«Subramaniam Ramanarayanan Anderson School of Management, UCLA Los Angeles, CA 90095 subbu Jason Snyder Anderson School of ...»
Information Disclosure and Firm Performance: Evidence from the Dialysis
Anderson School of Management, UCLA
Los Angeles, CA 90095
Anderson School of Management, UCLA
Los Angeles, CA 90095
Abstract: We study the impact of information disclosure policies on firm performance by
exploiting a policy change that provides plausibly exogenous “shocks” to firms’ reputations based on their allocation to coarse performance categories. We examine this issue in the context of dialysis facility ratings generated by Medicare, based on patient survival rates. We exploit the discrete nature of the performance ratings to implement a regression discontinuity design, and find that firms that receive negative quality ratings subsequently experience a reduction in patient mortality rates. We provide suggestive evidence that this improvement is associated with strategic patient selection. There is no impact of ratings on overall patient volumes, but facilities receiving poor grades treat fewer well-informed patients post-disclosure. We do not find comparable supply-side or demand-side effects for firms that receive positive ratings. The overall evidence is consistent with disappointing information being a significant motivator of firm behavior.
We are grateful for helpful comments by Simon Board, Vanessa Burbano, Christopher Carpenter, Barton Hamilton, Ginger Jin, Thomas Koch, Phillip Leslie, Steve Lippman, Moritz Meyer-Ter-Vehn, Lamar Pierce, Ashley Swanson, and seminar and conference participants at the American Economic Association Annual Meeting, American Society of Health Economists Meeting, Econometric Society Asian Meeting, International Industrial Organization Conference, NBER HealthCare program meeting, University of Texas Austin, UCLA, UC Irvine, UC Berkeley and the Olin School of Business at Washington University at St. Louis. We are thankful to Scott Klein and Jennifer LaFleur from ProPublica for their help with accessing the Dialysis Facility Compare data and to Leemore Dafny and Chris Ody for sharing the Medicare Cost Reports data with us.
Nicholas Ross provided outstanding research assistance. Special thanks to Al Roth’s Market Design blog for making us aware of the issue and the data.
I. Introduction Across many settings, persistent differences in firm performance are attributable to information asymmetries. In the absence of an informed marketplace, uncertainty over product quality enables inefficient (or low-quality) firms to continue to operate and prosper.
This dispersion in provider quality is often perpetuated by misaligned seller incentives, and lack of competition. In sectors such as healthcare where such variation in performance might be harmful to consumers, regulators have implemented information disclosure policies designed to increase product quality transparency. Prior research on these disclosure programs has mainly focused on how rank-ordered quality disclosure programs influence firm performance and consumer choice. However, in practice, many information disclosure programs are designed quite differently and focus on identifying sellers at the extremes of the quality distribution.1 In spite of their prevalence, there is little causal evidence in the literature that studies programs designed in this way and compares the relative effectiveness of positive and negative information disclosure. In this paper, we address this shortcoming by focusing on firm responses to positive and negative quality disclosure in the market for treating end-stage renal disease (ESRD), specifically, the dialysis industry.
ESRD expenditures in the United States account for nearly 1 percent of the entire federal budget2 and continue to grow every year. In spite of the importance of this industry, there is substantial unexplained variation in quality of the firms (dialysis facilities) responsible for treating ESRD. In response to this variation in quality, the Centers for Medicare and Medicaid Services (CMS) created a quality rating of dialysis facilities designed to expose the best and worst performing facilities. Since 2001, CMS has made available, through the Dialysis Facility Compare (DFC) tool available on its website, quality As an illustrative example of the type of program that targets the extremes of the quality distribution, consider the Blue Ribbon Schools Program administered by the Department of Education, which has recognized approximately 6,000 truly exceptional schools out of over 100,000 eligible schools. In contrast, the Environmental Working Group recently attempted to shame cereal makers responsible for the so-called “cookie breakfast” by compiling a list of the “Ten Most Sugary Cereals.” This figure comes from the fact that in 2009 Medicare was approximately 13% of the federal budget (The Henry J. Kaiser Foundation Medicare Premier (2009)) and ESRD was approximately 6% of the Medicare budget (United States Renal Data System (2011)). This is a lower bound since more federal money certainly goes to ESRD.
information on dialysis facilities in the form of three coarse categories based on a facility’s continuous risk-adjusted patient survival rate (averaged over the previous four years): worse than expected, as expected, and better than expected. The belief is that the revelation of this information will inspire quality improvements on the part of firms at the bottom of the quality ladder. We use this setting in this study to examine how firm performance responds to quality disclosure and try to identify the factors that mediate that response.
Estimating the causal impact of a disclosure program designed to highlight the firms at the tails of the quality distribution is challenging. Beyond endogeneity and omittedvariables-bias concerns, mean reversion undermines most attempts to study such programs.
For example, if a firm receives a negative rating (resulting from either poor underlying quality or bad luck),3 any subsequent improvement in its performance could simply be driven by mean reversion, as opposed to being attributable to disclosure.
To overcome these empirical challenges, we use a regression discontinuity design that exploits the fact that the coarse rankings were discontinuously assigned. Our empirical identification strategy relies on two unique institutional features. First, facilities with similar expected patient survival rates could receive drastically different quality ratings due to the discontinuous assignment process. Second, in 2008, CMS updated the statistical method used to classify facilities into performance categories. This revision, intended to better identify the best and worst performers, resulted in the percentage of facilities being classified as performing as expected dropping from 96 percent in previous years to 80 percent in 2008.4 Correspondingly, the number of facilities rated superior or inferior increased sharply. This change provides the statistical power to implement the regression discontinuity design. The underlying continuous scores that are used to construct the categorical rankings were made available to the public only in 2010. We observe the continuous scores and the coarse ratings after the fact, but at the time covered by our study, only the coarse ratings were visible to the market participants (both dialysis facilities and patients).
This is especially important when there is a random component to the measurement of product quality measurement.
This information was released on the DFC website towards the end of 2008. In Section IV.B, we expand on the timeline of events relevant to our analysis.
Our findings from the regression discontinuity design show that a facility receiving a worse than expected performance grade in 2008 (that is, based on patient survival data for 2004-2007) experiences an improvement in patient survival compared to a facility that is graded as expected. The magnitude of this improvement is economically meaningful, translating into a nearly 15-point improvement in the national risk-adjusted mortality ranking for the average facility rated worse than expected. Given an unconditional average death rate of 25 patients per 100 patient-years at such a facility, the quality disclosure would result in six fewer patient deaths per 100 patient-years on average. In contrast, we find that distinguishing between facilities that performed better than expected and as expected has little impact on patient survival rates. Further analysis suggests that the performance improvement of worse than expected facilities is largely a matter of strategic patient selection, with the greatest degree of improvement being registered among facilities prone to engaging in cherrypicking healthier patients.5 The improvement does not vary according to a facility’s ownership status—for-profit versus not-for-profit—but we find evidence suggesting that performance improvement is more likely in facilities operating in more competitive settings.
It is also found to be persistent only amongst firms with the strongest incentives to improve.6 We document demand-side responses to disclosure as well; facilities rated worse than expected subsequently treat fewer well-informed patients. Patients with prior dialysis experience are less likely to transfer into these facilities and patients that have seen a kidney specialist prior to dialysis are less likely to begin treatment at these facilities. All told, we find strong causal evidence that disclosure policies aimed at the extremes of the quality distribution impact both firm performance and consumer choice, but this impact is heavily skewed towards influencing the worst performers.
This study builds on an extensive body of research on the impact of mandatory information disclosure programs, such as ratings, rankings and report cards, across a variety Dranove, et al. (2003) makes a related point that the advent of surgical report cards is associated with physicians attempting to treat relatively healthier patients.
We identify such firms as those that had better performance in 2004 than in any other year in our 2004-2007 timeframe. Given that the performance measure for 2009 is computed using patient survival rates averaged across 2005-2008 (and excludes 2004), such firms will experience a mechanical decline in performance, conditional on their performance in 2008, when compared to firms for which 2004 was not the best year. We surmise that such firms face relatively stronger incentives to improve performance as a result.
of settings. Disclosure of quality-related information could be implemented voluntarily by sellers, carried out by third-party, or mandated by regulation. The purpose of voluntary disclosure is to create a norm-like institution to encourage “self-regulation” of firms (King and Lennox 2000; King, Lenox, and Terlaak 2005; Terlaak 2007; Pfarrer et al 2008).
Whether disclosure is voluntary or not, the objective is to reward top performers (Corbett et al, 2005), punish laggards (Short and Toffel, 2008), or both (Kleindorfer and Orts, 1998).
The literature examining the impact of disclosure showcases two primary results.7 First, at the consumer level, there is substantial evidence that quality disclosure leads to vertical sorting. In other words, consumers are more likely—all else being equal—to choose higherquality products after disclosure.8 The second major finding, which is also more closely related to this study, is that disclosure leads sellers to invest in product quality improvements. For example, Jin and Leslie (2003) find a significant decline in hospital admissions for food-borne illnesses after the adoption of restaurant hygiene grade cards in Los Angeles County. Seller response has also been shown to be heterogeneous with respect to dimensions such as the competitiveness of the local market. These findings have been shown across multiple sectors such as healthcare (Chen (2008); Cutler, Huckman, and Landrum, (2004); Pope (2009)), finance (Greenstone, Oyer, and Vissing-Jorgensen (2006); Knaup and Wolf Wagner (2012); Lau, Ng, and Zhang (2012)), and education (Elsbach and Kramer (1996)).
A small number of studies have focused on the effectiveness of good versus bad ratings at evoking the desired seller and consumer response. Chatterji and Toffel (2010) show that firms that receive poor ratings based on their toxic emissions exhibit a subsequent decline in their future emissions. In a study focusing on the introduction of hospital report cards in New York State in the early 1990s, Dranove and Sfekas (2008) find that hospitals that received negative news experienced a drop in demand, but positive reports brought no benefits. Scanlon et al. (2002) find a similar asymmetric consumer response to health plan ratings. However, in the absence of a natural experiment, one cannot rule out the possibility See Dranove and Jin (2010) for a comprehensive overview.
Examples include Hastings and Weinstein (2008) which shows evidence of such sorting among parents choosing public schools for their children while Jin and Sorenson (2006), Dafny and Dranove (2008), and Scanlon et al. (2002) show similar findings among enrollees choosing health plans.
that the results in some of these studies could be driven in part by mean reversion. As we discuss in Section III below, the research design employed in this study helps us overcome this identification challenge.
With respect to the existing literature discussed above, our study makes a number of important contributions. We study the effects of information disclosure in an industry of first-order importance. With chronic kidney disease becoming increasingly prevalent in the US, effective management of dialysis facilities is key to improving social well-being. Our study also has important implications that extend beyond the dialysis industry. First, we make use of a regression discontinuity design that allows for sharp identification of the causal impact of information disclosure while mitigating concerns about mean reversion.9 Second, the setting of the study enables us to examine the effect of disclosure programs that identify firms on both ends of the quality distribution. Our results thus shed light on the optimal design of disclosure programs, suggesting that negative information is stronger than positive information as a motivator of firm behavior and consumer choice. Finally because of the lag in the disclosure policy we are able to disentangle supply-side versus demand-side responses, which has not been done in the previous literature.