On Feb 11, 2008, at 12:37 PM, JBrown72073 at cs.com wrote:
> This article from Sunday's NY Times claims that the bottom fifth of
> households (by income) spend on average twice what they earn.
> Oddly, the authors don't include 'interest' as an expenditure
> category, but is this anywhere near right or does it just reflect a
> lot of underreporting of income when your income is fairly informal
> to begin with? Or is it just magnifying what it's reporting by
> ignoring assets? (This would help their argument that we're all
> fine and prosperous because we have AC and color TV's.)
These guys, Cox & Alm of the Dallas Fed, are serial hacks and apologists. They did some work on mobility that was extremely devious, starting their sample with people so young that their only contribution to mobility was growing up. The technique was so crude that the only rationale that makes any sense is that they knew what they wanted to prove and would manipulate the data accordingly.
Their source on this article is the BLS's Consumer Expenditure Survey. The BLS acknowledges that the income data are not as reliable as the expenditure data, and urges researchers interested in income to turn to other sources (excerpt below). Note the preposterously large gap between income and expenditures for the middle and top fifths - you have to wonder how our national savings rate is 0%.
It's also very devious to emphasize gadgets like VCRs and DVD players, whose prices have been dropping like rocks. You'll notice that the words "housing," "college tuition," and "health care" don't appear in their chart.
Doug
----
<http://www.bls.gov/cex/csxfaqs.htm#q20>
> 20. Why do average annual expenditures exceed income for some of
> the demographic groups? How can consumer units spend more than they
> earn?
>
> Data users may notice that average annual expenditures presented in
> the income tables sometimes exceed income before taxes for the
> lower income groups. For data prior to 2004, the primary reason for
> that is believed to be nonresponse to questions about income, a
> common problem in household surveys. The average incomes shown in
> the published tables for 2003 and earlier are derived from
> information provided by complete income reporters (consumer units
> that provide information for at least one of the major sources of
> their income, such as wages and salaries, self-employment income,
> or retirement income). However, even complete income reporters may
> not have provided a full accounting of all income from all sources.
> Research has shown that some complete reporter consumer units
> classified in the lower income classes have expenditure levels that
> are more typical of upper income consumer units. Their expenditures
> raise the average expenditure levels of the income class in which
> they are classified.
>
> Beginning in 2001 for the Interview Survey and 2004 for the Diary
> Survey, the income data include information collected from
> respondents using income ranges or brackets—for example $2,000-
> $2,499—in addition to discrete income amounts, as provided in the
> past. Respondents who are unable or unwilling to provide a specific
> dollar amount may be able or willing to estimate a range for their
> incomes. The use of bracketing in data collection provides more
> reliable income estimates to the extent that it increases the
> percentage of households providing income data.
>
> In addition, starting in 2004, the Consumer Expenditure Survey uses
> imputation to fill in missing values for income data. The published
> tables now include income data from all consumer units—not just
> complete reporters. (See FAQ 21.) Income imputation has reduced the
> gap between income and expenditures when negative, and increased it
> when positive. For example, in 2003 (the last year prior to
> imputation), expenditures exceed income on average for all complete
> reporters who report less than $40,000 in income. In 2004,
> expenditures exceed income on average for all consumer units for
> whom less than $30,000 is reported or imputed. Similarly, in 2003,
> income exceeds expenditures for total complete reporters by less
> than $8,400; in 2004, income exceeds expenditures for all consumer
> units by more than $11,000.
>
> However, there are reasons why expenditures exceed income for the
> lower income groups despite the use of imputed income data.
> Consumer units whose members experience a spell of unemployment may
> draw on their savings to maintain their expenditures. Self-employed
> consumers may experience business losses that result in low or even
> negative incomes, but are able to maintain their expenditures by
> borrowing or relying on savings. Students may get by on loans while
> they are in school, and retirees may rely on savings and investments.
>
> 21. I understand that, beginning with publication of the 2004
> tables, the Consumer Expenditure Survey results include imputed
> income data. What does this mean for the typical user?
>
> Nonresponse is a common problem in household surveys, particularly
> for questions regarding income. Nonresponse means that the
> respondent either does not know, or refuses to provide, the
> information requested. Prior to publication of the 2004 tables, the
> Consumer Expenditure Survey handled nonresponse to income questions
> by publishing income data for complete income reporters only. To be
> classified as a complete income reporter, the respondent had to
> provide a value for at least one major source of income for the
> consumer unit. However, not all “complete” reporters provided a
> full accounting of income for all sources for which receipt was
> reported.
>
> Starting in 2004, the Consumer Expenditure Survey introduced
> multiple imputation to fill in the blanks resulting from
> nonresponse to income questions. In this method several estimates
> are made each time the respondent reports the receipt of, but no
> value for, a particular source of income. The estimates are made
> based on characteristics of the member or consumer unit for which
> receipt is reported. The average of these estimates is used to
> provide the final figures shown in the tables. Because all consumer
> units now have actual or imputed values for income data available
> to produce means and other information, the old complete income
> reporter data are no longer published in tables.
>
> The introduction of multiply imputed data allows for use of the
> full set of income data from all consumer units, and therefore more
> accurate comparisons of income and expenditures. For example,
> instead of producing estimates for complete income reporters, some
> of whom are still missing income information, income data are now
> provided in tables for all consumer units with these blanks filled
> in, resulting in smaller gaps by which expenditures exceed income
> for low income consumer units. (See FAQ 20.) In addition, when
> examining income by demographic—such as age or composition of
> consumer unit—income data now are presented for all consumer units
> within that category rather than for complete reporters only.
> Consider, for example, consumer units whose reference person is
> under 25 years old. Prior to 2004, the age tables show average
> annual expenditures for all consumer units in this age group, but
> income is shown only for complete reporters in this age group.
> Starting in 2004, both expenditure and income data are shown for
> all consumer units within this age group. Therefore, the data are
> more appropriately compared. For additional information, see
> Special Notice Income Imputation Introduced With 2004 Data on CE
> home page.