Category: Data Source

Is scholarly refereeing productive at the margin?

No, basically:

In economics many articles are subjected to multiple rounds of refereeing at the same journal, which generates time costs of referees alone of at least $50 million. This process leads to remarkably longer publication lags than in other social sciences. We examine whether repeated refereeing produces any benefits, using an experiment at one journal that allows authors to submit under an accept/reject (fast-track or not) or the usual regime. We evaluate the scholarly impacts of articles by their subsequent citation histories, holding constant their sub-fields, authors’ demographics and prior citations, and other characteristics. There is no payoff to refereeing beyond the first round and no difference between accept/reject articles and others. This result holds accounting for authors’ selectivity into the two regimes, which we model formally to generate an empirical selection equation. This latter is used to provide instrumental estimates of the effect of each regime on scholarly impact.

That is from a new NBER paper by Aboozar Hadavand, Daniel S. Hamermesh, and Wesley W. Wilson.  This is exactly the kind of work — critical, data-driven self-reflection about science — what Progress Studies wishes to see more of.

How much did the bailouts cost?

Deborah Lucas has studied this question, and here is the core of her results:

This review develops a theoretical framework that highlights the principles governing economically meaningful estimates of the cost of bailouts. Drawing selectively on existing cost estimates and augmenting them with new calculations consistent with this framework, I conclude that the total direct cost of the 2008 crisis-related bailouts in the United States was on the order of $500 billion, or 3.5% of GDP in 2009. The largest direct beneficiaries of the bailouts were the unsecured creditors of financial institutions. The estimated cost stands in sharp contrast to popular accounts that claim there was no cost because the money was repaid, and with claims of costs in the trillions of dollars. The cost is large enough to suggest the importance of revisiting whether there might have been less expensive ways to intervene to stabilize markets. At the same time, it is small enough to call into question whether the benefits of ending bailouts permanently exceed the regulatory burden of policies aimed at achieving that goal

Here is the paper, via the excellent Kevin Lewis.

You will note that 3/4 of that sum comes from the bailouts of the government mortgage agencies.  I am myself uncertain how to think about this problem.  First, is it useful to think of the additional bailout expenditure as being monetized, if only indirectly through the mix of Fed/Treasury policy?  If yes (debatable), and the monetization itself limits a harmful further deflation, can it be said that this monetization is not a transfer away from citizens in the usual sense that an inflation in Zimbabwe might be?  But rather a net gain for citizens or at least a much smaller loss?  Is the interest paid on those monetized reserves the actual cost?

In any case, where exactly does the “3.5% of gdp” loss “come from”?

I do not know!

New evidence that YouTube doesn’t radicalize

The role that YouTube and its behind-the-scenes recommendation algorithm plays in encouraging online radicalization has been suggested by both journalists and academics alike. This study directly quantifies these claims by examining the role that YouTube’s algorithm plays in suggesting radicalized content. After categorizing nearly 800 political channels, we were able to differentiate between political schemas in order to analyze the algorithm traffic flows out and between each group. After conducting a detailed analysis of recommendations received by each channel type, we refute the popular radicalization claims. To the contrary, these data suggest that YouTube’s recommendation algorithm actively discourages viewers from visiting radicalizing or extremist content. Instead, the algorithm is shown to favor mainstream media and cable news content over independent YouTube channels with slant towards left-leaning or politically neutral channels. Our study thus suggests that YouTube’s recommendation algorithm fails to promote inflammatory or radicalized content, as previously claimed by several outlets.

That is from a new paper by Mark Ledwich and Anna Zaitsev.  That hardly settles the matter, but you may recall the last serious papers on this topic also indicated that YouTube does not radicalize.  So if you are still believing that YouTube radicalizes, you will need to come up with additional facts for your point of view.

Here is a Mark Ledwich tweet storm on the paper.

There is now a NIMBY index

Check out the new NBER paper by Joseph Grourko, Jonathan Hartley, and Jacob Krimmel:

We report results from a new survey of local residential land use regulatory regimes for over 2,450 primarily suburban communities across the U.S. The most highly regulated markets are on the two coasts, with the San Francisco and New York City metropolitan areas being the most highly regulated according to our metric. Comparing our new data to that from a previous survey finds that the housing bust associated with the Great Recession did not lead any major market that previously was highly regulated to reverse course and deregulate to any significant extent. Moreover, regulation in most large coastal markets increased over time.

One embedded lesson is that the number of veto points over new construction is increasing.  And “By our metric, about one half of all communities in the Regulation Change index increased regulation, one-third decreased, while only 18 percent showed no net change.”

Here is a graph of housing affordability vs. their index of restrictiveness:

Here is my earlier Bloomberg column calling for more indices — this is exactly what I wanted.

Which researchers really work long hours?

No, not work smart but put in what would appear to be lots of extra hours.  Why not measure who submits papers to journals in the off-work hours?:

Main outcome measures Manuscript and peer review submissions on weekends, on national holidays, and by hour of day (to determine early mornings and late nights). Logistic regression was used to estimate the probability of manuscript and peer review submissions on weekends or holidays.

Results The analyses included more than 49 000 manuscript submissions and 76 000 peer reviews. Little change over time was seen in the average probability of manuscript or peer review submissions occurring on weekends or holidays. The levels of out of hours work were high, with average probabilities of 0.14 to 0.18 for work on the weekends and 0.08 to 0.13 for work on holidays compared with days in the same week. Clear and consistent differences were seen between countries. Chinese researchers most often worked at weekends and at midnight, whereas researchers in Scandinavian countries were among the most likely to submit during the week and the middle of the day.

Emphasis added.  Get this, you lazy bastards:

The average probability of a manuscript being submitted at the weekend for both journals was 0.14, and for a peer review it was 0.18. Peer review submissions during holidays had average probabilities of 0.13 (The BMJ) and 0.12 (BMJ Open), which were higher than the probabilities for manuscripts of 0.08 (The BMJ) and 0.10 (BMJ Open).

For weekend paper submission, China appears to be at about 0.22, India at about 0.09, see Figure 1.  France, Italy, Spain, and Brazil all submit quite late in the afternoon, often a bit after 6 p.m.

That is from a new paper by Adrian Barnett, Inger Mewburn, and Sara Schroter.  They do not tell us when they submitted it, but I wrote this blog post a wee bit after 8 p.m.

Via Michelle Dawson.

Is the college wealth premium *zero*?

Now this one is a stunner:

The college income premium—the extra income earned by a family headed by a college gra duate over an otherwise similar family without a bachelor’s degree—remains positive but has declined for recent graduates. The college wealth premium (extra wealth) has declined more noticeably among all cohorts born after 1940. Among non-Hispanic white family heads born in the 1980s, the college wealth premium is at a historic low; among all other races and ethnicities, it is statistically indistinguishable from zero [emphasis added]. Using variables available for the first time in the 2016 Survey of Consumer Finances, we find that controlling for the education of one’s parents reduces our estimates of college and postgraduate income and wealth premiums by 8 to 18 percent. Controlling also for measures of a respondent’s financial acumen—which may be partly innate—, our estimates of the value added bycollege and a postgraduate degree fall by 30 to 60 percent. Taken together, our results suggest that college and post-graduate education may be failing some recent graduates as a financial investment. We explore a variety of explanations and conclude that falling college wealth premiums may be due to the luck of when you were born, financial liberalization and the rising cost of higher education.

That paper is by William R. Emmons, Ana H. Kent and Lowell R. Ricketts, and comes from the St. Louis Fed, not from some bunch of (college-educated) cranks.

Via the excellent Samir Varma.

Countries where the average person was richer in 2009

Libya, Yemen, Equatorial Guinea, Greece, the Central African Republic, Sudan, East Timor, Lebanon, Greece, and Trinidad & Tobago.  In Syria and Venezuela data collection has stopped altogether, but they would make the list too.

Ethiopia had the highest growth rate of the decade, with Nauru, Rwanda, Ghana, Mongolia, Turkmenistan, Laos, China, India, Bangladesh, Cambodia, and Myanmar as other winners too.  Note that the numbers for Turkmenistan are disputed, especially for the last few years of the decade, but still the country had a strong performance early on.

Here is the full FT piece by Steve Johnson.

China isn’t close to being the #1 economy

From my latest Bloomberg column:

The key point is the difference between income and wealth. GDP and related numbers measure income flows: namely, the quantity of goods and services produced in a given nation in a given year. Wealth is a measure of the total stock of resources in a nation and is much higher. Furthermore, the gap between wealth and income is usually higher for nations that have been wealthy and stable for a very long time, such as the U.S.

When it comes to national wealth, the U.S. has a big lead over China, possibly as much as three times greater. That is a very rough estimate by Michael Beckley of Tufts University, drawing on data from the World Bank and the United Nations.

For a relevant pointer to Beckley, I thank Evan Abramsky of AEI.

The gender gap in confidence, revisited — gender differences in research reporting

His team analyzed more than 100,000 medical studies and 6.2 million life sciences article that were published over a 15-year period, finding that women-authored studies were 12 percent less likely to contain at least one of a group of 25 positive terms, including “favorable,” “excellent” and “prominent.” In the most prestigious and influential journals, women were 21 percent less likely to describe their findings with such words.

Male authors deployed the word “novel” 60 percent more often than their female counterparts. “Unique” was used 44 percent more often by male authors, and “promising” was used 72 percent more often by male authors.

Here is the article, here is the unique study itself.

Sex Differences in Personality are Large and Important

Men and women are different. A seemingly obvious fact to most of humanity but a long-time subject of controversy within psychology. New large-scale results using better empirical methods are resolving the debate, however, in favor of the person in the street. The basic story is that at the broadest level (OCEAN) differences are relatively small but that is because there are large offsetting differences between men and women at lower levels of aggregation. Scott Barry Kaufman, writing at Scientific American, has a very good review of the evidence:

At the broad level, we have traits such as extraversion, neuroticism, and agreeableness. But when you look at the specific facets of each of these broad factors, you realize that there are some traits that males score higher on (on average), and some traits that females score higher on (on average), so the differences cancel each other out. This canceling out gives the appearance that sex differences in personality don’t exist when in reality they very much do exist.

For instance, males and females on average don’t differ much on extraversion. However, at the narrow level, you can see that males on average are more assertive (an aspect of extraversion) whereas females on average are more sociable and friendly (another aspect of extraversion). So what does the overall picture look like for males and females on average when going deeper than the broad level of personality?

On average, males tend to be more dominant, assertive, risk-prone, thrill-seeking, tough-minded, emotionally stable, utilitarian, and open to abstract ideas. Males also tend to score higher on self-estimates of intelligence, even though sex differences in general intelligence measured as an ability are negligible [2]. Men also tend to form larger, competitive groups in which hierarchies tend to be stable and in which individual relationships tend to require little emotional investment. In terms of communication style, males tend to use more assertive speech and are more likely to interrupt people (both men and women) more often– especially intrusive interruptions– which can be interpreted as a form of dominant behavior.

…In contrast, females, on average, tend to be more sociable, sensitive, warm, compassionate, polite, anxious, self-doubting, and more open to aesthetics. On average, women are more interested in intimate, cooperative dyadic relationships that are more emotion-focused and characterized by unstable hierarchies and strong egalitarian norms. Where aggression does arise, it tends to be more indirect and less openly confrontational. Females also tend to display better communication skills, displaying higher verbal ability and the ability to decode other people’s nonverbal behavior. Women also tend to use more affiliative and tentative speech in their language, and tend to be more expressive in both their facial expressions and bodily language (although men tend to adopt a more expansive, open posture). On average, women also tend to smile and cry more frequently than men, although these effects are very contextual and the differences are substantially larger when males and females believe they are being observed than when they believe they are alone.

Moreover, the differences in the subcategories are all correlated so while one might argue that even among the subcategories the differences are small on any single category when you put them all together the differences in male and female personalities are large and systematic.

Relatively small differences across multiple traits can add up to substantial differences when considered as a whole profile of traits. Take the human face, for example. If you were to just take a particular feature of the face– such as mouth width, forehead height, or eye size– you would have difficult differentiating between a male face and a female face. You simply can’t tell a male eyeball from a female eyeball, for instance. However, a look at the combination of facial features produces two very distinct clusters of male vs. female faces. In fact, observers can correctly determine sex from pictures with greater than 95% accuracy [4]. Here’s an interesting question: does the same apply to the domain of personality?

…There now exists four large-scale studies that use this multivariate methodology (see here, here, here, and here). All four studies are conducted cross-culturally and report on an analysis of narrow personality traits (which, as you may recall, is where most of the action is when it comes to sex differences). Critically, all four studies converge on the same basic finding: when looking at the overall gestalt of human personality, there is a truly striking difference between the typical male and female personality profiles.

Just how striking? Well, actually, really striking. In one recent study, Tim Kaiser, Marco Del Giudice, and Tom Booth analyzed personality data from 31,637 people across a number of English-speaking countries. The size of global sex differences was D = 2.10 (it was D = 2.06 for just the United States). To put this number in context, a D= 2.10 means a classification accuracy of 85%. In other words, their data suggests that the probability that a randomly picked individual will be correctly classified as male or female based on knowledge of their global personality profile is 85% (after correcting for the unreliability of the personality tests).

In other words, you can predict whether a person is male of female from their personality traits almost as well as by looking at their face. Overall, the big differences are as follows:

Consistent with prior research, the researchers found that the following traits are most exaggerated among females when considered separately from the rest of the gestalt: sensitivity, tender-mindedness, warmth, anxiety, appreciation of beauty, and openness to change. For males, the most exaggerated traits were emotional stability, assertiveness/dominance, dutifulness, conservatism, and conformity to social hierarchy and traditional structure.

I have also pointed out that gender equality magnifies differences in gender choices and behavior which is probably one reason why fewer women enter STEM fields in societies with greater equality. Consistent with this, personality differences between the sexes are large in all cultures but “for all of these personality effects the sex differences tend to be larger– not smaller– in more individualistic, gender-egalitarian countries.”

Addendum: See John Nye and co-authors on testosterone and finger length for some biological correlations.

We need more indices

That is the upshot of my latest Bloomberg column, as the Doing Business index, PISA scores, and the Corruption Perceptions Index have been highly influential.  Here are a few of my further requests:

These successes raise a question: Which other indexes might be useful? Think of the suggestions that follow as a kind of Christmas wish list.

How about a loneliness index? David Brooks has argued that America faces a crisis of loneliness, making us unhappy and impoverishing us spiritually. I find these claims plausible, especially since the median U.S. household size has been shrinking. Still, just how bad is this problem? One recent study found that American loneliness has not been rising lately, and that loneliness increases only after people reach their early 70s…

A stress index for Americans another related idea: Just how much do our lives focus our attention on our worries rather than on our joys and hopeful expectations?

There are less emotional concerns as well. How about an infrastructure speed index? I worry about bureaucratization and the slow pace of building important public works. Construction on Manhattan’s Second Avenue subway line, for example, started in 1972, paused, resumed in 2004, and was finally completed (the first phase, anyway) in 2017. In contrast, construction of the core New York City subway system, with 28 stations, began in 1900 and finished in 1904. Similarly, construction of the Empire State Building took only 410 days.

Why do so many U.S. infrastructure projects today take so long? And if the process of improving and reshaping the environment to further human progress is now so much slower, doesn’t it make sense to try to measure this decline for the purpose of eventual improvement? Given the need for a greener energy infrastructure, this is a matter of the utmost urgency.

Speaking of energy infrastructure, how about a severity index for climate change and associated problems?

There are further noteworthy suggestions at the link.  Which indices do you wish for?

The Old Boys’ Club

It is real, at least in one Asian data set, as these new NBER working paper results are brought to us by Zoë Cullen and Ricardo Perez-Truglia:

We use an event study analysis of manager rotation to estimate the causal effect of managers’ gender on their employees’ career progression. We find that when male employees are assigned to male managers, they are promoted faster in the following years than they would have been if they were assigned to female managers. Female employees, on the contrary, have the same career progression regardless of the manager’s gender. These differences in career progression cannot be explained by differences in effort or output. This male-to-male advantage can explain a third of the gender gap in promotions. Moreover, we provide suggestive evidence that these manager effects are due to socialization between male employees and male managers.

There is more to the abstract, including a discussion of the benefits of smoking together.  Here is an ungated copy.

Claims about polarization

…when it comes to moral issues, the prominent change is a partisan secular trend, in which both Democrats and Republicans are adopting more progressive views on moral issues, although at a different rate. While Democrats are early adopters of progressive views, Republicans adopt the same views at a slower pace. This secular change can be easily (mis)interpreted as a sign of polarization because, at the onset of the process, the gap between party supporters broadens due to faster pace at which Democrats adopt progressive views, and only toward the end, the gap between partisan supporters decreases.

That is from a new paper by Baldassari and Park, via Gaurav Sood.

Are we undermeasuring productivity gains from the internet?, part II

From my new paper with Ben Southwood on whether the rate of progress in science is diminishing:

Similarly, the tech sector of the American economy still isn’t as big as many people think. The productivity gap has meant that measured GDP is about fifteen percent lower than it would have been under earlier rates of productivity growth. But if you look say about the tech sector in 2004, it is only about 7.7 percent of GDP (since the productivity slowdown is ongoing, picking a more recent and larger number is not actually appropriate here). A mismeasurement of that tech sector just doesn’t seem nearly large enough to fill in for the productivity gap. You might argue in response that “today the whole economy is incorporating tech,” but that doesn’t seem to work either. For one thing, recent tech incorporations typically involve goods and services that are counted in GDP. Furthermore, there is a problem of timing, namely that the U.S. productivity slowdown dates back to 1973, and that is perhaps the single biggest problem for trying to attribute this gap mainly to under-measured innovations in the tech sector.

Other research looks at “worst case” scenarios from the mismeasurement of welfare adjustments in consumer price deflators and finds a similar result: a significant effect that nonetheless does not reverse the judgement that innovation has been slowing. 

The most general point of relevance here is simply that price deflator bias has been with productivity statistics since the beginning, and if anything the ability of those numbers to adjust for quality improvements may have increased with time. For instance, the research papers do not find that the mismeasurement has risen in the relevant period. You might think the introduction of the internet is still undervalued in measured GDP, but arguably the introduction of penicillin earlier in the 20th century was undervalued further yet. The market prices for those doses of penicillin probably did not reflect the value of the very large number of lives saved. So when we are comparing whether rates of progress have slowed down over time, and if we wish to salvage the performance of more recent times, we still need an argument that quality mismeasurement has increased over time. So far that case has not been made, and if you believe that the general science of statistics has made some advances, the opposite is more likely to be true, namely that mismeasurement biases are narrowing to some extent. 

You will find citations and footnotes in the original.  Here is my first post on whether the productivity gains from the internet are understated.

Claims about real rates of return

With recourse to archival, printed primary, and secondary sources, this paper reconstructs global real interest rates on an annual basis going back to the 14th century, covering 78% of advanced economy GDP over time. I show that across successive monetary and fiscal regimes, and a variety of asset classes, real interest rates have not been “stable”, and that since the major monetary upheavals of the late middle ages, a trend decline between 0.6-1.8bps p.a. has prevailed. A consistent increase in real negative-yielding rates in advanced economies over the same horizon is identified, despite important temporary reversals such as the 17th Century Crisis. Against their long-term context, currently depressed sovereign real rates are in fact converging “back to historical trend” – a trend that makes narratives about a “secular stagnation” environment entirely misleading, and suggests that – irrespective of particular monetary and fiscal responses – real rates could soon enter permanently negative territory. I also posit that the return data here reflects a substantial share of “nonhuman wealth” over time: the resulting R-G series derived from this data show a downward trend over the same timeframe: suggestions about the “virtual stability” of capital returns, and the policy implications advanced by Piketty (2014) are in consequence equally unsubstantiated by the historical record.

That is from a new paper by Paul Schmelzing, via the excellent Kevin Lewis.