Social Work Research and Endogeneity Bias

Abstract

This article discusses endogeneity bias as it applies to causal inference in social work research. Drawing upon on both traditional economic and econometric perspectives as well recent conceptual discussions within allied fields (e.g., public health), we define and discuss key sources of endogeneity bias, summarize a variety of methods that may ameliorate such bias, and discuss the applicability and appearance of these methods in social work research. Social work research may be especially prone to various sources of endogeneity bias given the nature of the research problems investigated, the emerging state of social work measurement, and the necessary reliance on nonexperimental methods to test for explanatory (e.g., intervention) effects. We argue for heightened attention to endogeneity bias given the ever-increasing calls within social work for change-intensive evidence-based programs and policies.

Does it pay to be a greenwasher or a brownwasher?

“…First, the two-stage least squares (2SLS) estimator involving instrumental variables has been adopted to alleviate possible endogeneity bias between CFP and firm-level green practice index (Stone & Rose, 2011). In this context, we rely on the set of instruments contained within a panel to instrument the GPI variable due to the absence of good external instrumental variables (Krafft, Qu, Quatraro, & Ravix, 2014;Miroshnychenko, Bozzi, & Barontini, 2018 This table presents the coefficients and t-statistics (in parentheses) using the 2SLS regressions with HAC standard errors.…”

Contemporary Quantitative Methods and “Slow” Causal Inference

“…Palinkas' article implicitly echoes this by arguing that causally informative quantitative and qualitative methods are essential to the development of our knowledge base. Stone and Rose (2011) have noted similar calls from other social work intervention scholars, which urge a transformation of the social work knowledge base to reflect the change-oriented goals of the profession (see Fraser, 2004), with a central, if not core, focus on ''control-capable'' knowledge (i.e., knowledge to directly inform practitioners' changerelated strategies; Rosen, Proctor, & Staudt, 1999).…”

Instrumental Variable Estimation in Social Work Research: A Technique for Estimating Causal Effects in Nonrandomized Settings

“…In the context of the Rubin causal model, approximating this random switch is the objective when researchers intend to make causal inferences from IVE. This approach is in contrast to most other approaches that address endogeneity (e.g., fixed effects, propensity scores, regression discontinuity) in which assignment to conditions of the endogenous causal variable is either directly modeled or controlled for (Stone & Rose, 2011). Thus, it is worth bearing in mind that the experimental approach is still considered the gold standard against which the IVE approach is judged.…”

“…These approaches include rigorous methods of handling missing data (Rose & Fraser, 2008), propensity score matching (Guo & Fraser, 2009), and handling clustered or longitudinal data (Guo & Hussey, 1999). We have limited our discussion to IVE for addressing the internal validity of a causal impact estimate, but IVE is designed to address endogeneity, which also implies that it can address measurement error (Stone & Rose, 2011). Historically, social work researchers have used latent approaches to address the measurement error problem; thus, the focus in this article is on the relationship between endogeneity and internal validity.…”

“…In this article, we first provide a brief review of the background assumptions and the concept of endogeneity (for detailed discussion, see our companion article [Stone & Rose, 2011] beginning on p. 54 of this issue). We then introduce and discuss IVE by describing the assumptions required in an instrumental variable analysis and providing an example showing how satisfying these assumptions can lead to unbiased causal estimation.…”