The Relationship Between Sarbanes–Oxley Policies and Donor Advisories in Nonprofit Organizations

Nezhina and Brudney (2012) document increased expenditure of both time and money, including longer and more frequent meetings, increased audit fees, board and executive training expenses, and re-allocation of resources to administrative expenses, etc.

Prior studies have either included the written policies as part of a broader governance index or governance construct or, in the case of Harris et al. (2017), did not include all three written policies (Harris et al. 2017 do not separately consider records retention policies). Nezhina and Brudney’s (2012) survey-based and qualitative research, in contrast, represented a valuable early set of univariate analyses of the impacts of SOX on the nonprofit sector.

While diminishing judgment is certainly not always a positive outcome, in certain instances—such as reducing organizational malfeasance—the benefits of removing judgment from the equation may outweigh the costs.

We began by downloading all available data from the Charity Navigator site, which provided information on donor advisories as well as basic financial and organizational data that were derived from the organizations’ IRS Form 990 returns. Organizations’ age was then determined by examining the IRS Business Master Files records, while data for the governance controls were gathered by accessing the IRS’ e-file 990 database (available at https://​aws.​amazon.​com/​public-datasets/​irs-990/​) using customized Python scripts (a detailed replication tutorial for downloading and re-creating the 990 e-file data is available at http://​social-metrics.​org/​sox). Governance data for non-e-filing organizations were gathered from the IRS Statistics of Income micro-data files (as in Yetman and Yetman 2012), available at https://​www.​irs.​gov/​statistics/​soi-tax-stats-charities-and-other-tax-exempt-organizations-statistics. We also gathered data on the state regulatory environment from the Multi-State Filer Project website (http://​multistatefiling​.​org). In the following sections we provide further details on the specific data sources and operationalization procedures for our variables.

The mean length of time for remediation in our dataset is 1.6 years. According to Charity Navigator, “…in order for a charity's rating to be restored, it must provide public domain documentation or similarly reliable and accessible information that demonstrates that the issues identified in the Donor Advisory have been resolved. Once the Donor Advisory has been lifted, a record of it will remain in the charity’s ‘Historical’ section of its profile page so that donors have an opportunity to learn about issues the charity previously faced and, if available, how it resolved those issues” (www.​charitynavigator​.​org/​index.​cfm?​bay=​search.​donoradvisories).

We also ran logit versions of these three models including all control variables and the three individual SOX policies as the independent variables. The results for the three SOX variables as well as the control variables were similar; as a result, for ease of presentation, we present the two-stage models shown in Table 4.

Specifically, Fig. 2 shows the predicted likelihood of an organization receiving a donor advisory (y-axis) across our sample’s empirical range of Program Efficiency Ratios (x-axis) for different values of the independent variable (the different lines for 0, 1, 2, and 3 SOX policies). The NTEE category is set as Human Services, and all remaining control variables from Model 5 in Table 5 are held constant at their mean values.

The figure shows a clear ordering in the predicted probabilities of a donor advisory with each successive SOX policy—as the number of SOX policies goes up, the probability of a donor advisory goes down. Similarly, there are stark differences in the predicted likelihoods based on the level of organizational efficiency, with greater efficiency being associated with lower probabilities of receiving a donor advisory. When efficiency and SOX adoption are viewed together, we see the differences in predicted likelihoods for the different numbers of SOX policies are more pronounced for organizations with lower efficiency ratios. For instance, moving from 0 to 1 SOX policy at the 0% program spending level results in a 1% drop (34% as percentage change) in the predicted likelihood of a donor advisory (from 2.8 to 1.8%), a much greater change than the 0.5% drop (from 1.45 to 0.94%) seen from moving from 0 to 1 SOX policy at the mean level of efficiency (80.1%) or the 0.26% drop (from 0.74 to 0.48%) for such a move at the 100% efficiency level. On average, if a wholly inefficient organization moves from 0 to 3 SOX policies, it will experience a 2% absolute decrease (from 2.8 to 0.79%) in the predicted likelihood of a donor advisory. In contrast, making the same policy change in an organization at the maximum efficiency level is associated with a decrease in predicted probability of 0.54% (with the predicted likelihood dropping from 0.74 to 0.2%).

Specifically, to analyze the potential influence of unobservable confounding variables, we used the Konfound package in Stata to calculate the ITCV impact statistic for our independent variable SOX Policies in a modified version of Model 5. The test showed that an omitted variable would have to be correlated at − 0.247 or greater with SOX Policies and at 0.247 or greater with our dependent variable Donor Advisory in order to render SOX Policies statistically insignificant. This corresponds to an ITCV impact for an omitted variable of − 0.061 (− 0.247 × 0.247). Given the difficulty in determining absolute standards for impact threshold, we employ the technique recommended by Larcker and Rusticus (2010, p. 203) and use “…the impact for the … control variables to provide a reasonable benchmark to the ITCV.” We thus calculated the impact (in line with Larcker and Rusticus, calculated as the product of the partial correlations of each control with Donor Advisory and SOX Policies) of all control variables shown in Model 5 (Table 5). None of the controls had an impact close to the threshold of -0.061; the controls with the largest impact from Model 5 are Program Efficiency, Total Revenues, Audit Committee, Complexity, and Audited Financial Statements, yet their ITCV impacts range only from .01 to − 0.04.

For the models shown in Table 6, we found only one of the donor advisories was for an organization that had reported a significant diversion, so re-running those results would show no important changes in results. In the sample used in Tables 4 and 5, in turn, we found 12 reported significant diversions; 10 of these had a donor advisory while 2 did not. We thus replicated the test shown in Model 4 omitting these 10 and 12 cases in two separate tests. In both cases, there were no changes in sign or significance in any of the model variables nor in the overall pseudo-R². The correlation between material diversions and donor advisories in this sample is 0.27 (p < 0.001).