A Stock-Flow Analysis of the Welfare Caseload

Jacob Alex Klerman, Steven J. Haider
2004 The Journal of human resources  
This paper reconsiders the methods used in previous studies to assess the welfare caseload movements during the 1990s. We develop a model in which the welfare caseload is the net outcome of past flows onto and off of the caseload and show that such a stock-flow model can explain some of the anomalous findings in previous studies. We then estimate the stock-flow model using California administrative data. We find that approximately 50 percent of the caseload decline in California can be
more » ... nia can be attributed to the declining unemployment rate. These estimates are more robust and larger than those obtained when applying more typical methods to the same California data. *During the 1990s, the welfare caseload peaked and then declined by half. 1 The 1990s also witnessed a robust economic expansion and a series of major welfare reforms. Applying conventional difference-in-difference models to aggregate welfare caseload data, many studies have estimated the relative importance of the economic expansion and welfare reforms in explaining the caseload decline (for example, CEA 1997, Wallace and Blank 1999, and Ziliak et al. 2000; see Blank 2002 for an extensive review). These studies reached widely varying conclusions regarding the cause of the decline. Several studies suggest that these different conclusions are due to differing specifications for the relationship between the current welfare caseload and lags of the explanatory and dependent variables (for example, Figlio and Ziliak 1999, CEA 1999). However, none of the studies explicitly consider the source of these relationships. This aggregate caseload literature has developed independently of a large literature that examines individual-level flows onto and off of welfare (for example, Hutchens 1981, Blank and Ruggles 1996, and Hoynes 2000). This paper attempts to combine the two literatures by considering the implications of viewing the aggregate caseload, a stock variable, as the net outcome of past flows onto off of welfare. This stock-flow approach suggests a source for the strong dependence of the caseload on lags of the explanatory variables found in previous studies. Furthermore, it suggests that the conventional models are mis-specified and that this misspecification can explain the disparate results across the studies. Beyond suggesting a critique of the existing literature, the stock-flow perspective also provides for an alternative estimation strategy. Specifically, we estimate the underlying flow relationships from the individual-level data and then simulate the implied impact on the caseload Klerman and Haider, p. 4 stock. This basic idea has been used to examine many behaviors, including welfare receipt (for example, Boskins and Nold 1975, Hutchens 1981), fertility (for example, Heckman and Walker 1992) and single motherhood (for example, Moffitt and Rendall 1995). Such an approach is not feasible with the available national data, and instead, we use California administrative data. Because these data are only for one state, we are not able to distinguish the impact of policy changes from more general time effects. However, there is sufficient variation to precisely estimate the impact of changing economic conditions. Our results suggest that approximately half of the caseload decline in California can be attributed to changing economic conditions, as measured by the unemployment rate. 2 These estimates are substantially larger than the 20 to 35 percent estimates that are obtained from more typical methods using the same California data. Although the aggregate regressions for California are very similar to their national counterparts, extrapolating from our California stockflow results to what would be obtained with national data should be done with much caution. The 1990s economic recovery was stronger and welfare reform was implemented later in California as compared to the rest of the nation. II. Explaining caseload changes Numerous studies have examined the recent declines in the welfare caseload. 3 These studies generally specify aggregate regressions in which the log of the caseload per capita st y is a function of various explanatory variables,
doi:10.2307/3559030 fatcat:b5zvnsb4kvecvmuyvwkp6k2tyi