Religious Values Motivating CSR: An Empirical Study from Corporate Leaders’ Perspective

Our definitions of religion, religiosity, religious value, and religious education are as follows, respectively: religion—the belief in and worship of God; religiosity—excessively or sentimentally religious; religious value—a set of ethical principles founded in religious traditions, texts, and beliefs; religious education—the teaching of a particular religion and its varied aspects, such as beliefs, customs, doctrines, etc.

Dordt College’s mission statement: “As an institution of higher education committed to the Reformed Christian perspective, Dordt equips students, alumni, and the broader community to work effectively toward Christ-centered renewal in all aspects of contemporary life.”

Usnews.com, niche.com, wikipedia.com, and 4icu.org/religious/.

BoardEx updates its database on a regular basis. We retrieved our initial sample from BoardEx only once as a snapshot as of the download date to ensure data consistency. Moreover, instead of providing data at firm-year level, BoardEx reports the start and the end date of each firm-manager pair. So, we first use this information to create a dataset of firm-manager-year observations. We then aggregate the data to construct our main variable of interests, i.e. Leadership RSA, at the firm-year level and merge with other datasets by same common fields (firm identifier and year).

In BoardEx, managers with following strings in their roles are excluded: Acting, Honorary, Emeritus, Deputy, Interim, Coordinator, Advisory, Alternate, Area, Administrative, Associate, Assistant, Brand, Designate, Editor, Staff, Zone, and Branch.

We hand-collect information on schools’ religious affiliations in the following four steps. First, we identify, for each school in our sample, whether it is a private or public school based on the information from usnews.com. Because all public schools in the U.S. are secular, we limit our second-step search to the pool of private schools. Second, for each private school in our sample, we screen school names for indicator of its religious affiliations. For instance, if school name contains “Trinity”, “Catholic”, or “Christian”, we flag it as a religious-affiliated school. Third, for those schools that we cannot tell religious affiliation from their names, we rely on schools’ official websites as our primary information source, specifically the “mission/history/heritage/facts/numbers” pages under the “about” tab. For instance, Georgetown University’s “Who We Are” page states, “… we’re the nation’s oldest Catholic and Jesuit university”. Accordingly, we identify Georgetown as Catholic. Lastly, we doublecheck all sample schools’ religious affiliations using information from, for example, the U.S. Department of Education website and Wikipedia.com.

We compare the descriptive statistics of the initial sample of 1484 firms and that of the final sample and find that all statistics are qualitatively similar. Actual sample sizes in different regressions vary due to data availability across different specifications.

In mathematical form, it equals to ln [1 + (number of top RSA managers/total number of top managers) ^ 2].

For brevity, the summary statistics of county-level religiosity are omitted in the paper but are available from the authors upon request.

The problem of the standard R-squared or adjusted R-squared under many fixed effects, three (state, industry, and year) in our case, is that its value will mostly be driven by the fixed effects and not by the regressor of interest. We use a built-in function for R-squared calculation in STATA so that when calculating R-squared, every variable has already been demeaned with respect to all the fixed effects. Specifically, Adjusted R-squared = 1–[RSS/(N-1)]/[TSS/N-K-1-kk)], where RSS is the residual sum of squares, TSS is the total sum of squares, N is the sample size, K is the number of explanatory variables in the model, and kk is the number of fixed effects (which addresses the inflating effect from multiple fixed effects).

We employ “reghdfe” package (Correia 2017) in Stata to include multiple levels of fixed effects and have double-clustered standard errors in all our regressions.

The propensity score is the predicted value from the regression of leadership RSA on all other control variables.

Results are omitted for brevity but are available from the authors upon request.