Association for Public Policy Analysis & Management

Nonprofits in the United States must comply with various state and federal regulations to maintain their tax-exempt status and to solicit funds from the public. Among the various types of regulations, one onerous obligation requires nonprofits to file independent audit reports with the state government. Nearly half of all states require nonprofits with a certain annual revenue size to be audited if they solicit funds from their states’ residents. However, in addition to its potential to achieve greater accountability and transparency, such regulatory instruments can impose significant burdens on nonprofits that lie just above the threshold where the costs incurred by an organization would be neither proportionate nor justifiable. This leads to the bunching behavior, reducing the usefulness of governance regulations in counteracting management’s reporting bias.

The present paper quantifies the nonprofit organizations’ bunching behavior to state-level audit thresholds in the United States. Using IRS 990 Returns from 1989 to 2015, we employ bunching estimation with state- and year-fixed effects to detect general patterns of the avoidance behavior and explore the heterogeneous aspects of it. We take advantage of variations in U.S. audit thresholds across states and time to estimate nonprofits’ bunching in the distribution of the total reported revenues or contributions.

Exploring the audit thresholds’ variations in states and time periods, the results consistently demonstrate that nonprofits adjust their revenues and contributions so that they remain below the audit threshold. The results report that the bunching estimates are much larger for revenue-based thresholds than contribution-based thresholds. Comparing the effects by threshold level shows larger estimates for lower level thresholds, suggesting that revenue is easily manipulated at the lower thresholds even in those of crossing from lower (review) to top tier requirements (audit). These results are substantiated by a series of falsification tests using placebo periods and placebo thresholds. In addition, by applying the normalizing-and-pooling strategy, the analysis reveals substantial heterogeneity in the degree of sorting by organizational characteristics that are associated with misreporting financial information. We also conduct tests for placebo effects on the lagged outcome variables and find strong evidence that those facing the audit threshold at which total revenues and contributions increase discontinuously were driving the results. Furthermore, to evaluate causal effects of the threshold implementation, we study and compare two alternative approaches that account for non-random selection, controlling for a rich set of observables. First, we consider a donut-RD in which we simply drop the observations closest to the threshold where sorting is present. Second, we apply a difference-in-differences design to analyze the effect of a relaxed (rising) audit threshold on organizational outcomes.

The contribution of this work is twofold. First, this study provides broad evidence, across multiple thresholds and across multiple states, on the extent to which noncompliant behaviors limit our ability to learn from RDD based on regulatory thresholds. Second, it also presents the practical sides of the subject by addressing the issues and challenges in managing regulatory compliance on public disclosure requirements.