Links May 2021

“Algorithm appreciation: People prefer algorithmic to human judgment” Even though computational algorithms often outperform human judgment, received wisdom suggests that people may be skeptical of relying on them (Dawes, 1979). Counter to this notion, results from six experiments show that lay people adhere more to advice when they think it comes from an algorithm than from a person. People showed this effect, what we call algorithm appreciation, when making numeric estimates about a visual stimulus (Experiment 1A) and forecasts about the popularity of songs and romantic attraction (Experiments 1B and 1C). Yet, researchers predicted the opposite result (Experiment 1D). Algorithm appreciation persisted when advice appeared jointly or separately (Experiment 2). However, algorithm appreciation waned when: people chose between an algorithm’s estimate and their own (versus an external advisor’s; Experiment 3) and they had expertise in forecasting (Experiment 4). Paradoxically, experienced professionals, who make forecasts on a regular basis, relied less on algorithmic advice than lay people did, which hurt their accuracy. These results shed light on the important question of when people rely on algorithmic advice over advice from people and have implications for the use of “big data” and algorithmic advice it generates. [source]

Underserved populations experience higher levels of pain. These disparities persist even after controlling for the objective severity of diseases like osteoarthritis, as graded by human physicians using medical images, raising the possibility that underserved patients’ pain stems from factors external to the knee, such as stress. Here we use a deep learning approach to measure the severity of osteoarthritis, by using knee X-rays to predict patients’ experienced pain. We show that this approach dramatically reduces unexplained racial disparities in pain. Relative to standard measures of severity graded by radiologists, which accounted for only 9% (95% confidence interval (CI), 3–16%) of racial disparities in pain, algorithmic predictions accounted for 43% of disparities, or 4.7× more (95% CI, 3.2–11.8×), with similar results for lower-income and less-educated patients. This suggests that much of underserved patients’ pain stems from factors within the knee not reflected in standard radiographic measures of severity. We show that the algorithm’s ability to reduce unexplained disparities is rooted in the racial and socioeconomic diversity of the training set. Because algorithmic severity measures better capture underserved patients’ pain, and severity measures influence treatment decisions, algorithmic predictions could potentially redress disparities in access to treatments like arthroplasty. [source]

“We report on the results of two online surveys—one with 231 security experts and one with 294 MTurk participants—on what the practices and attitudes of each group are. Our findings show a discrepancy between the security practices that experts and non-experts report taking. For instance, while experts most frequently report installing software updates, using two-factor authentication and using a password manager to stay safe online, non-experts report using antivirus software, visiting only known websites, and changing passwords frequently.” [source]

Online nudges sometimes work: “We empirically assess whether browser security warnings are as ineffective as suggested by popular opinion and previous literature. We used Mozilla Firefox and Google Chrome’s in-browser telemetry to observe over 25 million warning impressions in situ. During our field study, users continued through a tenth of Mozilla Firefox’s malware and phishing warnings, a quarter of Google Chrome’s malware and phishing warnings, and a third of Mozilla Firefox’s SSL warnings. This demonstrates that security warnings can be effective in practice; security experts and system architects should not dismiss the goal of communicating security information to end users. We also find that user behavior varies across warnings. In contrast to the other warnings, users continued through 70.2% of Google Chrome’s SSL warnings. This indicates that the user experience of a warning can have a significant impact on user behavior. Based on our findings, we make recommendations for warning designers and researchers.” [source]

Apparently color matters. “Based on user feedback, the [Bing ads] team estimated the best blue color could generate $80 million to $90 million in ad sales.” [link] But there is some question as the veracity of this claim. See this Substack. Also this.  

From Sunita Sah (NYT):

Disclosure can also cause perverse effects even when biases are unavoidable. For example, surgeons are more likely to recommend surgery than non-surgeons. Radiation-oncologists recommend radiation more than other physicians. This is known as specialty bias. Perhaps in an attempt to be transparent, some doctors spontaneously disclose their specialty bias. That is, surgeons may inform their patients that as surgeons, they are biased toward recommending surgery.

My latest research, published last month in the Proceedings of the National Academy of Sciences, reveals that patients with localized prostate cancer (a condition that has multiple effective treatment options) who heard their surgeon disclose his or her specialty bias were nearly three times more likely to have surgery than those patients who did not hear their surgeon reveal such a bias. Rather than discounting the surgeon’s recommendation, patients reported increased trust in physicians who disclosed their specialty bias.

Remarkably, I found that surgeons who disclosed their bias also behaved differently. They were more biased, not less. These surgeons gave stronger recommendations to have surgery, perhaps in an attempt to overcome any potential discounting they feared their patient would make on the recommendation as a result of the disclosure.

Surgeons also gave stronger recommendations to have surgery if they discussed the opportunity for the patient to meet with a radiation oncologist. This aligns with my previous research from randomized experiments, which showed that primary advisers gave more biased advice and felt it was more ethical to do so when they knew that their advisee might seek a second opinion.

Old Marginal Revolution posts of interest:

• Clinton welfare reform was not such a big deal, one way or the other.  And what it is like to be obsessed with mood affiliation.
• How small is the world really?, how networked are you anyway?, and why it matters.
• Peter Klein has an interesting Rand Journal piece (pdf) on conglomerates: “This paper challenges the conventional wisdom that the 1960s conglomerates were inefficient. I offer valuation results consistent with recent event-study evidence that markets typically rewarded diversifying acquisitions. Using new data, I compute industry-adjusted valuation, profitability, leverage, and investment ratios for thirty-six large, acquisitive conglomerates from 1966 to 1974. During the early 1970s, the conglomerates were less valuable and less profitable than standalone firms, favoring an agency explanation for unrelated diversification. In the 1960s, however, conglomerates were not valued at a discount. Evidence from acquisition histories suggests that conglomerate diversification may have added value by creating internal capital markets.” In other words, today’s Google announcement isn’t as crazy as it may sound.  Here is further positive evidence on conglomerates, and Glenn Hubbard also thinks the 1960s conglomerates were largely efficient.  Here is some evidence, however, that conglomerates tend to be less innovative.  Scharfstein and Stein are less positive more generally.  Here is some evidence that the non-Google divisions will receive favoritism in the allocation of capital within the conglomerate.  That all said, conglomerates are understudied in microeconomics, in part because they are hard to study.
• While cruising the internet I ran into this recent working paper (pdf) by Daniel Benjamin, James J. Choi, and Geoffrey Fisher: “We randomly vary religious identity salience in laboratory subjects to test how identity effects contribute to the impact of religion on economic behavior. We find that religious identity salience causes Protestants to increase contributions to public goods. Catholics decrease contributions to public goods, expect others to contribute less to public goods, and become less risk averse. Jews more strongly reciprocate as an employee in a bilateral labor market gift-exchange game. Atheists and agnostics become less risk averse. We find no evidence of religious identity-salience effects on disutility of work effort, discount rates, or generosity in a dictator game.” In the recent hullaballoo, it has been forgotten that perhaps the best paper on whether religion is good for you was written by Jonathan Gruber.
• If you discount everything back to today, the net present value of the Senate seat would be ~$6.2 million. Here are the calculations.  One way to dispute the numbers is to adjust for the opportunity cost of the talented labor of the would-be Senator.  Still, with finance falling apart, the job market slowing down, and the option of receiving bids from wealthy but non-talented people, I remain surprised that 500K would be viewed as a going offer.  The associated fame and power is fun.  Clearly the seat was worth more than that to its previous holder, no?
• “Which do you think takes a bigger toll on the environment, owning a dog, or owning an SUV? My bet would be on the dog. I’m thinking of all of the resources that go into dog food.” That is from Arnold Kling.  And if you believe in a zero or very low discount rate, don’t forget to count all those puppies too.
• Mark Aguiar and Erik Hurst write: “Using scanner data and time diaries, we document how households substitute time for money through shopping and home production. We find evidence that there is substantial heterogeneity in prices paid across households for identical consumption goods in the same metro area at any given point in time. For identical goods, prices paid are highest for middleaged, rich, and large households, consistent with the hypothesis that shopping intensity is low when the cost of time is high. The data suggest that a doubling of shopping frequency lowers the price paid for a given good by approximately 10 percent. [TC: is that all????]  From this elasticity and observed shopping intensity, we impute the shopper’s opportunity cost of time, which peaks in middle age at a level roughly 40 percent higher than that of retirees [emphasis added]. Using this measure of the price of time and observed time spent in home production, we estimate the parameters of a home production function. We find an elasticity of substitution between time and market goods in home production of close to 2. Finally, we use the estimated elasticities for shopping and home production to calibrate an augmented lifecycle consumption model. The augmented model predicts the observed empirical patterns quite well. Taken together, our results highlight the danger of interpreting lifecycle expenditure without acknowledging the changing demands on time and the available margins of substituting time for money.” Here is the paper, and thanks to Bruce Bartlett for the pointer.  We also learn that people with children pay higher prices (presumably they have less time to search) and people in their forties with children pay the highest prices of all, six to eight percent more than people in their twenties or sixties.
• • I also take these results to imply that poor households, which shop more frequently and pay lower prices, are better off in material terms than CPI-based measures of real income will imply.  That being said, they also have less time.  Fans of the “happiness literature,” which suggests more money above a certain level doesn’t make you better off, should favor less search.  After all, we are told that people enjoy time spent with friends more than either money or sex.  So does this view (not mine) suggest that we shut down discount outlets and induce more consumption of time?  Are single price monopolies better than price discrimination?  Is Marshall’s the true enemy of the middle class?
• Derive the conditions under which post-disaster looting is efficient. Hints: Start with a queuing problem, and then ask when rents will not be exhausted; that is, the resources spent obtaining the goods should not equal the value of the goods themselves.  The quest for looted goods therefore should be monopolized or somehow restricted, rather than competitive.  The goods should be perishable, available for subsequent resale, and the negative incentive effect on future production should be small.  The discount rate and the transactions costs of immediate sale by the (previous) owner should both be high. Extra credit: Does efficiency more likely rise or fall when we consider looting by the police?
• You need only 2,000 Facebook friends: You’ve heard of internet celebrities getting paid to mention a product in a tweet or shoot out an Instagram with a brand in the shot. Now a hotel in Sweden is taking social media marketing to a new level by offering a free stay to anyone with a serious online following. In the words of Stockholm’s Nordic Light Hotel, it “accepts personal social networks as currency.” Anyone with more than 2,000 personal Facebook friends or 100,000 followers on Instagram gets a free seven-night stay at the luxury hotel, which usually costs $360/night. All you have to do is post when you make the reservations, when you check in, and when you check out, all with the requisite hotel tags. (“If the guest does not shares the posts that are necessary to take part of the discount/ free nights, the guest will be charged full price for the stay,” the hotel warns.) The full article is here, and for the pointer I thank Bryan Lassiter, a loyal MR reader.

In recent decades, gentrification has transformed American central city neighborhoods. I estimate a spatial equilibrium model to show that the rising value of high-skilled workers' time contributes to the gentrification of American central cities. I show that the increasing value of time raises the cost of commuting and exogenously increases the demand for central locations by high-skilled workers. While change in the value of time has a modest direct effect on gentrification of central cities, the effect is substantially magnified by endogenous amenity change driven by the changes in local skill mix. [AEJ]

“Twitter’s decision to double its character count from 140 to 280 characters last year hasn’t dramatically changed the length of Twitter posts. According to new data released by the company this morning, Twitter is still a place for briefer thoughts, with only 1% of tweets hitting the 280-character limit, and only 12% of tweets longer than 140 characters. Brevity, it seems, is baked into Twitter – even when given expanded space, people aren’t using it. Only 5% of tweets are longer than 190 characters, indicating that Twitter users have been for so long trained to keep their tweets short, they haven’t adapted to take advantage of the extra room to write. Meanwhile, most tweets continue to be very short, Twitter says. The most common length of a tweet back when Twitter only allowed 140 characters was 34 characters. Now that the limit is 280 characters, the most common length of a tweet is 33 characters. Historically, only 9% of tweets hit Twitter’s 140-character limit, now it’s 1%.” [source]

“If clicking with someone feels like you’re “on the same wavelength,” it turns out there’s a good reason for that. In what’s called “interpersonal synchronization,” people click in an unspoken meeting of the minds about how long to linger before a museum painting or when to get up from the coffeehouse table. Such synchrony occurs when an overheard remark triggers in both of you a simultaneously raised eyebrow, when what you see on your companion’s face reflects the feelings and thoughts inside your own brain. Your body language matches, what catches your attention catches his, you become impatient at the same time about the same things.”