How to improve median income statistics

The Obama administration touted news reports like this one from the Washington Post: Middle class incomes had their fastest growth on record last year. The Census Bureau itself was a bit more circumspect because they recognize that there is considerable uncertainty in their median income statistics because very few resources go into measuring it. They only claimed that it was the fastest growth in median income since 1997-98 even though their data shows that the Washington Post is correct in saying that it is the fastest measured increase ever. The Census Bureau made a more modest claim because there is over a percentage point of uncertainty in their estimates.

It is rare for economic statistics to include estimates of uncertainty like this and the fact that it is easy to estimate uncertainty for median income demonstrates that median income is really easy to measure. It is also a demonstration of how unimportant median income is that we tolerate such an unprecise measure. As I explained in a WLIO TV interview last week, there are a number of simple things we could do to improve the measure.

First, rather than using a tiny Census sample, we could simply use income tax data and that would make the measurement extremely precise at approximately zero cost.

It would also eliminate the minor controversy about the effect of the changing Census survey methodology which also boosted the official measure. Most people don’t care about median income, so it hardly got any press, but the Census started changing their methodology in 2013 in a way that would tend to boost the measure in later years. The Census believes that they had been undercounting median income in previous years. In contrast, income records from the IRS would be much more comprehensive and consistent over time, so there would be less possibility for this kind of problem.

Secondly, we need to adjust for changes in household size which has been falling. That would have boosted real median individual income over time as household size shrinks.

Third, we should combine median income with life expectancy data to calculate median expected lifetime income (MELI). That would not only add richness by including our most powerful measure of health, but it would also correct the demographic effects of an aging population.

Fourth, we should use a better inflation index. The Census Bureau uses the rather odd CPI-U-RS for comparing growth in median income over time. Nobody else uses this. Sentier Research uses the CPI to adjust median income and this is a more standard measure that makes the growth in median income look worse. The CPI-U-RS has been essentially the same as the CPI since the late 1990s, but before that, there were substantial differences. But the best measure of inflation is probably the PCE price index which would show even more optimistic growth than the census bureau as Doug Short demonstrates:

Fifth, we should try to measure real income better. Ideally we should subtract off the value of taxes because that isn’t really income and we should add in the value of government benefits like education and healthcare, but the latter is really controversial to do, so either pre-tax income or after-tax income is probably all we are going to get for now because it is easy to measure cash flow and it is harder to measure non-cash benefits like the value of employer-provided health insurance. Also, homeowners get a large stream of economic benefit from living in their homes rather than paying rent. This should be included in income measures too, but it would be a bit more expensive to add this detail so it is rarely attempted.

Matthew Yglesias at Vox has ideas about how to improve median income measures that substantially overlap with mine.