Campbell (2007) describes the institutional theory on CSR with eight propositions. He answers the question: under what conditions do firms act in socially responsible ways? His research uses an institutional analysis and comparative political economy approach to answer that question. Our paper complements the theory with empirical research focusing on three of Campbell’s propositions.

Corporations will be less likely to act in socially responsible ways when they are experiencing relatively weak financial performance and when they are operating in a relative unhealthy economic environment where there is a limited possibility for near-term profitability.

There are few papers that empirically research the economic conditions on CSR adoption. Branca, Pina, and Catalao-Lopes (2012) show that firms expect to reduce CSR spending when the business cycle is unfavorable. However, their analysis only looks at the corporate giving dimension of CSR without taking into consideration other CSR activities such as environmental, human rights, diversity. Campbell (2007) argues that firms will not behave socially responsible when they are operating in an unhealthy economic environment, however he does not provide empirical evidence. Hsiang-Lin, Hsiang-Hsuan, and Tzu-Yin (2010) test Campbell’s argument by including the inflation rate, industrial production index and the consumer confidence index as the economic variables. They find that firms in countries with higher consumer confidence levels and lower inflationary levels will act in more socially responsible ways. They only estimate results for financial firms in a country level setting.

The closest research to our analysis in terms of the cyclicality of CSR is Branca, Pina, and Catalao-Lopes (2012). Although we look at related issues in a similar context, we have very different approaches. Branca, Pina, and Catalao-Lopes (2012) take a time series approach by estimating the cross-correlation pattern of CSR giving and firm revenues with real GDP. They find a significant co-movement between real GDP and CSR giving. In contrast, we use panel data to estimate the relationship between the growth in industry value added and the probability of a firm to engage in CSR activities.

Our approach has several advantages and provides new insights. We offer more precise and reliable estimates on the impact of the business cycle on CSR by using firm’s data. Branca, Pina, and Catalao-Lopes (2012) focus on giving as the only measure of CSR, using IRS data. But data limitations do not allow them to control for any other firm’s characteristics.

Even though research on CSR has been increasing, researchers know little about how CSR can influence or be influenced by peer firms or industry dynamics. To my knowledge, few studies have tried to determine the peer effects of corporate social responsibility (see Cao, Liang,& Zhan, 2015; Leary & Roberts, 2014; Liu & Wu, 2015; Servaes & Tamayo, 2014). This research presents evidence on how firms react to their peers’ adoption of CSR. They find that the effects are stronger amongst peers with higher competitive pressure and a more transparent information environment (Cao, Liang, & Zhan, 2015). Liu and Wu (2015) show that when a firm’s competitors exhibit a higher level of CSR, the firm is more likely to engage in CSR activities.

Our contribution looks at peer effects by analyzing the proportion of firms within the same industry that adopt socially responsible practices and the firm’s decision to engage in CSR for a particular firm in the same industry, using controls defined later on.

A potential problem is the probability of a firm’s CSR and the competitor’s CSR being driven by the same industry level shocks. It could lead to peer effects appearance. Similarly, if the firm and its competitors are geographically close, region-level shocks may also affect the interpretation. To deal with this issue, we include industry times year fixed effects and state times year fixed effects in the analysis (see Liu & Wu, 2015). This way, we control for any time-varying trends in the industry and region level.

From the workers’ preferences perspective,² the closest paper is Marquis, Glynn, and Davis (2007). They provide a framework to look at how the community and culture pressure firms to implement CSR. Marquis, Glynn, and Davis (2007) suggest that communities are important influencers on corporate action. There are two main reasons for this argument. First, the set of rules and norms in place in the community serve also as a benchmark for legitimizing CSR (see also Margolis & Walsh, 2003; Marquis, Glynn, & Davis, 2007). Second, CSR activities orient towards the community where the corporation’s executive lives (Marquis, Glynn, & Davis, 2007). However, Marquis, Glynn, and Davis (2007), and Campbell (2007) provide none empirical analysis of their framework. Several studies in the US have shown that this is the case by using small samples of firms across the country. Most of these researches are study cases, for example Galaskiewicz and Wasserman (1989) looks at local giving patterns of Minneapolis based companies and how network effects influence them.

To the best of our knowledge, there is no study that includes workers’ attitudes from ANES into the analysis of CSR and data on firms from COMPUSTAT and KLD for the period we consider (1998-2010).

An annual dataset on environmental, social, and governance ratings, published at the end of each calendar year. It identifies the company information (Name, Ticker, unique security identifiers), and provides 50 CSR indicators organized in seven categories: The environment, community, human rights, employees, diversity, production and corporate governance. Table 1 describes the seven areas covered by KLD and their components.⁴ The number of firms in the sample has a substantial increase in 2003, and from that year onward KLD added CSR ratings for all firms belonging to the Russell1000Index.⁵

Table 1

Indicator	Components		Indicator	Components
Environmental	Beneficial Products and Services		Diversity	Representation (women and minorities)
	Pollution Prevention			Board of Directors (women and minorities)
	Recycling			Work/Life Benefits
	Clean Energy			Women and Minority Contracting
	Management Systems			Gay and Lesbian Policies
	Other			Employment of Under represented Groups
Community	Charitable Giving			Other
	Innovative Giving		Product	Quality
	Community Engagement			Benefits to Economically
	Other			Disadvantaged
Human Rights	Indigenous Peoples Relations			Access to Capital
	Strength			Product Safety
	Human Rights Policies and initiative			Marketing/Contracting
Employee Relations	Union Relations			Concern
	Cash Profit Sharing			Antitrust
	Employee Involvement			Customer Relations Concern
	Health and Safety
	Supply Chain Policies, Programs and Initiative		Governance	Reporting Quality
	Other			Public Policy

Source: www.msci.com

It is an annual database of financial, statistical, and market information on publicly traded companies. Characteristics include indicators such as total assets, earnings, industry classification, sales, number of employees, among others. Such characteristics would most likely account for time varying conditions faced by the firm.

Data from the Bureau of Economic Analysis, which includes the value added by industry sector. Data is available from 1998 to 2010.

Data from American National Election Studies survey; a bi-annual survey on the state of the individual’s attitudes towards an array of issues. In particular, we use information about workers’ attitudes (on a 0 to 100 scale, 100 being positive) towards diversity, unions and environmentalists. ANES codes data by state and we match it to the state location of the firm. Although this might not directly capture the workers’ opinions each individual firm faces, it would provide a proxy for the attitudes prevailing in the American electorate.

To analyze how the proportion of firms within the same industry affect the decision to engage in CSR for a particular firm in the same industry, we estimate equation 2, where α_i and X_ijt are defined as before. We further include year fixed effects (γ), and interest lies in θ and Θ.

C S R i j t = α i + γ t + θ p f i r m s + Θ p f i r m s 2 + δ X i j t + U i j t (2)

We exploit state-level variation in workers attitudes (towards specific issues) to assess if these attitudes influence the likelihood of firms to engage in (overall and thematic) CSR. We estimate equation 3 using a fixed effects LMP.

C S R i s t = α i + δ X i s t + η D i v s t + λ U n i o n s s t + ρ E n v s t + γ t + U i s t (3)

Besides economic activity (measured here with the growth of value added), there are other factors that can influence CSR activities. We include these factors in the regression to control for their potential influence on CSR decisions. We consider several control variables (X_ijt) related to the firm’s characteristics. The specialized literature on CSR always include these variables. They include firm size; return on assets interacted by the number of employees, as another measure of the firm’s size, advertising intensity, R&D expenses, and the Hirschman-Herfindahl Index (HHI).

According to Belu and Manescu (2013) firm size is relevant for economic performance. Bigger firms will have higher variation on profitability. Market structure is important for CSR since it affects CSR adoption decision (Bagnoli & Watts, 2003). Results can vary significantly depending on the market structure. If markets are competitive, monopolistic or oligopolistic market power enables some firms to always to earn unusual returns.

Return on Assets (ROA) is a profitability measure that expresses the amount of net income, plus interest payments, but before preferred dividend per unit of average current and last year’s assets (Belu & Manescu, 2013). We measure firm size as the number of employees of the firm in logs; advertising intensity is the log advertising expenses; R&D intensity is the log of R&D expenses. We define the Hirschman-Herfindahl Index for each industry at the six-digit NAICS code level. Computed by adding the square of the market share of all players operating in an industry in a year.⁸ As displayed in Table 2, the control variables have values similar to the existing literature.

Table 2

Variable	Mean	Std. Dev.	N
ROA (Return on Assets in logs)	0.771	0.18	15,865
Emp (Number of employees in thousands)	16.355	56.96	23,063
ROA * Emp	17.219	57.916	15,743
HHI (Herfindahl Index)	0.187	0.177	23,368
R&D (log of R&D expenditures)	0.072	0.527	23,368
Advert (log of Advertisement expenditures)	4.113	7.275	23,368
Growthjt-1 (Growth in Value added)	0.03	0.111	19,809
Diversity (CSR diversity index)	0.5	0.07	14,442
Unions (Union relations index)	0.564	0.055	14,442
Environment (CSR environment index)	0.65	0.046	13,601
pfirms (# firms CSR/#firms industry)	0.560	0.172	23,368

Source: Own elaboration.

We assess the relationship between growth of value added and the probability of a firm engaging in CSR. Estimation results are in Table 3. Column 1 reports the results of a linear probability model with fixed effects and in the next columns we account for the possibility of mis-specification because of autocorrelation in data.

Table 3

Corporate Social Responsibility
	(1)	(2)	(3)	(4)	(5)
	OLS	Logit	AR(1)		AR(ρ)
Growthjt-1	0.16***	1.53***	0.075***	0.10***	0.16***
	(0.057)	(0.31)	(0.029)	(0.029)	(0.035)
Control Variables	Yes	Yes	Yes	Yes	Yes
Firm fixed effects	Yes	Yes	Yes	Yes	Yes
Observations	13,491	6,591	11,248	13,139	13,491
Number of id	2,243	1,003	2,144	2,123
Number of groups					2,243

Notes: Robust standard errors clustered at the industry level are in parentheses. The control variables are Number of employees (in logs), ROA*Employees, HHI Index, ln R&D expenditures, ln Advertisement expenditures, and time trend (year). All regressions control for firm fixed effects. *** is significant at the 1% level; ** is significant at the 5% level; * is significant at the 10% level.

We find that the coefficient on the growth of value added, Growth_jt-1, is positive and statistically significant even after accounting for autocorrelation. Column 2 reports the estimates of equation 1 with autocorrelation of order AR1. The coefficient on the growth of value added Growth_jt-1 is positive and statistically significant. Also, given that the confidence intervals of the coefficients on Growth_jt-1 overlap, we keep the most parsimonious model with no autocorrelation for the rest of the paper. For a further check of results in Table 3, we include the dependent variable difference by looking at the firm’s transitions to engage in CSR or stop doing CSR activities;⁹ the results hold with positive and significant coefficients as seen in the Appendix Table A1.

Even though the overall results reflect that the economic environment at the industrial level (i.e. Growth_jt-1) is important on a firm’s decision to engage in CSR, we estimate the heterogeneous treatment by the firm’s economic performance.¹⁰

In Table 4, industry’s Growth_jt-1 is insignificant for top performers “Gazelles” (top 25 percent) related to CSR. For “Elephants” (bottom 75 percent), even when controlling for the firm’s characteristics and a time trend, an increase in Growth_jt-1 leads to an increase in the probability to engage in CSR. The result coincides with the proposition that most corporations will be more likely to engage in CSR when they are experiencing a healthy environment (Campbell, 2007). However, for top performers, there could be other more significant factors that determine whether a firm engages in socially responsible activities.

Table 4

Corporate Social Responsibility
	(1)	(2)
	Elephants	Gazelles
Growthjt-1	0.23***	-0.018
	(0.063)	(0.071)
Control variables	Yes	Yes
Firm fixed effects	Yes	Yes
Observations	10,025	3,466
R-squared	0.016	0.020
Number of id	2,032	791

Notes: Robust standard errors clustered at the industry level are in parentheses. The control variables are Number of employees (in logs), ROA*Employees, HHI Index, ln R&D expenditures, ln Advertisement expenditures, and time trend (year). All regressions control for firm fixed effects. *** is significant at the 1% level; ** is significant at the 5% level; * is significant at the 10% level.

Table 5 shows marginal effects of a multinomial logit. The probability to engage in CSR when there is a ten-percentage increase in Growth_jt-1, increases the probability of adopting CSR by 1.7 percentage points (equivalent to 25 percent), regarding not adopting. Similarly, a ten-percentage increase in Growth_jt-1 decreases the probability of disengaging in CSR by 4 percentage points (equivalent to 62 percent).

Table 5

Corporate Social Responsibility
	(1)	(2)
	Engage	Disengage
Growthjt-1	0.17**	-0.40***
	(0.072)	(0.066)
Control variables	Yes	Yes
Firm fixed effects	Yes	Yes
Observations	3,605	3,605

Notes: Robust standard errors clustered at the industry level are in parentheses for the last two columns. The control variables are Number of employees (in logs), ROA*Employees, HHI Index, ln R&D expenditures, ln Advertisement expenditures, and year. All regressions control for firm fixed effects. *** is significant at the 1% level; ** is significant at the 5% level; * is significant at the 10% level.

Our results confirm the theory that firms will be more likely to engage in socially responsible activities when there is a healthy economic environment for each industry.