A couple of months ago, NBER released an working paper by Bruce Sacerdote disputing the finding of official statistics that median and lower income households haven’t had any growth in real wages over the last 40+ years. The measurement problem revolves around bias in the price indexes used for deflating nominal income.
There’s a lot to discuss in Sacerdote’s paper, so I plan to return to it in a future post. But for this post I’d like to look at a narrower question that was asked by Kevin Drum of Mother Jones. He looks at a chart showing median earnings of 25-to-34 year old men, deflated either by the BLS consumer price index for all urban consumers (CPI-U) or by the BEA personal consumption expenditures price index (PCE-PI):
Drum describes the two series:
The CPI story is grim: In the previous generation, young men earned about 8 percent more than their fathers. That’s not great, but it’s better than nothing. However, in this generation, millennial men earn 10 percent less than their fathers.
The PCE story is different. In the previous generation, young men earned 22 percent more than their fathers. That’s pretty good. In the current generation, millennial men earn about the same amount as their fathers. Stagnation like that is bad news, but at least millennials aren’t literally losing ground.
He then asks a critical question:
So which should we believe? There are arguments for both, and it’s a political hot potato too since inflation measures show up in all sorts of benefit calculations. It would be nice if the economic community could thrash out agreement on an overall best measure, and then make it available as a standard series going back 70 years, but if it turns out that the new measure leads to (for example) lower cost-of-living adjustments for Social Security benefits, you can expect a massive pushback…
All that conceded, we really should be able to agree on a good, general-purpose inflation measure. We can still have lots of different measures for specialized purposes, but the headline inflation rate should be something that, say, 90 percent of economists can agree about.
What would be a good, general-purpose inflation measure for household income? A couple of caveats—first, I’m looking at this strictly as an analytical question, separate from the political question of what measure should be used for escalating Social Security benefits. (I think the discussion of escalation policy also needs to address some broader questions, such as “what goals are we trying to reach through escalation?” and “do we want escalation to maintain absolute standard of living or relative standard of living?”) Second, I want to restrict my attention to statistics that are either available or could be readily calculated using official statistics, which means some long-standing issues in price measurement, such as new goods and quality adjustment, aren’t going to be fully resolved. The agencies are trying to address these problems, but they acknowledge that further work is required and that the available methods aren’t universally applicable.
I see four characteristics that I’d like my general-purpose inflation measure to have:
- The index should use chain-weighting. This method, which updates the weights each period to use current weights, ensures that the price index more accurately reflects the adjustments that consumers make as relative prices change. For example, 20 years ago, large flat-screen high-definition televisions were extremely expensive, and few consumers purchased them. Now they are relatively cheap and have become ubiquitous. Chain-weighting updates the weights to account for the fact that consumers are buying many more flat-screen, high definition TVs, and thus gives this spending category an appropriate weight in the index.
- The index should cover a long historical period, enabling the kind of long-term comparisons that Sacerdote, Drum, and other would like to make, and it should be based on methods that are as consistent as possible over time. That means the index will have to subject to revision, since price index methods have changed over time, and revising the historical data is the only way that we can attempt to maintain consistency.
- The weights should reflect household expenditures as accurately as possible. This issue is relevant because there are well-known problems with responses to the BLS consumer expenditure survey, which is used to calculate the weights for the CPI. A 2015 book from the Conference for Research in Income and Wealth, Improving the Measurement of Consumer Expenditures, delves into these problems. A chapter by William Passero, Thesia Garner, and Clinton McCully shows that after a detailed reconciliation with comparable personal consumption expenditures, the CE survey was only picking up about 74 percent of household expenditures in 2010, down from 84 percent in 1992. A chapter by Caitlin Blair shows that, relative to the reconciled PCE estimates, the the CE survey respondents tend to under-report items that respondents may have trouble recalling, such as apparel and other goods, as well as sensitive items like tobacco and alcoholic beverages. Conversely, items like housing that are well reported are over-represented in the CPI. The 2003 study by David Lebow and Jeremy Rudd, as well as Blair’s chapter, indicate that use of the less accurate CE survey weights results in small, systematic bias in the CPI.
- The weights should reflect spending directly attributable to households, rather than expenditures made “on behalf” of households. This issue is important because PCE, as part of GDP, includes several types of expenditures made on behalf of households in order for GDP to be a comprehensive measure of the nation’s production. In particular, PCE includes the final consumption expenditures of nonprofit institutions serving households (such as nonprofit universities, hospitals, and religious institutions). PCE also includes the expenditures made on behalf of households by certain government social insurance programs, such as Medicare and Medicaid. Because BEA’s PCE classification system now separately identifies spending by nonprofit institutions, it would be quite easy for them to calculate PCE excluding nonprofits. The exclusion of government social insurance benefits made on behalf of households is a bit trickier, but I know that BEA staff have done some work on the topic. (I’ll note that the international guideline, System of National Accounts 2008, recommends that these expenditures be counted as part of government final consumption expenditures, though it also provides for a separate presentation of “actual consumption,” which would show household actual consumption including expenditures made on behalf of households.)
How do these criteria match with the available indexes?
BLS’s featured CPI’s, the CPI for all urban consumers (CPI-U) and the CPI for urban wage earners and clerical workers (CPI-W) are not chain weighted, do not revise the historical data to provide consistency, and use the biased CE survey weights. While these indices play an important role in assessing current inflation and providing escalators for various programs, for longer-term assessments of the standard of living I’d recommend that analysts avoid using these indices. The BLS uses these indices to calculate real average hourly and weekly earnings, but given the availability of better indices, I think this choice should be reconsidered.
The BLS also produces the CPI research series using current methods (CPI-U-RS), which revises the historical CPI data to incorporate various methodological changes that BLS has made. These changes include the adoption of owners’ equivalent rent in the early 1980s, the adoption of the geometric mean formula to address lower-level substitution bias in the 1990s, and the increased use of hedonic methods for quality adjustment over time. The Census Bureau’s official estimates of real median income are deflated using the CPI-U-RS, and tend to produces estimates in between those based on the CPI-U and the PCE price index that were discussed by Drum and are shown in the chart above. The CPI-U-RS is preferable to the CPI-U, but because it is not chain weighted and relies on CE survey weights, it’s clearly possible to improve upon it.
The BLS also produces the chained CPI for all urban consumers, which a chain-weighted variant of the CPI. The use chained weights keeps the index current, though there is still some bias due to its reliance on the less accurate CE survey for the weights. The biggest downside to this index, however, is that unlike the CPI-U-RS, it hasn’t been back-cast to the historical periods, so estimates are only available beginning with December 1999. That is a critical flaw for using the index to study generations-long trends in the standard of living.
The BEA’s PCE price index is chain weighted and incorporates revisions intended to maintain substantial consistency in concepts and methods over time. Most of its component price indices come either from the CPI or from the components of the CPI-U-RS. The weights are regularly benchmarked (at 5 year intervals) to data from the economic census and BEA’s benchmark input-output accounts, which reconcile and balance the supply and use of detailed commodities. While no weights are perfect, in my opinion the benchmarked estimates coming from the balancing process are the best available estimates of total household spending. The downside for the PCE-PI is that its scope is too broad, including spending made by governments and nonprofit institutions on behalf of households.
So, the bottom line is that of the currently available indices, the CPI-U-RS, the C-CPI-U, and the PCE-PI each have their strength and weaknesses, but none meets all of my criteria. The good news is that I think it should be possible, with relatively modest effort, to derive an adjusted CPI/PCE price index that uses PCE weights, albeit adjusted to remove expenditures made by governments and nonprofit institutions.
Who could produce this index? It probably would make sense for BEA to take the lead on it, because they already produce estimates of most of the weighting information that the index would use. But I’d really prefer that the work be done as a joint project by all three agencies, with the agencies committing to use the resulting index for all of their estimates of real earnings and real household income. If the statistical agencies can come to an agreement on the appropriate index and then use it consistently, I think it would be an important signal for what researchers should use. Maybe this could be something that “90 percent of economists can agree about.”