Serving the Unemployed: Do More Generous Social Insurance Programs Provide Better Quality Service?

Introduction

Recent headlines about administrative deficiencies in unemployment insurance underline the importance of bureaucratic performance (McDermott & Cowan, 2020; Rosenberg, 2020). Social policies are designed to dampen the impact of economic shocks on household finances by investing in social insurance and temporary assistance to families, smoothing income across time and space. These policies, however, can only effectively provide economic security to the extent that public agencies deliver program benefits promptly and properly to households in need. Bureaucratic performance shapes more than the effectiveness of social policy, however. It has political consequences.

As the face of government and the most frequent venue for citizen-state interactions, citizens’ perceptions of their government are shaped by their interactions with bureaucracy, and especially social policy administration (e.g., Moynihan & Herd, 2010; Moynihan & Soss, 2014; Soss & Schram, 2007). In aggregate, individuals’ interactions with public policies, benefits, and their administration generate feedback to reinforce or undermine political support for the program (Béland, 2010; Compton et al., 2019; Compton & Lipsmeyer, 2019; Jacobs & Weaver, 2015; Pierson, 1993, 2000; Weaver, 2010). Disparities in the quality of such interactions can produce feedback effects that further ingrain racial and ethnic differences in public policy outcomes (e.g., Michener, 2019).

Administrative performance also matters for accountability. It is not only citizens and clients whose preferences are influenced by bureaucratic performance. Evidence shows that politicians are aware of and pay attention to administrative performance data. Politicians attribute responsibility for performance, and tend to do so with a negativity bias by placing responsibility on bureaucratic leaders for poor performance (Nielsen & Moynihan, 2017). If it is the case that poor bureaucratic performance can hamper the effectiveness of public policy, and contribute to negative feedback among citizens and politicians, it is critical to understand the factors contributing to that performance.

Why do some bureaucracies implement social policy relatively better than others? One factor may have to do with support, either in terms of administrative resources or in terms of political will. Given that governments with more generous social policies have demonstrated a willingness to put political (and financial) capital into social support programs, it might follow that bureaucracies in those environments will perform better. Adequate resources are a necessary condition for high performing public organizations, along with autonomy and clear goals (Compton & Meier, 2017; Meier, 1997). Elected officials must allocate a sufficient budget for a public organization to acquire the human, technological, or fixed capital resources necessary to get their job done. With more substantial investment in social policy benefits, there may be greater institutionalized capacity for successful implementation as well.

Governance, however, is rarely consistent or rational. Many policy attempts may be largely symbolic or designed to fail (McConnell, 2010). The agency responsible for implementing a given policy may also face goal ambiguity or goal conflict (Chun & Rainey, 2005), having been delegated multiple and potentially competing tasks by elected politicians. Such situations may lead to lower bureaucratic performance (Carrigan, 2018). It is also possible that governments manage to achieve budgetary wins, or enact generous statutory entitlements, but then lose the focus or political capital to follow through with administering those programs well (Patashnik, 2008). This may be especially problematic since once established, distributive public programs are exceedingly difficult to reform or rollback (Pierson, 1993). Instead, social programs may be slowly restructured, subjected to market-based reforms, or systematically obstructed by unnecessary administrative burdens (Hacker, 2004; Hurd & Moynihan, 2018; Pierson, 1994). Such efforts may break the link between a willingness to investment in generous social benefits and an interest in their continuous administrative performance.

As such, it shouldn’t be taken for granted that generous social benefits are also well implemented. Political and fiscal capital are in limited supply, and both must be expended to produce high quality administration. It’s a feat of government effort to produce well-functioning–let alone high performing–administration, and it remains an open empirical question whether those governments with generous policies are also more likely to put effort into administering those benefits well. That is, are more generous social welfare programs also better administered? If social insurance or social transfer programs are politically supported (as evidenced by demonstrated policy generosity), then the administration of those programs might also be more supported.

In the following sections I introduce my research design and methodology to explore this research question. The empirical evidence offered here points to a positive association between policy generosity and administrative quality: more generous states make fewer administrative errors and that relationship is driven by their making fewer underpayments. In the concluding sections, I discuss the behavioral mechanisms that may explain these results and I discuss the implications of these results for domains and policy contexts.

Research Design and Methodology

To examine whether governments with more generous social welfare programs also better implemented, I use administrative data from U.S. states’ unemployment insurance programs, from 2002-2015. The United States unemployment insurance (UI) program was established by the Social Security Act (SSA) in 1935. The primary goal of this program is to provide short-term economic relief to the involuntarily unemployed through weekly benefit payments to temporarily replace lost wages. As a result of political compromises at the time of passage, state governments were granted substantial autonomy over many aspects of UI program design and administration. Each of the 53 UI programs in the US today vary in their administrative structure, the sectors covered, qualifying requirements, eligibility rules, disqualification rules, weekly benefit amount, waiting period prior to first payment, duration of benefit payments, seasonal provision, and their financing structure (Blaustein, 1993).^[1]

Funding of regular UI benefits is done through a complex tax-credit scheme, paid mostly by employers and supplemented by federal funds under specific circumstances. Employers pay two taxes: one into to a state account at a tax rate determined by their “experience rating,” and one variable tax-rate into a federal account which provides administrative funds, grants and loans to states, and certain benefits payments in times of unusual demand. Funding for the administration of state UI programs is provided through a small federal payroll tax collected by the Internal Revenue Service, and then redistributed to state governments according to anticipated need.^[2]

In addition to meeting program benefit and funding rules, states must also meet federal performance standards. To monitor state-level administration, the U.S. federal Office of Unemployment Insurance, within the Department of Labor, implemented an improper payment detection system beginning in the 1980s, called the Benefit Accuracy Measurement (BAM) program. Subject to these federally mandated procedures, each state is required to randomly audit and re-investigate a sample of paid and denied claims, each week, and report aggregate performance indicators monthly.^[3] If selected for audit, every step of a claim determination process is re-investigated, and documentation is (re-)collected on the applicant’s prior work experience, socio-demographic background, and, crucially, on the presence, monetary consequence, and source of any errors made in the original processing of the claim application. The frequency of identified administrative errors determined to be the fault of a State Workforce Agency (SWA) employee is my primary outcome of interest.

Bureaucratic errors may be defined as “any deviation from an intended outcome that is mandated by either law or organizational rules” (Bullock, 2014). According to the Department of Labor, “administrative responsibility” for an error in Unemployment Insurance is defined as any error for which the SWA “was either solely responsible or shared responsibility with claimants, employers, or third parties, such as labor unions or private employment referral agencies.” This includes fraud, nonfraud recoverable overpayments, nonfraud nonrecoverable overpayments, official action taken to reduce future benefits, and payments that are technically proper due to finality or other rules (Office of Unemployment Insurance, 2018). Between 2002 and 2015, an agency-responsible error was identified in 7.4% of the 706,629 individual-level claims audited nationwide.^[4] To measure the frequency of bureaucratic errors in the processing of unemployment insurance in the US, I use three dependent variables. Summary statistics for all variables are included in Table 1.

First, the number of detected agency errors, aggregated annually, represents the overall accuracy of claims processing by a state’s UI agency. All else equal, fewer detected agency errors indicate relatively better public service. Not all errors, however, are equal in their impact on claimants or policy outcomes, so I also count the number of agent errors resulting in the overpayment of a UI claim (overpayment errors) and the number of agent errors resulting in underpayment of a claim (underpayment errors).^[5] An overpayment error means that a household in economic need received a larger benefit payment than state policy allows. The impacts of an over-payment on both households and the effectiveness of a policy may be very different than those of an under-payment error. When a household in need receives a smaller benefit payment than policy allows, it does not receive its entitled level of support from a social welfare program. Given that households applying for UI are economically vulnerable, widespread underpayment errors may undermine the capacity of the policy to provide economic stability. Similarly, the consequences for policy feedback may be quite different if an administrative error results in an over- versus under-payment to a client. It’s therefore important to examine the nature of the administrative errors that occur, as well as their overall frequency.

To assess the relationship between a state’s willingness to invest in a UI program and its administrative performance, a measure of each state’s UI program is needed. Social welfare “effort” is typically measured as the ratio of public expenditures on social policy (in aggregate or by program) divided by total government spending or GDP (for a discussion, see Olaskoaga et al., 2013). Social welfare “generosity,” on the other hand, is operationalized in terms of institutionalized entitlements: replacement rates (e.g., Pallage et al., 2013; Scruggs, 2006; Scruggs & Hayes, 2017; Wenzelburger et al., 2013), recipiency or take-up rates (e.g., Otto, 2018; Pfeifer, 2012), tax benefits (as tax expenditures, including deductions, credits, preferential rates, deferred tax obligations, and exclusions of income from taxation) (e.g., Hacker, 2004; Howard, 1997), or as a composite index of the above (see Kunißen, 2019; Kvist et al., 2013).

Generosity here is operationalized by states’ annual replacement rates, because they are direct indicators of the benefits issued to the target population.^[6] States’ annual replacement rates are measured as the ratio of the average weekly UI benefit paid to the average weekly wage of an insured worker. This variable has an in-sample range of .24 to .55, with a mean of .36. This means that, on average, state UI program benefits replace 36% of the average eligible worker’s prior wages.

Because accuracy in unemployment insurance processing is the product of many factors– economic, political, and organizational– several control variables are included in models presented here. To account for organizational factors, I first include a measure of administrative resources, which is the number of real US dollars spent on administrative costs divided by the number of regular state UI program claims per year. UI administrative costs are calculated annually and transferred by the Department of Labor according to the Federal Unemployment Tax Act. UI administrative costs include all personnel and support expenses, and excludes all postage costs, administrative costs associated with Employment Services, and administrative costs associated with veterans employment programs and Bureau of Labor Market Information programs. To further account for the wealth of a state, I include GSP per capita measured in thousands of real US dollars.

Next, a control is included for the overall work load, which is a count of all claims made in a state year, logged. These measures account for the overall size of the program, and the intensity of the current workload, all of which might affect individual bureaucratic processes.^[7] I include the rate of women claimants, female applicants, because previous work has shown that women experience lower levels of service quality in the administration of unemployment insurance (Ryu et al., 2012; Wenger & Wilkins, 2008). I also use an inverse normalized Herfindahl index to capture the racial and ethnic diversity of a state’s UI claimants in each year. Larger values of this measure represent more clientele diversity. This is included because bureaucratic errors are more likely in claims filed by minority clientele (Ryu et al., 2012), and because racial diversity has played a fundamental role in the design and administration of state social welfare institutions (Lieberman, 1998; Lieberman & Lapinski, 2001; Ryu et al., 2012; Soss et al., 2008).

To control for political influences that might affect policy or bureaucratic behavior, government liberalism is included to account for the role of government ideology in determining policy or administrative priorities (Hibbs, 1977; Hicks & Swank, 1984). This is measured as a weighted average of the ideology scores for each chamber of the state legislature and the governor (Berry et al., 1998, 2007, 2010), and is constructed on a zero to 1 scale, with greater values representing a more leftist or liberal ideology. This measure is lagged by one year, with the expectation that political or policy influences on outcomes will be observed only in subsequent, not concurrent, periods. Legislative professionalism is used to control for quality or competence of policy-making in a state. This measure is largely time invariant, and was measured in 2009 (Squire, 1992, 2017).

To estimate the relationship between annual state error counts, $y_{jt}$, where $j$ indicates the state and $t$ indicates time, and social welfare generosity, I use a Poisson regression model (King, 1988). Because these error counts are drawn from audit samples of heterogeneous size, all Poisson regressions here include an offset term— a logged count of all audited claims in the same period. State fixed effects are included in each model to account for time-invariant differences in (1) policy or political environment across states that might shape the probability of administrative errors occurring, and (2) the capacity or quality of the state-level auditing agency responsible for detecting and reporting claims.^[8]

Administrative performance is thus modelled here as a function of (1) a vector of state-level political and administrative variables observed annually, $\textbf{x}^{\prime}_{jt}$, (2) a vector of corresponding estimated regression coefficients $\mathbf{\beta}$, (3) a vector of binary indicators for each state $\textbf{z}^{\prime}_{j}$, (4) a corresponding vector of regression coefficients state fixed-effects $\mathbf{\gamma}$, and (5) a common disturbance term $\epsilon$. This model specification is represented as: $$\text{log}(\text{E}(y_{jt}| \textbf{x}_{jt},\textbf{z}_{j})) = \textbf{x}^{\prime}_{jt}\mathbf{\beta} + \textbf{z}^{\prime}_{j}\mathbf{\gamma} +\text{log}(audit \text{ } size_{jt}) + \epsilon.$$

Results

Results from my analysis are shown in Table 2. In column 1, labeled All Agency Errors, results from a model of the count of all detected agency-responsible errors are reported. These findings support the expectation that more generous states also process UI claims with greater accuracy. Specifically, the negative and significant coefficient on Replacement Rate indicates that state governments with greater investment in generous unemployment insurance benefits have also ensured better performing State Workforce Agencies— UI is administered with significantly fewer processing errors.

Interestingly, the availability of administrative resources, measured as in thousands of real US dollars per claim processed in a state-year, is not significantly associated with the frequency of agency errors. The coefficient on this estimate is negative, as might be expected, but it is not significant. Agencies also make more processing errors when facing a greater work load and where women constitute a larger percentage of clientele. Also, states with either greater legislative professionalism or a more liberal government make more errors overall.

Turning now to models 2 and 3, which respectively report models of agency-responsible overpayment errors and agency-responsible underpayment errors, additional results emerge. First, states with more generous UI benefits tend to make more overpayment errors and fewer underpayment errors. Although more generous states make fewer errors overall (as shown in the all agency errors results from model 1), all else equal, that relationship appears to be driven by those states making fewer underpayment errors (as model 3 reports). Interestingly, the relationship between administrative resources and over- or under-payments is again insignificant. Other control variables for organizational and administrative environment have similar effects on both over- and under-payment errors. Workload and the female claimant rate may reduce service quality, but it seems do so in an arbitrary way, with no bias towards over- or underpayment. Perhaps agents in these environments simply make more mistakes, of all type.

To better illustrate the substantive effect of these findings, simulated agency error rates are reported in Figures 1 and 2. Figure 1 represents the overall agency error rate, and Figure 2 reports both overpayment and underpayment error rates, as labelled. Key results from Table 2 are reinforced here. First, as shown in the first figure, states with more generous UI programs make significantly fewer errors. A move from the minimum to the maximum observed replacement rate (from 24 to 55% of prior wages replaced by benefits) is associated with a significant decrease in the rate of agent errors from .11 to .04. That’s an estimated decline in agency responsible errors of more than 60%. Put differently, a 1% increase in a state’s replacement rate is significantly associated with a 1.1% decrease in agency-responsible administrative errors.

Figure 1.Predicted Agency Error Rates, All Agency Errors

Note: Figure represents 95% confidence intervals for the predicted rate of agency-responsible errors in claims processing across the sample range of replacement rate. Simulation to produce these predictions uses model results reported in Model 1 of Table 2. These predicted incidence rates are generated with the $margins$ suite of commands in Stata 17 with all variables, except the offset term, allowed to vary as observed. Incidence rates are estimated based on the mean annual audit sample size, 894.

Figure 2.Predicted Agency Error Rates by Type

Note: Figure represents 95% confidence intervals for the predicted rate of agency-responsible errors in claims processing across the sample range of replacement rate. Simulation to produce these predictions uses model results reported in Models 2 and 3 of Table 2. These predicted incidence rates are generated with the $margins$ suite of commands in Stata 17 with all variables, except the offset term, allowed to vary as observed. Incidence rates are estimated based on the mean annual audit sample size, 894.

Figure 2 sheds more light on this relationship. Here, it’s shown that more generous states have slightly higher overpayment error rates and starkly lower underpayment rates. The solid upward sloping line in this figure represents the agency responsible overpayment error rate, which is estimated to increase from .033 to .041 over the in-sample replacement rate range, which is a 24% increase. Put differently, a 1% increase in a state’s replacement rate is significantly associated with a .28% increase in agency-responsible errors resulting in an overpayment.

The dashed downward sloping line in this Figure 2 represents the agency responsible underpayment error rate, which is estimated to decrease from .028 to .014. That’s equivalent to a 50% decline. Put differently, a 1% increase in a state’s replacement rate is significantly associated with a .85% decrease in agency-responsible errors resulting in an underpayment.

In sum, these results suggest that more generous states make fewer administrative errors. Despite a slight uptick in overpayments in more generous states, the more substantial decline in underpayments appears to drive an overall decrease in observed administrative errors. If UI replacement rates reflect an institutionalized commitment to more generously protecting individuals from economic insecurity, that commitment seems also to be evident in the types of administrative errors bureaucrats make. In the following sections I will discuss some possible behavioral explanations for these results.

Discussion & Conclusion

In sum, the evidence from this analysis suggests a positive association between social policy generosity and administrative performance, as measured by the frequency of bureaucratic errors. State governments which put forth greater effort in the generosity of unemployment insurance benefits also administer their programs with fewer errors and greater accuracy. The prioritization of generosity of social policy seems to go hand in hand with the prioritization of quality in administering those programs. In addition to the overall association between generosity and bureaucratic performance, there is an interesting pattern in the nature of the errors made, and their apparent impact on clientele. More generous states make fewer administrative errors, but underlying that trend is the tendency of more generous states to commit slightly more overpayments and substantially fewer underpayments. The persistent insignificance of administrative resources in these results may suggest that the link between generosity and performance is about more than money. It seems that priorities or values reflected in the design of unemployment insurance policy are also reflected in the administration of those programs.

What explains the pattern of administrative errors observed here? Is this pattern of administrative errors the result of intentional choices by street-level bureaucrats utilizing discretion (Lipsky, 1980), or is it the result of structural, organizational, or technological features in state governments or state workforce agencies (Peeters, 2020; Widlak & Peeters, 2020)? The research design and data available here are ill equipped to uncover the mechanisms driving these results. With the available data, I cannot say much about what behavioral, procedural, or other administrative differences or changes that may underlie the observe associations between policy generosity and administrative accuracy. I can, however, offer some possible explanations that should be considered in future work.

One possible explanation may have to do with a concern about efficiency or the misuse of public resources, which could lead to individuals or organizations valuing payment accuracy relatively more. Such values may be stronger in states less committed to generous UI benefits. Either due to ideological motivations or the realities of state budget constraints, street-level bureaucrats with a greater concern about efficiency or misuse of public resources may be more motivated to reduce any form of administrative error to ensure payment accuracy. If this were the only explanation, it would be expected that the rate of all error types would be greater in states with more generous policies. The results reported in Figure 1 do suggest that this is not the case, so this may not be the (only) explanation.

Second, street-level bureaucrats may not prioritize payment accuracy, per se, but may rather be motivated to reduce state expenditures on UI benefit payments. It’s plausible that state governments and administrations in areas with greater popular concern with reducing expenditures would also tend to enact less generous UI benefit rules. Perhaps due to ideology or state budget realities, employees in such contexts utilize available discretion to underpay or wrongly deny UI claims in an intentional or unintentional effort to reduce state expenditures. If this were the only explanation, I would expect to see higher overpayment and lower underpayment rates in more generous states compared to less generous states. The results in Figure 2 may support this explanation.

Third, rather than differences in concern about payment accuracy or reducing expenditures, street-level bureaucrats may view their relationship or service to clientele differently. In states with more generous program rules, it could be that street-level bureaucrats are more likely to sympathize with the problems and needs of claimants or see clients as deserving of assistance (e.g., Jilke & Tummers, 2018), or for a variety of reasons they may be more likely to identify with and represent their clients’ needs (e.g., Selden, 1997). If so, it might be expected that in more generous states, bureaucrats would be more likely to err on the side of generosity– they’d be more likely to erroneously overpay and less likely to erroneously underpay claims. Figure 2 is consistent with this explanation.

The above are just a few of the possible explanations for the results reported in this article, and there are surely additional rival explanations. Future work should endeavor to theorize on and empirically examine the behavioral, organizational, or political underpinnings of the results reported in this article, as well as the features of the administrative operating environment which may contribute to better outcomes. As an example, prior work has shown that the effectiveness of social policy spending is less in contexts with greater social capital (Compton, 2018) so, how is social context related to administrative errors? Prior work has also shown that process inclusivity is key in producing successful policy outcomes (Compton et al., 2019), so, does process inclusivity contribute to administrative accuracy? Because social capital or process inclusivity may plausibly shape both policy design and implementation, there are interesting questions to be explored.

Future work should also examine additional indicators of administrative performance, because accuracy in processing claims is just one indicator of successful governance or administration (Chudnovsky & Peeters, 2021; Compton et al., 2019). As an example, it’s well documented in the US that racial and ethnic minorities receive lower quality service from bureaucracies and government (see Ryu et al., 2012; Desante, 2013), so it is worthwhile to consider whether more generous social welfare systems are also more likely to ensure greater equity in public service.

Finally, future work should examine these questions in a broader set of institutional, political, or policy contexts. There are a few reasons why public administration in the US, and unemployment insurance in particular, may be a most likely context in which to observe politicized and value-driven explanations for administrative quality. First, compared to other industrialized democracies, the US has a more pluralist concept of society and a more politicized administration (Peters, 2021). In contexts with less politicized administration or stronger societal norms of reciprocity, for example, the determinants of administrative errors may be quite different.

Second, the way UI programs are funded in the US invites conflict between labor representatives, administration, and employers (Becker, 1981; O’Leary & Wandner, 1997). Because employers’ tax rates are directly determined by their experience with layoffs and unemployment in the past, they are incentivized to provide false evidence, to appeal UI determinations, and to reorganize businesses to evade accurate experience-rating. Labor groups have historically opposed experience rating, arguing that the system encourages employers to restrict employee benefit rights and to unjustifiably challenge claims to keep charges and tax rates down (O’Leary & Wandner, 1997). This conflict and the role that employers play in providing documentation relevant to claimants’ benefit payments may mean that unemployment insurance programs are more politicized and contentious than other social transfer programs run by state governments (like TANF or SNAP). Greater politicization may mean that street-level bureaucrats are more likely to hold differing levels of commitment or views on the value of unemployment insurance, which may contribute to comparatively greater variation in administrative quality. These features may mean that the US, and unemployment insurance specifically, is a more likely context in which to see this pattern of administrative errors.

Acknowledgements

This project received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement no. 694266). Many thanks to participants in the TAMU Bush School Quantitative Brownbag Series, the EPSA 2020 Politics of Bureaucracies panel, the 2019 CAGE Conference at Warwick, and the KJPS editor and anonymous reviewers for their helpful and constructive feedback. All remaining errors are my own.