Over the weekend I posted a response by John Komlos to an earlier column in which I critiqued his working paper on trends in living standards. I am grateful to Komlos for taking the time to engage my column, but I did want to take up his latest arguments before moving on from this debate. Perhaps unsurprisingly, I did not find the response persuasive. I believe it is clear that Americans have experienced real, substantial income gains over time—just not as large in percentage terms as Americans in the past.
In his response, Komlos disagrees with my criticism that he has treated taxes incorrectly, but I am pretty sure that he has done so. This gets a little wonky, and if you’re uninterested, skip to the paragraph beginning, “Let’s move on to Komlos’s justification for excluding health benefits as income.” I promise there are other more interesting rejoinders in this column.
Komlos notes that he subtracts corporate taxes borne by labor and by capital from income, saying that, “Market Income in column D [in Tab 6 of the Congressional Budget Office spreadsheet he used--SRW] includes these tax figures.” Thus, he says, he is “reducing income but just once.” But since he is looking at trends in post-tax income, then he has indeed subtracted this income twice.
Tab 6 of the CBO spreadsheet defines quantiles in terms of pre-tax income, which is the sum of market income (column D) and transfers (column E). As Komlos says, market income includes corporate taxes as income. The rationale is that corporate taxes make workers and investors poorer than they otherwise would be. Admittedly, this is an atypical way to think about pre-tax income, but given that CBO’s post-tax estimates–the ones Komlos is “improving”–subtract corporate income taxes from this broad definition of income that includes those taxes as income to be taxed away, it’s a wash.
Put it this way: pre-tax income, according to CBO equals corporate taxes borne (in columns M and Q) plus other market income (in columns I-L, P, and R-T) plus transfers (in columns W-AA). For short, pre-tax income= C+OM+T. Post-tax income subtracts from this corporate taxes (column AF) and other taxes (columns AD, AE, and AG). That is, taxes can be thought of as C+OT. In that case, post-tax income= C+OM+T-(C+OT) = OM+T-OT.
What Komlos has done is to subtract C twice. He doesn’t want C to be included in pre-tax income, so he naively subtracts it from post-tax income, computing his “improved” post-tax income as OM+T-OT-C. But that is equivalent to C+OM+T-(C+OT)-C, or C-C-C+OM+T-OT. “C” has been subtracted out twice, because it was already subtracted out of the post-tax estimates when Komlos subtracted it out again.
A similar problem recurs in his treatment of the employer’s share of payroll taxes. Here he discounts the income that would otherwise be received by workers if it was not taxed away as the employer’s share of payroll taxes. But he does not discount the employer’s share of payroll taxes when they are deducted from income. If pre-tax income is the employer’s share of payroll taxes plus the worker’s share of payroll taxes+other market income+transfers (E+W+OM+T) and taxes are the employer’s share of payroll taxes plus the worker’s share of payroll taxes+other taxes (E+W+OT), then post-tax income is E+W+OM+T-(E+W+OT)=OM+T-OT. Komlos instead computes E+W(1/r)+OM+T-E-W-OT = W(1/r-1)+OM+T-OT, which will be too small.
Note, too, that he has not treated the employee’s share of payroll taxes symmetrically. He can’t because CBO doesn’t separate it out from (pre-tax) wage and salary income or business income. But that’s not much of a justification for treating the employer’s share of payroll taxes differently. At any rate, because income from both the employee and the employer’s share of payroll taxes is taxed away, no discounting is required when one looks at post-tax income trends. One ends up with OM+T-OT because all taxes counted as income are subtracted out as taxes.
Now, a different question is how to think about the employer’s share of payroll taxes as pre-tax income. Komlos’s rationale for discounting is that $1 of this income isn’t worth $1 today because it can’t be touched until retirement. But this $1 might offset $1 that the worker might have saved herself, and so it allows her to spend $1 she otherwise would have saved. This is likely why CBO makes no effort at discounting.
Komlos would probably reply that the likelihood of the $1 in payroll taxes producing savings is much smaller than the likelihood of the $1 in private saving doing so. In that case, one would want to discount the $1 received as the employer’s share of payroll tax using the difference of the likely return on $1 in private savings and the likely return on $1 in savings from the payroll tax.
This isn’t actually what Komlos does, and it’s impossible to estimate the likely return from payroll taxes. Komlos doesn’t quite say that Social Security and Medicare won’t be there for today’s Millennials. Instead, he cites polling showing that Millennials doubt they will get any Social Security. There is no reason to believe that Millennial public opinion corresponds in any way to the likely future of Social Security, and at any rate payroll taxes fund Medicare and unemployment compensation too.
So Komlos’s “improvement” is no more than a guess, and once again, no discounting at all needs to be done because this income doesn’t show up in post-tax income. The whole debate is moot.
Let’s move on to Komlos’s justification for excluding health benefits as income. In his response, he justifies excluding Medicare benefits because—he “believes”—“Medicare payments have increased….mostly because of the ageing of the population.”
Again, he is “improving” CBO’s estimates based on a guess. It is certainly true that with an older population, more people get Medicare. Just as fewer have earnings. Perhaps we should leave out earnings too then? More productively, Komlos suggests standardizing income by age, which of course, was my purpose in showing separate estimates for elderly and non-elderly households. Komlos seems uninterested in that improvement.
He also says the CBO Medicare and Medicaid income estimates “make little sense,” citing CBO’s finding that the second and third income quintiles get more Medicaid and Medicare than the first and that even the top one percent gets some Medicaid. There are several moving parts here that clarify why these estimates are likely fine.
Part of the issue is that the CBO quantiles are defined based on comprehensive pre-tax income, including Medicaid and Medicare benefits. Komlos is obviously thinking about quantiles based on an income measure that excludes them. Tab 7 in the CBO spreadsheet provides the same estimates as Tab 6, except this time constructing quantiles based on pre-tax and -transfer income. In these estimates, the bottom quintile averaged $4,600 in Medicaid benefits in 2011, the second quintile $2,800, and the third quintile $1,250.
The difference between the Tab 7 and Tab 6 estimates is straightforward. Once Medicaid benefits are added to the income on which quantiles are based, some Medicaid recipients in the bottom fifth of pre-tax and -transfer income move up to the second fifth, and some non-recipients in the second fifth of pre-tax and -transfer income move down to the bottom fifth.
You get the same pattern for Medicare. Using pre-tax and -transfer income as the basis for rankings, the bottom fifth gets $6,550 in Medicare, the second quintile $3,800, and the third quintile $2,850.
Another issue is that these quantiles are defined with respect to all households, including those with and without children, with elderly and nonelderly heads. The bottom quintile includes a lot of elderly and otherwise childless households. These households are much less likely to receive Medicaid benefits than households with children. Tab 16 of the spreadsheet indicates that households with children in the (aggregate) bottom fifth have much more Medicaid income than elderly and nonelderly childless households in any quantile. In the same way, elderly households in the bottom fifth get more Medicare than nonelderly households in any other quantile.
As for why some upper-income households receive Medicaid benefits, it is possible that this reflects a flaw in the way that CBO statistically matches tax returns in the Statistics of Income data to households in the Current Population Survey. However, it is also possible that these are households where assets have been drawn down in order to qualify for Medicaid benefits.
In one scenario, a household in the top one percent with a disabled child might draw down its savings to the point where it qualifies for Medicaid. The household would start the year well-off and end it lower-income. Alternatively, an elderly household drawing down its savings to get long-term care Medicaid benefits might end up in the top one percent in the course of taking distributions from its savings or selling assets.
At any rate, this is all beside the point, because if Komlos wants to exclude the small amount of Medicaid income from the top fifth or top one percent, that will have no bearing on trends below the top.
Komlos also believes the increases in Medicaid and Medicare income over time are implausible. But his logic is badly mistaken here. Referring to the 340% increase in real income from Medicare benefits in the second fifth of households from 1979 to 2011 Komlos writes, “It is clear that age-standardized it cannot be the case that people in the 2nd quintile are getting 4 times as many x-rays and other treatments as they did in 1979.”
Of course, it is not that people in the second quintile are having the same procedures and treatments four times as much as they used to on a per person basis. First of all, there are more people receiving Medicare benefits. So the aging of the population would have increased income from Medicare—as it reduced income from earnings—even without medical utilization rates changing. Second, Medicare pays for a lot more procedures and treatments than it used to, reflecting the rising generosity of health insurance generally over the period. And third, people may have increased their utilization rates on top of all these other changes (though not by anything like a factor of four).
National Health Expenditure data indicate that between 1980 and 2011, real Medicare expenditures rose by a factor of 6.1. The CBO data indicate that income from Medicare benefits rose by a factor of 6.7 over the same period, and average benefits per household rose by a factor of 4.6 (4.8 from 1979 to 2011). There is nothing to see here—CBO’s estimates are perfectly consistent with national Medicare trends.
Turning to cost-of-living adjustment, Komlos rejects theconclusions of BEA economists that even the PCE deflator overstates health care price inflation…just because. He cites the price change of one item among hundreds in the PCE consumption basket—television sets—to conclude that what the PCE does is all “a lot of manipulation.” Regarding the implication from inflation-adjustment methods that a $400 TV in 1979 is worth $22 today, he says, “Well, tell that to a consumer whose income has declined since 2000 that he/she should not feel bad, because now he/she has a sharper image on his/her TV.”
But incomes are no worse today than in 2000, according to both the more-comprehensive CBO data and other sourcesusing narrower income definitions. Komlos is trying to justify his data-based analyses with empirically incorrect claims from outside the analyses.
Komlos inveighs against the mediocrity of our price indices, citing several technical reasons to prefer the CPI-U to the PCE and the CPI-U-RS along the way. These technical objections are beside the point, as I have written elsewhere, because the most technically sophisticated price index put out by the Bureau of Labor Statistics closely aligns with the PCE over time and is not subject to Komlos’s technical complaints.
The reader should note that Komlos is rejecting the conclusions of government economists and statistical agencies over the years in advocating continued use of the CPI-U. He simply disbelieves their conclusions. This is no way to attempt to get at the truth about what has happened to living standards.
He notes that the CPI-U is used to adjust the official poverty thresholds, which was a political decision made in 1969 by the old “Bureau of the Budget,” not an empirical one. CBO, BLS, the Census Bureau, and the Fed all use an index other than the CPI-U for research purposes—and have for over 20 years.
Komlos makes no effort to address my claim that even the PCE is likely to produce income growth trends that are understated. I came to this conclusion based on the arguments of technical experts on inflation measurement. Komlos feels no need to confront their arguments because of his gut feeling about things like hedonic pricing. Instead, he continues to cite his low-end estimates based on the discredited CPI-U.
So much for inflation adjustment.
Komlos wants me to address his arguments about the change in welfare over time, but estimating welfare trends is a Herculean task next to estimating income trends. Estimates of welfare levels and distribution depend heavily on the assumptions that go into the modeling. Komlos, for instance, lets relative standing affect welfare, so that apart from income levels, income concentration at the top affects welfare lower down.
That is a reasonable assumption on some level, but how much relative standing matters compared with absolute standing is unknown. Further, this sort of exercise leads down a path that isn’t productive for economics. It muddles positive questions about what the economy gives to whom with normative ones about what kind of subjective feelings of utility are legitimate for policy.
My level of welfare may be affected by a rise in the top one percent’s share. But it may also be affected by my sense of entitlement. If I feel worse off having to work because I believe my neighbors should subsidize my lifestyle, is it OK to take that into account in estimating welfare levels? If the American middle class is as rich as it has ever been and well up into the top one percent of global-historical income, is it legitimate to take dissatisfaction into account because we have gotten used to being rich and want more?
Of course, the extent of dissatisfaction Americans feel about their living standards is an empirical, positive question, and here Komlos doesn’t simply get the evidence wrong, he offers none for his claims of widespread dissatisfaction. He writes,
"If middle class were doing as well as you suggest there would not be so much malaise in the country according to all polls. There would not be so many suicides and drug-addiction epidemic, so many people in jail and so many disaffected voters. Your optimistic view just doesn’t jive [sic] with what’s going on in the country."
Let’s come back to middle-class malaise and tackle suicide, drugs, and incarceration. Suicide has been rising since 1999, but if economics were the main reason, we’d expect to see more of a long-run cyclical pattern than this, and we’d expect that African Americans would have much higher suicide rates than white (instead of lower rates).
Use of drugs other than marijuana has been rising since the early 1990s, but it remains below 1970s levels, as this chart based on data from the respected Monitoring the Future survey shows:
The net flow into incarceration (inflow less outflow) follows increases in violent crime very closely. The violent crime trend bears little relationship to economic trends.
Komlos presents no data to back up any of these assertions. It sure feels like he is trying to rationalize his conclusion with outside arguments. But either his income results show a struggling middle class or they don’t.
Komlos also presents next to no evidence in asserting that malaise is widespread. In fact, consumer sentiment is about as high as it has been in 35 years (except for during the late 1990s boom). Surveys indicate that apparently high levels of economic anxiety are actually not significantly higher than other types of anxiety about low-probability events, such as dying in a plane crash, or becoming a victim of violent crime or terrorism. It is difficult to interpret these kinds of responses without such context.
In addition, survey respondents consistently indicate they believe other people are doing worse than they themselves are, a phenomenon that is not specific to economics and that has been dubbed the “I’m OK, They’re Not Syndrome.”
For instance, a March Marist Poll of Americans found that 58% say their children will be better off than themselves, versus 33% who predict their children will be worse off. But when the Marist Poll asked whether “most children” will be “better or worse off than their parents,” a plurality—48%—said that most would be worse off, compared with just 43% saying better off. A logical interpretation of these results is that people have more information about their own lives and rely on overly-negative sources of information to assess how others are doing. (Sources of information like Komlos’s paper.)
In the same Marist Poll, 72% of Americans said they are better off than their parents, while just 20% said worse off. As the analyses I presented in my original critique of Komlos show, people are not wrong to think they are better off than their parents. The median fortysomething is better off by 93% than her parents at the same age. The Pew Economic Mobility Project, where I used to work, found that between 67% and 84% of fortysomethings have higher incomes than their parents did.
To this last bit of evidence, Komlos might say—as he wrote in his response to me—that this increase in income is primarily due to the greater number of two-earner families today compared with the past. He believes that this reflects increasing hardship, saying,
“Now it takes two [earners] per household. Yet, the [one-earner] family back then was able to save more than twice as much as today and had no credit card debt. Today, the credit card debt alone is some $7.5K per family of two adults and that even counts those who have zero debt. In addition, most households have two people working and have expenses associated with that including additional child-care expenses, travel, clothing, etc. that makes it much harder today.”
But there is no evidence that in general “it takes two” earners to achieve the same standard of living that used to take one. Men’s earnings have stagnated, but they have not declined. The biggest increases in hours of work have come from the wives of the best educated husbands. Hours worked have declined among husbands but not among single men, suggesting that the former have cut back in response to the help they get from a second earner (not that wives are increasingly compelled to work because of their husbands’ falling earnings).
Further, the rise in women’s labor force participation began in the 1940s, two decades before men’s earnings began to stagnate. Women have increasingly wanted to work, which is why, around the world, female educational attainment has risen while marriage and childbearing have been delayed or foregone and while fertility has fallen.
Komlos wants to deduct work-related expenses from the income of two-earner families today, just like Elizabeth Warren wanted to back when she was a law school professor making bad economic arguments (a topic for another column someday). But if we go down this road, we should add to “income” the psychic gains women enjoy from working. After all, if a mother joins the workforce and, on net, adds no income to the household because of work-related expenses, she must be working because it provides non-monetary gains.
(Note, by the way, that Komlos’s it-takes-two-when-it-used-to-take-one claim relies on an argument completely foreign to his paper—it questions not the rise in incomes but reinterprets the rise as not representing improvement.)
Komlos raises the specter of high credit card debt, though he has no citation for his $7,500 figure. According to theSurvey of Consumer Finances, in 2013, the average family with credit card debt had $5,700 in debt. But that does not include families without credit card debt. That means the average family (with or without credit card debt) has $2,200 in debt.
A small number of families with a lot of debt pull the mean upward. That’s why Komolos and the rest of us present medians when we talk about middle class incomes. Themedian family with credit card debt—that is, excluding zeroes—has $2,300 in debt. Because only 38% of families have credit card debt, the median debt across all families is…$0.
One last misinterpretation from Komlos: He claims that almost all the increase in the income of the middle class shown in the CBO data is due to transfers. This is purely an artifact of combining retirees and non-retirees. As the population ages, Social Security and Medicare become a bigger and bigger share of income in the middle fifth. I’ve addressed this at length elsewhere.
Ultimately if Komlos and others want to look at my numbers and say that having $10,000 to $15,000 above and beyond what our counterpart households had in 1979 constitutes “stagnation” or an immiserated middle class, they are welcome to do so. Different people can interpret a rise in income of that magnitude differently. But what they can’t do is insist their own bad numbers are correct. Doing so requires contortions that worsen—not improve—carefully estimated statistics like CBO’s.
This piece originally appeared in Forbes
This piece originally appeared in Forbes