
have much lower correlations with client size and performance compared to other measures
of audit quality, such as Big 4 membership, office size, and industry market leadership.
4. Results
All empirical models reported in Tables 4–8 are statistically significant at p < 0.01.
p-values for all coefficients are conservatively reported as two-tailed, even when we make
a directional prediction, and are based on robust standard errors clustered by each unique
firm (GVKEY). Results are robust to alternative clustering by unique audit office (clus-
ters = 961) and by unique CBSA (clusters = 120).
Earnings-quality regressions
Table 4 presents regression estimations for the three measures of earnings quality
(ABS_DA, ABS_WCACCR and ABS_DD) used to test Hypothesis 1. Overall, the results
are suggestive that as the average skill of the city’s labor force increases, earnings quality
improves for the clients of both non-Big 4 and Big 4 auditors. The coefficient on EDUCA-
TION is negative and significant at p < 0.01 for all the tests of earnings quality, except a
0.05 significance level for the non-Big 4 sample and ABS_DA. In contrast, the results are
not significant for ACCTLQ.22
In panel B, we present the results of the Chow test of equality across the two regres-
sions. The coefficient for EDUCATION is significantly different between the two
20. The mean and median values of this variable do not equal one because the national average of accoun-
tants per capita is calculated using all CBSAs, which is greater than the 120 CBSAs in our final sample.
21. Since some variables are relatively highly correlated such as COMMUTE, SECREOFFICE and
ACCCTLQ, we computed variance inflation factors (VIF) in all models to assess the degree of multi-
collinearity. No VIF in our models exceeded 4, which is below the level which generally causes concern
(e.g., Kennedy 2003 uses a rule of thumb of 10).
22. It is positively significant at the 0.10 level for non-Big 4 auditors when ABS_DA is the dependent variable,
but given that this is the only statistically significant coefficient in the table, we do not view this as a mean-
ingful result.

TABLE 2
Descriptive statistics at the firm-year level (3,804 unique firms in 120 CBSAs)
Variable N Mean Median SD 25th percentile 75th percentile
EDUCATION 24,944 0.332 0.328 0.057 0.285 0.371

ACCTLQ 24,944 1.142 1.127 0.238 0.991 1.284
ABS_DA 24,944 0.144 0.082 0.239 0.034 0.167
ABS_WCACCR 24,944 0.001 0.004 0.095 0.026 0.035
ABS_DD 24,944 0.010 0.025 0.078 0.049 0.005
LNAUDITFEE 24,944 0.052 0.000 0.223 0.000 0.000
GCERROR 24,944 6.374 6.363 1.439 5.247 7.378
NONBIG4 24,944 0.280 0.000 0.449 0.000 1.000
POPGROWTH 24,944 0.077 0.075 0.052 0.036 0.112
INCGROWTH 24,944 0.089 0.093 0.046 0.056 0.113
COMMUTE 24,944 0.180 0.170 0.070 0.123 0.211
SECREGOFFICE 24,944 0.295 0.000 0.456 0.000 1.000
LNOFFICE 24,944 16.400 16.680 1.944 15.160 17.920
INDEXPERT 24,944 0.308 0.000 0.462 0.000 1.000
INFLUENCE 24,944 0.125 0.044 0.206 0.015 0.128
TENURE 24,944 0.222 0.000 0.415 0.000 0.000
LNASSETS 24,944 5.730 5.738 2.243 4.077 7.330
LOSS 24,944 0.335 0.000 0.472 0.000 1.000
ROA 24,944 0.016 0.036 0.233 0.039 0.086
LEVERAGE 24,944 1.237 0.782 4.032 0.330 1.564
TOBINQ 24,944 1.908 1.450 1.472 1.083 2.145
SALE_GTH 24,944 0.104 0.065 0.351 0.042 0.183
OCF 24,944 0.059 0.083 0.178 0.016 0.144
STD_OCF 24,944 0.086 0.056 0.095 0.032 0.099
ACCUMDEF 24,944 0.392 0.000 0.488 0.000 1.000
LIFECYCLE 24,944 0.011 0.003 0.025 0.001 0.009
LAGABSACCRUALS 24,944 0.111 0.070 0.344 0.035 0.125
ISSUE 24,944 0.410 0.000 0.492 0.000 1.000
IPO 24,944 0.005 0.000 0.071 0.000 0.000
Note: All variables are as defined in the text and in the Appendix.
subsamples at the 0.01 level for two measures of earnings quality and at the 0.10 level with
ABS_DA, and the directional sign is larger (more negative) as expected for non-Big 4
offices. For completeness, we also report the F-statistic for the Chow test of ACCTLQ,
which is only statistically significant (p < 0.10) with ABS_DA.
The results are economically significant as well. Moving from Tampa, FL (where
EDUCATION = 0.255, 28th percentile) to Chicago, IL (where EDUCATION = 0.328,
80th percentile) reduces the estimate of abnormal accruals for Big 4 auditors by 0.008, or
0.8 percent of total assets, compared to 0.025, or about 2.5 percent of total assets, for
non-Big 4 auditors. Similar effects are observed for working capital accruals and accrual
estimation errors. Moving from Tampa, FL to Chicago, IL reduces predicted accrual esti-
mation errors (absolute working capital accruals) by 0.2 percent and 1.1 percent (0.3 per-
cent and 1.2 percent) of total assets for clients of Big 4 and non-Big 4 auditors,
respectively.
Alternative specifications to equation (1)

Table 5 presents several alternative specifications to equation (1) that address different
potential endogeneity threats. As noted in the introduction, it is possible that human

TABLE 3
Pearson correlation matrix
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)
(1) EDUCATION 1.00

(2) ACCTLQ 0.50 1.00
(3) ABS_DA 0.02 0.03 1.00
(4) ABS_WCACCR 0.01 0.02 0.02 1.00
(5) ABS_DD 0.03 0.02 0.29 0.19 1.00
(6) GCERROR 0.01 0.00 0.20 0.13 0.28 1.00
(7) NONBIG4 0.05 0.07 0.15 0.04 0.22 0.17 1.00
(8) LNAUDITFEE 0.03 0.14 0.17 0.01 0.22 0.18 0.52 1.00
(9) POPGROWTH 0.05 0.11 0.01 0.00 0.01 0.00 0.02 0.04 1.00
(10) INCGROWTH 0.21 0.25 0.03 0.02 0.02 0.02 0.02 0.04 0.22 1.00
(11) COMMUTE 0.31 0.54 0.04 0.00 0.03 0.02 0.05 0.14 0.27 0.16 1.00
(12) SECREGOFFICE 0.06 0.40 0.04 0.00 0.04 0.04 0.04 0.19 0.42 0.03 0.68 1.00
(13) LNOFFICE 0.26 0.17 0.14 0.04 0.20 0.16 0.64 0.70 0.06 0.03 0.20 0.02 1.00
(14) INDEXPERT 0.03 0.01 0.05 0.01 0.09 0.05 0.31 0.28 0.01 0.01 0.01 0.00 0.31 1.00
(15) INFLUENCE 0.21 0.13 0.03 0.01 0.04 0.04 0.01 0.32 0.03 0.05 0.14 0.04 0.60 0.11
(16) TENURE 0.00 0.00 0.05 0.00 0.07 0.05 0.20 0.26 0.01 0.00 0.05 0.06 0.19 0.10
(17) LNASSETS 0.04 0.03 0.23 0.07 0.29 0.25 0.87 0.58 0.02 0.03 0.02 0.08 0.58 0.29
(18) LOSS 0.09 0.02 0.16 0.23 0.20 0.28 0.21 0.18 0.01 0.01 0.03 0.03 0.14 0.06
(19) ROA 0.07 0.03 0.36 0.24 0.29 0.41 0.22 0.20 0.00 0.03 0.04 0.04 0.16 0.06
(20) LEVERAGE 0.04 0.03 0.03 0.00 0.00 0.04 0.08 0.05 0.00 0.04 0.01 0.01 0.04 0.03
(21) TOBINQ 0.11 0.04 0.22 0.02 0.15 0.14 0.11 0.08 0.01 0.03 0.02 0.02 0.03 0.01
(22) SALE_GTH 0.03 0.03 0.13 0.23 0.01 0.02 0.01 0.03 0.02 0.04 0.01 0.00 0.01 0.01
(23) OCF 0.07 0.04 0.27 0.07 0.26 0.35 0.23 0.23 0.03 0.02 0.05 0.05 0.17 0.07
(24) STD_OCF 0.08 0.04 0.32 0.02 0.29 0.23 0.32 0.28 0.01 0.07 0.04 0.04 0.22 0.11
(25) ACCUMDEF 0.14 0.06 0.21 0.09 0.20 0.27 0.25 0.23 0.00 0.02 0.06 0.05 0.16 0.08
(26) LIFECYCLE 0.11 0.07 0.32 0.08 0.33 0.38 0.29 0.27 0.01 0.06 0.05 0.05 0.20 0.10
(27) LAGABSACCRUALS 0.02 0.01 0.30 0.28 0.26 0.24 0.19 0.20 0.03 0.04 0.03 0.04 0.16 0.06
413

414
TABLE 3 (continued)
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)
(28) ISSUE 0.05 0.03 0.08 0.06 0.03 0.06 0.07 0.03 0.04 0.01 0.01 0.01 0.02 0.02
(29) IPO 0.02 0.01 0.02 0.00 0.04 0.06 0.06 0.08 0.01 0.01 0.01 0.01 0.07 0.03
(15) (16) (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) (27) (28) (29)

(15) INFLUENCE 1.00
(16) TENURE 0.05 1.00
(17) LNASSETS 0.03 0.20 1.00
(18) LOSS 0.01 0.08 0.33 1.00
(19) ROA 0.00 0.06 0.35 0.62 1.00
(20) LEVERAGE 0.01 0.01 0.10 0.00 0.03 1.00

(21) TOBINQ 0.05 0.00 0.18 0.04 0.26 0.08 1.00
(22) SALE_GTH 0.00 0.00 0.00 0.10 0.00 0.00 0.19 1.00
(23) OCF 0.01 0.08 0.37 0.50 0.78 0.03 0.21 0.02 1.00
(24) STD_OCF 0.01 0.09 0.41 0.26 0.41 0.05 0.36 0.18 0.40 1.00
(25) ACCUMDEF 0.00 0.11 0.41 0.51 0.43 0.00 0.16 0.04 0.40 0.32 1.00
(26) LIFECYCLE 0.00 0.09 0.45 0.38 0.60 0.07 0.40 0.02 0.56 0.45 0.46 1.00
(27) LAGABSACCRUALS 0.01 0.13 0.27 0.25 0.36 0.02 0.23 0.06 0.27 0.41 0.31 0.36 1.00
(28) ISSUE 0.00 0.00 0.07 0.06 0.15 0.06 0.11 0.14 0.15 0.07 0.08 0.07 0.08 1.00
(29) IPO 0.02 0.01 0.07 0.04 0.05 0.00 0.03 0.03 0.04 0.06 0.03 0.02 0.02 0.02 1.00
Notes: This table presents the pairwise Pearson correlations. Bold face type indicates statistical significance at p < 0.01. Variable definitions are provided
in the Appendix.
TABLE 4
Earnings-quality regressions
Panel A: Regressions DV = ABS_DA DV = ABS_DD DV = ABS_WCACCR
Non-Big 4 Big 4 Non-Big 4 Big 4 Non-Big 4 Big 4
Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
EDUCATION 0.349 0.013** 0.108 0.000*** 0.150 0.000*** 0.032 0.008*** 0.164 0.000*** 0.042 0.001***
ACCTLQ 0.054 0.087* 0.001 0.851 0.009 0.277 0.003 0.304 0.009 0.332 0.002 0.608
POPGROWTH 0.099 0.296 0.063 0.037** 0.045 0.185 0.023 0.028** 0.030 0.441 0.032 0.006***
INCGROWTH 0.061 0.555 0.044 0.163 0.050 0.132 0.001 0.957 0.065 0.092* 0.000 0.995
COMMUTE 0.197 0.013** 0.018 0.563 0.075 0.011** 0.004 0.780 0.065 0.052* 0.001 0.948
SECREGOFFICE 0.018 0.193 0.002 0.692 0.011 0.018** 0.003 0.126 0.010 0.052* 0.005 0.014**
LNOFFICE 0.001 0.820 0.002 0.242 0.001 0.607 0.000 0.473 0.001 0.547 0.001 0.327
CITYINDLEADER 0.013 0.430 0.001 0.691 0.008 0.081* 0.001 0.296 0.009 0.108 0.001 0.425
INFLUENCE 0.011 0.657 0.002 0.836 0.006 0.418 0.008 0.038** 0.006 0.510 0.009 0.018**
TENURE 0.020 0.027** 0.004 0.214 0.005 0.101 0.002 0.287 0.003 0.382 0.001 0.700
LNASSETS 0.012 0.009*** 0.004 0.000*** 0.008 0.000*** 0.005 0.000*** 0.010 0.000*** 0.006 0.000***
LOSS 0.085 0.000*** 0.020 0.005*** 0.003 0.332 0.005 0.004*** 0.000 0.949 0.008 0.000***
ROA 0.308 0.000*** 0.287 0.000*** 0.016 0.144 0.036 0.000*** 0.022 0.129 0.032 0.002***
LEVERAGE 0.001 0.259 0.001 0.159 0.000 0.633 0.000 0.066* 0.000 0.802 0.000 0.018**
TOBINQ 0.007 0.138 0.002 0.350 0.004 0.005*** 0.001 0.219 0.004 0.008*** 0.001 0.166
SALE_GTH 0.084 0.000*** 0.034 0.000*** 0.008 0.046** 0.006 0.051* 0.012 0.011** 0.007 0.022**
OCF 0.074 0.269 0.160 0.000*** 0.015 0.259 0.018 0.072* 0.001 0.934 0.020 0.054*
STD_OCF 0.302 0.000*** 0.242 0.000*** 0.130 0.000*** 0.094 0.000*** 0.130 0.000*** 0.105 0.000***
ACCUMDEF 0.010 0.260 0.017 0.000*** 0.004 0.171 0.003 0.025** 0.005 0.136 0.003 0.066*
LIFECYCLE 0.747 0.006*** 0.703 0.001*** 0.421 0.000*** 0.376 0.000*** 0.525 0.000*** 0.278 0.006***
LAGABS-
ACCRUALS 0.098 0.004*** 0.057 0.055* 0.013 0.027** 0.009 0.117 0.021 0.011** 0.007 0.145
ISSUE 0.013 0.143 0.010 0.000*** 0.005 0.119 0.001 0.565 0.006 0.106 0.001 0.493
IPO 0.041 0.155 0.070 0.021** 0.025 0.045** 0.025 0.009*** 0.021 0.099* 0.024 0.009***
415

416
TABLE 4 (continued)
Panel A: Regressions DV = ABS_DA DV = ABS_DD DV = ABS_WCACCR

Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value Coefficient p-value
Year and
industry FEs Yes Yes Yes Yes Yes Yes
N 6,990 17,954 6,990 17,954 6,990 17,954
R2 0.248 0.173 0.214 0.160 0.225 0.171
Panel B: Chow test of non-Big 4 = Big 4
F-stat. p-value F-stat. p-value F-stat. p-value
EDUCATION 2.83 0.093* 8.34 0.004*** 6.73 0.009***

ACCTLQ 3.01 0.083* 0.48 0.487 0.59 0.442
Notes: All variables are as defined in the text and in the Appendix. Year and industry (2-digit SIC) fixed effects are included but not reported. Two-tailed
p-values are presented beside the coefficient estimates and are calculated using robust standard errors clustered at the firm level (1,436 and 2,945 clusters
for non-Big 4 and Big 4 audited clients, respectively). *, **, and *** denote two-tailed statistical significance at the 0.10, 0.05, and 0.01 levels, respectively.
TABLE 5
Earnings-quality regressions—alternate specifications
DV = ABS_DA DV = ABS_DD DV = ABSWCACCR
Panel A: Different CBSAs
EDUCATION 0.655 0.007 0.193 0.021 0.140 0.021

(0.070)* 0.923 (0.019)** 0.572 (0.140) (0.618)
EDUCATION_CLIENT 0.364 0.058 0.004 0.017 0.045 0.029
(0.073)* 0.314 (0.940) (0.474) (0.558) (0.266)
ACCTLQ 0.066 0.009 0.010 0.005 0.002 0.001
(0.338) 0.534 (0.571) (0.521) (0.928) (0.855)
ACCTLQ_CLIENT 0.005 0.027 0.001 0.003 0.013 0.002
(0.911) (0.067)* (0.955) (0.600) (0.381) (0.805)
N 2,153 3,269 2,153 3,269 2,153 3,269
R2 0.549 0.428 0.248 0.209 0.249 0.262
H0: EDUCATION [Non-Big 4] = EDUCATION [Big 4] 3.20 0.074* 3.77 0.052* 1.35 0.245
H0: ACCTLQ [Non-Big 4] = ACCTLQ [Big 4] 0.66 0.418 0.08 0.779 0.02 0.878
Panel B: Propensity score matched samples (NONBIG4)
EDUCATION 0.220 0.080 0.213 0.051 0.206 0.069

(0.029)** (0.672) (0.000)*** (0.188) (0.000)*** (0.069)*
ACCTLQ 0.037 0.015 0.029 0.015 0.215 0.019
(0.126) (0.541) (0.003)*** 0.189 (0.042)** (0.122)
N 3,544 3,544 3,544 3,544 3,544 3,544
R2 0.4153 0.5356 0.2493 0.2154 0.2742 0.2431
H0: EDUCATION [Non-Big 4] = EDUCATION [Big 4] 2.06 0.151 7.63 0.006*** 4.21 0.040**

417

TABLE 5 (continued)
418
DV = ABS_DA DV = ABS_DD DV = ABSWCACCR
Panel C: Propensity score matched samples (HIGH_HC)
EDUCATION 0.520 0.082 0.185 0.042 0.219 0.046

(0.023)** (0.021)** (0.001)*** (0.025)** (0.001)*** (0.025)**
ACCTACCTLQ 0.078 0.006 0.000 0.007 0.006 0.001
(0.123) (0.489) (0.989) (0.118) (0.714) (0.755)

N 4,310 10,456 4,310 10,456 4,310 10,456
R2 0.5270 0.3568 0.2755 0.1872 0.2758 0.2339
H0: EDUCATION [Non-Big 4] = ACCTLQ [Big 4] 3.58 0.058* 5.56 0.018** 6.12 0.013**
Panel D: 2SLS
EDUCATION 0.327 0.099 0.105 0.026 0.122 0.038

(0.060)* (0.003)*** (0.025)** (0.048)** (0.025)** (0.005)***
ACCTLQ 0.051 0.002 0.004 0.002 0.004 0.001
(0.122) (0.737) (0.674) (0.400) (0.719) (0.679)
H0: EDUCATION [Non-Big 4] = EDUCATION [Big 4] 3.37 0.067* 3.70 0.054* 3.55 0.060*
Panel E: Multilevel modeling
EDUCATION 0.349 0.114 0.173 0.029 0.212 0.039

(0.004)*** (0.000)*** (0.000)*** (0.044)** (0.000)*** (0.020)**
ACCTLQ 0.045 0.004 0.012 0.000 0.018 0.004
(0.098)* (0.615) (0.193) (0.896) (0.091)* (0.242)
H0: EDUCATION [Non-Big 4] = EDUCATION [Big 4] 5.06 0.004*** 10.50 0.001*** 18.31 0.000***
H0: ACCTLQ [Non-Big 4] = ACCTLQ [Big 4] 4.16 0.042** 0.063 0.800 2.79 0.100*
Notes: All variables are as defined in the text and in the Appendix. This table presents five alternate specifications to the OLS estimation presented in Table 4. All
controls from Table 4 are included in the model but not presented. *, **, and *** denote two-tailed statistical significance at the 0.10, 0.05, and 0.01 levels, respectively.
capital is acting through the client, rather than the auditor, to improve earnings quality.
Clients may experience the same benefits of high human capital as auditors, which might
improve the quality of pre-audited earnings and create less need for high-quality external
auditors. While this is possible, the relationship between having more knowledgeable per-
sonnel in the financial reporting department and the quality of pre-audited earnings is
ambiguous ex ante since the incentives of clients for high-quality reporting are less clear
cut than those of auditors. For example, higher-quality client personnel might be more
skillful in opportunistically managing earnings without being detected, which would
decrease (rather than increase) earnings quality.
To further investigate this issue, we reestimate the earnings-quality regressions in
Table 4 using just the 28 percent of the sample for which the auditor and client are
located in different cities and directly control for the average human capital of the client’s
city (EDUCATION_CLIENT). Despite the significantly reduced sample size, the results in
panel A of Table 5 are consistent with the results for non-Big 4 auditors, though not for
Big 4 auditors. The coefficient for EDUCATION is negative in all regressions, but is only
statistically significant for the non-Big 4 subsample (for ABSWCACCR it is significant at
the p < 0.10 level in a one-tailed test). The coefficient magnitudes on EDUCATION are
generally larger in Table 5 than in Table 4 for both non-Big 4 and Big 4 auditors,
although a formal test (untabulated) indicates the differences between Table 4 and Table 5
are never statistically significant.
As in Table 4, ACCTLQ is again statistically significant only in one of the regres-
sions at the 0.10 level or better. We also find little support for the alternative explana-
tion that human capital works through the client instead of the auditor, since
EDUCATION_CLIENT and ACCTLQ_CLIENT are each significant in only one of the
six regressions. Thus, based on this test we conclude that the relationship between
human capital and earnings quality from Table 4 is more likely working through the
auditor than the client.
Panel B of Table 5 addresses the endogeneity threat of auditor self-selection bias.
Recent studies argue that differential audit quality is explained by differences in client
characteristics that are not adequately controlled for in cross-sectional models (Lawrence,
Minutti-Meza, and Zhang 2011; Minutti-Meza 2013). These studies advocate propensity
score matching as a way to create pseudo-random samples between Big 4 and non-Big 4
auditors. In panel B, we report the results after estimating a propensity score matching
equation, using all of the client control variables. We match each non-Big 4 observation
to a Big 4 observation with replacement, using a caliper distance of 0.001. The results are
consistent with those presented in Table 4, and in fact, the coefficients increase compared
to the full sample results in Table 4 for two of the three measures.23
In addition to clients self-selecting their auditors, they also self-select the cities in
which they place their headquarters, presenting another potential source of endogeneity.
23. We used a caliper of 0.001 because this is the largest caliper that resulted in no statistically significant
(p < 0.10) difference in means on any control variables, two-tailed with standard errors clustered by firm.
This was our main process for analyzing balance and was performed before viewing results. We applied
the same process and caliper when matching by HIGH_HC. Given recent research emphasizing the sensi-
tivity of results to various PSM choices (e.g., DeFond, Erkens, and Zhang 2016; Lawrence, Minutti-Meza,
and Zhang 2016; Shipman, Swanquist, and Whited 2017), we also conducted sensitivity tests by reestimat-
ing the propensity score models, using four alternative specifications: no replacement, two-neighbor match-
ing, three-neighbor matching, and nearest neighbor matching with a smaller caliper (.0001). These four
tests, times three different accrual dependent variables, results in twelve different models for Big 4 and
non-Big 4. For non-Big 4, EDUCATION was negative and significant (p < 0.05) in all of them. For Big 4,
EDUCATION was negative and significant (p < 0.10) in five of the models. For the GCERROR tests,
non-Big 4 and Big 4 have four models each. EDUCATION is negative and significant (p < 0.10) in all four
for non-Big 4 auditors, but is never significant for Big 4 auditors.

To address this, we estimate an additional propensity score matching equation, where we

use a logistic model to calculate the likelihood that clients are located in a high human
capital city. HIGH_HC is an indicator variable equal to one when EDUCATION is above
the sample median. We then match each client in a low human capital city to a client in a
high human capital city. Similar to the NONBIG4 matching, we estimate the propensity
score using all client characteristics and match with replacement, using a caliper of 0.001.
The matching is done separately within the non-Big 4 and Big 4 subsamples. These results,
presented in panel C of Table 5, are similar to those reported in Table 4, and again, the
coefficients mostly increase compared to Table 4.
Finally, there could be unobserved city-level omitted variables that are driving the
results for EDUCATION in Table 4. To address this concern, we estimate a 2SLS model
where we use LANDGRANT which is an indicator variable equal to one when a land
grant university is present in the city, and lagged values of educational attainment from
1980, as instruments for EDUCATION.24 Since we only find results for EDUCATION in
Table 4, we did not attempt to identify an instrumental variable for ACCTLQ. The 2SLS
results are presented in panel D of Table 5, and are consistent with Table 4 which
increases confidence that EDUCATION is capturing the intended human capital construct
in the empirical models.25
Finally, we also reperform all tests using multilevel mixed modeling, instead of OLS,
as an alternative way to account for the multilevel grouping of the data (e.g., Hox 2002).
Instead of correcting for correlation in the residuals after estimating the coefficients, multi-
level modeling accounts for correlation while solving the model using maximum likelihood
estimation. For each of these models, we allowed for random slopes and intercepts at the
CBSA level for each of the city-level variables, and random slopes and intercepts at the
firm level for all other variables. Results, presented in panel E of Table 5, are similar to
Table 4.26
Going-concern errors
Table 6 presents results using GCERROR as the proxy for audit quality. Similar to the
earnings-quality regressions in Table 4, EDUCATION is negatively associated with non-
Big 4 audit quality, as indicated by the statistically significant coefficient of 4.929 (p-
value < 0.01). In addition, we also find that ACCTLQ is negative and marginally signifi-
cant for non-Big 4 auditors (p < 0.10). However, for Big 4 auditors, the coefficients of
both EDUCATION and ACCTLQ are statistically insignificant at conventional levels.
In terms of economic significance, these results indicate that for a typical non-Big 4
client (i.e., median values of all control variables), the going-concern error rate decreases
from 35 percent to 28 percent (or a reduction of 20 percent) when going from a city like
Tampa, FL (low human capital) to Chicago, IL (high human capital).
24. The presence of a land grant university has been used previously in the economics literature to instrument
for city-level education (e.g., Moretti 2004a) and has been shown to have increased average education since
1900 (Shapiro 2006). In addition, because the decision to locate land grant universities was made in the
1800s, it is unlikely to be correlated with any of our dependent variables.
25. An important caveat to these instrumental variables is that, while they are correlated with our test vari-
ables of interest, and unlikely to have a causal effect on our dependent variables, they may not correct for
endogeneity arising from the self-selection of clients into certain cities. For example, if clients select head-
quarters based on the level of educational attainment, they might also be more likely to select certain cities
because of the presence of a land grant university, or to have made their location decision based on histor-
ical education levels.
26. The Chow test of coefficients across regressions was performed, using seemingly unrelated regression (suest
in Stata). As this command in Stata does not support mixed modeling, test of coefficients across regres-
sions was estimated by combining both subsamples and including an interaction term with NONBIG4 for
all variables. In all other models, this creates results that are very similar to the tabulated results.

TABLE 6
Going-concern errors
Panel A: Logit regression
DV = GCERROR
Non-Big 4 Big 4
Coefficient p-value Coefficient p-value
EDUCATION 4.929 0.008*** 2.209 0.250

ACCTLQ 0.793 0.072* 0.141 0.723
POPGROWTH 1.894 0.333 0.928 0.572
INCGROWTH 3.687 0.038** 0.733 0.639
COMMUTE 1.773 0.239 1.316 0.427
SECREGOFFICE 0.153 0.501 0.340 0.173
LNOFFICE 0.050 0.586 0.132 0.158
INDEXPERT 0.167 0.413 0.143 0.344
INFLUENCE 0.730 0.051* 0.805 0.227
TENURE 0.027 0.838 0.013 0.942
LNASSETS 0.472 0.000*** 0.132 0.042**
LOSS 1.014 0.000*** 0.989 0.000***
ROA 0.901 0.000*** 1.805 0.000***
LEVERAGE 0.007 0.539 0.023 0.049**
TOBINQ 0.073 0.043** 0.355 0.000***
SALE_GTH 0.111 0.197 0.440 0.017**
OCF 0.505 0.073* 1.472 0.007***
STD_OCF 0.752 0.149 1.065 0.212
ACCUMDEF 1.749 0.000*** 1.627 0.000***
LIFECYCLE 6.268 0.001*** 9.111 0.000***
LAGABSACCRUALS 0.080 0.319 0.044 0.654
ISSUE 0.057 0.645 0.101 0.489
IPO 0.774 0.018** 0.312 0.624
Year and industry fixed effects Yes Yes
N 4,825 8,553
Pseudo R2 0.284 0.238
F-stat. p-value
EDUCATION 1.46 0.227

ACCTLQ 2.90 0.089*
Panel C: Alternate specifications
Non-Big 4 Big 4 Chow test
Different CBSAs:
EDUCATION 4.743 1.61 0.43
(0.076)* (0.689) (0.510)
ACCTLQ 0.543 0.592 1.52
(0.322) (0.425) (0.218)

TABLE 6 (continued)
Panel C: Alternate specifications
Non-Big 4 Big 4 Chow test
Propensity score matching (NONBIG4):

EDUCATION 5.077 2.197 1.95
(0.007)*** (0.659) (0.163)
ACCTLQ 0.711 0.699 0.00
(0.108) (0.558) (0.992)
Propensity score matching (HIGH_HC):
EDUCATION 2.937 2.43 0.03
(0.182) (0.352) (0.874)
ACCTLQ .808 .540 0.09
(0.225) (0.362) (0.762)
IV regressions (2SLS):
EDUCATION 3.712 2.02 0.27
(0.082)* (0.429) (0.604)
ACCTLQ 0.938 0.135 3.01
(0.04)** (0.749) (0.083)*
Notes: The dependent variable GCERROR is coded one if there is either a Type 1 (false negative) or
Type 2 (false positive) going-concern reporting error, and zero otherwise. All variables are as defined
in the text and in the Appendix. Year and industry (2-digit SIC) fixed effects are included but not
reported. Two-tailed p-values are presented beside or in parentheses below the coefficient estimates
and are calculated using robust standard errors clustered at the firm level. *, **, and *** denote
two-tailed statistical significance at the 0.10, 0.05, and 0.01 levels, respectively.
In panel C of Table 6, we present the same alternative specifications that we per-

formed for the earnings-quality regressions to test the robustness of the human capital
variable EDUCATION.27 The tests show that the coefficient for EDUCATION is negative
and statistically significant in the non-Big 4 sample in three out of the four alternative
model specifications, but never statistically significant in the Big 4 sample. Overall, these
tests indicate the non-Big 4 effect is greater than the Big 4 effect, as in prior tests. How-
ever, in contrast to Tables 4 and 5, the formal Chow test fails to reject the null hypothesis
that the coefficients are equal across the two subsamples, despite the coefficient always
being significant in the non-Big 4 regressions and anywhere from 1.2 to 2.9 times larger
than for Big 4 auditors.
Auditor choice regressions

The result of estimating equation (3) is presented in Table 7. The coefficient on EDUCA-
TION is 2.825 (p-value = 0.028) and the coefficient on ACCTLQ is 0.443 (p-
value = 0.064). These results support the prediction in Hypothesis 3 that a firm is signifi-
cantly more likely to choose a non-Big 4 auditor as the level of human capital in the city
increases, both in terms of labor market size and average skill level.
In order to demonstrate the economic significance of this result, we construct a hypo-
thetical firm that has the median value of all firm- and city-level control variables based
on a subsample of firms with total assets less than the largest non-Big 4 client. This is
intended to create a hypothetical client firm with a relatively equal chance of choosing
27. The only exception is the multilevel modeling. We allowed the program to run continuously for one week,
but the multilevel model was unable to converge in the logistic regression.

TABLE 7
Logistic regression of auditor choice
Coefficient p-value
EDUCATION 2.825 0.028**

ACCTLQ 0.443 0.064*
POPGROWTH 1.116 0.306
INCGROWTH 2.263 0.052*
COMMUTE 13.381 0.000***
SECREGOFFICE 0.056 0.726
LNOFFICE 1.323 0.000***
INDEXPERT 0.323 0.004***
INFLUENCE 1.315 0.000***
TENURE 0.916 0.000***
LNASSETS 0.842 0.000***
LOSS 0.377 0.000***
ROA 0.325 0.169
LEVERAGE 0.012 0.095*
TOBINQ 0.122 0.000***
SALE_GTH 0.040 0.591
OCF 0.007 0.981
STD_OCF 0.488 0.285
ACCUMDEF 0.198 0.093*
LIFECYCLE 8.784 0.000***
LAGABSACCRUALS 0.017 0.819
ISSUE 0.065 0.417
IPO 0.364 0.321
Year and industry fixed effects Yes
N 23,913
Pseudo R2 0.685
Notes: The dependent variable is an indicator variable equal to one when the firm’s auditor in year
t+1 is a non-Big 4 audit firm, and zero otherwise. All variables are as defined in the text and in
Appendix. Year and industry (2-digit SIC) fixed effects are included but not reported. Two-tailed p-
values, based on robust standard errors clustered at the firm level (3,580 unique firms), are presented
beside the coefficient estimates. *, **, and *** denote two-tailed statistical significance at the 0.10,
0.05, and 0.01 levels, respectively.
between a Big 4 and non-Big 4 audit firm. If the audit office is located in a city with relatively
low levels of human capital, such as Tampa, FL, the predicted odds of choosing a non-Big 4
auditor is 21 percent. However, if the audit office is located in a city with relatively high levels
of human capital, such as Chicago, IL, the predicted probability rises to 60 percent.
Thus, the association between human capital within a city and the probability of
choosing a non-Big 4 auditor is both statistically and economically large. This suggests
that the supply of public company audits by non-Big 4 firms is significantly affected by
the city’s level of human capital, both educational attainment and the number of accoun-
tants, and that non-Big 4 firms can better compete with the Big 4 in cities with higher
levels of human capital.
Turning to the city-level control variables, POPGROWTH and SECREGOFFICE are
insignificantly different from zero. COMMUTE is positive and significant at p < 0.01, and
INCGROWTH is negative and significant at p < 0.10. Overall, four of the six city-level
variables (including human capital) in the regression are significant at the 0.10 level or

TABLE 8
Audit fees
Panel A: Regressions
DV = LNAUDITFEE
Non-Big 4 Big 4
Coefficient p-value Coefficient p-value
EDUCATION 0.823 0.001*** 0.079 0.660

ACCTLQ 0.227 0.000*** 0.005 0.897
POPGROWTH 0.247 0.319 0.369 0.040**
INCGROWTH 0.232 0.337 0.026 0.890
COMMUTE 0.457 0.040** 1.061 0.000***
SECREGOFFICE 0.023 0.493 0.033 0.235
LNOFFICE 0.417 0.000*** 0.177 0.000***
INDEXPERT 0.115 0.000*** 0.112 0.000***
INFLUENCE 1.646 0.000*** 1.715 0.000***
TENURE 0.043 0.010** 0.016 0.365
LNASSETS 0.388 0.000*** 0.480 0.000***
LOSS 0.095 0.000*** 0.048 0.001***
ROA 0.016 0.642 0.105 0.036**
LEVERAGE 0.000 0.936 0.002 0.089*
TOBINQ 0.013 0.043** 0.017 0.013**
SALE_GTH 0.030 0.020** 0.084 0.000***
OCF 0.042 0.335 0.169 0.004***
STD_OCF 0.024 0.793 0.126 0.246
ACCUMDEF 0.116 0.000*** 0.125 0.000***
LIFECYCLE 2.036 0.000*** 2.446 0.000***
LAGABSACCRUALS 0.000 0.960 0.033 0.007***
ISSUE 0.008 0.623 0.042 0.000***
IPO 0.022 0.674 0.132 0.158
LAG_GC 0.075 0.003*** 0.172 0.000***
Year and industry fixed effects Yes Yes
N 6,656 16,989
R2 0.829 0.863
F-stat. p-value
EDUCATION 6.48 0.011**

ACCTLQ 11.95 0.000***
Notes: The dependent variable is the natural log of audit fees. All variables are as defined in the
text and in the Appendix. Year and industry (2-digit SIC) fixed effects are included but not reported.
Two-tailed p-values are presented beside the coefficient estimates and are calculated using robust
standard errors clustered at the firm level (1,385 and 2,846 clusters for non-Big 4 and Big 4 clients,
respectively). *, **, and *** denote two-tailed statistical significance at the 0.10, 0.05, and 0.01
levels, respectively.
better. This further demonstrates that auditor choice in the public company audit market
is influenced by broad factors at the city level, including the two human capital variables
in our study. In untabulated analyses, we also compared the area under the ROC curve,

with and without the city-level human capital variables. The area under the ROC curve
was larger when city-level variables were included in the model (p < 0.01).
Audit fee regressions

Table 8 presents the results for the audit fee regression used to test Hypothesis 4. The
coefficient for EDUCATION is only positive and statistically significant for non-Big 4
auditors (p < 0.01). In addition, we find that coefficient for ACCTLQ is negative rather
than positive, and is statistically significant for non-Big 4 auditors (p < 0.01), but it is not
significant for Big 4 auditors. Both Chow tests in panel B reject the null that the coeffi-
cients are equal between the two subsamples.
The positive and significant coefficient on EDUCATION indicates higher fees for non-
Big 4 auditors with greater levels of EDUCATION, and is consistent with the evidence in
Tables 4, 5, and 6 that human capital increases the quality of non-Big 4 audits. Going
from Tampa, FL to Chicago, IL, audit fees increase by 6 percent, all else equal, due to
the effect of EDUCATION. In other words, as audit quality improves in high human capi-
tal cities, non-Big 4 auditors can charge higher fees.
However, Table 8 indicates that audit fees decrease for non-Big 4 auditors as
ACCTLQ increases, all else equal, decreasing by 7.2 percent when going from Tampa, FL
to Chicago, IL. Thus, the net effect on audit fees is effectively a wash for cities like Tampa
and Chicago with similar percentile values for EDUCATION and ACCTLQ. However, for
cities with differing percentile values, audit fees could be either higher because of the effect
of EDUCATION, or lower due to the supply of accountants.
The evidence in Table 7 indicates that in cities with larger labor pools companies are
more likely to select a non-Big 4 auditor. At the same time, there is no consistent evidence
that labor pool size affects audit quality (Tables 4–6). Together, these results suggest that
larger accounting labor pools may lower costs, and in this way help non-Big 4 auditors to
better compete with Big 4 firms in the public company audit market. However, the evi-
dence in Tables 4–6 is also important because it collectively suggests that non-Big 4 offices
are also better able to compete on quality in cities with higher human capital.
5. Additional analysis
Analyzing EDUCATION and ACCTLQ separately
Table 2 reveals that the correlation between EDUCATION and ACCTLQ is 0.50. To
determine if multicollinearity between these two variables affects inferences, we performed
the following procedures. First (as noted previously), we calculated VIFs, none of which
exceed 4. Second, we reestimated our main models including EDUCATION and ACCTLQ
individually, rather than in the same model.
When we estimate our models with only EDUCATION included, the coefficient for
EDUCATION remains similar to those presented, with the exception that when LNAU-
DITFEE is the dependent variable, p = 0.13 (compared to p < 0.01 in Table 8 when
ACCTLQ is in the model). When we estimate the models only including ACCTLQ, the
coefficients for ACCTLQ when GCERROR and NONBIG4 are the dependent variables
are both statistically significant at the p < 0.01 level, compared to p = 0.08 and p = 0.058
in our main analysis. When LNAUDITFEE is the dependent variable, ACCTLQ remains
negative and statistically significant at the p < 0.05 level (compared to p < 0.01 in
Table 8). For the accruals regressions, ACCTLQ is never statistically significant at the
p < 0.10 level, irrespective of whether we control for EDUCATION. Overall, the stability
of the coefficients in this additional analysis provides comfort that multicollinearity does
not materially affect the inferences drawn from our main analysis.
Finally, we analyzed a separate model after changing the definition of ACCTLQ to be
scaled by the number of business establishments in the CBSA (ACCT_PER_ESTAB),

instead of total employment.28 EDUCATION is less correlated with ACCT_PER_ESTAB

(0.15) than with ACCTLQ (0.50). All of our main results were qualitatively the same as
reported in the tables with two exceptions. First, in Table 6, the p-value for ACCT_PER_-
ESTAB is 0.112 for Big 4 auditors, whereas ACCTLQ was significant at the 10 percent
level. Second, in Table 8, ACCT_PER_ESTAB is negative and significant at the p < 0.05
level for Big 4 auditors, whereas ACCTLQ is not statistically significant. The Chow test
comparing non-Big 4 to Big 4 remains statistically significant at p < 0.01 in Table 8.
Other auditor variables

In Tables 7 and 8, LNOFFICE and INDEXPERT have the expected signs and are statisti-
cally different from zero. However, in Table 4, both variables are insignificant in all regres-
sions, and in Table 6, both have unexpected positive signs and are statistically significant.
Since the results in Tables 4 and 6 are inconsistent, we investigated further. With regards to
LNOFFICE, if we drop all city-level control variables from the regression, LNOFFICE
becomes consistently negative and statistically different from zero (p-value < 0.10) for mod-
els testing ABS_DD, ABS_WCACCR, and GCERROR, which is comparable to the results
in Francis and Yu (2009) and Choi et al. (2010). When measuring INDEXPERT, we did
not require a minimum number of observations per industry-year-city combination to deter-
mine auditor industry expertise so that we could maximize the number of cities included in
our final sample. If we require at least three observations per city to calculate INDEXPERT,
similar to prior studies such as Minutti-Meza (2013), it is no longer statistically significant
in any of the regressions, a result consistent with Francis and Yu (2009).
Alternative definition of EDUCATION

As an additional robustness test, we reestimated all models with EDUCATION redefined as
the percentage of the CBSA’s population with a master’s or a professional degree as an alter-
native measure of average skill level, instead of a bachelor’s degree. It is possible the kind of
knowledge spillovers that might affect auditors are more likely to come about through inter-
actions with individuals having advanced degrees. We find similar results in this analysis.
Dropping small cities and large cities

Some of the cities in the final sample do not have both Big 4 and non-Big 4 auditor
offices. To investigate whether the results are sensitive to the inclusion of these cities, we
drop all observations audited by offices in cities where there is not at least one Big 4 and
non-Big 4 office. We also dropped cities with fewer than 10 observations, and the five lar-
gest cities. We find similar results in these subsamples.
Signed abnormal accruals

In our main analysis of earnings quality, we focused on the absolute value of accruals,
because firms can have incentives to manage earnings either up or down. However, prior
research suggests that auditors are more likely to resist upwards earnings management by
insisting that clients adjust their earnings down (DeFond and Jiambalvo 1993; Nelson,
Elliott, and Tarpley 2002). Using signed abnormal accruals, EDUCATION is negative and
significant for both non-Big 4 and Big 4, but the Chow test for the difference between the
coefficients was not significant at p < 0.10.
Analysis of Grant Thornton and BDO

As discussed in section 2, we expect that larger auditing firms are less tied to local labor
conditions than smaller firms. In our main empirical analysis, we focused on the
28. Data on the number of business establishments was obtained from the Department of Labor’s website.

commonly used Big 4/non-Big 4 dichotomy. Although Grant Thornton and BDO are sig-
nificantly smaller than the Big 4, they have a substantial national presence, and therefore,
may be less tied to local labor markets than smaller, regional firms. Ultimately, it is an
empirical question as to whether their national networks are large enough to reduce reli-
ance on local labor markets, so we reperformed our main empirical analyses by creating
subsamples based on membership in the Big 6, instead of the Big 4. These results were
similar to the tabulated results, but differences between the two groups were generally
smaller and less statistically significant.
Analysis of accountants’ wages

Finally, several papers in urban economics test for the existence of human capital external-
ities by regressing individual wages on city-level educational attainment, controlling for
individual fixed effects (e.g., Moretti 2004a). Since we do not have individual wage data
for accountants, we performed a similar analysis at the CBSA level (n = 120), using the
natural log of accountants’ median wages in the CBSA. We found a positive and statisti-
cally significant relationship (p < 0.01). This result was robust to including all of the city-
level control variables included in our main analysis.29
6. Conclusion and limitation

The urban economics literature theorizes and documents that cities with greater human cap-
ital are associated with a variety of positive economic outcomes. In this paper, we extend
this literature to show that human capital also appears to affect the supply and quality of
public company audits. A city’s human capital appears to be an important and previously
unrecognized factor affecting the production of audits and the quality of audited earnings.
Specifically, we show that a city’s level of educational attainment is associated with better
audits (higher-quality earnings and more accurate going-concern reports), with a stronger
effect for non-Big 4 offices that are more tied to local labor markets.
In addition, a public company is more likely to choose a non-Big 4 auditor when the
level of human capital in the audit office’s city is greater (both higher educational level
and larger labor pool). We also provide evidence that human capital, measured by educa-
tional attainment, is associated with an increase in audit fees for non-Big 4 auditors, but
negatively associated with audit fees when measured by the size of the accounting labor
pool in the city. Overall, the results indicate that locally tied non-Big 4 auditors benefit
more from higher city-level human capital and are better able to compete with the Big 4
firms in the listed-company audit market.
An important caveat to these results it that we do not have data on the human capital
of individual auditors, and therefore cannot empirically distinguish between two (non-
mutually exclusive) explanations for these results. The first explanation is that the associa-
tions we document are driven by human capital “externalities” (Moretti 2004b). That is,
increasing the quantity and skill level of labor improves productivity through better
matching between employees and firms and the creation of knowledge spillovers in a city.
The second explanation is that high-ability individual auditors self-select into cities with
high overall levels of human capital.30 In the urban economics literature, several research-
ers have used data at the individual level to determine if human capital externalities exist,
and they generally have concluded that it does (e.g., Fu 2007; Moretti 2004a,b,c; Rauch
1993). We do not have such data for our sample, so a limitation of our study is that we
29. We thank an anonymous reviewer for suggesting this analysis.

30. Even if we were able to obtain human capital data for individual auditors, it is not clear that there would
be any meaningful variation in measureable characteristics of auditors, since we would expect almost all
auditors performing public company audits to have a college degree, all else equal.

cannot directly test the specific channel that gives rise to the associations that we docu-
ment. However, our research question is also different in the sense that we are only con-
cerned with the consequences of a “given” level of a city’s human capital on audit
markets and audit quality, not on the underlying channel or source of human capital. We
believe that investigating the consequences of city-level human capital on the audit market
is important in its own right, makes several contributions to the research literature, and
identifies avenues for future research to explore, including the underlying mechanisms that
give rise to human capital and its effect on auditing.
Thus, we believe the study is important in analyzing how a given set of labor charac-
teristics in a city affects the public company audit market. While there is an extensive liter-
ature on the characteristics of auditors and their clients that influence earnings quality,
relatively little attention has been paid to external forces that influence the ability of audit
firms to supply public company audits. In this respect, our study is a first step in exploring
the ways in which city characteristics (and geographic location more generally) affect vari-
ous capital market participants. Future research may be able to more fully test how the
underlying mechanisms work, which remains something of a “black box” in our study and
in the larger urban economics literature (Jovanovic and Rob 1989).
There are also implications for practitioners. For Big 4 accounting firms, the evidence
suggests that it may be more difficult to conduct high-quality audits by offices located in
cities with lower levels of human capital, in which case the firm may want to rely more
heavily on its national network when staffing engagements in such locations. For non-Big
4 firms, the evidence points to an enhanced ability to compete with Big 4 auditors in cities
with higher human capital. Smaller firms could use this knowledge to better promote their
quality in such audit markets.
Finally, the effects of geography and location on the behavior of economic agents
have received considerable attention in economics and finance. In this spirit, a potentially
fruitful area for future accounting research would be to examine the effects of location
and city-specific human capital on the behavior of firms and managers, and on other capi-
tal market participants such as investors, creditors, and financial analysts.31
Appendix
Variable descriptions
Variable Description Source
City-level variables:
EDUCATION The percentage of people within a CBSA with a bachelor’s U.S. Census
degree or higher
ACCTLQ The labor quotient for accountants and auditors Bureau of
(Occupation 13-2011) for the CBSA. Calculated as the Labor
number of accountants/auditors per capita at the CBSA Statistics
level divided by the number of accountants/auditors per (Occupation
capita at the national level. Calculated on a yearly basis 13-2011)
(The Appendix is continued on the next page.)
31. To illustrate, Gunn (2015) documents a positive association between city location and the performance of
sell-side analysts. Specifically, when analysts are located in cities with high human capital, analysts’ earn-
ings forecasts are more accurate and the market reaction to revisions in their forecasts and stock recom-
mendations is larger, both of which are indicative of improved labor productivity by analysts located in
cities with higher human capital.

Appendix (continued)
POPGROWTH The percentage change in population for each CBSA from U.S. Census
2005 to 2011
INCGROWTH The percentage change in median income for each CBSA U.S. Census
from 2005 to 2011
COMMUTE The percentage of the CBSA total population aged U.S. Census
16 years and above that worked outside of the home that
commuted more than 45 minutes per day
SECREGOFFICE An indicator variable equal to one if the audit office is Kedia and
located in the same CBSA as an SEC regional office and Rajgopal
zero otherwise (2011)
LANDGRANT An indicator variable equal to one if the city contains a Moretti (2004a)
land grant university. Cities containing a land grant
university are Albany, NY; Athens, GA; Baton Rouge,
LA; Boston, MA; Champaign-Urbana, IL; Columbia,
MO; Columbia, SC; Columbus, OH; Des Moines, IA;
Fargo, ND; Fayetteville, AR; Fort Collins, CO;
Gainesville, FL; Greensboro-Winston-Salem, NC;
Hartford, CT; Honolulu, HI; Knoxville, TN; Lafayette,
IN; Lansing, MI; Lexington, KY; Lincoln, NE; Macon,
GA; Madison, WI; Minneapolis, MN; Nashville, TN;
Pine Bluff, AR; Providence, RI; Raleigh, NC; Reno, NV;
Richmond, VA; Riverside-San Bernadino, CA;
Sacramento, CA; San Francisco, CA; State College, PA;
Tallahassee, FL; Tucson, AZ; Washington, DC
Firm-level variables:
LAGABSACCRUALS The absolute value of total accruals from year t1. Total COMPUSTAT
accruals are defined as the difference between net income
and operating cash flows, scaled by total assets
ABS_DA The absolute value of abnormal accruals estimated from COMPUSTAT
the following equation:
TOTACC ¼ b0 þ b1 ð1=ASSETSÞ þ b2 ðSALES ARÞ
þ b3 PPE þ b4 ROA þ
The equation is estimated by industry and year,
requiring at least 10 observations
ABS_DD The difference between the absolute value of the residual COMPUSTAT
from a cross-sectional regression by industry and year of
working capital accruals on past, present, and future cash
flows, and the industry average of the absolute value
ABS_WCACCR Absolute value of working capital accruals, defined as the COMPUSTAT
yearly change in current assets, excluding changes in cash
and equivalents, less the change in current liabilities,
excluding changes in the current portion of long term
debt and taxes payable, scaled by total assets
ACCUMDEF An indicator variable equal to one if the firm has negative COMPUSTAT
retained earnings and zero otherwise
(The Appendix is continued on the next page.)

Appendix (continued)
GCERROR A dichotomous indicator variable coded one if the auditor Audit Analytics
makes either a Type 1 (false positive) or Type 2 (false
negative) going-concern reporting error, and zero
otherwise
INDEXPERT A dichotomous variable equal to one if the auditor is the Audit Analytics
city-industry leader, measured as total audit fees within a
particular 2-digit SIC code, and calculated using all firm-
years within Audit Analytics before data restrictions
INFLUENCE The percentage that each company’s audit fees represents Audit Analytics
of the auditor’s total office fees
IPO A dichotomous variable equal to one if the firm first COMPUSTAT
appeared in COMPUSTAT within the last three years
ISSUE An indicator variable equal to one if the firm issued debt COMPUSTAT
or equity greater than 5 percent of total assets and zero
otherwise
LEVERAGE The ratio of total liabilities to common equity COMPUSTAT
LIFECYCLE Total common equity less retained earnings, scaled by total COMPUSTAT
assets
LNASSETS The natural logarithm of total assets COMPUSTAT
LNAUDITFEE The natural log of total audit fees Audit Analytics
LNOFFICE The natural logarithm of total audit office fees, calculated Audit Analytics
for all firm-years within before data restrictions
LOSS A dichotomous variable equal to one if the company COMPUSTAT
reported negative income before extraordinary items and
zero otherwise
NONBIG4 An indicator variable equal to one if the company’s auditor Audit Analytics
is not one of the “Big 4” and zero otherwise
ROA Net income scaled by total assets COMPUSTAT
OCF Operating cash flows, scaled by total assets COMPUSTAT
SALE_GTH The yearly percentage change in sales over prior year COMPUSTAT
STD_OCF The rolling three-year standard deviation of operating cash COMPUSTAT
flows
TENURE A dichotomous variable equal to one if audit tenure is COMPUSTAT
equal to three years or less and zero otherwise
TOBINQ Tobin’s Q, calculated as the market value of equity, less COMPUSTAT
the book value of equity, plus total assets, and scaled by
lagged total assets
References
Acs, Z. J., and C. Armington. 2004. The impact of geographic differences in human capital on ser-
vice firm formation rates. Journal of Urban Economics 56 (2): 244–78.
Antle, R., and B. Nalebuff. 1991. Conservatism and auditor-client negotiations. Journal of Account-
ing Research 29 (Supplement): 31–54.
Ball, R., S. Jayaraman, and L. Shivakumar. 2012. Audited financial reporting and voluntary disclo-
sure as compliments: A test of the confirmation hypothesis. Journal of Accounting and Eco-
nomics 53 (1–2): 136–66.
Becker, G. S. 1962. Investment in human capital: A theoretical analysis. Journal of Political Economy
70 (5): 9–49.

Bills, K., Q. Swanquist, and R. Whited. 2016. Growing pains: Audit quality and office growth. Con-
temporary Accounting Research 33 (1): 288–313.
Choi, J., C. Kim, J. Kim, and Y. Zang. 2010. Audit office size, audit quality, and audit pricing.
Auditing: A Journal of Practice and Theory 29 (1): 73–97.
Chow, G. C. 1960. Tests of equality between sets of coefficients in two linear regressions. Economet-
rica 28 (3): 591–605.
Christensen, B. E., S. M. Glover, T. C. Omer, and M. K. Shelly. 2016. Understanding audit quality:
Insights from audit partners and investors. Contemporary Accounting Research 33 (4): 1648–84.
Craswell, A. T., D. J. Stokes, and J. Laughton. 2002. Auditor independence and fee dependence.
Journal of Accounting and Economics 33 (2): 253–75.
DeAngelo, H., L. DeAngelo, and R. Stulz. 2006. Dividend policy and the earned/contributed capital
mix: A test of the life-cycle theory. Journal of Financial Economics 81 (2): 227–54.
Dechow, P., and I. Dichev. 2002. The quality of accruals and earnings: The role of accrual estima-
tion errors. The Accounting Review 77 (Supplement): 35–59.
Dechow, P., W. Ge, C. Larson, and R. Sloan. 2011. Predicting material accounting misstatements.
Contemporary Accounting Research 28 (1): 17–82.
Dechow, P. M., R. G. Sloan, and A. P. Sweeney. 1995. Detecting earnings management. The
Accounting Review 70 (2): 193–225.
DeFond, M., D. Erkens, and J. Zhang. 2016. Does PSM really eliminate the Big N audit quality
effect? Management Science (forthcoming).
DeFond, M., and J. Jiambalvo. 1993. Factors related to auditor-client disagreements over income-
increasing accounting methods. Contemporary Accounting Research 9 (2): 415–31.
DeFond, M., K. Raghunandan, and K. Subramanyam. 2002. Do non–audit service fees impair audi-
tor independence? Evidence from going concern audit opinions. Journal of Accounting Research
40 (4): 1247–74.
DeFond, M., and J. Zhang. 2014. A review of archival audit research. Journal of Accounting and
Economics 59 (2–3): 275–326.
Ettredge, M., M. Sherwood, and L. Sun. 2014. Metro area client bases, metro area auditor bases,
and client auditor choice. Working paper, University of Kansas.
Ferguson, A., J. R. Francis, and D. J. Stokes. 2003. The effect of firm-wide and office-level industry
expertise on audit pricing. The Accounting Review 78 (2): 429–48.
Fiolleau, K., K. Hoang, K. Jamal, and S. Sunder. 2013. How do regulatory reforms to enhance
auditor independence work in practice? Contemporary Accounting Research 30 (3): 864–90.
Francis, J. R. 2011. A framework for understanding and researching audit quality. Auditing: A Jour-
nal of Practice & Theory 30 (2): 125–52.
Francis, J. R., P. N. Michas, and M. D. Yu. 2013. Office size of Big 4 auditors and client restate-
ments. Contemporary Accounting Research 30 (4): 1626–61.
Francis, J. R., D. J. Stokes, and D. J. Anderson. 1999. City markets as a unit of analysis in audit
research and the re-examination of Big 6 market shares. Abacus 35 (2): 185–206.
Francis, J. R., and M. D. Yu. 2009. Big 4 office size and audit quality. The Accounting Review 84
(5): 1521–52.
Fu, S. 2007. Smart Cafe Cities: Testing human capital externalities in the Boston metropolitan area.
Journal of Urban Economics 61 (1): 86–111.
Glaeser, E. L. 1999. Learning in cities. Journal of Urban Economics 46 (2): 254–77.
Glaeser, E. L., and D. C. Mare. 2001. Cities and skills. Journal of Labor Economics 19 (2): 316–42.
Glaeser, E. L., and A. Saiz. 2004. The rise of the skilled city. Brookings-Wharton Papers on Urban
Affairs 5: 47–105.
Glaeser, E. L., J. A. Scheinkman, and A. Shleifer. 1995. Economic growth in a cross-section of
cities. Journal of Monetary Economics 36 (1): 117–43.
Gunn, J. L. 2015. City location and sell-side analyst research. Working paper, University of Pitts-
burgh.

Hay, D. C., W. R. Knechel, and N. Wong. 2006. Audit fees: A meta-analysis of the effect of supply
and demand attributes. Contemporary Accounting Research 23 (1): 141–91.
Henderson, J. V. 2003. Marshall’s scale economies. Journal of Urban Economics 53 (1): 1–28.
Hox, J. J. 2002. Multilevel analysis: Techniques and applications. Mahwah, NJ: Lawrence Erlbaum.
Johnson, E., I. K. Khurana, and J. K. Reynolds. 2002. Audit-firm tenure and the quality of financial
reports. Contemporary Accounting Research 19 (4): 637–60.
Jones, J. 1991. Earnings management during import relief investigations. Journal of Accounting
Research 29 (2): 193–228.
Jovanovic, B., and R. Rob. 1989. The growth and diffusion of knowledge. Review of Economic Stud-
ies 56 (4): 569–82.
Kedia, S., and S. Rajgopal. 2011. Do the SEC’s enforcement preferences affect corporate miscon-
duct? Journal of Accounting and Economics 51 (3): 259–78.
Kennedy, P. 2003. A guide to econometrics, 6th edn. Malden, MA: MIT Press; WileyBlackwell.
Kothari, S. P., A. J. Leone, and C. E. Wasley. 2005. Performance matched discretionary accrual
measures. Journal of Accounting and Economics 39 (1): 163–97.
Lawrence, A., M. Minutti-Meza, and P. Zhang. 2011. Can Big 4 versus non-Big 4 differences in
audit-quality proxies be attributed to client characteristics? The Accounting Review 86 (1): 259–
86.
Lawrence, A., M. Minutti-Meza, and P. Zhang. 2016. The importance of client size in the
estimation of the Big 4 effect: A comment on DeFond, Erkens, and Zhang. Management Science
(forthcoming).
Lennox, C. 1999. Are large auditors more accurate than small auditors? Accounting and Business
Research 29 (3): 217–27.
Lopez, D., and G. Peters. 2012. The effect of workload compression on audit quality. Auditing: A
Journal of Practice and Theory 31 (4): 139–65.
Lucas, R. E. 1988. On the mechanics of economic development. Journal of Monetary Economics 22
(1): 3–42.
Marshall, A. 1920. Principles of economics. London: Macmillan and Co., Ltd Library of Economics
and Liberty [Online] available from http://www.econlib.org/library/Marshall/marP1.html; Inter-
net.
Mas, A., and E. Moretti. 2009. Peers at work. American Economic Review 99 (1): 112–45.
McCracken, S., S. Salterio, and M. Gibbins. 2008. Auditor-client management relationships and
roles in negotiating financial reporting. Accounting, Organizations and Society 33 (4): 362–83.
Minutti-Meza, M. 2013. Does auditor industry specialization improve audit quality? Journal of
Accounting Research 51 (4): 779–817.
Moomaw, R. L. 1981. Productivity and city size: A critique of the evidence. The Quarterly Journal
of Economics 96 (4): 675–88.
Moretti, E. 2004a. Estimating the social return to higher education: Evidence from longitudinal and
repeated cross-sectional data. Journal of Econometrics 121 (1–2): 175–212.
Moretti, E. 2004b. Human capital externalities in cities. In Handbook of regional and urban eco-
nomics, vol. 4, eds. J. V. Henderson and J. F. Thisse, 2243–91. North-Holland, Amsterdam:
Elsevier.
Moretti, E. 2004c. Workers’ education, spillovers, and productivity: Evidence of from plant-level
production functions. The American Economic Review 94 (3): 656–90.
Nelson, M. W., J. Elliott, and R. Tarpley. 2002. Evidence from auditors about managers’ and audi-
tors’ earnings management decision. The Accounting Review 77 (Supplement): 175–202.
Numan, W., and M. Willekens. 2012. An empirical test of spatial competition in the audit market.
Journal of Accounting and Economics 53 (1–2): 450–65.
Pearson, T., and G. Trompeter. 1994. Competition in the market for audit services: The effect of
supplier concentration on audit fees. Contemporary Accounting Research 11 (1): 115–35.

Public Company Accounting Oversight Board (PCAOB). System of quality control for a CPA firm’s
accounting and auditing practice. QC Section 20.12.
Rauch, J. E. 1993. Productivity gains from geographic concentration of human capital: Evidence
from the cities. Journal of Urban Economics 34 (3): 380–400.
Reichelt, K. J., and D. Wang. 2010. National and office-specific measures of auditor industry exper-
tise and effects on audit quality. Journal of Accounting Research 48 (3): 647–86.
Reynolds, K., and J. Francis. 2000. Does size matter? The influence of large clients on office-level
auditor reporting decisions. Journal of Accounting and Economics 30 (3): 375–400.
Rosenthal, S. S., and W. C. Strange. 2004. Evidence on the nature and sources of agglomeration
economies. In Handbook of regional and urban economics, vol. 4, eds. J. V. Henderson and J. F.
Thisse, 2119–71. North-Holland, Amsterdam: Elsevier.
Rosenthal, S. S., and W. C. Strange. 2008. Agglomeration and hours worked. The Review of Eco-
nomics and Statistics 90 (1): 105–18.
Shapiro, J. 2006. Smart cities: Quality of life, productivity, and the growth effects of human capital.
The Review of Economics and Statistics 88 (2): 324–35.
Shipman, J. E., Q. T. Swanquist, and R. L. Whited. 2017. Propensity score matching in accounting
research. The Accounting Review 92 (1): 213–44.
Starr, R. 2011. PwC: Investment in people and infrastructure strengthens growth in Africa. Big4.-
com, August 13, 2011, accessed at http://www.big4.com/pricewaterhousecoopers/pwc-investme
nt-in-people-and-infrastructure-strengthens-growth-in-africa/.
Wallman, S. 1996. The future of accounting, Part III: Reliability and auditor independence. Account-
ing Horizons 10 (4): 76–97.

Public Company Audits and City-Specific Labor Characteristics

Uploaded by

Copyright:

Available Formats

You might also like

Public Company Audits and City-Specific Labor Characteristics

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Public Company Audits and City-Specific Labor Characteristics

Uploaded by

Copyright:

Available Formats

Public Company Audits and City-Speciﬁc

MATTHEW J. BECK, Michigan State University

JERE R. FRANCIS, University of Missouri–Columbia and University of

JOSHUA L. GUNN, University of Pittsburgh

Audits des societes cotees et caracteristiques de la main-d’œuvre

CAR Vol. 35 No. 1 (Spring 2018)

CAR Vol. 35 No. 1 (Spring 2018)

2. Background and hypothesis development

Human capital and cities

CAR Vol. 35 No. 1 (Spring 2018)

CAR Vol. 35 No. 1 (Spring 2018)

Human capital and auditing

CAR Vol. 35 No. 1 (Spring 2018)

Hypothesis 1. The quality of audited earnings improves (smaller abnormal accruals,

Hypothesis 2. The accuracy of auditor going-concern opinions increases as the level of

CAR Vol. 35 No. 1 (Spring 2018)

CAR Vol. 35 No. 1 (Spring 2018)

CAR Vol. 35 No. 1 (Spring 2018)

CAR Vol. 35 No. 1 (Spring 2018)

CAR Vol. 35 No. 1 (Spring 2018)

NONBIG4 is an indicator variable equal to one when a company is audited by a non-Big

CAR Vol. 35 No. 1 (Spring 2018)

CBSA listing and descriptive statistics

Panel A: City-level variable summary statistics

EDUCATION ACCTLQ POPGROWTH INCGROWTH COMMUTE SECREGOFFICE

Mean 0.289 0.896 0.094 0.090 0.117 0.050

CAR Vol. 35 No. 1 (Spring 2018)

N POP- INC- SECREG-

Boulder, CO 12 0.566 0.867 0.101 0.194 0.105 0

Panel B: CBSA listing

N POP- INC- SECREG-

San Diego-Carlsbad-San Marcos, CA 457 0.340 0.980 0.112 0.056 0.129 0

Los Angeles-Long Beach-Glendale, CA 1,779 0.300 0.984 0.019 0.093 0.197 1

CAR Vol. 35 No. 1 (Spring 2018)

Panel B: CBSA listing

N POP- INC- SECREG-

Columbia, SC 17 0.299 1.272 0.190 0.051 0.108 0

CAR Vol. 35 No. 1 (Spring 2018)

Sioux City, IA-NE-SD 2 0.282 0.970 0.156 0.196 0.041 0

Panel B: CBSA listing

N POP- INC- SECREG-

Virginia Beach-Norfolk-Newport News, VA-NC 9 0.271 0.934 0.058 0.123 0.110 0

Davenport-Moline-Rock Island, IA-IL 17 0.242 0.966 0.034 0.066 0.061 0

CAR Vol. 35 No. 1 (Spring 2018)

Panel B: CBSA listing

N POP- INC- SECREG-

Springﬁeld, MO 2 0.242 0.643 0.145 0.104 0.086 0

CAR Vol. 35 No. 1 (Spring 2018)

Toledo, OH 78 0.221 0.663 0.020 0.017 0.066 0

CAR Vol. 35 No. 1 (Spring 2018)

Variable N Mean Median SD 25th percentile 75th percentile

EDUCATION 24,944 0.332 0.328 0.057 0.285 0.371

Alternative specifications to equation (1)

CAR Vol. 35 No. 1 (Spring 2018)

(1) EDUCATION 1.00

CAR Vol. 35 No. 1 (Spring 2018)

CAR Vol. 35 No. 1 (Spring 2018)

(20) LEVERAGE 0.01 0.01 0.10 0.00 0.03 1.00

Audits des societes cotees et caracteristiques de la main-d’œuvre

EDUCATION 2.83 0.093* 8.34 0.004* 6.73 0.009*