A flaw regarding the chance of “out-earning” your parents

When Raj Chetty publishes a paper, it generally comes with a splash. The last one is no exception. His paper (co-authored), picked up by David Leonhardt at the New York Times and Justin Wolfers on Twitter, basically measures the American dream : what are your chances to do better than your parents. The stunning conclusion is that someone born in 1940 had a 90%+ chance of “out-earning” his parents compared with a few points above 50% for those born in the 1980s. I am not convinced. Well, when I am not convinced, I am saying I am not convincing about how big the drop is! I think the drop is smoother (the slope of decline is gentler) and the starting point for the 1940 cohort is too high.  As a big fan of Chetty, I must press this point.

More precisely, I am saying that the bar (income threshold) over which someone had to jump in 1940 is underestimated and overestimated in 1980. Setting the bar too low (high) means very high (low) chances of “out-earning” your parents. To set the bar too low, you must underestimate (overestimate) the income of the parents.  This could occur if household economies of scale are not accounted for.

An income of 30,000$ for 3 persons is not the same as an income of 60,000$ for 6 peoples. On a per capita basis, the income is the same. But, if you adjust for economies of scale in housing and furnitures, there are differences (the simplest is square root).  This gives you income per adult equivalent. Chetty et al. are aware of that and they provided a sensitivity analysis which is not mentioned by those who are relaying the article. Since household size has tended to fall over time, the growth in per capita income is faster than the growth in income per adult equivalent (a better measure). Any correction for this long-term demographic trend would attenuate the slope of the decline of the chance to out-earn your parents. And indeed, once Chetty et al. make the correction, the decline is much more modest (but still present – see below).


Simultaneously, Chetty et al. also present other important sensitivity checks. All of them relevant. But, in a strange decision, Chetty et al. decided to isolate each of the sensitivity checks rather than compile them. Taken individual, they all seem minor – except adjusting for family size. But compound this with the other sensitivity check proposed by Chetty et al.: price deflators. Using the well-known bias in the the CPI that overestimates inflation by 0.8%, Chetty et al. find that, by the end of their perod, there is roughly a ten percentage point difference between the baseline uncorrected CPI and the corrected CPI (see below). Compound this with the corrections for family and you still get a decline – but again the slope of the decline is much more modest. If you add panel B from figure 3 in Chetty et al – which includes taxes and transfers – you probably get a few extra points up. There will still probably be a decline, but a moderate one.


Finally, at footnote 19, Chetty et al. also point out that they do not account for in-kind transfers prior to 1967 (there were some).  And, on page 13, they point out that “one may be concerned that levels of absolute mobility for recent cohorts may still be understated because of increases in fringe benefits, nonmarket goods, or under-reporting of income in the CPS”. Add in all these little extra problems to the family size, the transfers and the inflation correction and I am not sure how big the drop from 1940 to the end of the studied period is. Finally, I would also add that an understudied point in economic history is what the distribution of in-kind payments according to income was. From studying the British industrial revolution, I have generally to see that it is the poorest workers who receive in-kind payments (which are not measured) and the richest receive much fewer of those in proportion of their incomes. One of the few to note that distributional was the hardcore left-leaning scholar Gabriel Kolko who mentioned this issue in Dissent back in the 1950s.  If Kolko is correct, then the income of “poor parents” in 1940 is underestimated. As a result, the bar over which the children of said parents must jump is set mildly too low. If that is the case, the odds for the 1940 birth cohort are overestimated.

Combine all of these things together and I am not sure that the drop is as dramatic as many are making it out to be. I would be very satisfied if Chetty et al. would publish all the corrections they did and do a sensitivity check with hypothetical regarding a sliding-scale of in-kind payments in 1940 according to income (10% of income for poorest to 0% for the richest). I would just like to see how much it matters.

15 thoughts on “A flaw regarding the chance of “out-earning” your parents

  1. I haven’t read the paper but it appears that the study ignores the most important question: what will your income buy? We have so many options that were unavailable to our parents and so many quality improvements that overshadow all these numbers.

  2. That is a utility argument that is impossible to “factor in” a study like Chetty (the fact that an iphone provides you much more utility than the TV you bought 25 years ago).

    • GDP is merely a vehicle to utility,. But utility is in the eye of the beholder, so Gross Domestic Utility is difficult to measure.

      There’s also the low-hanging fruit problems. I mean, eradicating smallpox had an awesome impact on gross utility, but you can only do it once. It’s doubtless true we’ve experienced less marginal impact from medical progress than in the age major vaccines were developed (still 90% of the overall healthcare benefit), but that’s just a result of the diminishing returns on investment as more resources are required to solve less impactful problems.

      I remain willing to perform a GDU survey using scalpels, but NBER has yet to admit the merit of my proposal.

  3. I’m neither an economist nor an historian, but ran across the Chetty paper and this discussion. The part that confuses me is the use of 1940 as the baseline. (Here, I am mostly dependent on the news reports and the web site as I haven’t fully read the paper itself.) A child born in 1940 has parents born in 1920 or a bit earlier and lived through the effects of the great agricultural displacement and the Great Depression. It seems very likely to me that the child of 1940 would be very likely to surpass their parents income as their parents would have been deeply affected by the economic turmoil of their teen years.

    Once you move out of the cohort of children born of parent of the Depression, effect size seems to decrease.

    It seems an interesting question about the after effects of large economic displacements. Is that relevant though to the general issue of decreasing opportunity?

    I would expect that parent age for children born close to 1940 would have a large effect size. And that seems to be true (figure S8 and 3C).

  4. Being of an age to know these things (78), I should point out that there are a lot of differences between life in the early post war world. The cost of health care relative to income was far lower back then. Like a local hospital room rate of $10 a day. Of course the hospital back then is more comparable to a skilled care nursing home today. They had oxygen, antibiotics, x-ray machines, iv drip. Basic wage level was $.75 an hour or $30 a week. Keep that in mind when you make the comparison. Unionized industrial workers earned considerably more, but the work was hard and you “earned” every dollar you made. Your new car was “crude” by today’s standards. Manual 3 speed transmission, low performance low compression engine. Fortunately they were simple to maintain because they required a lot of more service than today’s cars do.
    Gasoline was of course “cheap”, but 18 mpg highway (55 mph) was about all you got. Leaded gas of course. Manual steering, manual brakes, vacuum windshield wipers. Tire life wasn’t very long either back then. No air conditioning. Heaters recycled air from inside the car so windows steamed up in cold weather. Of course they were 100% “Made in the USA”. It was a much different society than the one that we now live in.

    TV sets were “small screen” B&W tube sets. Also quite expensive relative to income. The cheapest were in the $200 range (1948 dollars) and went up from there. At $30 a week wages, it would take 7 weeks of income to buy one. Even a basic radio wasn’t that “cheap”. Telephone was “party line” and long distance calling was “expensive”. No air conditioning. Some people were still burning coal in their furnaces. Most adults smoked cigarettes. Probably just as well as blood pressure and cholesterol meds hadn’t been invented yet. Kids all had the standard childhood diseases. Mump, measles, chicken pox, whooping cough. Polio crippled many. Cancer was 100% fatal. Doctors only had an “office nurse” as staff. You could get my stepfather (a dentist) to pull your tooth for $6.50. I had an ear wax build up in one of my ears. Cost me $8 to have a doctor wash it out. This was in 1960.

    Were things better back then? I don’t think very many people of today would want to go back to that time. My first job was in 1959. I worked 48 hours a week for $50. Salaried clerk in a camera store. Of course we didn’t miss stuff people take for granted now because it hadn’t been invented yet or was limited to rich people.

    • Great comment! Do you live in Muskegon? I work in the old Kelvinator plant (speaking of bygone eras) in Wyoming, if you recall that.

  5. Some other misleading parts here stem from the fact that this is not real longitudinal data about families and children. It doesn’t appear to account for things like immigrants and differing birth rates at different income levels. I am no statistician, but I don’t find that the copula is going to be able to capture things like real data would.

    1940 is a very misleading year at which to start.

  6. I don’t understand the impulse to adjust for family size. People’s preferences for children seem rather irrelevant for what we’re trying to determine here. Taken to an extreme example, has society improved 20% if it was originally composed entirely of couples making $10/hr and having 2 kids each but now the couples are making $6/hr and having no kids? If anything, the amount people spend on their kids is an important revealed preference that is insufficiently being accounted for in the inflation measures.

    Another massive revealed preference that’s seen a great deal of change is neighborhood, especially social quality and access to a good labor market. I’m not making any judgments for nor against those preferences, merely observing those preferences have gotten extraordinarily expensive and the difference might be large enough to impact inflation measures, but in my imperfect awareness this isn’t accounted for the way economists attempt to measure the quality of a car.

  7. […] 上記は,『ニューヨークタイムズ』のデイヴィッド・レオンハートの記事から引用した.紹介されている元ネタは Raj chetty による新研究だ.Jim Tankersley の文章にもっと詳しく書かれている.『ブルッキングス』での報道はこちら.さらに,ここで Vincent Geloso がデータのいろんな調整法について考察している. […]

Please keep it civil (unless it relates to Jacques)

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s