Comments and Discussion

Rebecca M. Blank: This paper adds to the growing literature that attempts to assess the impact of the welfare reforms of the mid-1990s. A key problem in this literature is how to parameterize the reforms in a way that allows them to be clearly distinguished from other changes that occurred at the same time, including a major economic boom and changes in closely related policies (such as the minimum wage and the earned income tax credit). Hanming Fang and Michael Keane claim to do a better job than previous papers by using a more complete specification to measure the impact of policy changes, and their results suggest they are fitting the data very well. In these comments I will primarily focus on this specification and its interpretation.

Before turning to more critical comments, however, let me stress the things that this paper does well. First, Fang and Keane have done an excellent job of coding detailed state-level policy changes, with much more sophistication than previous papers. For instance, rather than use a dummy variable to indicate whether a time limit is in effect, they try to find variables that differentiate between the anticipatory behavioral effects of time limits (those that occur before the limits actually hit) and the effects that occur once enough time has passed that some women are facing benefit losses due to time limits.

My own experience suggests that coding specific state policies is dif fi cult and that the information available for some states and some policies is open to multiple interpretations. Different observers code these policies in different ways, because there is no standardized interpretation (and often there is not even a wholly reliable source of information about what policies are in effect each year). I suspect future researchers will make some modifications to Fang and Keane's coding, but I suspect that they will borrow some of their ideas as well. I hope that the authors plan at some point to make information available on the coding they use and their sources of information, so that future researchers can both learn from and improve upon their data.

Second, Fang and Keane study a longer time period than do most previous papers, and they estimate the effects of reform through 2002. The addition of 2001 and 2002 to the data is potentially important, since these are years of more economic variation and thus provide information on how reformed welfare policies operate in a more sluggish economic environment.

Third, the paper provides a very useful description of trends in welfare and work participation. Most researchers have focused on aggregate changes in welfare participation and labor force involvement among single mothers. Fang and Keane expand this and look at these trends by the mother's educational attainment, race, marital status, age of children, and number of children. There is much to be learned by looking at the comparative changes among these different groups.

One minor concern about these descriptive trends (and about the data more generally), however, is whether these trends on program participation correlate well with the administrative data from government agencies on actual program recipiency. The CPS data have always undercounted welfare participants, but my understanding is that this undercount appears to have increased in the late 1990s.¹ This is a problem for all researchers who use CPS data to study welfare reform and deserves more attention from the research community. The paper's analysis of specific trends by demographic group led to me wonder whether there is any way to compare the administrative data by demographic group with these data to see whether the undercount trend is stronger among some groups than others.

Let me focus the remainder of my comments on my concerns about how to interpret the paper's results. Fang and Keane make a very strong claim about their specification and their results. They note that, ". . . the key identifying assumption is that we have adequately controlled for all the important time-varying factors that influenced the welfare participation and work decisions of single mothers over the 1980-2002 period." They argue that they have therefore avoided...