The Census Department reports the number of people in poverty increased 1.3 million to 35.9 million. In the same publication, Median Incomes declined 0.1% (the press release considers this "unchanged"). The number of uninsured rose 1.4 million to 44.961 million and the proportion of uninsured rose to 15.6% from 15.2%. Based on the anecdotal evidence and the continuing rise in inflation and health care costs, the same or greater increase should be expected when the 2004 data comes out next year.
Here's a pretty good primer on government economic data and its potential bias. I'd particularly recommend the final paragraphs that deal in more detail with the unemployment and employment data published by the Bureau of Labor Statistics. I hope that further installments in this series come out soon.
A couple of additional notes:
Inflation statistics are the most prone to bias since their manipulation has the greatest effect on other key statistics as well as the greatest effect on government budgets. Both GDP and productivity statistics use inflation statistics as key building blocks for their own data, so downward bias in inflation statistics leads to upward bias in both GDP and productivity statistics.
Economists acquiesce to this bias by accepting that higher prices will lead to substitution from more expensive goods to cheaper alternatives, but not the reverse. They also assume that over time there are quality improvements in goods but not quality reductions.
Having bias does not mean that the statistics are not useful, as long as the bias is consistent. A consistent bias means that marginal changes in the data series are still meaningful. An increase in the over-the-year inflation rate from 1% to 4% still indicates rising inflation, even if the true numbers were 4% and 7% respectively. However, the longer the period of time between data points, the less likely the bias is consistent and the less comparable the data. For example, it's probably no longer useful to compare CPI inflation from the 1960s with the present rate.
Finally, the main reason the BLS establishment survey is "more accurate" than the household survey is due to its benchmark. Each year, the level of the employment series ard adjusted by the census of quarterly unemployment insurance records. The population estimates for the household survey are benchmarked to population control updates from the Census, but the estimates for employment, unemployment and labor force are never benchmarked. State and Local estimates from the household survey actually use establishment survey data and unemployment insurance data as part of the estimation process.
The benchmark revision is published twice a year for U.S. data and once a year for State and Metro Area data and is the indication of how truly accurate the estimates have been. The latest benchmark revision in February 2004 lowered total nonfarm employment by 122,000 or 0.1%.