Strategic motivations for corporate social responsibility: profitability or legitimacy?

Abstract

Management research has broadly categorized strategic motivations for corporate social responsibility (CSR) as profit-seeking or legitimacy, which are at times conflicting and complementary. The nature of firm motivations has significant implications for firm-level and societal outcomes yet is not directly observable. In this study I theorize that strategic motivations may be inferred based on observed performance relative to context-specific groups of relevant peer firms. I integrate economics and strategic management traditions to develop both a theoretical framework and an empirical model of peer effects and their underlying strategic motivations, and employ a novel instrumental variables estimation that exploits unique aspects of a recent innovation in dynamic industry classification. Results indicate that firms select referent peers based on strategic context and shed new light on the multi-dimensional nature of CSR strategies, revealing subtleties of firm behavior often masked by traditional empirical approaches. Contributions to theory and opportunities for future research are highlighted.

Appendix—analytical model

Consider a firm i which in period t observes its institutional and stakeholder context, the CSP of relevant peer firms, and its own level of CSP. The firm then determines the level of CSP to implement in the following period, e.g., via implementation or disclosure of additional CSR activities. The researcher does not observe the firm’s detailed CSR strategy nor its industry peers’ strategies but does observe the resulting measurement of CSP for all firms. This process then repeats itself in the subsequent period. The following simple linear-in-means (LIM), dynamic panel data model describes this process:

$${y}_{it}=\alpha {y}_{it-1}+\beta {\overline{y} }_{-it-1}$$

(3)

The scalar outcome variable y_it is the current year CSP, and the lagged dependent variable y_it−1 is included to capture the path dependence of CSP. y̅_−it−1 represents the mean outcome for firm i’s reference peer group in the previous period, where − i index indicates the firm’s own outcome is excluded from the peer group mean calculation. To model strategic change, the dependent variable measure employed in this study is the change in performance relative to the previous period, or ∆y_it = y_it − y_it-1. Subtracting y_it-₁ from each side of equation 3 leads to:

$$\Delta {y}_{it}=\left(\alpha -1\right){y}_{it-1}+\beta {\overline{y} }_{-it-1}$$

(4)

According to equation 4, in each period the firm faces a decision whether and how much to increase CSP. The firm forms a strategic performance target for its own CSP based on observations of the performance of its relevant peers as well as its own past performance. Such performance targets (or “aspiration levels”) form the reference point by which firms evaluate their current performance. Aspiration levels are often assumed by researchers to take one of several alternate functional forms based on peer performance (the “social” component) and on realizations of aspiration levels from prior periods (the “historical” component) (Washburn and Bromiley 2012). One common approach prescribes a single aspiration level calculated as the weighted average of social and historical levels, with the relative weights either predetermined by the researcher or estimated using grid search methods to find the best model fit (Greve 2003a). In these studies, the historical component is based on a recursive function that effectively incorporates all past realizations of own performance into the current level. Furthermore, this grid search approach often indicates that the social component tends to dominate historical in terms of explanatory power (Greve 2003a). Other models (e.g., Mezias et al. 2002) have included these two components separately in the same regression model to account for potentially additive aspirations on different performance comparisons. Finally, other researchers have tested “switching models” where firms are hypothesized to change behavior upon performance surpassing various thresholds. Different thresholds may apply in different strategic contexts, whether applied at the peer group average (Bromiley 1991) or at multiple levels representing different regimes of financial stability (Chen and Miller 2007). I augment the Bromiley (1991) switching model with firm performance relative to the social aspiration level driving firm behavior. Inclusion of the lagged dependent variable accounts for the possibility that historical firm performance influences behavior above or below the peer-based target. Note that the lagged dependent variable, by construction, includes effects of all past firm performance levels through recursion.

Returning to the analytical model, the firm sets its current period CSP strategy based on the level of CSP realized in the prior period relative to its peer performance-based strategic target. Adding and subtracting own lagged performance to the right-hand side of equation 4 and rearranging terms yields the following equation:

$$\Delta {y}_{it}=\left(\alpha +\beta -1\right){y}_{it-1}-\beta \left({y}_{it-1}-{\overline{y} }_{-it-1}\right)$$

(5)

Equation 5 represents the basic descriptive model of this study in terms comparable to the aspiration levels literature; it describes the influence on firm strategy by the firm’s prior performance relative to a performance target based on the observed prior performance of relevant peers (often referred to as “attainment discrepancy”). For notational simplicity, let P_it ≡ y_it signify a firm’s own performance, and S_it ≡ y̅_−it signify the “social” performance target based on average performance of relevant peer firms. Also, let γ ≡ α + β − 1 and δ ≡ − β, such that:

$$\Delta {P}_{it}=\gamma {P}_{it-1}+\delta \left({P}_{it-1}-{S}_{it-1}\right)$$

(6)

The empirical models described by Eqs. 3 and 6 are mathematically equivalent, and either model may be used to represent peer effects and path dependency in CSP. The base LIM parameters (equation 3) are thus easily derived from the empirical results, i.e., β =  − δ and α = 1 + γ + δ, where α represents the autoregressive coefficient on firm performance and β the direct peer effect. Equation 6 is perhaps more recognizable with respect to empirical studies on aspiration levels and better describes the theoretical response shapes in Table 1. However, I use equation 3 for empirical analysis given the simpler interpretation of the main regression coefficients of interest, the prevalence of this functional form in the large literature of peer effects, and for compatibility with available instrumental variables (see following subsection). Rewriting equation 3 in terms of the equation 6 variables then yields the following:

$${P}_{it}=\alpha {P}_{it-1}+\beta {S}_{it-1}$$

(7)

Equation 7 represents the core analytical model used to empirically test the study hypotheses (Eq. 1 in main manuscript). To test hypotheses 1–3 I further derive the switching model (Eq. 2) in the main manuscript.