Stimulus Keeping 6 Million Americans Out of Poverty in 2009, Estimates Show
 The figures start with detailed Census survey data collected in the Current Population Surveys for March 2004, March 2005, and March 2006. As described in Appendix A, we project these data forward to match the number and employment situation of households as of April, May, and June 2009. We also assign food stamps to additional eligible households so that food stamp levels for each state match the actual May 2009 levels as reported by the Agriculture Department.
 The following data are less precise than the overall 6.2 memo estimate because we are unable to fully account for variations in the recession’s effects on different age, racial-ethnic, and other groups. The following numbers thus should be viewed as approximations.
 We start with Census data on a nationally-representative sample of about 90,000 households, each one of which is presumed to represent a larger number households in the nation as a whole at the time it was surveyed. Researchers and government analysts use such data to calculate national estimates multiplying the data for each sample household by a sample “weight” provided by the Census Bureau and summing the results. Thus, our data, based on about 278,000 household records, yields a “weighted” total of about 113 million households for 2003-2005. Our method adjusts those weights upward (or, where appropriate, downward) in order to approximate the population size and household employment conditions for 2009.
For example, raw “unweighted” Census survey data from the March 2004-2006 CPS show an average of 2,059 households where household members reported 15 or more weeks of unemployment prior to the survey. Following standard Census procedures, these households would represent a weighted total of 2.6 million households nationwide in March 2004-2006. However, recent monthly data indicate that there are now 6.1 million such households nationwide. Therefore, we increase the weights accordingly until they reach the new total. (As noted in the text, we do this adjustment separately by state. Thus, in Massachusetts we multiply the weights for all households with household members unemployed 15 weeks or more by 1.67; in Michigan, we multiply by 2.67.)
 Another potential source of imprecision relates to the timing of the survey data. The March CPS asks households about their annual income, poverty status, and program participation as of the previous year. When we “weight up” unemployed and underemployed households, however, we do so with respect to their March employment status, not their status during the previous year. Because there is only a partial correspondence between March employment status and employment status during the previous year, we may not always be “weighting up” the families with the greatest risk of poverty during the previous year.
 For a description of the HHS/Urban Institute TRIM model, see footnote 8 below.
 To the extent that the increase in food stamp participation has been caused by changes in eligibility we may not accurately represent the income and demographic information of the additional food stamp participants.
 National Research Council, Measuring Poverty: A New Approach (National Academy Press: 1995).
 TRIM is developed and maintained by the Urban Institute under contract with the Office of the Assistant Secretary for Planning and Evaluation at the Department of Health and Human Services. Documentation of the TRIM Model is at http://trim3.urban.org/T3Technical.php. While the model was developed in large part to allow users to compare current policies with proposed policies, we use data only for TRIM’s “baseline” (i.e., current-policy) scenario.
 In producing the CPS files, the Census Bureau, like TRIM, also assigns benefits for some people with missing data. Unlike TRIM, however, Census does not use this process of assigning missing benefits to try to match the actual number of recipients shown in program records.
 Because medical expenditure data are not available directly from the CPS, they must be estimated using a formula. In this analysis, we estimate each family’s out-of-pocket medical spending by starting with the medical share of the poverty line (specifically, a version of the NAS poverty line referred to in Census publications as “MIT” or “Medical-out-of- pocket-expenditures In the Thresholds”) for a standard family of two adults and two children. We adjust that amount up or down using a set of ratios provided by the Census Bureau that depend on the family’s size, health insurance status, and age. That yields the family’s own estimated medical out-of-pocket expenses. We subtract those expenses from family income prior to comparing family income to the poverty line. (The NAS poverty line we use is the appropriate one for this comparison, namely, the MIT line minus the medical share. The result is by definition identical to using the Census Bureau’s MIT method. As a mechanical matter, however, we follow the original NAS recommendations by treating these medical expenses as an item to be subtracted from income rather than as part of the bundle of basic needs.)
Our estimates of work expenses also rely on formulas provided by the Census Bureau. The formulas are based on data on median weekly out-of-pocket child care expenses and other work-related expenditures from the Survey of Income and Program Participation. Values vary depending on year, weeks worked, number and ages of children (in the case of the child care formula), and other family characteristics. The estimated value of work expenses is capped at the value of the worker’s earnings. A couple’s child care expenses are further capped based on the earnings of the lower-earning spouse (although, in practice, few couples have estimated child care expenses large enough to be affected by this cap).
 Following the approach used in the TRIM model, we also treat unrelated household members younger than 15 living with no relatives as de facto family members of the head of household. This affects fewer than 200,000 poor children.
 As described in footnote 10, the version of the NAS threshold used here is essentially what Census publications call the MIT line with the medical share removed.