Valentin Klotzbücher

Economist; now statistician in clinical research

[I4R @I4Replication]

1/ What is the prevalence of personally identifiable information (PII) in author-collected data publicly shared in the social sciences?

We audited 327 replication packages from 11 leading journals in economics, political science, and psychology

Answer: surprisingly high 🧵

[Tim Hua 🇺🇦 @Tim_Hua_]

Never underestimate the power of faith.

[Rex "garbage in" Douglass Ph.D. @RexDouglass]

So I like Data Colada, but I want to point out something:
-The studies are so bad they should have been ridiculed independent of sluething
-People think this is how social science checks itself. It's not. We have no real mechanism at all.
-Smoking guns are memes, not science

[I4R @I4Replication]

The Institute for Replication and University of Freiburg are jointly organizing the Replication Games on Saturday, September 19th.

The Games will focus on replicating papers on the causes of political violence.

[Kevin A. Bryan @Afinetheorem]

This kind of thing is *everywhere* in the corporate and academic sector. If you are an exec and you see this, call your IT head or CIO and fire them because they do not understand the correct balance of risk and benefit of AI and are going to kill your company.

[Philippe Lemoine @phl43]

Once again, if people really did systematic robustness analyses that take seriously researcher's degrees of freedom, most results in economics and other quantitative social sciences would disappear.

Some of the results based on quasi-experimental methods would survive, but they rarely have much relevance to policy because the effects they estimate don't tell you much about the effects that are policy-relevant.

You need to make a ton of additional assumptions to get from the effect your quasi-experimental method is allowing you, in the best case scenario, to estimate somewhat reliably to the effect that is policy-relevant.

But economists generally ignore that and make sweeping claims based on those results and pat themselves on the back while jerking off about the so-called "credibility revolution in economics" and how good we're having it compared to the dark ages of economics.

This is the right criticism of economics if you ask me, one of them anyway, but instead people go into those insane rants about "neoliberalism" or whatever and implicitly criticize what is actually valuable about economics, i. e. the fact that it encourages people to think in terms of tradeoffs, to consider general equilibrium effects, etc.

[Rex "garbage in" Douglass Ph.D. @RexDouglass]

Correct. Nothing about this makes sense if you actually know how to use AI to do reviewing.

[Luis Garicano 🇪🇺🇺🇦 @lugaricano]

I think the problem is that the piece conflates two separate things:

1. As an academic economist, the measures of research productivity and impact are quite objective and the reality is hard to dispute: no scholar alive is as influential as @DAcemogluMIT across so many fields. So the answer to the question below is a resounding yes: his reputation is justified by his scholarship.

2. As a public intellectual, everyone is fair game. I loved "How Nations Failed", did not agree with "Power and Progress", as well as with much of what Daron has written on AI. No one (and certainly Acemoglu does not) should expect others to agree with one's own views just because one is a brilliant economist.

Now, separately: as a profession, we cannot do more to confirm the prejudice about status behaviour in our field than start asking for the cancellation of @TheEconomist or engage in dumb mob behaviour towards it.

[Andy Masley @AndyMasley]

@TomTiffanyWI @DavidCCrowley Complete bullshit.

Data centers aren't draining our lakes. The AI water issue is fake. https://t.co/QOYLoDhxYg

And by 2030 will take up as much land as Disney World, or 1/15th of the land we use for Christmas trees.
https://t.co/ELhPDPP9em

[Luis Garicano 🇪🇺🇺🇦 @lugaricano]

🧵 Not many things fill me with rage. One of them is the way scientists create deadly viruses and, even after COVID, governments do nothing to avoid this.
Today's Times describes the state of play. A thread
1. Lab leaks are common.
https://t.co/0Znok2beuA

[Tom Nuttall @tom_nuttall]

It's grimly amusing how even when it comes to high-intensity low-stakes culture wars, Germany feels about five years behind the rest of the world.
https://t.co/vHw8oaCOGH

[Rex "garbage in" Douglass Ph.D. @RexDouglass]

You'd think that I'd be happy about the Refine announcement but seriously this should be an open source pipeline that's interogatable. There was a brief moment you needed more than prompt engineering for this - but that's not true anymore. This is already one shot.

[Khoa Vu @KhoaVuUmn]

Publish fake data and get tenure at a top business school in the US.

[François Valentin @Valen10Francois]

French MP argues we should give extra money to French pensioners retired in Italy to pay for their tax lawyers

The doom loop marches on

[Jesús Fernández-Villaverde @JesusFerna7026]

A few weeks ago, Daron Acemoglu (@DAcemogluMIT), David Autor (@davidautor), Keelan Beirne, and Andrew Scott circulated a provocative paper, “Baby Busts and Growth Booms: Demographic Change and the Macroeconomy,” which reports that lower birth rates are associated with higher growth in GDP per working-age adult across countries and with higher wage growth across US commuting zones, with no negative impact on aggregate GDP or earnings.

Since I have been working on fertility and growth for some time, I read the paper carefully. I did not find the evidence compelling.

My reason fits in two words: terra incognita. The fertility collapse now underway has no historical precedent. The seven decades of data the paper analyzes, however carefully, contain nothing remotely comparable to the ultra-low fertility rates we observe today.

To their credit, the authors know this. The conclusion states: “Our findings describe the past; whether they provide a reliable guide to the economic consequences of coming demographic transitions is not yet established. … Changes this rapid are outside the support of the historical evidence, and could theoretically yield different adjustment dynamics from those we document.” I agree word by word. I simply would have given this caveat more prominence because it is the heart of the policy question.

Let me explain why by discussing the evidence across countries and leaving the evidence from US commuting zones for another day. I want to keep this post short.

In the introduction, the paper presents as its key finding a negative and quantitatively large relationship between 1950 birth rates and economic growth from 1970 to 2020.

But the birth and fertility rates of 1950 provide little information about the consequences of the ultra-low birth and fertility rates we are experiencing today. Observations from a very different range of values are unlikely to capture the nonlinear cumulative effects of ultra-low birth and fertility rates.

I include a table listing the bottom ten countries in 1950, 1980, and 2025, according to the UN World Population Prospects, consistent with the authors’ sample exclusions for microstates, territories, and tax havens. In 1950, Austria had the lowest birth rate, at 15.5 births per 1,000 people, with a total fertility rate (TFR) of 2.09, roughly the replacement rate. Most of the negative effects I have emphasized in my writings stem from rapid population decline, as will be the case for South Korea, with a TFR of 0.75. If you ask me whether I expect problems for a country with a TFR of 2.09, my answer is no: no shrinking cities, no empty schools, no social security dilemmas. In fact, I believe that TFRs between 2 and 2.4, such as those in Western Europe in the early 1950s, are probably the best-case scenario in terms of economic growth: you have either a stable population or a gentle increase, and you do not suffer the pressures of TFRs of 3.0 or higher. Thus, I do not find extrapolating from Austria’s growth experience from 1970 to 2020 (which is effectively what the regression does) very informative about South Korea’s growth experience from 2045 to 2095: the former was close to an ideal demographic situation, the latter not so much.

Moreover, the range of TFRs (which say more than birth rates about the future behavior of the population) at the bottom of the cross-section is not wide enough to learn much about the nonlinearities that appear below 2.1. In summary, I cannot see the external validity of this regression.

The authors, with outstanding craftsmanship, address this problem as best they can by using later-vintage birth rate and growth data. If I read Appendix Table A1 (Panel A, countries) correctly, they run the same specification with 30-year growth windows for each starting decade. The coefficients on the t−20 birth rate (measured, as in the paper, per 100 people) are: 1950 vintage (growth 1970–2000): −0.13 (s.e. 0.10), not significant; 1960 (1980–2010): −0.26 (0.09); 1970 (1990–2020): −0.32 (0.06); and 1980 (growth 2000–2020, only a 20-year horizon): −0.26 (0.05). They also report robustness to using log birth rates, but taking logs of the regressor does nothing about the nonlinearities that matter, namely those well below replacement.

The problem is that even in 1980, birth and fertility rates were not as low as they are today. In the table, you can see that Germany, the country with the lowest birth rate at the time, had a TFR of 1.56 (and no country was below Denmark’s 1.54). But keep two facts in mind.

First, less importantly, Germany, like the other countries in the 1980 panel, has experienced large waves of immigration over the past few decades. By comparison, countries with high fertility in 1980, such as Mexico, have had large emigration flows. Therefore, the reported coefficient averages effects that include each country’s endogenous migration response, and the low-birth-rate observations come precisely from countries with positive net migration. The paper argues that lagged birth rates are orthogonal to subsequent flows. But orthogonality is not the same as informativeness about countries that will have little migration to speak of.

Second, more importantly, let me return to my central point about nonlinearities. A TFR of 1.56 means each generation is 76% the size of the last: the population declines about 0.9% per year, so after 100 years you retain roughly 40% of the initial population. A TFR of 0.75 means each generation is 37% of the last: about a 3.35% decline per year, leaving roughly 3.5% after a century. Hence, we are dealing with a factor of about 11. A TFR of 1.56 and a TFR of 0.75 both count as “low fertility” in a regression, but they are qualitatively different regimes. The 1980 variation the paper identifies lives in the gentle-decline world; nothing in their data resembles the second.

My favorite part of the paper is the evidence linking lower birth rates to more labor-saving patents, larger shares of high-tech exports, and faster TFP growth, with a WWII casualty exercise presented as an out-of-sample test of the labor-scarcity channel. I take this evidence seriously. But the mechanism does not extrapolate any better than the regression. In any semi-endogenous growth model in the tradition of Chad Jones, fewer people eventually mean fewer ideas. With a TFR of 0.75, the pool of potential innovators shrinks by almost two-thirds in each generation, and that is likely to overwhelm any gain from developing labor-saving technology. In comparison, the war shock was bad for one cohort, but then it was over.

I have some further minor quibbles about identification in the cross-section, but those are better left to a seminar discussion than to X.

I want to close by acknowledging that the assessment of empirical exercises is ultimately subjective. I do not find the evidence from Austria in 1950 or from Germany in 1980 persuasive regarding what will happen to South Korea. Instead, I find the evidence I reported in Table 1 of “The Wealth of Working Nations,” with Gustavo Ventura (@King_ofSweden) and Wen Yao, relating GDP growth in G7 countries plus Spain to demographic forces, to speak much more directly to the issue at hand and to imply much more negative outcomes, yes, including for Germany:

https://t.co/fclL5aozqL

Other readers might find the evidence in “Baby Busts and Growth Booms” more compelling, or at least intriguing. But we should not fool ourselves: we are sailing into demographic terra incognita, and in these waters, the pretense of knowledge is the most dangerous temptation. I am keenly aware of the limits of what I can determine.

[Luis Garicano 🇪🇺🇺🇦 @lugaricano]

These rent control laws, and the disability laws, and the migration,... all of them, the unifying feature is to entirely disregard incentives. Postulate that people are angels, and then legislate. We are not angels. Incentives work. The legislation makes things worse.

[Santi Ruiz @rSanti97]

Your project should be more ambitious. Intelligence is cheaper than ever. Increasingly, money is cheap. And the world needs new institutions. Go harder!

[Luis Garicano 🇪🇺🇺🇦 @lugaricano]

Having been involved in budgetary negotiations in Spain, I agree completely: investment is the residual once pensions and healthcare are funded. That is where the political pressure is, and the rest is leftovers. Policy in the West comes down to demographics.

[Rex "garbage in" Douglass Ph.D. @RexDouglass]

Activist scholarship doesn't have to be bad it just is. Almost to the person. The reason is collider bias- you only need one route to money and being good work and being political propaganda are both routes that pay and tend to trade off with one another.