Category: Data Source

The new phrenology?

Ubiquitous facial recognition technology can expose individuals’ political orientation, as faces of liberals and conservatives consistently differ. A facial recognition algorithm was applied to naturalistic images of 1,085,795 individuals to predict their political orientation by comparing their similarity to faces of liberal and conservative others. Political orientation was correctly classified in 72% of liberal–conservative face pairs, remarkably better than chance (50%), human accuracy (55%), or one afforded by a 100-item personality questionnaire (66%). Accuracy was similar across countries (the U.S., Canada, and the UK), environments (Facebook and dating websites), and when comparing faces across samples. Accuracy remained high (69%) even when controlling for age, gender, and ethnicity. Given the widespread use of facial recognition, our findings have critical implications for the protection of privacy and civil liberties.

Here is more from Michal Kosinki, published in a journal which is an offshoot of Nature.

The traffic impact of Covid

For the first time since 2007, preliminary data from the National Safety Council show that as many as 42,060 people are estimated to have died in motor vehicle crashes in 2020. That marks an 8% increase over 2019 in a year where people drove significantly less frequently because of the pandemic. The preliminary estimated rate of death on the roads last year spiked 24% over the previous 12-month period, despite miles driven dropping 13%. The increase in the rate of death is the highest estimated year-over-year jump that NSC has calculated since 1924 – 96 years.

Here is the full story, the Great Psychometric Test continues.  Via Nick A.

Causes of the sex drought

Among young men, declines in drinking frequency, an increase in computer gaming, and the growing percentage who coreside with their parents all contribute significantly to the decline in casual sex. The authors find no evidence that trends in young adults’ economic circumstances, internet use, or television watching explain the recent decline in casual sexual activity.

Here is more from Scott J. South and Lei Lei, via the excellent Kevin Lewis.

The Gregory Clark working paper

I wanted to present this paper to you all, here is the abstract:

Economics, Sociology, and Anthropology are dominated by the belief that social outcomes depend mainly on parental investment and community socialization. Using a lineage of 402,000 English people 1750-2020 we test whether such mechanisms better predict outcomes than a simple additive genetics model. The genetics model predicts better in all cases except for the transmission of wealth. The high persistence of status over multiple generations, however, would require in a genetic mechanism strong genetic assortative in mating. This has been until recently believed impossible. There is however, also strong evidence consistent with just such sorting, all the way from 1837 to 2020. Thus the outcomes here are actually the product of an interesting genetics-culture combination.

The title is “For Whom the Bell Curve Tolls: A Lineage of 400,000 English Individuals 1750-2020 shows Genetics Determines most Social Outcomes.”  For those of you who don’t know, Greg’s home page provides some context.

Higher education sentences to ponder

We find that the WTP [willingness to pay] for in-person instruction (relative to a remote format) represents around 4.2% of the average annual net cost of attending university, while the WTP for on-campus social activities is 8.1% of the average annual net costs. We also find large heterogeneity in WTP, which varies systematically across socioeconomic groups. Our analysis shows that economically-disadvantaged students derive substantially lower value from university social life, but this is primarily due to time and resource constraints.

That is from a new NBER working paper by Aucejo, French, and Zafar.

Why are atheists so unpopular?

Here is a new research paper from Tomas Ståhl:

There is a widespread cross-cultural stereotype suggesting that atheists are untrustworthy and lack a moral compass. Is there any truth to this notion? Building on theory about the cultural, (de)motivational, and cognitive antecedents of disbelief, the present research investigated whether there are reliable similarities as well as differences between believers and disbelievers in the moral values and principles they endorse. Four studies examined how religious disbelief (vs. belief) relates to endorsement of various moral values and principles in a predominately religious (vs. irreligious) country (the U.S. vs. Sweden). Two U.S. M-Turk studies (Studies 1A and 1B, N = 429) and two large cross-national studies (Studies 2-3, N = 4,193), consistently show that disbelievers (vs. believers) are less inclined to endorse moral values that serve group cohesion (the binding moral foundations) [emphasis added by TC]. By contrast, only minor differences between believers and disbelievers were found in endorsement of other moral values (individualizing moral foundations, epistemic rationality). It is also demonstrated that presumed cultural and demotivational antecedents of disbelief (limited exposure to credibility-enhancing displays, low existential threat) are associated with disbelief. Furthermore, these factors are associated with weaker endorsement of the binding moral foundations in both countries (Study 2). Most of these findings were replicated in Study 3, and results also show that disbelievers (vs. believers) have a more consequentialist view of morality in both countries. A consequentialist view of morality was also associated with another presumed antecedent of disbelief-analytic cognitive style.

Via the excellent Samir Varma.

A Majority of Doses are Now Second Doses

As of Feb. 18 (last day of full data) we gave out 817,708 second doses and just 702,426 first doses. In other words, a majority of doses are now second doses. As Daniel Bier writes this means that we are boosting some people from ~85% to ~95% protected when we could be vaccinating more first timers and getting them from 0% protected to ~85% protected.

If we followed the British rule and delayed the booster to 12 weeks, we could immediately more than *double* the number of people going from 0% protected to to ~85% protected. More first-doses would be great for the newly protected and more people at ~85% protection would also reduce transmission so there would be fewer new infections and less threat to the non-vaccinated.

The opportunity cost of not delaying the booster is measured in lives lost.

Cool, green Britannia

Over the summer of 2020, as coronavirus cases fell and life in Britain felt briefly normal, something very abnormal was happening to the country’s electricity supply. No coal was burned to generate any portion of it for a period of more than two months, something that had not happened since 1882. Britain’s four remaining coal-burning power plants are zombies, all but dead. Within a couple of years they will be closed and Britain will probably never burn coal for electricity again.

The elimination of power stations that burn coal has helped Britain cut its carbon emissions faster than any other rich country since 1990 (see charts). They are down by 44%, according to data collected by the Department for Business, Energy and Industrial Strategy (BEIS) during a period when the economy grew by two-thirds. Germany’s emissions, in contrast, are down by 29%; coal is still burned to generate some 24% of its electricity. Britain has made cuts to its emissions 1.8 times larger than the EU average since 1990. In America, emissions over the same period are up slightly.

Here is the full article from The Economist.  I’ll say it again, whether it is AI, the Oxford/Astrazeneca vaccine, the speed of the current vaccination program, this switch to greener energy, the reemergence of Oxbridge, the new Dominic Cummings-inspired DARPA-like science funding plan, or London being the world’s best city — current Great Britain remains grossly underrated.

Profile of Youyang Gu, data scientist

In mid-April, while he was living with his parents in Santa Clara, Calif., Gu spent a week building his own Covid death predictor and a website to display the morbid information. Before long, his model started producing more accurate results than those cooked up by institutions with hundreds of millions of dollars in funding and decades of experience.

“His model was the only one that seemed sane,” says Jeremy Howard, a renowned data expert and research scientist at the University of San Francisco. “The other models were shown to be nonsense time and again, and yet there was no introspection from the people publishing the forecasts or the journalists reporting on them. Peoples’ lives were depending on these things, and Youyang was the one person actually looking at the data and doing it properly.”

The forecasting model that Gu built was, in some ways, simple. He had first considered examining the relationship among Covid tests, hospitalizations, and other factors but found that such data was being reported inconsistently by states and the federal government. The most reliable figures appeared to be the daily death counts. “Other models used more data sources, but I decided to rely on past deaths to predict future deaths,” Gu says. “Having that as the only input helped filter the signal from the noise.”

The novel, sophisticated twist of Gu’s model came from his use of machine learning algorithms to hone his figures.

Here is the full Bloomberg piece by Ashlee Vance, I am especially pleased because Youyang was an Emergent Ventures winner.  Here is Youyang Gu on Twitter.

Diversity in policing

In the wake of high-profile police shootings of Black Americans, it is important to know whether the race and gender of officers and civilians affect their interactions. Ba et al. overcame previous data constraints and found that Hispanic and Black officers make far fewer stops and arrests and use force less than white officers, especially against Black civilians. These differences are largest in majority-Black neighborhoods in the city of Chicago (see the Perspective by Goff). Female officers also use less force than male officers. These effects are supportive of the efficacy of increasing diversity in police forces.

That is a new paper in Science by Bocar A. Ba, Dean Knox, Jonathan Mummolo, and Roman Rivera.  Via Anecdotal.

Half Doses of Moderna Produce Neutralizing Antibodies

A new phase II study from Moderna shows that half-doses (50 μg) appear to be as good as full doses (100 ug) at generating correlates of protection such as neutralizing antibodies.

In this randomized, controlled phase 2 trial, the SARS-CoV-2 vaccine candidate mRNA-1273, administered as a two-dose vaccination regimen at 50 and 100 μg, exhibited robust immune responses and an acceptable safety profile in healthy adults aged 18 years and older. Local and systemic adverse reactions were mostly mild-to-moderate in severity, were ≤4 days of median duration and were less commonly reported in older compared with younger adults. Anti-SARS-CoV-2 spike binding and neutralizing antibodies were induced by both doses of mRNA-1273 within 28 days after the first vaccination, and rose substantially to peak titers by 14 days after the second vaccination, exceeding levels of convalescent sera from COVID-19 patients. The antibodies remained elevated through the last time point assessed at 57 days. Neutralizing responses met criteria for seroconversion within 28 days after the first vaccination in the majority of participants, with rates of 100% observed at 14 and 28 days after the second vaccination. While no formal statistical testing was done, binding and neutralizing antibody responses were generally comparable in participants who received the 100 μg mRNA-1273 and the 50 μg dose at all time points and across both age groups. Overall, the results of this randomized, placebo-controlled trial extend previous immunogenicity and safety results for mRNA-1273 in the phase 1 study in an expanded cohort including participants older than 55 years of age [16, 19].

[These data] confirm that a robust immune response is generated at both 50 and 100 ug dose levels.

As I wrote earlier, halving the dose is equivalent to instantly doubling the output of every Moderna factory.

See my piece in the Washington Post on getting to V-day sooner for an overview of dose stretching strategies.

Addendum: France says one dose is sufficient for previously COVID infected.

What are the most important statistical ideas of the past 50 years?

We argue that the most important statistical ideas of the past half century are: counterfactual causal inference, bootstrapping and simulation-based inference, overparameterized models and regularization, multilevel models, generic computation algorithms, adaptive decision analysis, robust inference, and exploratory data analysis. We discuss common features of these ideas, how they relate to modern computing and big data, and how they might be developed and extended in future decades. The goal of this article is to provoke thought and discussion regarding the larger themes of research in statistics and data science.

By Andrew Gelman and Aki Vehtari, via Lampros Tzontsos.

Monopsony it ain’t, rather output loss

We use highly consistent national-coverage price and wage data to provide evidence on wage increases, labor-saving technology introduction, and price pass-through by a large low-wage employer facing minimum wage hikes. Based on 2016-2020 hourly wage rates of McDonald’s Basic Crew and prices of the Big Mac sandwich collected simultaneously from almost all US McDonald’s restaurants, we find that in about 25% of instances of minimum wage increases, restaurants display a tendency to keep constant their wage ‘premium’ above the increasing minimum wage. Higher minimum wages are not associated with faster adoption of touch-screen ordering, and there is near-full price pass-through of minimum wages, with little heterogeneity related to how binding minimum wage increases are for restaurants. Minimum wage hikes lead to increases in real wages (expressed in Big Macs an hour of Basic Crew work can buy) that are one fifth lower than the corresponding increases in nominal wages.

That is a new paper from Orley Ashenfelter and Štěpán Jurajda.  I will ask again my standard question: don’t we have ways of helping poorer individuals that boost output rather than harming it?  Why don’t we focus on those?

Fortunately roadblocks are arising.

Via Ilya Novak.