Second-Order Devolution and the Implementation of TANF in the U.S. States

Abstract

Welfare reform gave the U.S. states the opportunity to engage in second-order devolution (SOD), which allows local governments to exercise more discretion in the implementation of the Temporary Assistance for Needy Families (TANF) program. Proponents of welfare decentralization insist that local governments better understand the needs of the poor and are therefore able to implement TANF more effectively. Nevertheless, opponents argue that decentralization could lead to a "race to the bottom" and, thus, SOD might lead to more restrictive TANF implementation. We investigate these competing claims by examining how differences in decentralization affect (1) TANF caseload decline, (2) the use of sanctions, and (3) several work-related outcomes among recipients. Based on a series of state-level analyses, we find that SOD states experienced a greater degree of caseload decline than non-SOD states. In addition, SOD states were more likely to use punitive policy tools, such as TANF sanctions. However, we also find that SOD states display marginally better TANF performance, as reflected in higher rates of employment exits and earnings gains among TANF recipients. Thus, we find support for both sides of the decentralization debate. [PUBLICATION ABSTRACT]

Headnote

ABSTRACT

Welfare reform gave the U.S. states the opportunity to engage in second-order devolution (SOD), which allows local governments to exercise more discretion in the implementation of the Temporary Assistance for Needy Families (TANF) program. Proponents of welfare decentralization insist that local governments better understand the needs of the poor and are therefore able to implement TANF more effectively. Nevertheless, opponents argue that decentralization could lead to a "race to the bottom" and, thus, SOD might lead to more restrictive TANF implementation. We investigate these competing claims by examining how differences in decentralization affect (1) TANF caseload decline, (2) the use of sanctions, and (3) several work-related outcomes among recipients. Based on a series of state-level analyses, we find that SOD states experienced a greater degree of caseload decline than non-SOD states. In addition, SOD states were more likely to use punitive policy tools, such as TANF sanctions. However, we also find that SOD states display marginally better TANF performance, as reflected in higher rates of employment exits and earnings gains among TANF recipients. Thus, we find support for both sides of the decentralization debate.

INTRODUCTION

THE PASSAGE OF NATIONAL welfare reform legislation in 1996 resulted in a significant transfer of policymaking authority from the federal government to state governments. This process has often been referred to as "first-order" devolution. However, scholars have increasingly begun to recognize that in designing their new welfare systems under the Temporary Assistance for Needy Families (hereafter, "TANF") programs, many states have engaged in second-order devolution (hereafter, "SOD") by shifting important policymaking discretion further downward to local governments (Whitaker and Time 2001; Gainsborough 2003). Currently, fourteen states devolve significant TANF authority and responsibility to local governments or regional boards.

As with first-order devolution, the practice of SOD under TANF has been the subject of debate. Proponents of welfare decentralization insist that local governments better understand the needs of the poor and are therefore better able to provide more appropriate services to their welfare clients. Opponents argue that decentralization to the local level could lead to a "race to the bottom" - much as is claimed to have occurred at the state level with respect to welfare benefit levels - because of the actual or perceived need by local officials to avoid the immigration of the poor and the loss of tax revenue (Bailey and Rom 2004; Peterson and Rom 1989). Rather than affecting benefit levels, for which state governments remain responsible, SOD might lead to stricter TANF implementation among local governments. This possibility is enhanced due to the fiscal relationship between states and local implementing jurisdictions in SOD states, as we explain below.

Existing scholarship on SOD under TANF has focused on the increase in discretion to local government and how this might enhance variation in policy outcomes or contribute to policy success or stringency across local jurisdictions (Kelleher and Yackee 2004; Cho et al. 2005; Fording, Soss, and Schräm 2007; Soss, Fording, and Schräm 2008). However, these studies are limited to single state studies. To date, no systematic analysis has been conducted of the impact of administrative structure on the implementation of welfare policy that compares centralized states with SOD states. We conduct an analysis of the effects of SOD across the states by exploring how differences in administrative structure due to SOD affect the implementation of five important TANF outcomes often addressed in the welfare literature: TANF caseload decline, the use of sanctions for noncompliance with TANF work requirements, welfare exits through employment, job retention, and the earnings of TANF clients.

CENTRALIZATION VERSUS DECENTRALIZATION

As U.S. governments have grown in response to the demands of growing populations, they have often shared authority and responsibility with other levels of governments (Kettl 2000; Agranoff and McGuire 2001). In redis - tributive policy, the question of which government is best suited to provide welfare services has been one of the most prominent debates since the passage of national welfare reform in 1996, when the federal government gave more discretion to state governments in designing and implementing welfare policies.

Proponents of devolution argue that the problems and demands of constituencies vary across states and are conditioned by unique state environments. Therefore, federal government cannot easily identify the needs of a state and design appropriate policies for meeting a states unique demands. States benefit from the flexibility offered by decentralization because state leaders likely better understand their own problems and demands of their residents, thereby providing more appropriate services (Whitaker and Time 2001; Dye 1990).

On the other hand, opponents of devolution argue that giving more discretion and responsibility to state or local governments in welfare policy design and implementation could result in a race to the bottom phenomenon. In other words, state or local governments would compete to reduce welfare generosity in benefit levels, eligibility, and penalties for noncompliance with TANF rules to avoid the in-migration of the poor and the loss of tax revenue. Many studies have investigated inter-state competition and the race to the bottom at the state level, and while there is disagreement concerning the magnitude of the effect, most studies find some evidence of a race to the bottom (Peterson and Rom 1989; Enchautegui 1997; Figlio, Kolpin, and Reid 1999; Volden 2002; Berry, Fording, and Hanson 2003; Bailey and Rom 2004; Bailey 2005; Dejong et al. 2006). If the race to the bottom works as expected, devolution might result in undesired outcomes.

Second-Order Devolution

Although many studies and debates have focused on first-order devolution, relatively few studies have examined SOD. Despite its association with welfare reform, it is important to note that SOD existed prior to TANF under the Aid to Families with Dependent Children program in the administration of welfare policy.1 Prior to 1996, fifteen states administered the AFDC program locally while thirty-five states administered AFDC at the state level (Gainsborough 2003). Under TANF, states have the option to grant administrative authority and responsibility to local governments just as the federal government grants authority and responsibility to state governments in designing and implementing welfare policy (Adkisson 1998). After the passage of national welfare reform in 1996, SOD accelerated (Fording, Soss, and Schram 2007).

Among the fifteen states that administered AFDC locally, seven states seem to have changed little in the division of authority between states and localities. These localities serve as functional administrative tools of states with relatively little formal policymaking discretion. On the other hand, eight of these fifteen states grant significant authority and responsibility to localities under TANF: California, Colorado, Maryland, Minnesota, New York, North Carolina, Ohio, and Wisconsin (Pandey and Collier- Tenison 2001; Gainsborough 2003). In California, counties were given more opportunities to design the rules of the program, including such policies as the extension of time limits, the number of required work hours, and exemptions from work requirements. In Colorado, counties receive block grants from the state and the state enforces a maintenance-of-effort (MOE) requirement for counties. Counties also have a great deal of discretion in deciding other aspects of programs. In Maryland, counties are responsible for assessing applicants for additional benefits and submitting the candidate list of local directors of social services. They are also rewarded for savings in their welfare program. In Minnesota, counties determine education and training services and case management strategies, including sanction policies and time limits. In New York, counties have discretion in designing work requirements and other rules affecting individual behavior. In North Carolina, some counties follow a state plan, but some can develop their own plan and define eligibility and benefit levels. In Ohio, counties receive block grants and can design their own programs, including services and eligibility for services. In Wisconsin, responsibility for welfare devolved to the local level with a competitive bids system. In most counties, the county governments administer welfare reform (Gainsborough 2003). In sum, although these states display variation in the degree of devolution, they enjoy considerable authority over spending and establishing some aspects of policies, and they enjoy significant discretion over work requirements, time limits, additional benefit eligibility, and sanctions (Fording, Soss, and Schräm 2007).

Among the thirty-five states that administered AFDC directly, twentythree states continue to administer TANF without devolution. Six states have engaged in a small degree of devolution to the local level, but six other states granted significant TANF authority to local or regional advisory boards rather than devolving responsibility to local governments. These regional boards are usually composed of representatives from local government, local business, community groups, and service providers, and they are responsible for designing and delivering TANF services. Like states devolving significant authority to counties, although variation exists in the power these boards have over TANF policy, they have considerable authority over designing and delivering TANF services (Gainsborough 2003; Fording, Soss, and Schräm 2007).

Thus, combining the eight states that devolve significant authority to county governments, along with the six states that devolve significant policy discretion to regional workforce development boards, there are fourteen states that currently devolve significant discretion in TANF implementation to local jurisdictions of some kind. These fourteen states are indicated in Figure 1 and are the states that we refer to as "SOD states" throughout this article. As can be seen, SOD states are not concentrated in any particular region of the country, nor do they appear to be dominated by ideologically conservative or liberal states. However, SOD states do tend to be larger states and include the four largest states (in population) among their ranks.

Both supporters and opponents of devolution predicted that granting more discretion and responsibility to state governments would result in SOD (Whitaker and Time 2001; Gainsborough 2003). But has this affected policy outcomes? The academic literature suggests that local factors affect policy outcomes more strongly in a decentralized structure because first-order devolution increases state discretion and SOD increases county discretion (Francis 1998). Local factors should produce more variation in policy outcomes both across states and within states through increased discretion because local factors are different from state to state and county to county.

In addition to providing more variation, there is reason to expect that TANF implementation might take a fundamentally different form in SOD states as well. Yet, despite the potential significance of SOD under welfare reform, only a few studies directly examine the potential effects of SOD on TANF implementation. Both Cho et al. (2005) and Kelleher and Yackee (2004) examined the potential beneficial impact of SOD on TANF implementation in North Carolina. This state provides a unique opportunity to evaluate SOD because its counties had the option to design their own TANF program or rely on the state plan. The results of the two studies provide mixed support for the benefits of SOD. Although officials in SOD counties were more likely to perceive that TANF implementation was effective (Cho et al. 2005; Kelleher and Yackee 2004), comparisons on objective measures of program success between SOD and non-SOD counties were much less conclusive (Kelleher and Yackee 2004).

In their study of SOD in Florida, Fording, Soss, and Schräm (2007) examined the potential punitive effects of SOD by examining local variation in TANF sanctioning practices. They found that TANF clients in Florida were sanctioned at a relatively high rate compared to other states, although there was considerable variation in sanctioning rates across counties. In addition, the local socioeconomic and political environments were found to exert strong effects on sanctioning outcomes. The authors attribute both the degree of punitiveness in implementation and the variation in implementation outcomes to the fact that Florida's TANF program is characterized by significant decentralization. Nevertheless, this conclusion is merely speculative because they do not explicitly examine sanctioning outcomes in other non-SOD states. In sum, there is good reason to believe that SOD might have a significant effect on TANF implementation, but relatively little systematic evidence exists that directly examines this question.

COMPETING THEORIES OF THE EFFECTS OF SECOND-ORDER DEVOLUTION: EFFICIENCY VS. STRINGENCY

Past research suggests that SOD might influence the implementation of TANF through the increased discretionary power of local administrators and case managers. The key to understanding exactly how this discretion can affect TANF outcomes lies in identifying the differences between SOD and non-SOD states in the implementation environment. The literature on intergovernmental relations suggests that several distinguishing features of SOD might lead welfare implementation to take a different form than in non-SOD states.

The first consideration is the fact that in SOD states, local actors have greater flexibility in designing local programs. As proponents of decentralization argue, local governments better understand the needs of their poor population as well as the social and economic environments that surround them. Policymakers in SOD states can provide more appropriate services for welfare recipients and perhaps modify program requirements in such a way to maximize their community's resources. Hence, the increased discretionary power granted local administrators and case managers might lead to greater program success in SOD states compared to non-SOD states. We term this potential effect of SOD the efficiency thesis and test this hypothesis by examining the impact of SOD on several TANF outcomes commonly studied as indicators of successful implementation in the TANF literature.

Critics of decentralization in welfare administration argue that rather than fostering program success, SOD is likely to lead to a more punitive approach to TANF implementation (Fording, Soss, and Schräm 2007; Soss, Fording, and Schräm 2008). We refer to this possibility as the stringency hypothesis. This outcome could occur for two reasons. The first factor to consider is the fiscal relationship between states and local implementing jurisdictions in SOD states. Under SOD, some local jurisdictions receive block grants to fund TANF implementation, much as the federal government now disperses TANF block grants to the states. Block grants exert financial pressure on welfare spending because the county (or regional board) must pay 100 percent of each dollar of welfare costs beyond the state contribution (Brueckner 2000). In this way, block grants impose a financial burden upon counties because counties are responsible for all spending if the TANF caseload increases beyond some point. Counties in some SOD states are also rewarded for savings in their welfare program. Hence, based on these incentives, counties in SOD states might be more likely to establish tougher TANF policies and implement policies more strictly to reduce the TANF caseload.

A second reason that SOD counties might implement sanctions in a stricter fashion is the political pressure and competition among counties to avoid the in-migration of the poor and the loss of business and tax revenue. Migration of the poor across counties within states might occur more often and more easily than migration across states due to proximity. Even if it does not occur, the fear of welfare migration can still impact policy decisions (Bailey and Rom 2004). If this is the case, counties in SOD states should produce more punitive implementation outcomes due to the influence of such concerns on the exercise of local discretionary power of local administrators and case managers.

Hypotheses

To test for the effects of SOD on TANF implementation, we examine the impact of SOD on five different implementation outcomes. First, we hypothesize that SOD states will experience greater caseload decline. This is a natural implication of both the efficiency thesis and the stringency thesis presented above. According to the efficiency thesis, SOD might result in higher rates of successful employment exits from TANF by welfare recipients due to the enhanced ability of local administrators to tailor implementation styles to local labor markets. If this is the case, then it naturally follows, ceteris paribus, that SOD states will experience greater caseload decline.

The stringency thesis also leads to a prediction of greater caseload decline under SOD, but for very different reasons. According to the stringency thesis, local administrators and case managers have greater incentive to rely on formal policy tools, such as sanctions, to discourage current welfare recipients from continuing to receive TANF benefits. In addition, if SOD does indeed result in a more punitive style of TANF implementation, it is likely to be reflected not just in sanctions, but in the inherent use of discretion by frontline welfare staff, who rely on less formal strategies to discourage TANF receipt. If this is the case, then in addition to the indirect effects of SOD on caseload decline predicted by the first two hypotheses, we might also expect SOD to have a direct and negative impact on the size of the TANF caseload by encouraging welfare exits for reasons other than sanctions or employment. Based on this logic, we test the caseload decline hypothesis, which predicts that SOD states will have higher caseload decline, on average, than non-SOD states under welfare reform.

The Caseload Decline Hypothesis: SOD States are expected to have higher caseload decline than states with centralized administration.

We also examine the impact of SOD on four additional TANF implementation outcomes that represent different causal paths for the possible effect of SOD on caseload reduction. The stringency thesis predicts that under SOD, local TANF administrators are more likely to rely on punitive policy tools to reduce TANF caseloads. To test this proposition, we examine the impact of SOD on the use of TANF work sanctions, which represent one of the most significant punitive policy tools available to case managers under welfare reform. Federal law requires that TANF clients be subject to penalty, or sanction, if they fail to meet TANF work participation requirements. States have some flexibility in determining the severity of TANF sanctions, and these penalties can range from a minimum of a partial benefit reduction for the head of the household to a temporary cessation of all TANF benefits (and optionally Food Stamps) for the entire TANF family. We test what we term the sanction stringency hypothesis, which predicts that SOD states will sanction TANF clients at a higher rate than centralized states.

Sanction Stringency Hypothesis: SOD States are expected to have higher sanction rates than states with centralized administration.

According to the efficiency thesis, we should expect TANF implementation to be more successful in SOD states than in states utilizing a more centralized implementation process. Although scholars have conceptualized TANF program success in a variety of ways, by far the most common outcome of interest is successful job placement. We therefore test what we term the employment improvement hypothesis, which predicts that SOD states will have higher rates of successful welfare exits through (1) a higher rate of TANF exits due to employment, (2) a higher level of job retention for TANF clients, and (3) a larger gain in earnings among TANF clients, on average, than non-SOD states.

Employment Improvement Hypothesis: SOD states are expected to experience higher welfare exit rates through employment, higher levels of job retention, and larger earnings gains, on average, than states with centralized administration.

SECOND-ORDER DEVOLUTION AND CASELOAD DECLINE

We begin by testing the caseload decline hypothesis, which predicts that SOD states will exhibit higher rates of TANF caseload decline than centralized states. We test this hypothesis by utilizing state panel data for forty-seven states over the period 1980-2003, estimating the coefficients for equation 1 below.2 As we utilize state-panel data, the model is estimated by relying on panel corrected standard errors and includes fixed effects for states and years.

(ProQuest: ... denotes formula omitted.) [1]

The dependent variable, welfare recipient rate, is based on the AFDC/ TANF recipient rate and is measured in two ways. For our first model, we rely on the simple recipient rate, which is calculated as the number of AFDC/ TANF cases per 1000 state population. As an alternative, we also estimate equation 1 using the log of the recipient rate as the dependent variable (Blank 2001; Ziliak et al. 2000). As we describe below, this choice does little to alter our substantive conclusions.

Our hypothesis of interest concerns the impact of SOD on the welfare caseload. We measure SOD using an indicator of significant SOD, as defined by Gainsborough (2003) and reflected in Figure 1. For the fourteen SOD states, SOD is defined as the proportion of the year during which TANF was implemented. Consequently, this variable takes on a value of 1 for all stateyears during which TANF was in place throughout the entire year, a value of 0 for all state-years during which AFDC was in place the entire year, and a value between 0 and 1 for state-years during which TANF was implemented in mid-year.3

Although several states devolved some financial responsibilities to counties prior to welfare reform, our measurement strategy effectively assumes that SOD is unlikely to have had significant caseload effects in the absence of the policy tools (e.g. sanctions, time limits, etc.) available to local administrators and frontline staff under TANE Therefore, the remaining thirty-six centralized non-SOD states are coded as 0 for the entire period of the analysis. As our model includes state and year fixed-effects, we are thus able to control for state-specific trends in caseload levels, as well as the effects of national forces affecting caseload dynamics in SOD and centralized states alike. This lends a quasi-experimental dimension to our design, thereby mitigating concerns over causality.

The remainder of equation 1 consists of variables that control for various social, economic, and political determinants of caseload dynamics (e.g. Blank 2001; Fording 2001; Ziliak et al. 2000). To capture state economic need, we include the unemployment rate, and state per capita income. For each of these indicators, we expect that as the state's economic health improves, caseload decline should accelerate. To control for the state political environment, we include a measure of state government ideology, as constructed by Berry et al. (1998). Consistent with past research on welfare generosity, we expect that government liberalism will have a positive effect on the recipient rate.4

Although TANF reflected a significant departure from the AFDC program, during the early 1 990s, many of the reforms implemented under TANF were being implemented in states through federally-granted AFDC waivers. While the magnitude of their effect has been debated, past studies have consistently found that the implementation of AFDC waivers led to a decrease in caseloads (e.g. U.S. Council of Economic Advisors 1997; Ziliak et al. 2000). Therefore, we include AFDC restrictive waivers, which is measured for all states as the cumulative number of restrictive AFDC waivers implemented in that state (until the termination of AFDC in 1996). For our analysis, restrictive waivers include changes to AFDC that (1) limited exemptions for work requirements, (2) introduced time limits for welfare receipt, (3) strengthened sanctions for violations of program rules, (4) introduced a family cap on benefits for children born while in welfare, or (5) strengthened work requirements. We also include a separate dummy variable earnings disregard waiver) for waivers designed to strengthen work incentives by increasing the earnings disregard for AFDC clients. We expect that restrictive waivers should have a negative effect on the caseload, while the effect of earnings disregard waivers should be positive.5

Finally, we include the state minimum wage and a lag of the dependent variable welfare recipient ratej (]). The former takes on the value of the state minimum wage for state-years in which a state minimum wage is in place and is equal to the federal minimum wage for state-years in which a state minimum wage does not exist. We expect the variable to be negatively related to the welfare recipient rate. A lag of the dependent variable is included to model dynamics, as we expect that the effects of the independent variables might persist through time (e.g. Ziliak, et al. 2000).

Estimation and Results

Preliminary tests confirmed that our dependent variable (the AFDC/TANF recipient rate) is nonstationary. Indeed, the coefficient for the lagged dependent variable is approximately 0.90 when the model is estimated in levels (results not shown). Our concerns were also validated by more formal tests of stationarity designed for panel data.6 To deal with nonstationarity, we follow Ziliak et al. (2000) and estimate equation 1 in first-differences. We retain our fixed effects for states and years, which in the first-differenced version of the model now account for trends in welfare caseloads that are either state-specific or driven by national forces. After converting our model to first-differences and adding a lagged dependent variable, the estimation period now encompasses the years 1982-2003.

The coefficient estimates for this modified version of equation 1 are presented in Table 1. AU coefficient estimates are generated by OLS with panel corrected standard errors in parentheses (Beck and Katz 1995). The coefficients in the second column of results are multiplied by 100 for ease of interpretation. As the dependent variable for this model is the log of the recipient rate, these coefficients represent the expected percentage change in the dependent variable given a one-unit increase in the independent variable of interest.

As expected, caseloads appear to have been greatly affected by economic conditions, as both indicators of state economic health (unemployment rate, per capita income) are significantly related to the welfare recipient rate. Caseload dynamics also appear to be sensitive to the state political environment - as government ideology is positive and significantly related to caseload change - but only when the dependent variable is measured as the log of the recipient rate.

The effects of the policy measures are generally significant as well and in the predicted direction. As expected, the implementation of restrictive AFDC waivers had a negative impact on the AFDC caseload, confirming the findings of past studies of caseload decline (Ziliak et al. 2000). The stringency of state TANF policies has also had an important effect on caseload dynamics. This is reflected in the effects of TANF eligibility and TANF flexibility, each of which is found to be statistically significant and in the predicted direction. In contrast, neither of the policies designed to increase work incentives earnings disregard waiver and state minimum wage) had a significant effect on caseload change. However, the coefficients are in the predicted direction.

We now move to the hypothesis of interest - the effect of SOD. Regardless of how the dependent variable is measured, we reach the same conclusion: the implementation of SOD has had a negative effect on the AFDC/TANF recipient rate, thus contributing to a greater degree of caseload decline in SOD states. When we examine the simple (unlogged) recipient rate, the estimated effect of SOD is -0.749. This suggests that holding other variables constant, the implementation of SOD in a state resulted in a decrease of 0.749 in the number of welfare recipients per 1,000 state residents. When we examine the model that utilizes the logged recipient rate, the estimated effect of SOD is -5.701, which translates to a 5.7 percent reduction in the TANF recipient rate. To help put this effect into perspective, based on the results reported in Table 1, it would take an increase of approximately 3.5 percentage points in the state unemployment rate, or a decrease of nearly $3,000 in a state's per capita income, to produce a decrease in the welfare recipient rate equal to the decrease attributed to SOD.

Of course, the degree of decentralization (i.e., SOD) is only one dimension of TANF policy. States were responsible for many other important decisions in designing their TANF programs, as reflected in the variation in the eligibility and flexibility indexes included in equation I.7 Thus, it is difficult to discern the net impact of TANF, and the contribution of SOD to that net impact, from a mere visual inspection of the results in Table 1. In addition, it is important to remember that the coefficients in Table 1 represent the immediate effects of the TANF variables (due to the presence of the lagged dependent variable). The total effect is significantly larger. We therefore present Figure 2, which provides a more detailed picture of the total impact of SOD and the TANF program more generally.

Figure 2 presents the predicted net effect of the implementation of TANF on a state's welfare caseload (based on the results for the logged version of the dependent variable in Table 1). The net impact, as we define it, is the sum of the effects of the three TANF variables in our model (SOD, TANF eligibility index, and TANF flexibility index). We calculate the net impact for three hypothetical states, each of which reflects a different approach to welfare reform. A "restrictive" state is defined as a state with a flexibility value at the 10th percentile (4) and an eligibility value at the 90th percentile (19). Examples of such states in our data include Georgia, Oklahoma, and Wyoming. A "typical state" is defined as a state with a flexibility value at the 50th percentile (7) and an eligibility value at the 50th percentile (14). A number of states fit this description. Finally, a "generous state" is defined as a state with a flexibility value at the 90th percentile (11) and an eligibility value at the 10th percentile (8). New York, Massachusetts, and Rhode Island are examples of states that come close to these values. For each of these hypothetical states, we calculate the total effect of the implementation of TANF (taking into account the effects distributed through the lagged dependent variable), assuming either SOD or centralized administration.

The results of this simulation provide some interesting insights into the ultimate impact of SOD and the TANF program on the welfare recipient rate. For the hypothetical restrictive state, TANF had a statistically significant negative impact on the caseload, resulting in either a 26 percent or 35 percent decrease in the caseload, depending on the state's choice regarding SOD. For a typical state, the impact of TANF was significant and negative (a decrease of 16 percent) if a state chose SOD, yet the effect of TANF was insignificant if a state chose to implement TANF in a centralized fashion. For a generous state, the implementation of TANF is actually predicted to have resulted in an increase in the TANF recipient rate, but even in the case of centralized administration, the effect is not statistically significant (p = . 12).8 In summary, this illustration suggests that while SOD states experienced greater caseload decline than non-SOD states, the overall impact of the TANF program on the caseload varied greatly, depending on a state's approach to welfare reform.

SECOND-ORDER DEVOLUTION AND TANF IMPLEMENTATION

Based on the preceding analysis, we find strong evidence that SOD has contributed to a reduction in welfare caseloads in the welfare reform era. However, recall that both the stringency thesis and the efficiency thesis predict that such an effect should exist. Thus, while the caseload analysis suggests that one of these theories might be correct, it does little to help determine the causal mechanisms underlying the SOD effect. That is, does SOD reduce caseloads due to greater stringency in TANF implementation? Or does caseload decline result from greater success (efficiency) in moving TANF clients into the workforce?

To answer this question, we move to an analysis of state TANF outcomes. We begin with a state-level analysis of TANF sanctioning as our test of the stringency thesis. We then estimate the effects of SOD on three measures of TANF performance in employment outcomes: employment exits, job retention, and earnings gains.

The Sanction Stringency Hypothesis

If SOD influences the implementation of sanctions as expected, sanction rates in SOD states should be higher than those in centralized (non-SOD) states. We test this hypothesis by estimating the coefficients for equation 2 below. As we utilize state-panel data, the model is estimated by relying on panel corrected standard errors and a GLS correction to overcome autocorrelation across panels. We also include fixed effects for years and region (based on the four-region Census definition).

... [2]

The dependent variable is the state sanction rate, calculated for each state and year as the number of closed cases due to work- related sanctions, divided by the average monthly TANF caseload. We use data extracted from the Annual Report to Congress from 2000 to 2003. A number of states (Alaska, Arkansas, California, Indiana, Maine, Minnesota, Missouri, Montana, New Hampshire, New York, Rhode Island, Texas, Vermont, and Washington) do not terminate a case for non-compliance with work requirements. In addition, Wisconsin implements a pay-for-performance sanction policy (perhour reduction). Hence, we exclude these states from the analysis because it is not possible for sanction exits to occur. We also exclude Oregon due to a lack of sanctions data. Thus, we are able to observe 33 of 50 states over a four-year period to test the sanction stringency hypothesis.

We examine the effect of administrative decentralization on sanctions by using a dummy variable to measure SOD, where 0 = a state administered TANF program (centralized), and 1 = a locally administered TANF program with significant devolutionary power (states with relatively strong SOD, as described above). If the stringency hypothesis is correct, we anticipate that SOD states will have higher sanction rates, on average, than centralized states.

As with our recipient rate model, we include a number of controls. We include sanction severity as a measure of the state's sanction policy; we expect that the more severe a state's sanction policy, the more likely TANF recipients will leave the program. This variable is measured as the percentage of the TANF benefit deducted due to the first instance of noncompliance. We also include other sanction rates, which is defined as the sanction rate for violations of other (non-work-related) TANF rules. These include child support sanctions, sanctions for teen parents failing to meet requirements, and sanctions issued for failure to meet an individual responsibility plan. Our expectation is that as the use of other sanctions increases, there will be fewer opportunities to sanction clients for violations of work requirements. This is due to the likelihood that recipients who have a high risk to be sanctioned due to work-related requirements are also likely to have a high risk to be sanctioned due to other reasons. We anticipate that sanction severity will be positively associated with state sanction rates, while other sanction rates will be negatively associated with state sanction rates.

We also consider various dimensions of a state's political, economic, and social environment. To capture the effects of political ideology, we once again rely on the measures of state government and citizen ideology developed by Berry et al. (1998). As both variables are measured so that higher values reflect greater liberalism/Democratic strength, we expect that state partisanship and citizen ideology will be negatively associated with state sanction rates.

To capture the effects of race on sanction policy outcomes, we include nonwhite caseload %, defined for each state-year as the percentage of the TANF caseload that is either black or Hispanic. Past research has shown that racial resentment has played a key role in shaping whites' attitudes toward welfare, and racial resentment of whites has been found to influence redistributive policy choices (Gilens 1999; Schräm, Soss, and Fording 2003; Soss et al. 2001). Thus, we expect to find stricter sanction implementation in states where blacks and Hispanics comprise a higher percentage of the TANF caseload.

To control for "paternalistic pressure" (Mead, 1997; Soss et al., 2001; Fellowes and Rowe, 2004), we include the recipient rate (lagged one year) and unmarried birthrate (of TANF recipients). We expect that as paternalistic pressure increases (due to rising caseloads and increasing unmarried births), so too will the stringency of sanction implementation because local administrators and case managers are under more pressure to achieve the dual goals of decreasing caseloads and unmarried birthrates among TANF families.

To capture the effects of economic conditions, local employment opportunities, and local labor markets, we also consider the effects of state unemployment rates, per capita income, and state poverty rates. In states where there are numerous employment opportunities and relatively prosperous economic environments, recipients are likely to have an easier time finding work and will therefore be more likely to avoid sanctions imposed for non-compliance for work requirements. Yet, in states with higher poverty rates and generally poorer economic health, local administrators and case managers might be more sympathetic to TANF recipients because they recognize recipients' environments and understand their need for help (Fording, Soss, and Schräm 2007). Thus, while we expect that state economic environments should matter, we have no clear expectation regarding the direction of the effect.

Results

The coefficient estimates for equation 2 are presented in Table 2 below. Consistent with our expectations, we find that as the percentage of TANF clients who are black or Hispanic non-white caseload %) increases, sanction rates also increase. Unexpectedly, the size of the recipient rate is negatively associated with sanctions, despite the fact that we measure the caseload in the year prior to our observation of the dependent variable. One explanation for this finding is that during this period, states were not experiencing caseload pressures due to unprecedented caseload decline that occurred after the passage of PRWORA. Thus, by the early 2000s, the size ofthe TANF caseload could be more reflective of states with higher numbers of "hard to serve" TANF clients, who would qualify for an exemption from TANF work requirements and, thus, be immune from work-related sanctions.

The use of sanctions for other reasons (other sanction rate) is significantly related to a states use of work-related sanctions and in the anticipated negative direction. Sanction severity is positively related to the sanction rate as expected, but it is not significant. Although a stronger sanction policy is more likely to lead to a sanction exit, it might be less likely to be used by case managers due to the anticipated consequences for clients. This could serve to diminish the effect, leading to the lack of significance we observe in Table 2.

As for economic conditions, we find that per capita income is negatively related to the sanction rate, while our other indicators of state economic health (poverty rate and unemployment rate) are unrelated to sanctions. Finally, and most importantly for our purposes, we find strong support for the sanction stringency hypothesis. SOD states display significantly higher sanction rates than centralized states, even after controlling for several dimensions of state political and socioeconomic environments. Based on the results, the sanction rate in SOD states is estimated to be nearly 6 percentage points higher in SOD states than in centralized states. Given that the mean sanction rate across all state years is only 9.49 (with a standard deviation of 12.00), the SOD effect would appear to be substantively significant as well.

The Employment Improvement Hypothesis

Our final set of analyses test the efficiency thesis by examining the possible beneficial effects of SOD on TANF. Specifically, we examine variation in TANF success by testing the employment improvement hypotheses. In these models, we examine three dependent variables, defined as follows:

Employment Exit Rate: The number of welfare exits through employment, divided by the average monthly TANF caseload (x 100).

Job Retention Rate: The percentage of employed adult recipients in a given performance year who were employed for two consecutive quarters.

Average Earnings Gain: The rate of change in earnings of employed adult recipients who were employed for two consecutive quarters.

Data for the employment exit rate are obtained from the TANF Annual Report to Congress and are available for 49 states (excluding Oregon) from 2000 through 2003. Data used to measure the job retention rate and the average earnings gain are also taken from the ARC and are available for years 1 998 through 2002.9 For each of these three dependent variables, we estimate the same model, as represented below in equation 3.

... [3]

The independent variables in our employment performance models are defined as above for the sanction stringency model. However, for this model we do not include a measure of non- work sanctions. Our expectations are as follows. We expect that sanction severity will be positively related to employment performance because stringent sanction policies might provide TANF recipients with greater incentives to comply with work requirements, thus improving employment outcomes. We expect that the recipient rate will be positively associated with employment performance due to increased pressure to decrease the caseload through employment in states where the caseload is relatively high. Citizen and government ideology are expected to be negatively associated with employment performance due to the (presumably) less stringent style of welfare implementation in liberal states, which in turn could result in longer welfare spells for TANF clients. State economic prosperity (as measured by unemployment rate, per capita income, and poverty rate) is expected to be positively associated with employment performance because a robust economy provides more opportunities to be employed. Unmarried birth rates and non-white caseload % should be negatively associated with welfare exits through employment and job retention rates because single mothers and black and Hispanic clients face greater barriers to successful employment. Finally, TANF programs in SOD states are expected to display superior employment outcomes compared to centralized states because of their ability to better match services to client needs.

Results

The coefficient estimates for each of the three measures of employment performance are presented in the last three columns of Table 2. As before, the effects of several control variables conform to our expectations. For example, stricter sanction policies lead to increased performance for two of the three employment measures: employment exits and job retention. Higher percentages of black and Hispanic clients on welfare lead to lower rates of employment exits and lower average earnings gains, but lead to higher rates of job retention. One possible explanation for this combination of results is that white clients might have more opportunities to obtain jobs, increase their earnings, and then finally leave welfare through employment. Non-white clients, on the other hand, might have more difficulty finding good-paying jobs, but once they find a job, they might be more likely to keep it and stay on TANF, due to either a lack of opportunities for advancement or a greater dependence on the supportive services available through TANF.

The recipient rate is positively related to the job retention rate (as predicted), but as in the model of state sanction rates, the recipient rate displays an (unexpected) negative relationship with two measures of employment performance, employment exits and average earnings gain, despite the fact the caseload is measured in the year prior to our observation of the dependent variable. As in the previous analysis, we attribute this to the lack of caseload pressure due to declining caseloads and the likely association between caseload size and "hard to serve" TANF recipients. Unmarried birthrates have no impact on employment exit rates, and citizen ideology and state government ideology are likewise insignificant. Finally, our measures of state economic prosperity (income, poverty, and unemployment) are largely insignificant.

Moving to effects of SOD, overall we find moderate support for the employment improvement hypothesis. After controlling for several socioeconomic and political variables, we find that TANF programs in SOD states display significantly better employment outcomes for two of our three measures of performance: employment exits and average earning gain. The effect of SOD on job retention rates is estimated to be positive (as predicted), but the effect is not statistically significant. Thus, it would appear that the relationship between SOD and caseload decline, to some degree, could be accounted for by each of the competing causal mechanisms outlined above - greater stringency, as well as greater efficiency.

CONCLUSION

Since the passage of PRWORA in 1996 and the subsequent implementation of TANF, the practice of SOD under TANF has been the subject of considerable debate. Proponents of SOD have consistently argued that the flexibility offered by decentralization to local governments will ultimately lead to improved TANF performance because local leaders better understand their labor markets, as well as the needs of their poor, and therefore can provide more appropriate services (Whitaker and Time, 2001; Dye, 1990). On the other hand, opponents of devolution have argued that giving more discretion and responsibility to state or local governments in welfare policy design and implementation could result in a "race to the bottom" phenomenon in which local governments compete to reduce welfare generosity to avoid the immigration of the poor and the loss of tax revenue. Thus, critics of SOD argue that for these reasons, SOD may lead to a more punitive welfare system.

Based on the results of our analyses, we find that decentralization has led to a nontrivial reduction in the welfare caseload in SOD states, as predicted by both critics and supporters of SOD. This reduction in the caseload appears to be driven by different forces. On the one hand, we find that SOD is associated with a greater reliance on punitive policy tools in the form of TANF sanctions. Although we cannot pinpoint the causal mechanism driving this relationship, this result is consistent with either a race to the bottom or fiscal incentives introduced by shifting some of the financial responsibility for TANF implementation to the local level. Therefore, this study makes a valuable contribution to the literature on the race to the bottom by suggesting that decentralization can lead to a race to the bottom not just in policy choice - such as benefit levels or other TANF rules - but also that it might have an important impact on the implementation of redistributive policies.

At the same time, we also find SOD to be associated with higher levels of success for two important TANF employment outcomes, welfare exits due to employment and earnings gains among TANF recipients. As these are two of the more important goals of TANF, these results are consistent with arguments made by proponents of decentralization who argue that SOD should result in more successful implementation. However, as with the case of the SOD effect on sanctions, we cannot be entirely sure of the causal mechanism underlying this effect. It might well be due to greater efficiency in implementation in a decentralized environment, as proponents of SOD would claim. Nevertheless, it is also possible that, like the sanction effects, the employment and earnings effects also result from greater stringency in implementation due to SOD. This possibility is mitigated to some extent because we directly control for both sanction stringency and government ideology in our employment models. But the possibility remains and thus additional research is needed to unpack the causal linkages between SOD and TANF performance.

Footnote

ENDNOTES

1. Under AFDC, states could choose between state administration and state-supervised administration. The former indicated that the state administered the policy directly while the latter indicated that local governments administered the policy under the supervision of a statewide welfare agency. However, federal funds were given to state governments and states had the responsibility to insure that the policy was administered according to federal guidelines under both structures (Adkisson, 1998).

2. We exclude Alaska, Hawaii, and Nebraska due to a lack of data for one or more variables.

3. Implementation dates for TANF are provided by Crouse (1999).

4. For states in which TANF was implemented midyear, we multiply the eligibility or flexibility score by the proportion of the year that TANF was in place.

5. As with our measures of TANF policies, our waiver variables are measured as the proportion of the year in which a waiver was implemented. Waiver implementation dates are provided by Crouse (1999).

6. We utilize the test introduced by Hadri (2000).

7. Descriptive statistics for these variables, as well as all the variables in our analyses are provided in the appendices.

8. When we replicate Figure 2 using the results from the version of the model that uses the unlogged welfare recipient rate, we find a very similar pattern, however the positive net effects of TANF on the caseload are statistically significant. This is not terribly surprising and suggests that in the absence of strong punitive policies, poor families might have been attracted to TANF by the provision of childcare, job training, and other supportive services that were more likely to be provided under TANE For an example of such an argument in the context of waivers implementation, see Moffitt (1996).

9. Alabama, California, Nebraska, New Mexico, New York, and Virginia are excluded due to a lack of data.

References

REFERENCES

Adkisson, Richard V. 1998. "Multi-Level Administrative Structure and the Distribution of Social Service Expenditures: A Nebraska Example." The Social Science Journal 35:303-18.

Agranoff, Robert, and Michael McGuire. 2001. "American Federalism and the Search for Models of Management." Public Administration Review 61:671-81.

Bailey, Michael A., and Mark Rom. 2004. "A Wider Race? Interstate Competition Across Health and Welfare Programs." The Journal of Politics 66:326-47.

Bailey, Michael A. 2005. "Welfare Migration and the Multifaceted Decision to Move." American Political Science Review 99:1-1 1.

Beck, Nathaniel, and Jonathan Katz. 1995. "What to Do (and Not to Do) with Time-Series Cross-Section Data." American Political Science Review 89:634-47.

Berry, William D., Evan J. Ringquist, Richard C. Fording, and Russell L. Hanson. 1998. "Measuring Citizen and Government Ideology in the American States." American Journal of Political Science 42:337-48.

Berry, William D., Richard C. Fording, and Russell L. Hanson. 2003. "Reassessing the 'Race to the Bottom' in State Welfare Policy." The Journal of Politics 65:327-48.

Blank, Rebecca M. 2001. "What Causes Public Assistance Caseloads to Grow?" Journal of Human Resources 36:85-1 18.

Brueckner, Jan K. 2000. "Welfare Reform and the Race to the Bottom: Theory and Evidence." Southern Economic Journal 66:505-25.

Cho, Chung-Lae, Christine A. Kelleher, Deil S. Wright, and Susan Webb Yackee. 2005. "Translating National Policy Objectives into Local Achievements across Planes of Governance and among Multiple Actors: Second-Order Devolution and Welfare Reform Implementation." Journal of Public Administration Research and Theory 15:31-54.

Crouse, Gil. 1999. "State Implementation of Major Changes to Welfare Policies, 19921998." Office of Human Services Policy, Assistant Secretary for Planning and Evaluation, Department of Health and Human Services. http://aspe.hhs.gov/HSP/Waiver-Policies99/ policy_CEA.htm

De Jong, Gordon F., Deborah Roempke Graefe, Shelley K. Irving, and Tanja St. Pierre. 2006. "Measuring State TANF Policy Variations and Change after Reform." Social Science Quarterly 87:755-81.

Dye, Thomas R. 1990. American Federalism: Competition among Governments. Lexington, MA: Lexington Books.

Enchautegui, Maria E. 1997. "Welfare Payments and Other Economic Determinants of Female Migration." Journal of Labor Economics 15:529-54.

Fellowes, Matthew C, and Gretchen Rowe. 2004. "Politics and the New American Welfare States." American Journal of Political Science 48:362-72.

Figlio, David N., Van W. Kolpin, and William E. Reid. 1999. "Do States Play Welfare Games?" Journal of Urban Economics 46:437-54.

Fording, Richard C. 2001. "The Political Response to Black Insurgency: A Critical Test of Competing Theories ofthe State." American Political Science Review 95:1 15-30.

Fording, Richard C, Joe Soss, and Sanford F. Schräm. 2007. "Devolution, Discretion, and the Impact of Local Political Values on TANF Sanctioning." Social Service Review 81:285-316.

Francis, Richard M. 1998. "Predictions, Patterns, and Policymaking: A Regional Study of Devolution." Publius: The Journal of Federalism 28:143-60.

Gainsborough, Juliet F. 2003. "To Devolve or Not to Devolve? Welfare Reform in the States." Policy Studies Journal 31:603-23.

Gilens, Martin. 1999. Why Americans Hate Welfare: Race, Media, and Politics ofthe AntiPoverty Policy. Chicago, IL: University of Chicago Press.

Hadri, Kaddour. 2000. "Testing for Stationarity in Heterogeneous Panel Data." The Econometrics Journal 3:148-61.

Kelleher, Christine A., and Susan Webb Yackee. 2004. "An Empirical Assessment of Devolutions Policy Impact." Policy Studies Journal 32:253-70.

Kettl, Donald F. 2000. "The Transformation of Governance: Globalization, Devolution, and the Role of Government." Public Administration Review 60:488-97.

Mead, Lawrence M. 1997. The New Paternalism: Supervisory Approaches to Poverty. Washington, DC: Brookings Institution Press.

Moffitt, Robert. 1996. "The Effect of Employment and Training Programs on Entry and Exit from the Welfare Caseload." Journal of Policy Analysis and Management 15:32-50.

Pandey, Shanta, and Shannon Collier-Tenison. 2001. "Welfare Reform: An Exploration of Devolution." Social Justice 28:54-75.

Peterson, Paul E., and Mark Rom. 1989. "American Federalism, Welfare Policy, and Residential Choices." American Political Science Review 83:71 1-28.

Schräm, Sanford E, Joe Soss, and Richard C. Fording. 2003. Race and the Politics of Welfare Reform. Ann Arbor, MI: University of Michigan Press.

Soss, Joe, Sanford F. Schräm, Thomas P. Vartanian, and Erin O' Brien. 2001. "Setting the Terms of Relief: Explaining State Policy Choices in the Devolution Revolution." American Journal of Political Science 45:378-95.

Soss, Joe, Richard C. Fording, and Sanford F. Schräm. 2008. "The Color of Devolution: Race, Federalism, and the Politics of Social Control." American Journal of Political Science 52:536-53.

U.S. Council of Economic Advisers. 1997. "Explaining the Decline in Welfare Receipt, 1993-1996." Technical Report. URL:http://www.whitehouse.gov/WH/EOP/CEA/ Welfare/Technical-Report.html.

Volden, Craig. 2002. "The Politics of Competitive Federalism: A Race to the Bottom in Welfare Benefits?" American Journal of Political Science 46:352-63.

Whitaker, Ingrid Phillips, and Victoria Time. 2001. "Devolution and Welfare: The Social and Legal Implications of State Inequalities for Welfare Reform in the United States." Social Justice 28:76-90.

Ziliak, James P., David Figlio, Elizabeth Davis, and Laura Connolly. 2000. "Accounting for the Decline in AFDC Caseloads: Welfare Reform or the Economy?" Journal of Human Resources 35:570-86.

Appendix

(ProQuest: Please refer to full text PDF for images of the Appendix.)