nents include different samples, it is only possible to use the Interview Survey for the kind of microlevel analysis that we required.⁴²

BLS prepared a large number of tabulations for us from the 1991 Interview Survey and the 1989-1991 surveys combined. For processing convenience and to meet our timetable, these tabulations treated each quarterly interview falling within a calendar year as a separate observation, inflating the amounts by four to obtain annual figures. This procedure increases sample size because it uses all of the available data and not just the data for consumer units who responded to all interviews within a year.⁴³ For actual use in updating the reference family poverty threshold, however, we believe it would be preferable to aggregate quarterly amounts for those units with complete data, making an appropriate adjustment to the weights to account for other units.

We began our analysis by looking at the distribution of expenditures on the basic bundle of food, clothing, and shelter (including utilities). BLS arrayed consumer units by their expenditures on these four categories, separately and combined, and, in each instance, determined the dollar values corresponding to the spending level for every 5 percent of units, from the lowest 5 percent to the highest 5 percent.

In examining spending patterns on food, clothing, and shelter, we found it convenient to look at the distribution in terms of the dollar values that demarcated every 5th percentile of the distribution. However, for purposes of calculating the reference family poverty threshold, whatever percentile value is chosen must be reexpressed as a percentage of median expenditures on food, clothing, and shelter for the same reason that relative thresholds are expressed as a percentage of median income or expenditures rather than as a percentile value. That is, if the thresholds are expressed as, say, the 25th or 30th percentile of income or expenditures, then, by definition, 25 or 30 percent of families are always poor; however, if the thresholds are expressed as, say, 40, 50, or 60 percent of median income or expenditures, then changes that affect the distribution of income or expenditures below the median can increase or decrease the poverty rate. As an example, a recession could move some families in the lower half of the income distribution from above to below 50 percent of the median, so that the poverty rate increased whether median income itself stayed the same or fell. Conversely, an income assistance program could move families from below to above 50 percent of median income, so that the poverty rate decreased whether median income stayed the same or