Journal of Accounting and Public Policy xxx (xxxx) xxxxxx

Contents lists available at ScienceDirect

Journal of Accounting and Public Policy

journal homepage: www.elsevier.com/locate/jaccpubpol

Full length article

Corporate social responsibility and geographic dispersion

Guifeng Shia, Jianfei Sunb, , Li Zhangc, Yufang Jind
a
Antai College of Economics and Management, Shanghai Jiao Tong University, China
b
Institute for Social and Economic Research, Nanjing Audit University, 86 Yushan West Road, Nanjing, Jiangsu Province 211815, China
c
School of Business and Economics, Thompson Rivers University, Canada
d
School of Management, Fudan University, China

AR TI CLE I NF O AB S T R A CT

Keywords: Using a sample of U.S. listed rms from 1994 to 2015, we examine how the geographic dispersion
Corporate social responsibility of a rm aects corporate social responsibility (CSR). We nd that corporate geographic dis-
Geographic dispersion persion is negatively associated with CSR scores. We further nd that this robust negative re-
Geographic concentration lationship is more pronounced for rms located in the small communities, which is consistent
Social interaction
with the social interaction explanation. A Heckman test shows that the results hold when we
JEL classication: control for endogeneity between geographical dispersion and CSR. Our paper provides evidence
G30 that local rms may better protect their stakeholders than geographically dispersed rms.
M41
M14

1. Introduction

In the past decade or so, the relationship between rm value and corporate diversication has become an important issue in
economics and nance. Relatively earlier research have paid attention to global diversication and industrial diversication.1
However, observing that some corporations of the U.S. maintain operations in more states than others, a few recent studies started to
stress the signicance of geographic diversication and examine how it aects rms in various ways such as stock returns (Garcia and
Norli, 2012), corporate decision-making (Landier et al., 2009) and earnings management (Shi et al., 2015). This recent focus,
however, has neglected the issue of how geographic diversication aects corporate social responsibility (CSR). This issue is crucial
because not only do rms decisions on geographic expansion inuence shareholders benets but they also impact stakeholders
benets. As a result, corporate geographic expansion may exacerbate the inherent conict between shareholders and stakeholders.
Our paper seeks to ll in this gap, by oering empirical evidence on whether geographically dispersed or concentrated rms (also
called local rms) have higher CSR scores and thus may provide more benets to stakeholders.
By computerized parsing and extracting state name counts of the U.S. listed companies from annual Form 10-K reports that are
led with the Securities and Commission,2 we dier geographically dispersed rms from geographically concentrated rms. Using
this data of corporate geographic dispersion, together with the CSR data obtained from the independent social choice investment

We thank Diego Garcia for the help on the paper. We thank conference participants at the 2016 JAPP Conference held at the IE Business School for their
comments. We also thank the referees and the editors, Martin Loeb and Lawrence Gordon, for their comments on the paper. Shi acknowledges funding from the
Humanities and Social Sciences Foundation of the Chinese Ministry of Education (16YJA630045). All errors are our own.

Corresponding author.
E-mail address: sunjianfei@nau.edu.cn (J. Sun).
1
E.g., Denis et al.(2002), Doukas and Pantzalis, (2003), and Jiraporn et al. (2006).
2
We follow Garcia and Norli (2012) to collect the data of geographic dispersion for the period of 20082015. We are very grateful to Diego Garcia for sharing with
us the data on geographic dispersion from 1994 to 2007.

http://dx.doi.org/10.1016/j.jaccpubpol.2017.09.001

0278-4254/ 2017 Elsevier Inc. All rights reserved.

Please cite this article as: Shi, G., Journal of Accounting and Public Policy (2017),
http://dx.doi.org/10.1016/j.jaccpubpol.2017.09.001
G. Shi et al. Journal of Accounting and Public Policy xxx (xxxx) xxxxxx

advisory rm Kinder, Lydenberg, Domini (hereafter KLD),3 we examine the relationship between corporate geographical dispersion
of rms and their CSR scores.
Our main results show that the geographically dispersed rms have lower CSR scores than geographically concentrated rms, or
local rms. This nding is consistent with the prediction of three dierent explanations: social interaction, agency cost, and customer
and investor recognition. However, our further analysis shows that the signicantly negative association between geographic dis-
persion and CSR scores exists only when rms headquarters are located in the small communities, which supports the social in-
teraction explanation. In addition, a Heckman's model shows that the results are robust when we control for endogeneity between
geographical dispersion and CSR.
Our paper contributes to the literature in the following two ways. First, while there exist many papers that examine the de-
terminants of CSR (McGuire et al., 2012; Herbohn et al., 2014; Monteiro and Aibar-Guzmn, 2010; Lee et al., 2017; Li et al., 2017),
our paper argues that there exists another factor, geographic dispersion, that might aect CSR. One related paper is by Landier et al.
(2009) who study how corporate geographic dispersion aect corporate decision-making including corporate attitudes to employees
and the dismissals of employees.4 Our paper, however, studies the association between geographic dispersion and CSR which has
seven categories, including community, environment, diversity, corporate governance, employee relations, human rights and pro-
duct. Focusing on employees only, while providing social responsibility information from corporate employment side, might not
reect the overall scope of CSR. Second, our paper also contributes to the literature of geographic dispersion. The literature has
documented evidence of the signicant role of geographic dispersion on stock returns (Garcia and Norli, 2012), rm value (Gao et al.,
2008), earnings management (Shi et al., 2015) and corporate management and nance policy making (Landier et al., 2009). Our
paper nds that corporate geographic dispersion also have a signicant impact on CSR, and thus provides evidence that local rms
may better protect their stakeholders than geographically dispersed rms.
The remainder of the paper is organized as follows. In Section 2, we discuss the literature and develop our hypotheses. In Section
3, we provide a description of the data sources and the sample selection procedure. In Section 4, we report the main results. In Section
5 we test which of the three dierent explanations can account for our results. Section Section 6 provides endogeneity test and other
robustness check. Concluding remarks are provided in Section Section 7.

2. Literature review and hypotheses development

The literature has extensively studied CSR and documented various factors that would have the impact on CSR. While some
papers focus on the dierences between countries, which could be further categorized as cultural dierences (Scholtens and Dam,
2007) and/or legal distinctions in terms of required versus voluntary disclosures (Ioannou and Serafeim, 2011), a lot more other
papers study what kind of rm-specic characteristic variables would inuence CSR, including rm size, corporate business or
industry, rm performance, gender diversity, corporate sustainability performance, and asset age (Monteiro and Aibar-Guzmn,
2010; Ciocirlan and Pettersson, 2012; Huang, 2013; Herbohn et al., 2014; Lee et al., 2017; Li et al., 2017). Monteiro and Aibar-
Guzmn (2010) document that rm size, corporate business, protability, foreign ownership, quotation on the stock market and
environmental certication would explain CSR disclosure. Ciocirlan and Pettersson (2012) nd that work force gender diversity
aects CSR disclosure, and Huang (2013) nds that CSR disclosure is also aected by either board gender diversity or CEO's gender.
Using the Australian data, Herbohn et al. (2014) nd that both corporate sustainability performance and asset age are strongly
associated with CSR disclosure. Recent attention on religiosity also shows that social norm such as local religiosity can also impact
CSR (McGuire et al., 2012). Using a sample of China's listed rms, Lee et al. (2017) nd that state subsidies have a material inuence
on CSR disclosure choice, and this eect is concentrated among the non-state-owned enterprises rather than the state-owned en-
terprises and especially when subsidies are granted through non-tax based rather than tax-based channels. Studying how the market
responds to expected regulatory costs related to haze in China, Li et al. (2017) also show that higher expected haze-related regulatory
costs may aect CSR disclosure.
Geographic dispersion has also been proven to be a crucial factor for a number of questions in both economics and nance.
Literature has documented that the corporate geographic dispersion aects stock returns, corporate decision-making, earnings
management, rm value and et al. (Garcia and Norli, 2012; Landier et al., 2009; Shi et al., 2015; Gao et al., 2008). Using a sample of
U.S. listed rms in the period from 1994 to 2008, Garcia and Norli (2012) nd that returns on stocks of local rms exceed those of
geographically dispersed rms by 70 basis points per month. Landier et al. (2009) nd that local rms are more employee hostile and
dismissals of divisional employees are more common in divisions located farther away from corporate headquarters. Shi et al. (2015)
nd that geographically dispersed rms have lower accrual based earnings management but higher real earnings management than
geographically concentrated rms. Gao et al. (2008) document that rms with subsidiaries located in dierent regions of the United
States experience a valuation discount of 6.2%. There is one related paper by Boeprasert (2012) who studies how corporate head-
quarter locations aect CSR. Boeprasert (2012) nds that rms whose headquarters are located further from major metropolitan
areas tend to have higher CSR scores. In other words, Boeprasert (2012) looks at corporate headquarters locations, but we focus on
corporate geographic dispersion.

3
Now owned by MSCI ESG Research, KLD is an independent rating agency specializing in assessing corporate social performance for a large sample of publicly
traded companies in the US since 1991. It collects information from a variety of sources including company lings and direct communications with the company,
governments and other organizations, as well as media, and it rates rms using a proprietary framework of positive and negative indicators.
4
Unlike Garcia and Norli (2012), Shi et al. (2015) and us, Landier et al. (2009) dene geographically dispersed rms as rms below the sample median for the
proportion of divisions in the same state as headquarters, while concentrated rms are rms above the median.

2
G. Shi et al. Journal of Accounting and Public Policy xxx (xxxx) xxxxxx

In this paper we argue that corporate geographic dispersion can aect CSR. Our review of the literature shows that three possible
arguments can help us understand the relationship between geographic dispersion and CSR. The rst argument, which is the primary
focus of this paper, is based on the social interaction argument. This argument contends that for geographically concentrated rms, or
local rms, the local investors have better access to soft information through the channels of simple familiarity and social interaction
(Hong et al., 2004; Barber and Odean, 2008; Landier et al., 2009). Accordingly, social interactions can more easily and more
frequently inuence these rms decision-making related to stakeholder relationships, and social disclosures can be more closely
associated with social environment and public-pressure faced by the rms (Patten, 1991). On the other hand, for geographically
dispersed rms, the social interaction eect would decrease sharply, since they operate the business out of the places where their
headquarters are located so the main stakeholders, including clients and employees, are out of their sight. As a result, the social
interaction argument predicts that geographically dispersed rms would tend to have lower CSR.
The second argument is the agency costs argument. It predicts that rms expansion on geographically diverse operations can
increase agency costs. It has been documented in the literature that both global dispersion and industrial diversication are asso-
ciated with a signicant rm value discount (Denis et al., 2002; Doukas and Pantzalis, 2003). Analyzing dispersed rms, Landier et al.
(2009) nd that managers of these geographically dispersed rms made decisions that may not have aligned with the companys
shareholders. Gao et al. (2008) also argue that agency cost provides a plausible explanation for why geographically dispersed rms
are worth less than their geographically concentrated counterparts. While geographic dispersion may lead to higher agency costs,
CSR may be a proxy for the rm value. This is because, through having higher CSR scores, rms can obtain a better reputation and
image, which may strengthen their competitive advantage in product markets (Porter and Kramer, 2006) and translate into lower
nancing costs (Dhaliwal et al., 2011; Ghoul et al., 2011) and thus increase rm values (Jiao, 2010). Based on this theory, one might
expect the geographic dispersion to be negatively associated with CSR.
The third argument is based on the customer and investor recognition story (Merton, 1987; Garcia and Norli, 2012). According to
this argument, local rms might prefer to have higher CSR scores since superior CSR performance would seem to be a good ad-
vertisement to gain recognition in both product and nancial markets. The recognition achieved via CSR signals the local rms
quality and helps reduce information asymmetry between the product and capital markets and the local rm. However, this is not the
case for dispersed rms, since they spread business throughout the country and they tend to be already well known. Thus, the
customer and investor recognition argument predicts that the geographic dispersion will be negatively associated with CSR.
In summary, consistent with the three arguments of social interaction, agency cost, and customer and investor recognition, our
rst hypothesis is as follows:
H1. Geographically dispersed rms tend to have lower CSR scores than local rms.
While all three explanations propose that geographic dispersion is negatively associated with CSR, which one is the driving force
is an interesting question. To address this issue, we further develop three more hypotheses as follows.
We rstly try to test the social interaction argument. Social interactions can often inuence corporate decision-making and this
interaction would be more signicant for local rms residing in small communities. The reasons are the following. Managers of
geographically concentrated rms or local rms within small communities will be more concerned about stakeholders because
stakeholders are likely to interact with them much more often. Since local rms that currently implement stakeholder-friendly
policies are likely to have more frequent interaction with their stakeholders within a small community, they tend to have strong
incentives to have higher CSR scores. In contrast, managers of geographically dispersed rms would care less about the stakeholders
from small communities because of the out of sight eect as discussed before. If the social interaction argument is the cause of the
negative association between CSR and geographic dispersion, we should expect the negative association to be more pronounced for
rms located in the small communities. Thus, our second hypothesis is as follows:
H2a. The negative impact of geographical dispersion on CSR is greater for rms in small communities than for those in big
communities.
We then attempt to study the agency costs argument. The argument suggests that, like industry diversication, geographic
expansion gives rise to agency problems, so such decisions and policies may not align with the companys shareholders (Denis et al.,
1997; Doukas and Pantzalis, 2003; Landier et al., 2009; Gao et al., 2008). As a result, geographic dispersion leads to a lower rm
value. Furthermore, as Jiao (2010) points out, a higher CSR is associated with higher shareholder welfare and rm value and lower
agency costs. If this agency costs hypothesis holds, we should expect that the negative association between CSR scores and geographic
dispersion to be more pronounced for rms with worse governance because as literature has suggested these rms should have more
severe agency problems. Our third hypothesis is then as follows:
H2b. The negative impact of geographical dispersion on CSR is greater for rms with worse governance than for rms with better
governance.
Lastly, we examine the customer and investor recognition argument. This hypothesis suggests that local rms would have higher
CSR in order to capture the attention of customers and investors and thus sell better and earn more. In contrast, geographically
dispersed rms would not need to do so as they are already better known, and doing so would not bring them enough benets. If this
customer and investor recognition hypothesis holds, we should expect that the negative association between CSR and geographic
dispersion to be more pronounced for rms with lower customer and investor recognition. Thus, we state our fourth hypothesis as
follows:

3
G. Shi et al. Journal of Accounting and Public Policy xxx (xxxx) xxxxxx

H2c. The negative impact of geographical dispersion on CSR is greater for rms with lower customer and investor recognition than
for those with higher customer and investor recognition.

3. Samples and variables

We obtained CSR data from KLD Research and Analytics, which is widely used in empirical studies (Benson and Davidson, 2010;
Jiao, 2010; Bae et al., 2011; Kim et al., 2012).5 The database provides a well-established measure of corporate social responsibility.
Firms are rated with a strength score and a concern score in seven major categories: Community, Diversity, Corporate governance,
Employee relations, Human rights, Environment, and Product. We use all seven categories to dene a CSR proxy, which is computed
as total strengths (positive ratings) minus total concerns (negative ratings) in KLDs seven social rating categories.6 It is important to
note that for each of these CSR scores, higher numbers correspond to better levels of corporate social responsibility.
Our data on geographic dispersion ranges from 1994 to 2015. We rst obtain the data between 19942007 from Garcia and Norli
(2012). Then, for the period from 2008 to 2015, we follow Garcia and Norli (2012) and get, through a computerized parsing of all 10-
K's led with the SEC during that period, a count of the number of times each 10-K mentions a U.S. state name. We also follow Garcia
and Norli (2012) in dening the degree of geographic dispersion of a rms business operations as the number of dierent operating
states mentioned in its 10-K lings. As Garcia and Norli (2012) describe, the counted number is comprised of the occurrence of
dierent state names in the 10-K sections Item 1: Business, Item 2: Properties, Item 6: Consolidated Financial Data, and Item 7:
Managements Discussion and Analysis. We eliminate rm observations that did not mention any U.S. state names in their 10-K
Forms. Therefore, geographic dispersion for rm i for scal year t is an integer in {1, 2, , 50}. See the details of variable con-
struction in the Appendix A.
Following DiGuili and Kostovetsky (2012), we controlled for a series of rm-specic variables that were likely to be associated
with CSR: size, ROA, cash, dividend, and debt. This is because rms with larger assets, higher ROAs, more cash, more dividends and
less debt are associated with higher KLD scores as documented in the literature. ROA, cash, dividend and debt are winsorized at the
top and bottom 1 percent of their distributions. We addressed possible industry and year eects through the control of the two-digit
SIC industry dummies and year dummies.7,8 All variables denitions are presented in the variables Appendix A. The nancial data are
obtained from Compustat database. Finally, we deleted the nance and utility industries and made available data on all variables
used in regression analyses. This provides us with an initial sample of 24,864 rm-year observations during the period 19942015.

4. Empirical results

4.1. Univariate analysis

Table 1 reports descriptive statistics. The average number of dierent states mentioned in corporate SEC ling was 10.02 (with a
median of 7), indicating that on average rms operated business in 10 states within the United States. The average CSR score from
KLD was 0.20, meaning the average CSR concerns are slightly larger than the average CSR strengths. Firms in the sample have
average assets of $5958 million, an average ROA of 0.02, and an average cash holding ratio of 0.20. The dividend and debt ratios
were 0.01 and 0.22, respectively.
If geographic dispersion has an impact on CSR, we should expect CSR to vary across local rms and dispersed rms. To test this,
we rst divide the sample into ve groups by the quintile of geographic dispersion and then compare how the groups dier for certain
variables. Based on the number of states mentioned on the annual reports, rms below the 20th percentile are referred to as local
rms and rms over the 80th percentile as dispersed rms. In our sample, local rms do business in four states or less and
dispersed rms in fteen or more states (not reported).
Table 2 shows the distribution of geographic dispersion and CSR across the ve quintiles. We can easily see that CSR is sig-
nicantly lower for dispersed rms than that for local rms. The average CSR score in the rst quintile of geographic dispersion (local
rms) is 0.07, while that in the fth quintile of geographic dispersion (dispersed rms) is 0.62. Moreover, consistent with our
hypothesis, the CSR score shows a clear decreasing trend from the rst quintile to the fth quintile. The results from Table 2 also
show that the dierences in both the number of states and CSR score between local rms and dispersed rms are statistically
signicant, with a t-statistic of 180 and 14.94, respectively. This suggests that dispersed rms tend to have lower CSR scores.

4.2. Multivariate regression analysis

In this section, we examine whether the geographic dispersion is negatively associated with CSR after controlling for rm specic

5
An alternative proxy, the Business Ethics 100 Best Corporate Citizens ratings also serve to measure CSR. Its ranking standards are similar to KLDs. One major
dierence is that the KLD database is more comprehensive and covers more rms.
6
CSR is usually considered as a proxy for social performance. As the manual provided by KLD Research & Analytics Inc. explains (on the rst page), KLD STATS
(STATISTICAL TOOL FOR ANALYZING TRENDS IN SOCIAL AND ENVIRONMENTAL PERFORMANCE) is a data set with annual snap-shots of the environmental,
social, and governance performance of companies rated by KLD Research & Analytics, Inc.
7
Our results do not change when we use Fama-French 48 industry dummies classication.
8
We use industry instead of rm xed eects because of insucient within-variation over time in CSR scores for each rm, which otherwise would force iden-
tication of the CSR-related coecients from only these changes. This is similar to the GIM index by Gompers et al. (2003) and Giroud and Mueller (2011).

4
G. Shi et al. Journal of Accounting and Public Policy xxx (xxxx) xxxxxx

Table 1
Descriptive statistics of selected variables.

Variable N Mean Std. Dev. Bottom 25% Median Top 25%

CSR 24,864 0.20 2.41 2.00 0.00 1.00

Geographic dispersion 24,864 10.02 8.97 4.00 7.00 12.00

Firm characteristics
Assets ($, M) 24,864 5958.00 25,924.12 376.58 1113.71 3515.15
ROA 24,864 0.02 0.15 0.01 0.05 0.09
Cash 24,864 0.20 0.21 0.04 0.11 0.28
Dividend 24,864 0.01 0.02 0.00 0.00 0.02
Debt 24,864 0.22 0.20 0.02 0.19 0.33

This table documents some characteristics of rm-level variables based on our samples that are used in at least one of our multivariate tests. ROA, cash, dividend and
debt are winsorized at 1st and 99th percentiles. See Appendix A for variable denitions.

Table 2
Univariate statistics by local rms versus dispersed rms.

Geographic Local Mean Quintile 2 Mean Quintile 3 Mean Quintile 4 Mean Dispersed Mean Dierence = Local -
dispersion (Median) (Median) (Median) (Median) (Median) Dispersed (t-value)
2.92 5.46 7.92 11.69 25.01 -22.09***
(3.00) (5.00) (8.00) (12.00) (21.00) ( 1 8 0)

CSR 0.07 0.07 0.20 0.28 0.62 0.69***

(0.00) (0.00) (0.00) (0.00) (1.00) (14.94)
Assets ($, M) 4,919.47 4,308.28 4,839.35 6,979.24 9,182.79 4,263.32***
(728.57) (753.88) (968.68) (1479.08) (2201.96) (7.93)
ROA 0.01 0.01 0.02 0.03 0.04 0.03***
(0.05) (0.05) (0.05) (0.05) (0.05) (11.95)
Cash 0.27 0.25 0.20 0.14 0.10 0.17***
(0.19) (0.18) (0.12) (0.08) (0.05) (42.99)
Dividend 0.013 0.013 0.011 0.012 0.011 0.002***
(0.000) (0.000) (0.000) (0.002) (0.003) (4.12)
Debt 0.18 0.19 0.22 0.25 0.27 0.09***
(0.13) (0.14) (0.19) (0.23) (0.25) (23.34)

This table tests the dierences of certain variables between local rms and dispersed rms. See Appendix A for variable denitions. T-statistics are in parentheses.
Statistical signicance at the 10%, 5%, and 1% level is indicated by *, **, and ***, respectively.

variables. Table 3 reports regression coecients for dierent models that capture the inuence of geographic dispersion on CSR. For
simplicity and brevity, Table 3 does not report regression coecients of the constant, industry and year dummy variables. Con-
sidering that the residuals can be correlated across rms and/or over time for all multivariate analyses, we report test statistics and
signicance levels based on the standard errors double-clustered by rm and year (Petersen, 2009). Model 1 is a univariate regression
of CSR on the geographic dispersion. Size is an important determinant of CSR since larger rms tend to have more visibility than
smaller rms. Model 2 controls for size. Furthermore, Hong et al. (2012) nd that large rms with enough resources and nancial
slack are likely to have higher CSR. Following them, we used the KZ score as a measure of nancial slack in Model 3. Firms with lower
KZ values are identied as those with more nancial slack. 9 Model 4 includes other rm-specic control variables such as ROA, Cash,
Dividend, and Debt. Prior research also suggests that bigger, better known, and more mature rms tend to engage more in socially
responsible activities, and that product market competition is an important factor for CSR (Jiao and Shi, 2012). We further mitigate
these concerns by including as control variables into Model 5 a dummy variable for SP500 rms, a variable of rm age dened as the
log of the number of months since the rms stock appears in CRSP, and a variable of product market competition dened as the
Herndahl-Hirschman index (HHI) based on the two-digit SIC industry classication.
The results of Table 3 show that the coecients of geographic dispersion in all models are negative and signicant, indicating that
geographically dispersed rms tend to have lower CSR scores. These results are consistent with Hypothesis H1. With the addition of
dierent control variables, the coecients on the variable of interest show slight changes, but they are still statistically signicant
determinants of CSR score. The statistically signicant negative coecient of the KZ score in Model 3 shows that less nancially
constrained rms spend more on CSR. The coecients of other rm specic variables in Model 4 and Model 5 indicate that rms with
more prots, rms with more cash and less debt and rms that pay more dividends, tend to have higher CSR scores.
Table 4 presents results using 5 alternative measures of geographical dispersion. In Model 6 we exclude Delaware, Washington,
and New York since many companies are incorporated or registered in Delaware, Washington is the name of the United States
capital, and the New York is related to the name of stock exchange. In Model 7, we dene a dummy variable for a dispersed rm. It is

9
Our method of computing KZ score is identical to that used by Hong et al. (2012). For brevitys purpose, we omit the detailed description of the construction of KZ
scores.

5
G. Shi et al. Journal of Accounting and Public Policy xxx (xxxx) xxxxxx

Table 3
Multivariate regression of CSR on geographic dispersion.

OLS (1) OLS (2) OLS (3) OLS (4) OLS (5)

Geographic dispersion 0.011 *

0.023 ***
0.018 ***
0.020 ***
0.020***
(1.732) (3.947) (3.243) (3.555) (3.612)
Size 0.314*** 0.303*** 0.343*** 0.188**
(2.856) (2.775) (2.927) (2.012)
KZ score 0.115***
(5.697)
ROA 0.379* 0.495**
(1.688) (2.452)
Cash 0.745*** 0.550***
(4.000) (3.164)
Dividend 7.311*** 5.678***
(4.835) (4.214)
Debt 0.530*** 0.341**
(3.313) (2.187)
SP500 0.832***
(3.465)
Firm age 0.001
(0.019)
Product market competition 2.544***
(2.718)
# of obs. 24,864 24,864 23,041 24,864 24,830
Adj R-squared 0.119 0.155 0.160 0.165 0.175

This table presents the association between geographic dispersion and CSR by multivariate regression analysis. The dependent variable is CSR. See Appendix A for
variable denitions. OLS (1) is a univariate regression while OLS (2) adds size as controls. OLS (3) includes the KZ score, and OLS (4) includes other rm specic
control variables: ROA, Cash, Dividend and Debt. OLS (5) adds SP500, rm age, and product competition as control variables. The dummies for industry and year xed
eects are included (as well as constant) in all the columns, but not reported. Standard errors are double-clustered by rm and year (Petersen, 2009) and t-values are
presented in parentheses. Statistical signicance at the 10%, 5%, and 1% level is indicated by *, **, and ***, respectively.

Table 4
CSR and alternative measures of geographic dispersion.

OLS (6) OLS (7) OLS (8) OLS (9) OLS (10)

Excluding DE, Washington, and NY 0.020***

(3.511)
Disperse dummy for top quintile 0.376***
(3.507)
Herndahl *
(1) 0.775***
(4.826)
Census division 0.055***
(3.136)
Three-year moving average 0.021***
(3.635)
Size 0.342*** 0.340*** 0.335*** 0.337*** 0.365***
(2.914) (2.908) (2.871) (2.880) (3.010)
ROA 0.385* 0.373* 0.393* 0.390* 0.351
(1.717) (1.659) (1.787) (1.747) (1.442)
Cash 0.739*** 0.772*** 0.693*** 0.743*** 0.785***
(3.971) (4.089) (3.758) (3.962) (4.153)
Dividend 7.362*** 7.482*** 7.384*** 7.458*** 7.708***
(4.851) (4.847) (4.831) (4.902) (4.852)
Debt 0.532*** 0.529*** 0.502*** 0.523*** 0.522***
(3.325) (3.298) (3.152) (3.266) (3.191)
# of obs. 24,864 24,864 24,864 24,864 23,144
Adj R-squared 0.165 0.164 0.165 0.163 0.168

This table presents the association between alternative measures of geographic dispersion and CSR by multivariate regression analysis. The dependent variable is CSR.
See Appendix A for variable denitions. In OLS (6), geographic dispersion is replaced by the number of dierent states (excluding Delaware, Washington, and New
York). In OLS (7), geographic dispersion is constructed using disperse dummy variable measured by the top quintile of geographic dispersion. In OLS (8), geographic
dispersion is replaced with the Herndahl index based on the number of dierent states. In OLS (9), geographic dispersion is replaced with the number of dierent
Census divisions using the state names mentioned in their 10-K statements. In OLS (10), geographic dispersion is constructed using a 3-year moving average of the state
counts from the 10-K statements. The dummies for industry and year xed eects are included (as well as constant) in all the columns, but not reported. Standard
errors are double-clustered by rm and year (Petersen, 2009) and t-values are presented in parentheses. Statistical signicance at the 10%, 5%, and 1% level is
indicated by *, **, and ***, respectively.

6
G. Shi et al. Journal of Accounting and Public Policy xxx (xxxx) xxxxxx

Table 5
CSR individual ratings categories.

CSR COM GOV DIV EMP ENV HUM PRO

Geographic 0.019 ***

0.003 ***
0.001 0.006**
0.007 ***
0.002 *
0.001 0.001
dispersion (3.816) (2.908) (0.606) (2.275) (3.860) (1.722) (1.512) (0.773)
# of Business Segment 0.049** 0.004 0.004 0.004 0.014* 0.019** 0.005** 0.017**
(1.975) (0.701) (0.836) (0.401) (1.684) (2.159) (2.278) (2.550)
Size 0.363*** 0.074*** 0.140*** 0.364*** 0.104*** 0.067* 0.019*** 0.087***
(3.233) (6.096) (4.293) (8.323) (3.415) (1.751) (3.269) (6.643)
ROA 0.258 0.049 0.178*** 0.324*** 0.331*** 0.054 0.002 0.066**
(1.241) (1.562) (3.750) (3.916) (3.961) (1.010) (0.075) (1.995)
Cash 0.700*** 0.073** 0.200*** 0.472*** 0.192** 0.127** 0.013 0.049
(3.779) (2.391) (3.901) (5.420) (2.372) (2.109) (0.931) (1.493)
Dividend 6.766*** 1.466*** 0.690* 2.955*** 1.650*** 1.397** 0.156 1.237***
(4.914) (3.859) (1.910) (4.493) (2.789) (2.107) (1.350) (2.795)
Debt 0.634*** 0.103*** 0.017 0.407*** 0.163*** 0.064 0.003 0.088**
(4.117) (3.702) (0.474) (4.890) (2.792) (1.324) (0.237) (2.343)
# of obs. 22,550 22,550 22,550 22,550 22,550 22,550 22,550 22,550
Adj R-squared 0.163 0.125 0.204 0.313 0.171 0.140 0.098 0.154

This table presents the associations between geographic dispersion and CSR and each individual CSR rating category. The seven individual categories of KLD ratings
are CSR, COM, GOV, DIV, EMP, ENV, HUM, and PRO, which represent community, diversity, employee relations, environment, human rights and product, respec-
tively. See Appendix A for denitions of other variables. The dummies for industry and year xed eects are included (as well as constant) but not reported, and t-
values are calculated using robust standard errors and presented in parentheses. Statistical signicance at the 10%, 5%, and 1% level is indicated by *, **, and ***,
respectively.

equal to 1 if a rm is in the top quintile of geographic dispersions and 0 otherwise. In Model 8, we calculated the Herndahl index
based on the number of dierent states. The Herndahl index is more likely to classify a rm as local when many states are mentioned
in its 10-K Form but one state (maybe the headquarters state) receives a large number of counts. To be consistent with ideas of
geographical dispersion, we time Herndahl by minus one.10 In Model 9, we used the nine U.S. census geographic divisions as the
measure of location, rather than the fty U.S. states. A rm that operates in California and Maine is arguably more geographically
dispersed than one that operates in Arizona and Montana (both part of the Mountain division). In Model 10, we simply computed a 3-
year moving average of the state counts from the 10-K Form statements. Results are reported in Table 4. We can see that our main
results still hold when we use these 5 alternative measures of geographic dispersion. All ve coecients are negative and signicant.
We also used the Fama-MacBeth method to address the time-series clustering of dependent variables stemming from the limited
changes in a rms CSR scores over time. Fama-MacBeth regressions, along with Newey-West corrections, impose a structure on the
xed eects. Our analysis (untabulated) shows the robustness of the results using this alternative estimation method. We further
examined the relationship between the change in geographic dispersion and the change in CSR. The coecient is insignicant
(untabulated). This is mainly due to the fact that there are not enough variations in these two variables in our sample.

4.3. Individual CSR ratings categories

Although the main results show a negative association between geographic dispersion and the total CSR scores, it is natural to ask
what results would be between geographic dispersion and the seven individual CSR ratings categories. Previous studies (Jiao, 2010;
Kim et al., 2012) examine both aggregated and disaggregated subscores from KLD data as a proxy for CSR. Following the literature,
we replace aggregated CSR net scores with individual KLD ratings categories, that is, qualitative issue areas dened by KLD as
community (COM), governance (GOV), diversity (DIV), employee relations (EMP), environment (ENV), human rights (HUM) and
product (PRO). Each category score is obtained by subtracting its total concerns from its total strengths.
Then we run regressions of CSR and each category score on geographic dispersion. Table 5 reports the results. Note that we
include the number of business segments to control for the impact of segments dispersion.11 The patterns do not change for CSR. For
individual CSR ratings categories, we nd that for four (COM, DIV, EMP, and ENV) out of seven subscores, corporate geographic
dispersion is negatively and signicantly associated with the individual CSR rating category score. Note that the coecients of
geographic dispersion for COM and EMP regressions are all signicant at 1 percent level, those for DIV and ENV are signicant at 5
and 10 percent level, respectively.
Note that in the modern world, technology development would make corporate geographic expansion or dispersion easier than
before, which may bring more sales, inuences and et al. along the way. However, as every coin has two sides, rms being local can
also provide benets. It has been shown in the literature that local rms have higher stock returns (Garcia and Norli, 2012) and
higher rm values (Gao et al., 2008). Our paper shows that local rms have higher CSR, which is also consistent with Landier et al.
(2009).

10
Otherwise, large Herndahl index would mean smaller geographical dispersion, or local rms.
11
As literature (Denis et al., 2002; Schmid and Walter, 2012) pointed out, segment data measure global diversication. Our measure instead captures geographic
dispersion within the country (USA). Our main results do not change when we control geographic segment from Compustat database in the regressions.

7
G. Shi et al. Journal of Accounting and Public Policy xxx (xxxx) xxxxxx

Table 6
The eect of social interactions.

Panel A: Small and large communities

OLS (1) OLS (2) OLS (3)

Full Sample Small Community Large Community

Geographic dispersion 0.020*** 0.030*** 0.010

(3.559) (4.080) (1.313)
Size 0.341*** 0.288** 0.406***
(2.911) (2.378) (3.323)
ROA 0.381* 0.477* 0.316
(1.694) (1.744) (1.170)
Cash 0.742*** 0.648*** 0.830***
(3.991) (2.727) (3.422)
Dividend 7.265*** 6.180*** 8.141***
(4.866) (3.315) (3.824)
Debt 0.535*** 0.664*** 0.332
(3.330) (3.290) (1.525)
# of obs. 24784 12486 12298
Adj R-squared 0.164 0.155 0.193

Panel B: Rural and urban rms based on headquarter locations

OLS (4) OLS (5) OLS (6) OLS (7)

Non populous 20 MSAs Populous 20 MSAs < 3 million population in MSA > 3 million population MSA

Geographic dispersion 0.029*** 0.014** 0.019*** 0.009

(3.679) (2.073) (2.620) (1.376)
Size 0.413*** 0.317*** 0.163 0.094
(3.293) (2.606) (1.634) (0.955)
ROA 0.278 0.351 0.662* 0.629***
(0.818) (1.572) (1.885) (2.718)
Cash 0.797*** 0.672*** 0.714** 0.462**
(2.881) (3.002) (2.529) (2.027)
Dividend 8.010*** 7.989*** 10.167*** 3.588*
(3.691) (3.803) (4.380) (1.864)
Debt 0.775*** 0.446** 0.540* 0.447**
(2.761) (2.454) (1.692) (2.121)
# of obs. 9806 14,267 7775 9533
Adj R-squared 0.183 0.174 0.145 0.127

Panel C: High and low informal socializing score

OLS (8) OLS (9)

High socializing score Low socializing score

Geographic dispersion 0.022*** 0.014*

(2.885) (1.851)
Size 0.310** 0.499***
(2.554) (3.842)
ROA 0.767** 0.219
(2.368) (0.844)
Cash 0.679** 1.036***
(2.161) (4.341)
Dividend 8.967*** 7.218**
(4.022) (2.102)
Debt 0.547* 0.526**
(1.911) (2.432)
# of obs. 8918 9755
Adj R-squared 0.168 0.229

This table presents the eect of social interactions on the association between the geographic dispersion and CSR. See Appendix A for variable denitions. In Panel A, we divide
the whole samples into two groups based on the county population. We use the median of the population of the counties in which rms headquarters are located in order to
divide the sample into two subsamples, low and high population, representing small and large communities. Model 1 is based on the full sample. Model 2 and Model 3 present
the results of rms headquarters located in small- and large- community counties, respectively. In Panel B, we divide the sample into two groups based on corporate
headquarter locations. In Model 4 and 5, rms are dened as rural rms if their headquarters are located in the non populous 20 MSA, and as urban rms otherwise. In Model
6 and 7, rms are dened as rural rms if their headquarters are located in the MSA of less-than-three-million population, and as urban rms otherwise. In Panel C, we divide
the whole sample into two groups based on the informal socializing score from the Social Capital Benchmark Survey. Model 8 and 9 present the results of high and low
informal socializing score, respectively, based on the median value of states. The rm-specic controls variables (Size, ROA, Cash, Dividend and Debt) are included. The
dummies for industry and year xed eects are included (as well as constant) in all the columns, but not reported. Standard errors are double-clustered by rm and year
(Petersen, 2009) and t-values are presented in parentheses. Statistical signicance at the 10%, 5%, and 1% level is indicated by *, **, and ***, respectively.

8
G. Shi et al. Journal of Accounting and Public Policy xxx (xxxx) xxxxxx

5. Tests of three dierent explanations

In the previous section, we show that geographic dispersion is negatively associated with CSR scores. This nding is consistent
with the prediction of three alternative explanations: social interactions, agency costs, and customer and investor recognitions. In this
section, we study which of the three is the driving force of our results. Specically, we examine the validity of Hypotheses H2a, H2b,
and H2c in explaining the negative linkage between geographic dispersion and CSR scores.

5.1. Social interaction argument

We rst test the social interaction argument. It predicts that the impact of geographic dispersion on CSR within a small com-
munity will be larger than that within a big community. To capture this argument of social interactions and visibility in the com-
munity, we use the median of the population of the county in which rms headquarter is located in order to divide the sample into
two subsamples, low and high population, representing small and large communities. Then we reexamine the relationship between
geographic dispersion and CSR on these two subsamples. Population gures are taken from those reported by the US Census Bureau
for the years 1990, 2000 and 2010 (interleaving years are extrapolated using the annual growth rate between these years). If the
social interaction argument drives the results, then we should expect to see that our main results of Table 3 are more pronounced for
the rms that are located in small-community counties, i.e., counties with low population. Panel A of Table 6 presents such results. As
can be seen, the signicant association between geographic dispersion and CSR scores exists only when rms headquarters are
located in small-community counties. The coecient is -0.030, signicant at 1 percent level. In other words, we fail to reject
Hypothesis H2a that social interaction argument explain the negative association between the geographic dispersion and CSR.
Boeprasert (2012) studies the relationship between corporate headquarter locations and CSR and nds that rms whose head-
quarters are located further from major metropolitan areas have higher CSR scores. His ndings, while related to what this paper
studies, does not capture geographic dispersion, because the fact that a rm's headquarter is located further from major metropolitan
areas does not necessarily mean that this rm is geographically concentrated, or vice versa. His method, however, is similar to what
we do Panel A of Table 6, where we use headquarter county's population to dene small and large communities. For the purpose of
robustness, similar to Boeprasert (2012) we divide the sample into two subsamples, rural and urban rms, based on headquarter
locations and then repeat what we do for Panel A of Table 6. Results are reported in Panel B of Table 6. As we can see, the results are
more pronounced for the rms that are located in rural areas than urban areas, which is similar to Panel A of Table 6.12
Besides the above method, another way to test Hypothesis H2a is to look at informal socializing measures. For this we use data
from the Social Capital Benchmark Survey.13 There are many dierent facets of social capital that emerge from the Social Capital
Community Benchmark Survey: social trust, inter-racial trust, conventional politics, protest politics, civic leadership, associational
involvement, informal socializing, diversity of friendships, giving and volunteering and equality. We use the informal socializing
score to measure the social interaction at the state level in order to capture the interaction relation between managers of rms and
their stakeholders in the neighborhood. The informal socializing dimension measures the degree to which residents had friends over
to their home, hung out with friends in a public place, socialized with co-workers outside of work, played cards or board games with
others, and visited with relatives. Since informal socializing enhances the social interaction between managers and local stakeholders,
we should expect our main results in Table 3 to be more pronounced for the rms that are located in states with a higher informal
socializing score. As we can see, Panel C in Table 6 shows that main results only hold as expected using informal socializing scores.14
The coecient is -0.022, signicant at 1 percent level.

5.2. Agency cost argument

To test Hypothesis H2b, we rst use the median of corporate Gompers et al. (2003) Governance Index (GIM) scores to divide the
sample into two subsamples, low and high GIM, representing rms with better and worse governance. Then we reexamine the
relationship between geographic dispersion and CSR on these two subsamples. The Chow test suggests that there is no statistically
signicant dierence (untabulated) between two coecients of geographic dispersion of these two subsamples. Thus we reject
Hypothesis H2b that agency costs drive our main results.
We also include a few more proxies for agency costs into the main regressions of Table 3. These variables are institutional
ownership, the number of block holders, and insider ownership measured by percent of common stock held by all the ocers and
directors of the company.15 This, however, would not generate a coecient on geographic dispersions signicantly dierent from
before (results untabulated).

12
We also regress each individual CSR ratings categories on this Section 5.1, the patterns still hold. Same for subsequent s 5.2,5.3 and 5.4. The results are not
reported.
13
The data is obtained by the Saguaro Seminar at Harvards Kennedy School of Government and is disseminated by the Roper Center for Public Opinion Research at
the University of Connecticut. The survey posed hundreds of social-capital questions to a nationwide sample of over 30,000 respondents. The dataset of informal
socializing scores is available only by state. The detailed information can be found at http://www.hks.harvard.edu/programs/saguaro/measurement.
14
Note that the survey does not cover all states, and thus we lose some observations.
15
A block is dened as 5% ownership or more.

9
G. Shi et al. Journal of Accounting and Public Policy xxx (xxxx) xxxxxx

5.3. Customer and investor recognition argument

To test Hypothesis H2c, we rst follow Lehavy and Sloan (2008) to use breadth of ownership to capture the recognition of a stock
among informed investors and then the median of this variable to divide the sample into two subsamples, low and high breadth,
representing rms with lower and higher recognition. Then we reexamine the relationship between geographic dispersion and CSR
on these two subsamples. Contradictory to Hypothesis H2c, we nd that the negative impact of geographical dispersion on CSR is
greater for rms with higher recognition than for those with lower recognition (untabulated). Thus we reject Hypothesis H2c that the
customer and investor recognition drives the results
We also adopt a couple of more rm-level proxies for the customer and investor recognition into the main regressions of Table 3.
These variables are advertisement intensity and analyst coverage. Firm-level advertisement is regarded as a proxy for customer
awareness about the rm that also increases investor awareness of a rm (Grullon et al., 2004). Following the literature, we com-
puted advertising intensity as advertising expenditures divided by sales, setting it equal to zero when advertising expenditures are
missing. The number of analyst followings is obtained from Thompson Institutional Brokers Earnings Services (I/B/E/S). We include
these two variables as the additional independent variables in the regressions. The result (untabulated) is not dierent from the result
for Hypothesis H1.
Finally, we follow Garcia and Norli (2012) to use the countys rm density as a measure of investor recognition. This county-level
proxy is computed as the ratio of the number of listed rms to the population of the county where rm is located. The results remain
unchanged (not reported).

5.4. Other possible explanations

Boeprasert (2012) nds that rms locating further from major metropolitan areas have higher CSR scores. In contrast, our paper
nds that geographically concentrated rms (or local rms) have higher CSR scores. These two papers study how corporate geo-
graphical features may aect CSR from two dierent perspectives, but the results are complementary. Boeprasert (2012) further
argues that when rms locate further away from major metropolitan areas, they tend to use CSR as a complementary channel to
improve transparency. It is possible that this information environment story could play an important role in our results. However,
after we use the mean of the bid-ask spread to further divide small communities subsample into high and low spread groups, our
patterns still hold for both groups and there is no signicant dierence between two geographic dispersion coecients (results
untabulated).
Prior literature (Boh et al., 2007) also nds that geographic dispersion creates barrier to collaboration across sites, thus increase
coordination costs. Dawkins (2005) suggests that internal communication is a potential powerful channel for enhancing CSR.
Therefore, an increase in coordination costs may reduce CSR. To examine this, we use the average distance between rm's head-
quarter and each mentioned state in 10-k report as a proxy for internal coordination and communication cost. Then we divide the
sample into two subsamples of long and short distance. Our patterns still hold for both subsamples and there is also no signicant
dierence between two geographic dispersion coecients (results untabulated).16

6. The endogeneity between geographic dispersion and CSR

Now we examine the possible endogeneity between the geographic dispersion and CSR scores. This might stem from the possi-
bility that rms that have lower CSR scores might choose to operate business in more states. We address this issue through the two-
stage Heckman model (Heckman, 1976, 1979; Maddala, 1983).
Table 7 presents the results. In the rst stage, we use rm market value to capture whether rms have CSR scores or not. We use
this variable in the rst stage of the Heckman model because KLD only reports CSR scores of large rms.17 As we can see, in the rst
stage of Heckman model, the coecient of total market capitalization is 0.497, signicant at 1 percent level. In the second stage
regression, the coecient of geographic dispersion is 0.022, still signicant at 1 percent level. Also note that in the second stage the
coecient of Inverse Mills ratio is 1.340, signicant at 1 percent level. This implies that while sample selection bias does exist, our
pattern still does not change when we control for this endogeneity concern.

7. Conclusion

We examined whether local rms or geographically dispersed rms have higher CSR scores. We argue that, ceteris paribus, local
rms have higher CSR scores than geographically dispersed rms. Using a sample of 24,864 rm-year observations of US rms from
1994 to 2015 and controlling for other rm-specic determinants as well as industry and year xed eects, we nd the geographic
dispersion to be signicantly and negatively associated with CSR scores. We also examine the validity of three alternative arguments
to explain the negative link between CSR scores and geographic dispersion: social interaction, agency cost, and customer and investor

16
We thank the referee for pointing out these two possible explanations.
17
For each year beginning with 1991, KLD provides CSR data of approximately 650 companies that comprise the Domini 400 Social SM Index and S & P 500. KLD
expanded its coverage to include all companies on the Russell 1000 in 2001 and Large Cap Social SM Index in 2001. In 2003, KLD added full coverage of the Russell
3000, that has about 3100 companies for each year.

10
G. Shi et al. Journal of Accounting and Public Policy xxx (xxxx) xxxxxx

Table 7
Heckman two stage regression of CSR on domestic geographic dispersion.

1 st Stage 2nd Stage

***
Log of total market capitalization 0.497
(25.356)
Geographic dispersion 0.001 0.022***
(0.415) (4.932)
Size 0.212*** 0.613***
(8.728) (9.819)
ROA 0.134*** 0.673***
(2.712) (5.441)
Cash 0.046 0.858***
(0.665) (4.917)
Dividend 1.288*** 5.005***
(2.885) (5.038)
Debt 0.279*** 0.756***
(3.890) (4.876)
Inverse Mills ratio 1.340***
(8.485)
# of obs. 73,575 24,489
Pseudo or Adj R-squared 0.594 0.179

This table presents results from Heckman two stage regression that examine the endogeneity between domestic geo-
graphic dispersion and CSR. See Appendix A for other variable denitions. The dummies for industry and year xed
eects are included (as well as constant) in all the columns, but not reported. Standard errors are double-clustered by
rm and t-values are presented in parentheses. Statistical signicance at the 10%, 5%, and 1% level is indicated by *, **,
and ***, respectively.

recognition. Our evidence supports the notion of social interaction, but not the arguments of agency cost and customer and investor
recognition. Our results are robust after we control for the possible endogeneity.
Our paper contributes to the literature as follows. First, our paper documents a new factor, geographic dispersion, that negatively
aects CSR. Recently studies on CSR has attracted more and more attentions from both researchers and practitioners, and this new
nding can help people understand the determinants of CSR and thus may be able to suggest on how to improve CSR and benet all
stakeholders.
Second, our paper also contributes to the literature from the perspective of the geographic dispersion. While there exists a few
studies on the roles of geographic dispersion on stock returns (Garcia and Norli, 2012), rm value (Gao et al., 2008), earnings
management (Shi et al., 2015) and corporate management and nance policy making (Landier et al., 2009), our paper shows that
corporate geographic dispersion can also have a negative signicant impact on the debtholders or stakeholders, rather than just
shareholders. Our paper provides evidence that local rms may better protect their stakeholders than geographically dispersed rms.

Appendix A. Variables denition

Variables Denition

Geographic dispersion The number of dierent states mentioned in corporate SEC lling (10-K Form)
CSR CSR score = Total strength score Total concern score. The strength score is the points a rm receives on
the community, diversity, corporate governance, employee relations, human rights, environment, and
product strength in the KLD database; the concern score is the points on the community, diversity,
corporate governance, employee relations, human rights, environment, and product concern
Size Log of the book value of assets
ROA The ratio of income before extraordinary items over total assets
Cash The ratio of cash balances over assets
Dividend The ratio of cash dividends over assets
Debt The ratio of total debt over assets
KZ score The ratio is identical to that used by Baker et al. (2003) and Hong et al. (2012)
SP500 Dummy variable equal to one if the rm is a member of the SP500 and zero otherwise
Log age Log of the number of months since a rms stock rst appears in CRSP
Product market Herndahl-Hirschman index (HHI) based on the two-digit SIC industry classication
competition

11
G. Shi et al. Journal of Accounting and Public Policy xxx (xxxx) xxxxxx

References

Bae, K.H., Kang, J.K., Wang, J., 2011. Employee treatment and rm leverage: a test of the stakeholder theory of capital structure. J. Financ. Econ. 100, 130153.
Baker, M., Stein, J.C., Wurgler, J., 2003. When does the market matter? Stock prices and the investment of equity-dependent rms. Q. J. Econ. 118 (3), 9691005.
Barber, B., Odean, T., 2008. All that glitters: the eect of attention and news on the buying behavior of individual and institutional investors. Rev. Financ. Stud. 21,
785818.
Benson, B.W., Davidson, W.N., 2010. The relation between stakeholder management, rm value, and CEO compensation: a test of enlightened value maximization.
Financ. Manage. 39, 929963.
Boeprasert, A., 2012. Does geographical proximity aect corporate social responsibility? Evidence from US market. Int. Bus. Res. 5 (9), 138147.
Boh, W.F., Ren, Y., Kiesler, S., Bussjaeger, R., 2007. Expertise and collaboration in the geographically dispersed organization. Organ. Sci. 18 (4), 595612.
Ciocirlan, C., Pettersson, C., 2012. Does workforce diversity matter in the ght against climate change? An analysis of fortune 500 companies. Corp. Soc. Responsib.
Environ. Manage. 19, 4762.
Dawkins, J., 2005. Corporate responsibility: the communication challenge. J. Commun. Manage. 9 (2), 108119.
Denis, D.J., Denis, D.K., Yost, K., 2002. Global diversication, industrial diversication, and rm value. J. Finance 57, 19511979.
Denis, D.J., Denis, D.K., Atulya, S., 1997. Agency problems, equity ownership, and corporate diversication. J. Finance 52, 135160.
Dhaliwal, D., Li, O.Z., Tsang, A., Yang, Y.G., 2011. Voluntary non-nancial disclosure and the cost of equity capital: the case of corporate social responsibility
reporting. Account. Rev. 86, 59100.
DiGuili, A., Kostovetsky, L., 2012. Are red or blue companies more likely to go green? Politics and corporate social responsibility. J. Financ. Econ. 111 (1), 158180.
Doukas, J.A., Pantzalis, C., 2003. Geographic diversication and agency costs of debt of multinational rms. J. Corpor. Finance 9, 5992.
Gao, W., Ng, L., Wang, Q., 2008. Does geographic dispersion aect rm valuation? J. Corpor. Finance 14, 674687.
Garcia, D., Norli, O., 2012. Geographic dispersion and stock returns. J. Financ. Econ. 106, 547565.
Ghoul, S.E., Guedhami, O., Kwok, C.C.Y., Mishra, D., 2011. Does corporate social responsibility aect the cost of capital? J. Bank. Finance 35, 23882406.
Giroud, X., Mueller, H.M., 2011. Corporate governance, product market competition, and equity prices. J. Finance 66, 563600.
Gompers, P., Ishii, J., Metrick, A., 2003. Corporate governance and equity prices. Quart. J. Econ. 118, 107155.
Grullon, G., Kanatas, G., Weston, J., 2004. Advertising, breadth of ownership, and liquidity. Rev. Financ. Stud. 17, 439461.
Heckman, J., 1976. The common structure of statistical models of truncation, sample selection and limited dependent variables and a simple estimator for such models.
Ann. Econ. Soc. Measur. 5, 475492.
Heckman, J., 1979. Sample selection bias as a specication error. Econometrica 47 (1), 153161.
Herbohn, K., Walker, J., Loo, H., 2014. Corporate social responsibility: the link between sustainability disclosure and sustainability performance. ABACUS 50,
422459.
Hong, H., Kubik, J., Stein, J., 2004. Social interaction and stock market participation. J. Finance 59, 137163.
Hong, H., Kubikz, J., Scheinkman, J., 2012. Financial Constraints on Corporate Goodness. Working Paper.
Huang, S., 2013. The impact of CEO characteristics on corporate sustainable development. Corp. Soc. Responsib. Environ. Manage. 20, 232244.
Ioannou, I., Serafeim, G., 2011. The impact of corporate social responsibility on investment recommendations. Strateg. Manage. J. 36 (7), 10531081.
Jiao, Y., 2010. Stakeholder welfare and rm value. J. Bank. Finance 34, 25492561.
Jiao, Y., Shi, G., 2012. Social Preference, Product Market Competition, and Firm Value. Working Paper.
Jiraporn, P., Kim, Y.S., Davidson, W.N., Singh, M., 2006. Corporate governance, shareholder rights and rm diversication: an empirical analysis. J. Bank. Finance 30,
947963.
Kim, Y., Park, M.S., Wier, B., 2012. Is earnings quality associated with corporate social responsibility? Account. Rev. 87, 761796.
Landier, A., Nair, V.B., Wulf, J., 2009. Tradeos in staying close: corporate decision-making and geographic dispersion. Rev. Financ. Stud. 29, 11191148.
Lee, E., Walker, M., Zeng, C., 2017. Do Chinese state subsidies aect voluntary corporate social responsibility disclosure? J. Account. Public Policy 36 (3), 179200.
Lehavy, R., Sloan, R., 2008. Investor recognition and stock returns. Rev. Acc. Stud. 13, 327361.
Li, C.K., Luo, J., Soderstrom, N.S., 2017. Market response to expected regulatory costs related to haze. J. Account. Public Policy 36 (3), 201219.
Maddala, G.S., 1983. Limited-Dependent and Qualitative Variables in Econometrics. Cambridge University Press, New York.
McGuire, S., Newton, N., Omer, T., Sharp, N., 2012. Does Local Religiosity Impact Corporate Social Responsibility? Working Paper.
Merton, R.C., 1987. A simple model of capital market equilibrium with incomplete information. J. Finance 42, 483510.
Monteiro, S., Aibar-Guzmn, B., 2010. Determinants of environmental disclosure in the annual reports of large companies operating in Portugal. Corp. Soc. Responsib.
Environ. Manage. 17, 185204.
Patten, D.M., 1991. Exposure, legitimacy, and social disclosure. J. Account. Public Policy 10, 297308.
Petersen, M.A., 2009. Estimating standard errors in nance panel data sets: comparing approaches. Rev. Financ. Stud. 22, 435480.
Porter, M., Kramer, M., 2006. Strategy and society: the link between competitive advantage and corporate social responsibility. Harvard Bus. Rev. 84 (12), 7692.
Schmid, M.M., Walter, I., 2012. Geographic diversication and rm value in the nancial services industry. J. Empir. Finance 19, 109122.
Scholtens, B., Dam, L., 2007. Cultural values and international dierences in business ethics. J. Bus. Ethics 75, 273284.
Shi, G., Sun, J., Luo, R., 2015. Geographic dispersion and earnings management. J. Account. Public Policy 34 (5), 490508.

12