Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
POVERTY THRESHOLDS 135 caution. Research has found that subjective poverty thresholds vary significantly with the type of question and other differences in methodology. In the Netherlands, Flik and Van Praag (1991) developed estimates for several subjective poverty thresholds that varied by more than 200 percent. Some variation may be appropriate, to the extent that different questions carry different meanings, but research has also found significant variation with small modifications in question wording (see below). In general, little is known about how respondents interpret the questionsâfor example, whether they exclude taxes or include in-kind benefits in their responses. Another problem is that estimates are often based on small sample sizes, which carry large standard errors. Although the standard errors can be reduced by increasing the sample size, the responses also often show wide variation around the mean. For example, a question in the 1993 General Social Survey about the weekly amount of a poverty line for a two-adult/two-child reference family (see below) elicited responses that averaged $341 per week, but they varied from as low as $25 to as high as $1,500 per week. The standard deviation was $167, or 49 percent of the meanâa high variation. (The range excludes two clear outlying responses of $5,000 and $7,000 per week.) Because of these characteristics of survey responses, it may be difficult to set an actual threshold using them with any confidence. A quite different problem might arise if survey responses are known to be used to set official poverty thresholds: respondents might give different answers because of knowledge that the poverty line affects eligibility levels in a number of government assistance programs. More broadly, subjective responses may reveal more about underlying differences in expectations and current circumstances than about relative needs. For example, O'Hare et al. (1990) found that Hispanics gave answers to a question about the poverty line that were substantially lower than the answers of other groups. This result may have occurred simply because this group is constrained in income and consequently has lower expectations. Research Findings There has been extensive work on the development of subjective poverty thresholds, particularly by analysts in Europe (see, e.g., Flik and Van Praag, 1991; Goedhart et al., 1977; Hagenaars, 1986; Hagenaars and de Vos, 1988; Hagenaars and Van Praag, 1985; Van Praag, 1968; Van Praag, Dubnoff, and Van der Sar, 1988; Van Praag, Goedhart, and Kapteyn, 1980).33 Analysts have sometimes used a single question on minimum income: "What do you consider an absolute minimum net income for a household such as yours?" 33 Maritato (1992) provides a detailed review of the literature on subjective poverty measurement in Europe, Canada, and the United States.
POVERTY THRESHOLDS 136 Sometimes they have used a question evaluating income at multiple levels: "Under our conditions, I would call a net household income per week [or month or year] of about x very bad, bad, insufficient, sufficient, good, very good." One method uses a minimum income question together with a question on whether the household can, with its current income, make ends meet "with great difficulty, with difficulty, with some difficulty, rather easily, easily, or very easily." Analysts have also used different econometric techniques to estimate subjective poverty thresholds (or thresholds at various levels, including a poverty level and higher levels) from the survey responses. Typically, the methods try to take account of the influence of family size and the respondent's own income on these responses. Sometimes the estimation uses the data from only a subset of respondents, such as those who report that they can only make ends meet with their own income with some difficulty. Work on subjective measures of poverty has also been done in the United States and Canada (see, e.g., Colasanto, Kapteyn, and Van der Gaag, 1984; Danziger et al., 1984; De Vos and Garner, 1991; Kilpatrick, 1973; Michalos, 1989; Morissette and Poulin, 1991; Poulin, 1988; Rainwater, 1974, 1992; Vaughan, 1993). The questions used in some of these studies asked respondents about the income needed for families similar to theirs to "make ends meet." But different question wordings have been used. For example, the question used by De Vos and Garner (1991) asked specifically about income needed before deductions, while the one used by Colasanto, Kapteyn, and Van der Gaag (1984) asked about after-tax income. The question used by Danziger et al. (1984) did not specify whether respondents were to answer in before-tax or in after-tax terms. Although the variations in question wording were minor, the resulting estimated thresholds differ substantially.34 De Vos and Garner (1991) estimated a poverty threshold (1982 CEX data) of $32,530 in 1992 dollars, or 229 percent of the official 1992 two-adult/two-child poverty threshold. Danziger et al. (1984) estimated a four-person family poverty threshold (with 1980 data from the 1979 Income Survey Development Program Research Panel) of $24,680 in 1992 dollars, or 173 percent of the official 1992 threshold. In contrast, Colasanto, Kapteyn, and Van der Gaag (1984) (with data from the 1981 Wisconsin Basic Needs Study) estimated a four-person family subjective threshold of only $12,160 in 1992 dollars, or 85 percent of the official 1992 threshold. The question analyzed by Colasanto, Kapteyn, and Van der Gaag specifically asked about after-tax income; also, their data source was limited to a single state (Wisconsin). It seems clear that a good deal more work is needed before the approach of using survey responses to derive poverty thresholds could be seriously considered for an official measure. If such responses were available over time 34 There were also differences in estimation methodology.