Scott Olesen

Frank Boyden: What makes for singular leaders?

2021-10-13T00:00:00-04:00

This past holiday weekend I read John McPhee’s The Headmaster, a very short biography of Frank Boyden, who was headmaster of Deerfield Academy, which is now one of the top prep schools in the US.

Founded in 1797, the academy had risen in prominence through the 19th century. Many New England governors and Congress members went there. But by 1900, Deerfield had declined. There were only about 10 students, and the trustees were very near to closing the school.

Frank Boyden became headmaster of Deerfield in 1902, having just graduated from college. He was hoping to hold the job for a few years to save money for law school. The school’s board trustees, who were very near to closing up the school altogether, took a long shot on Boyden, an unknown and unremarkable member of the latest graduating class from Amherst College, about 20 miles from Deerfield. They had previously invited multiple of the more notable members of his graduating class to take the job. All had declined.

Here’s the amazing part: Frank Boyden was headmaster of Deerfield for 64 years, from 1902 to 1966. In that time it went from a purely local institution of about 10 students to a highly-ranked and internationally-known school. Today, there are over 600 students, more than 10% of which are foreign nationals, and the school has an endowment of $600 million.

What did Frank Boyden do? How did he achieve this incredible feat? McPhee’s book is not an analysis —he calls it a “sketch” of Boyden— and so does not offer any hypotheses. But I found this story remarkable. What made Boyden great? How did he turn this institution around? Are there common threads with other people who had held long tenures at great institutions? Fauci’s 20 year tenure at National Institute of Allergy and Infectious Diseases comes to mind.

In the absence of a more careful analysis, I’ll list some of the features of Boyden as described in The Headmaster that I found remarkable:

Energy. Boyden gets up early, works all day, and goes to sleep late. He sleeps much less than 8 hours, and all the remaining time appears to be “work”. (Boyden died in 1972, at age 92, but I talk about him in the present tense.)
Emotional intelligence. Boyden can remember many people’s names. He appears to know when he can tell a white lie and get away with it. He can predict boys’ behaviors and personalities very well. He took bets on marginal characters, and was sometimes proved wrong, but often was proved right with incredible success. Tom Ashley is a Deerfield legend: Boyden talked an unsuccessful local boy, Ashley, into coming to Deerfield. Ashley shined, going on to Amherst, returning to Deerfield as a faculty member, and instituting key reforms at the school before being killed in World War I at age 24. These kinds of bets bred fierce loyalty.
He developed a cult of personality. He earned the respect of his early students by some crazy stunts, like playing baseball with them and completely throwing himself into the sport. In later years, he earned respect by knowing everyone, being everywhere, and having good advice for people. Deerfield started to take off not in small part because influential people thought Boyden was special, and they wanted their children to be near him.
He bet that sports would be an educational tool. Boyden decided that sports would be a way to educate his male student body and keep their energies in line. Sports in education were considered somewhat gauche in 1902, so Boyden took a bet that paid off.
He kept doing the same kind of thing for a long time. Educational institutions are generally conservative. Sure, charter schools are everywhere where there is a struggling public education system. But the children of the wealthy and powerful mostly attend the most conservative, long-standing, highly respected institutions, rather than the most new-fangled. By keeping up his style of education, his cult of personality, the sense that Deerfield was doing something long-running and special, Boyden attracted a certain kind of family.
He was good at fundraising. This one doesn’t seem interesting or unique enough to me to merit further discussion.

There are a lot of pieces to this story I don’t understand. First, I didn’t go to a prep school, so I don’t know what it’s like to have a fierce loyalty to a headmaster, although I have seen similar dynamics among athletes, who come to respect and believe in their coaches in a way I’ve rarely felt about anyone.

Second, I’m not sure what we should use as the counterfactual for any analysis. What other institutions on the brink of failure hired essentially unqualified people who became brilliant leaders?

Finally, I’m not sure if anything I could learn about Frank Boyden would be relevant for the current age. The idea of long-standing leadership at an institution and the cult of personality seems to have died out. Or maybe it is only that, once an institution becomes important enough, it is unwilling to put so much power into the hands of a single person? So while Boyden grew in influence as Deerfield did, his successors could never “own” the whole institution in quite the same way.

I still find something incredibly Romantic about Boyden’s story. Here was a man that, it appears, was uniquely suited to be a prep school headmaster, and he was in exactly the right time, place, and circumstances to make it happen. There was no wandering about: he essentially did one, very big thing in his life.

Greek time

2021-09-08T00:00:00-04:00

My wife’s parents were born in Greece, and all her blood relatives, except her sister, were raised in Greece. I was raised in a white, upper-middle class American family in New Jersey. As you might expect, I struggled to understand my wife’s family’s conception of time.

Here’s the way I originally understood “Greek time”: Greeks are always late. The Greeks even make fun of themselves for this. I once heard this joke: Two Greek bump into one another in the market and make a plan to meet up that evening. One friend will come to the other’s house, which is a ten minute walk away, sometime around 6. By 6:30, the visitor still hasn’t arrived, so the host calls her friend, who says on the phone, “Yes, yes, I’m coming now!” At 6:45, the visitor still hasn’t arrived, so the host calls again, and the friend says, again, “Yes, yes, I’m coming now!” By 7, the visitor isn’t there, so the friend calls a third time, and this time the visitor loses her temper: “How do you expect me to get over to your house when you are calling me all the time?”

I don’t think the Greeks are alone in being perceived this way by people like me. (To be fair, I didn’t really perceive of the Greeks at all for my early life; I thought of all Greeks as ancient until I had a crush on one.) I’ve heard of “Mediterranean time”, “island time”, etc. etc. Apparently, in Brazil, you can refer to “English time” if you want to be clear that a rendezvous should be at the literal clock time that was discussed.

But I wasn’t ungenerous in thinking that Greeks were late. I understand that practices differ in different cultures. I thought that Greek time was a difference in understanding about what it meant to be politely “on time”. When my wife and I host parties, we always joke about who will arrive at the party’s nominal start time (when we are still finishing cooking), who will show up a courteous 15 minutes late, an acceptable 30 minutes late, or a lackadaisacal one hour late. I appreciated that “on time” depends on things like what the social event is. You should definitely not be an hour late for a sit-down dinner party, but you can be an hour late for a standing-room-only cocktail party (back in the misty ages of 2019 when we had such events).

Greek time, I reasoned, was mostly the same as my time, just delayed by somewhere between 30 and 120 minutes. When my in-laws planned parties, I noticed that they would tell the Americans one start time, and they would tell the Greeks a time about an hour earlier, to make sure that everyone showed up together. (The Greek-Americans were unpredictable.)

Over time, I’ve come to understand that Greek time is a much more complex, rich, and rewarding thing than I first understood. I’ll try to explain using an analogy that I came up with but which Greeks might say is hugely incorrect. Regardless, it helped me understand, so I hope it might help you understand too.

The Greek language has at least three words for “time”: chronos, ora, and kairos. Chronos, from which we get English words like “chronological” and “chronometer”, refers to the literal and sequential passage of time. In modern Greek, chronos can also mean “one year”. Ora, literally “the hour” or “one hour”, also refers to clock time. In Greek, “what time is it?” is literally “what is the ora?”

Kairos is sometimes interchangeable with chronos and ora, but it has a more metaphorical flavor. It refers to time in the sense of being the opportune moment. If you say it’s the right ora to call your friend, you probably mean that you had an agreement to call at a certain clock time, or that you booked time in your daily calendar for this task. But if you say it’s the right kairos to call, you mean something more. Maybe you’re now in the mood to talk to your friend. Maybe you have a story that you really want to tell them. (Confusingly, kairos also means “weather”, but I think this overall fits in my picture. It’s the right ora to go to the beach if you’re following an itinerary; it’s the right kairos to go to the beach if it’s sunny.)

I was raised to believe that the “time” you should go to a party or go to dinner or call a friend has everything to do with the clock. If you agreed on a clock time, that’s the clock time you must be there. But I came to appreciate that, for the Greeks, there are many factors affecting when it’s the right “time” to do something, and these things only have a rough relation to clock time. The right time for dinner is surely some time in the evening, but it has to do with how hungry we’re feeling, whether we’ve come up with a good plan, if people are feeling antsy or relaxed, whether we’ve gotten everything else done, etc. etc.

Since this revelation, clock time feels like an insipid, one-dimensional way to address the multifaceted concept of when is the right “time” to do things. It’s like being asked to rate every food on a scale of 1 to 10. I hate ham, so that’s a 1, and I love (real horitatiki) Greek salad, so that’s a 10. But will I eat some ham on (Western) Easter? Yes. And will I not want to eat Greek salad every day, at every meal? No (but just barely). How can a 1 to 10 scale express the fact that every food has a time and place, and that what I want to eat has to do with what I’ve eaten recently, where we’re eating, and who I’m eating with? And how can a single clock time express a family’s mutual desire to embark on some activity?

I don’t disagree that Greek time is, roughly speaking, 60 minutes behind American time, any more than I deny that the mean height of US men is 14 cm higher than the mean height of US women. But just as mean height ignores all variability in height, and all the other interesting variations in human forms, and all the complexity of what we mean by “sex”, so also it pains me to say that Greek time is just American time, but delayed by 60 minutes. And I no longer think it’s that the Greeks are unable to mentally adjust to American time, it’s more that they don’t want to give up this richer conception of time and exchange it for a bleak and unforgiving mathematical datum.

Using wastewater to estimate real-world vaccine effectiveness

2021-08-27T00:00:00-04:00

Cross-posted from Biobot’s blog

In an epidemic, better knowledge can lead to wiser policy and can save lives. Can wastewater-based measurements of SARS-CoV-2 activity help us understand the effects of Covid-19 vaccination, and how those effects might change with new variants like Delta?

To understand how wastewater can help us better understand the role of vaccines during this Delta wave, we should first look back to how we originally came to understand the Covid-19 vaccines’ effects on disease transmission.

Early on, it wasn’t clear if vaccines prevented disease transmission via asymptomatic carriers

Vaccines for Covid-19 were developed in record time, thanks to a combination of decades of pre-existing research, expedited review processes, and an abundance of qualifying clinical trial participants. Clinical trials showed that the Pfizer, Moderna, and J&J vaccines dramatically reduced vaccinated people’s risk of hospitalization and death. This has made vaccines an incredibly important part of the Covid-19 response.

But when the first vaccine data was released, epidemiologists had a further question: do the vaccines also prevent asymptomatic infections? A substantial fraction, something like one-third to two-thirds, of Covid-19 infections are asymptomatic, and these asymptomatic infections play an important role in Covid-19 transmission. If a vaccine prevents severe Covid-19, it will blunt the effects of the spreading virus. But if a vaccine also prevents asymptomatic infections, then mass vaccination can also slow the spread of the virus itself.

While the clinical trials showed that the vaccines protects against hospitalization and death, they were not designed to test whether vaccinated individuals could still become asymptomatic carriers of the virus. If vaccinated individuals still got infected with the virus but did not experience any symptoms, and thus were asymptomatic carriers, then they could still act as silent spreaders of the virus. Knowing whether or not the vaccines prevented asymptomatic virus carriage, in addition to preventing death or hospitalization, was of critical importance for predicting how vaccines would change how Covid-19 spreads.

Large ad hoc studies were needed to study how vaccines affect asymptomatic infection

Now, we are confident that Covid-19 vaccination also prevents asymptomatic infection in addition to outcomes like hospitalization. However, this knowledge was not easy to come by. The current confusion about how effective the vaccines are against the Delta variant only emphasizes how hard it is to measure this aspect of vaccine efficacy.

To investigate how vaccines protect against asymptomatic infection, a clinical trial must follow vaccinated and unvaccinated people and check whether fewer of the vaccinated people develop symptomatic Covid-19. When a participant in the trial gets sick, it triggers a chain of events: the participant gets a Covid-19 test, and if the test is positive, the test result is logged in the trial’s database.

By contrast, to determine if vaccines also prevent asymptomatic carriage of the virus, a research study needs to test asymptomatic people, both vaccinated and unvaccinated. In this kind of study, there is no “trigger” for testing; participants must be regularly tested even if they have no symptoms. For example, some studies compiled the results from tests on healthcare workers at a few large medical centers who were routinely screened for Covid-19, regardless of their vaccination status or whether they had any symptoms. These studies are useful and important for understanding the effect of vaccination on asymptomatic carriage, but they require compiling testing records from thousands or tens of thousands of people.

Can wastewater be an independent measure of vaccine’s effectiveness against asymptomatic carriage?

Researchers have suggested that wastewater-based epidemiology could be used to inform the distribution of vaccines. Could wastewater measurements of SARS-CoV-2 also have provided information on whether vaccines are effective against asymptomatic carriage of the virus early on?

New research suggests that a measurement called the wastewater-case ratio, or WC ratio, might help provide this information. The WC ratio —the wastewater concentration of SARS-CoV-2 virus divided by the number of reported Covid-19 cases— correlates with testing availability: when it is hard to get a Covid-19 test, such as early in the pandemic, the number of confirmed cases was low relative to the number of people who were actually infected. Because wastewater, roughly speaking, measures the number of infected people regardless of whether they seek a test, the WC ratio is high when testing is scarce, and it is low when testing is abundant.

Vaccination should also affect the WC ratio, but in a way that depends on the effects of vaccination. For example, if vaccination prevents symptomatic disease but not asymptomatic carriage, the WC ratio would not change: for every reported case, mostly representing symptomatic individuals, there would be no change in the relative number of asymptomatic carriers. But if vaccination prevents asymptomatic carriage, then the WC ratio could potentially fall: for every reported case, there would be fewer asymptomatic shedders. In other words, if the ratio of asymptomatic carriers to symptomatic infected people changes with vaccination, then the WC ratio might also change with vaccination rates.

There are other possibilities. For example, imagine that vaccination prevented symptomatic infection by “downgrading” symptomatic infected to asymptomatic ones. In this scenario, vaccination would prevent symptomatic cases but not ongoing transmission of the virus. There would be the same number of total infections, but more asymptomatic infections for every symptomatic infection, such that the WC ratio would rise with increasing vaccination rates.

In reality, vaccination downgrades some symptomatic infections to asymptomatic carriage and also prevents many people from becoming asymptomatic carriers. In this situation, increasing vaccination rates could lead to proportionally fewer asymptomatic infections compared to symptomatic infection rates, so the wastewater-case ratio would actually fall with increasing vaccination rates.

During a future epidemic, wastewater-based epidemiology could provide clues about vaccine effectiveness

Many questions remain unanswered. What can the WC ratio tell us, quantitatively, about vaccine effectiveness? How many samples, across how many locations, over what time scale would be required? How soon after vaccination programs began could conclusions about vaccine effectiveness be derived from wastewater data?

Nevertheless, in an epidemic, good data saves lives. As wastewater monitoring becomes better integrated into our public health surveillance systems, and as research into wastewater-based epidemiology advances, we may be able to more quickly gather life-saving information, like about the effects of vaccination, using wastewater.

Deal or No Deal: decision-making under uncertainty

2021-07-10T00:00:00-04:00

This post represents a joint effort with Tony Kulesa

Imagine you are in one of these situations:

You are going to pass $N$ gas stations on your way to return a rental car, and you want to minimize theprice you will pay for gas before returning the car. (But you must fill up the car before returning it!)
You are at a speed dating event, where you will meet $N$ people, and you’d like to “lock in” your preferred match by proposing to immediately leave the event with that person after you meet them. (This assumes that you much rather go on any date, even with the last person you meet, than to leave alone.)
You need to buy a gift for someone, and you have limited time to shop. You know you’ll get to see about $N$ options before you have to pick one and go home.

How do you figure out which option to pick? If you’re like me, your instinct is to look at some fraction of options –maybe a quarter, or a half– and then pick the best one after that first exploration.

While this is a reasonable type of strategy, it’s usually the wrong one. In fact, you usually need to look at many fewer options before you have enough information to make a good choice.

To see why, let’s look at the game show Deal or No Deal. The show’s format is a bit convoluted. There are a number of briefcases, each with a certain amount of money inside. The contestant picks a case at the beginning of the show and holds it, unopened. They then open other cases, getting successive offers for a prize. They must take an offer, or pass on it forever. If they reject all offers, they go home with the amount of money that was in the original case they picked.

Here’s a slightly simplifed version of the game: there are $N$ cases, each with some reward. The contestant opens each case, one at a time, and can either go home with that case or reject it forever. If they reject all cases, they must take home the reward from the last case. You might recognize this game as analogous to the situations I posed at the start!

What is the optimal strategy for this game? It seems pretty clear we should open some number of cases to get a sense of what the range of rewards looks like and then choose a case that looks good. There is a tension between explore and exploit: if you explore with too few cases, you’ll take a sub-par case because you don’t know better, but if you explore with too many cases, you’re end up rejecting a case you should have taken.

If we clarify the problem a little more, we can work out the optimal explore/exploit balance mathematically. Say there are $N$ cases, and you will open $M<N$, then take the first case larger than all the $M$ you initially explored. (If you reach the last case, you must take it.) What is the optimal $M$?

It turns out, the optimal $M$ is much smaller than I expected. It’s tempting to think we should use some constant fraction, maybe a quarter or a half of all cases, for exploration. In fact, the optimal $M$ is $\sqrt{N}-1$. For 3 to 6 cases, you should only open 1 before picking the best one you’ve seen. For 7 to 11 cases, you should open 2. For 12 to 20 cases, you should open 3. For 100 cases, you should open only 9.

Can you imagine hearing that you have 20 options, and saying to yourself, “When I’ve seen 3, that will be enough to have a fair sense of the distribution of rewards”? I certainly couldn’t, but that is in fact the right thing to say to yourself!

And the result of this strategy is surprisingly good: the expected value for the rank of the result is $N - \sqrt{N}$. So if you have 100 cases, you can expect to get a case in the 90th percentile or better. (The distribution of results is actually right-skewed, so you can generally expect a result better than the expected value.)

Now, there are two big caveats. The first is that this game might not actually be the one you’re playing. In the game described here, it is always better to take some case than to take no case. In real life, we usually have some alternative that might be better than a middling case. If, say, you were looking for a business venture to invest in, then you always have the alternative of just holding on to your money, or putting it in another investment, rather than being forced to take one of the $N$ ventures offered to you.

The second caveat is that this strategy gets better with larger $N$, that is, with more cases. Stated another way, when $N$ is low, the difference between choices of $M$ is smaller. For example, for $N=20$, the optimal $M$ is 3, giving an expected value of 16.8 (i.e., you might expect one of the 3 best cases), and choosing instead $M=10$ only lowers the expected value to 14.8 (i.e., you will still likely get one of the top 5 cases). That being said, if the game described is an accurate characterization of your situation, it is still irrational to explore with the number of cases that might seem more natural.

Wastewater for equitable Covid-19 monitoring

2021-07-08T00:00:00-04:00

Cross-posted from Biobot’s blog

Systemic inequities in our healthcare system have led Covid-19 to take a disproportionate toll on different groups of Americans, notably different races and ethnicities. Can wastewater-based monitoring of Covid-19 help promote health equity?

Wastewater-based monitoring serves everyone connected to a sewer system. This means that, if a wastewater monitoring program includes wastewater treatment plants that serve a representative population, then the monitoring program automatically serves a representative population.

In contrast, an equitable monitoring system based on clinical testing requires not only placing testing sites in the right places, but also that a representative population seeks testing. In other words, even if a testing site is located near a group of people, there is no guarantee those people will be served by that site. The same holds true for vaccination: even if a vaccination site is near people that could benefit from the site, there is no guarantee that they will benefit.

To visualize wastewater-based monitoring’s equitability, we compared the demographics of the US population against the demographics of the approximately 200 counties served by wastewater treatment plants currently participating in the US Department of Health and Human Services (HHS) national wastewater monitoring project. We also compared the demographics of the counties served by the wastewater monitoring program against the demographics of people who have been fully vaccinated.

We found that wastewater monitoring better represents the age structure of the US, compared to the vaccinated population. Younger people are underrepresented among the vaccinated population, while older people are overrepresented. Prioritizing access to vaccines for people at greater risk of severe Covid-19, such as older adults, helped prevent the most deaths and hospitalizations. However, all age groups can spread the virus, so it is important for disease surveillance to have an even representation across age groups, which wastewater monitoring can provide.

Wastewater monitoring also more equitably represents Black and Hispanic Americans, who are disproportionately affected by Covid-19 and are vaccinated at lower rates than white Americans. For example, while 12% of the US population is Black, only 8% of the vaccinated population is. Among counties served by the HHS project, however, 12% of residents are Black. Similarly, Hispanics make up 19% of the US population and 23% of the population served by the HHS project but only 15% of the vaccinated population.

Native American populations, too small to be easily seen on a bar chart, are an important part of the HHS project. Nine tribal lands spread over six different states are participating in the project. Native Americans make up 0.8% of the US population and 0.6% of the population served by the project.

Overall, wastewater-based monitoring is an important tool for tracking Covid-19, and the US government’s investment in a national monitoring program is an exciting step for this new, equitable way of tracking disease.

RAGE: an incident response framework

2021-04-30T00:00:00-04:00

A few years ago, I was on a family vacation, when we got into a situation. We had rented boats in the morning, when weather conditions were good. By the afternoon, conditions had deteriorated, but everyone was feeling happy after lunch. There were some family dynamics at play that made it hard to for us to realize that conditions were no longer good, and we all trotted along on our plan or re-embarking. The situation steadily became more heated, until a boat capsized, at which point it was clear that was a true emergency.

We spent a fair amount of time thinking, after this happened, about how we could have avoided it. If we had recognized the situation and spoken calmly about it, we may have made some decisions that would have avoided the incident. I wanted some kind of framework for how to deal with this sort of thing, but I couldn’t find any “emergency response for dummies”. So we developed a four-part framework, called RAGE:

Recognize
Assess
Generate options
Execute

RAGE isn’t an exhaustive list of everything you need to do, but it makes sure that you don’t miss anything truly important. (If it’s too hard to remember, you could say ROX: Recognize, Options, eXecute. But I like RAGE better).

Recognize. One failure in our family vacation fiasco was that there was a never a moment where anyone said, “This is, or has the potential to become, a serious situation.” I think the recognition step is particularly hard in these kinds of slowly-building situations, but I don’t think it’s necessarily easy otherwise. I don’t think it’s necessarily instinctual to shout “Fire!” when you see a fire.

When my wife and I use RAGE, we often say “Recognize!” or “RAGE!” or we physically point and say, “This is a problem.” This kind of very awkward statement is important for me, so I snap out of my day-to-day “everything is fine, let’s slide along” kind of thinking. Hearing “RAGE!” is how I prime myself to go through the next steps.

Assess. Quickly analyze the situation. What is going on? The point of this step is to slow you down, even a little bit. Slowing down seems like the last thing you want to do in a crisis, but the point is to make sure you haven’t missed anything obvious. In the easy case of a fire, it’s best to not just rush in, but to take the two seconds to ask, what’s the overall situation? Are other people safe? Is there a much larger fire, of which I’m only seeing a small part?

Generate options. This is another slowing step. It’s easy to imagine that there’s only one possible course of action and get to it. If you’re on your own, then generating options is a way to make sure that you engage your brain and don’t do something stupid. Importantly, there are often options that don’t have to do with mitigating the crisis itself. If there’s a fire, it may be better to call 911, run away, or make sure other people stay away, than it is to rush in yourself.

Execute. Take action. Unlike the previous two steps, which slow you down, this one is about acceleration. Although running away may be the right option, doing truly nothing is usually the worst. The execution step reminds you to not be a bystander and to avoid paralysis. One way to avoid paralysis is to know what to do, and just the RAGE framework itself might be enough.

And I’ll give some examples of using the RAGE framework. They are not particularly dramatic, but I think that’s what it’s for!

Leaving town. In March 2020, my wife and I were living in a 350 sq. ft. studio in Washington, DC. Covid restrictions were making our life less pleasant, but being able to walk around outside meant it was still bearable. One Saturday morning, my wife came back from a run while I was groggily brooding at the kitchen table. She started talking to me about her run. I was mostly tuned out, until she told me, “This is a situation.” (Recognize!) Essentially, the DC police had cordoned off the area around our apartment. I put on shoes and we went for a walk outside. We had to essentially explain ourselves right as we walked out our front door. (Assess!) We put out a few options, one of which was to drive 26 hours from DC to my in-law’s house in Texas. (Generate options!) It was something we had talked about a little, mostly as a joke, but with the situation as it was, we decided to do it. We packed bags and were on the road by noon on Saturday, arriving late Sunday night, in time for us to be at virtual work on Monday morning. (Execute!)

Medical situation. A family member took her blood pressure, and it was high, high to the point that you’re supposed to call a doctor. She took her blood pressure again, a few minutes later, and got a similarly high number. This was the hard part: do you say it’s not that high, and the fact that the second reading is high is likely the stress from seeing the high first number? The moment that we saw that this could be a situation requiring hospitalization was when we activated RAGE. (Recognize! Assess!) Our options were to do nothing, or to go to a doctor. There were a few choices for doctors, including a small clinic nearby and a hospital further away. We got on the phone, to see if the small clinic could definitively diagnose our family member. (Generate options!) Only at this point, when we had a primary option, the small clinic, and a backup, the hospital, did we get in the car. (Execute!) In this example, it may seem like recognition and assessment were mixed up, but I’d say they just happened very quickly, one after the other. The “oh-shoot” moment, that this could be a problem, was the critical one to get us to do a serious assessment and generation of options. And ultimately, generating options and calling clinics, which seem like a delay, saved us time in the end, since we could go to the local clinic and be seen right away. We avoided jumping right in the car and speeding off to the hospital, which would have been suboptimal.

A hike. My wife and I decided to do an ambitious hike in Canyonlands National Park in August. We carried the recommended amount of water, but we didn’t anticipate how hot it would be, and how much water we would need. At a critical point in our trip, we had the “recognize” moment and ultimately decided to turn back. Other people have other mechanisms for making sure they break the trudge, or the blinders, or the “summit fever”, or whatever you want to call it. RAGE is ours.

Economics of healthcare

2021-04-19T00:00:00-04:00

I recently read Case’s and Deaton’s Deaths of Despair, a book about how America’s working class (more specifically: white, non-college-educated, middle age people) are suffering an increase burden of “deaths of despair” like alcoholism and drug overdose.

The authors cite a 1963 article, Kenneth Arrow’s “Uncertainty and the Welfare Economics of Health Care”, as mathematical proof that a free market cannot deliver “socially acceptable” healthcare. I found these claims very interesting, especially because I knew of Arrow from his impossibility theorem which says that there’s essentially no perfect (ranked) voting system. (I don’t pretend to understand it all, but I found it comforting, in a sense, that there’s no perfect way to do voting.)

Arrow’s article on healthcare focuses on how healthcare is different from other goods traded in a marketplace, and why a true free market for healthcare will never work in a way that we think is reasonable.

Why do we care about free markets?

Free markets are Pareto optimal. A state of affairs (e.g., a distribution of resources that arises from a choice of economic system, such as a free market) is Pareto optimal if there is no change in the state of affairs that would make anyone worse off. Arrow points out that this is a kind of optimality, but it’s not necessary the best kind of optimality. We might be very happy to make a few people a little less happy if it makes many people much more happy.

Free markets are also appealing because they separate allocation and distribution. In Arrows’ terminology, “allocation” is what the market system does: based on demand, it sets prices and optimizes supply. “Distribution” is what social policy does: it taxes and gives subsidies to ensure that purchasing power is distributed in a fair way. In a market that meets the technical definition of “free”, a government can confine itself to ensure that purchasing power is fairly distributed and leave the market to itself.

Is healtcare in America not a free market?

This is the curious part: American healthcare is not fully free, nor does it seem “unfree” enough to get the full benefits of being unfree. Case and Deaton hit this point squarely:

[Americans] like to believe that the [healthcare] system is a free-market one, in spite of the fact that the government is paying half of the costs, is paying the prices demanded by pharmaceutical companies without negotiation […], is granting patents for devices and drugs, is permitting professional associations to restrict supply, and is subsidizing employer-provided healthcare through the tax system.

That list is bewildering, so I think it’s worth some recapitulation and explanation (which you can find elsewhere in Case’s and Deaton’s book):

Federal and state governments pay about half of healthcare costs in America, largely through Medicare and Medicaid. The government isn’t “interfering” in a free market; it is the key player in that market.
The 2003 law that established Medicare Part D (which pays for prescription drugs) had a “noninterference” clause that prevented Medicare from negotiating on drug prices with pharmaceutical companies. There’s no way one could call this a “free” market.
The Federal Patent Office issues patents, which essentially prevent companies from making profits off of another company’s innovations for some time, and the FDA creates periods of exclusivity during which a drug too similar to an existing, approved drug will not be approved. Patents and exclusivity certainly help innovation, but they aren’t part of a truly “free” market, and they are examples of government “interference” in a market that is generally welcomed.
Federal and state governments essentially allow the medical profession (i.e., associations of physicians) to control medical education, licensure, and certification. In other words, the government allows doctors to decide how many doctors there are. It would be like the government allowing US car manufacturers to collude in deciding how many cars to sell each year (thus increasing the price) and preventing foreign cars from being imported.
Employer-provided healthcare is tax free. If your employer increases your compensation by \$1 in wages, then you and your employer may tax on that dollar, approaching half of it. If your employer gives you that \$1 as healthcare coverage, then neither you nor your employer pays tax. This is an enormous tax break for healthcare: we don’t get, say, office furniture tax-free.

Healthcare can’t be a free market anyway

Arrow, writing some 60 years ago, did not have his eyes on the healthcare system we have today, but he still saw how Americans’ overall approach to healthcare, which hasn’t changed fundamentally in that time, could never be a “free” market in the technical, economics-jargon sense.

Unmarketability

An important feature of a free market is that things that are valuable are bought and sold. This is not the case in health. Arrows gives the example of communicable diseases, which feels very pertinent during Covid. In a free market, if someone is not vaccinated (or doesn’t wear a mask) and presents risks to those around them, they would need to pay others for the privilege of putting them at risk (or the others would pay the risky individual to get vaccinated). This is clearly not how we think about healthcare.

Another critical thing that would be marketable in a free, competitive market, the kind that can qualify for optimality as described above is risk-bearing. Ideally, you could get free-market insurance against the costs of illness. However, no plausible healthcare insurance policy could distinguish between unavoidable and avoidable risks, which creates moral hazards, i.e., the temptation for covered individuals to take less care of their health because they know someone else will bear the financial cost. We feel more confident in something like car insurance that we can assign blame because car insurance doesn’t cover the unavoidable wear and tear on your car, while health insurance does.

Information

The trouble with insurance and risk-bearing is principally about information. If we knew what was wrong with us and why (and what will become wrong with us), then the situation would look very different. But as it is, the purchased good in healthcare is, to a large degree, information, such as a doctor’s diagnosis or recommendation about what action to take. It’s very hard to know how much to pay for information. In other words, it’s hard to know how much what you will learn from a doctor is worth. If you knew that going to a doctor would provide information that would relieve your pain, then you could put a price on that, since you know your pain. But if you don’t know if you will get relief, or even the probability that you will get information that will give you relief, how can there be a sensible market?

One of the reasons this problem with information arises is because, unlike other goods, we cannot test healthcare many times to determine our preferences. I buy groceries many times, and I can compare different foods side by side, and so I can get a sense of what I like and don’t. But if I’m sick and need some operation or treatment, I can’t run many tests, or try a little bit of two kinds of healthcare and decide which one would suit me better for that situation. For this reason, we demand an element of trust in our relationships with physicians: they should be recommending what is best for us in a way that we don’t expect a car sales person to do. Or, stated the other way, we find it clearly immoral for a doctor to try to upsell up or whatever in the way we expect that a car sales person might.

Range of goods offered

When I buy something off of Amazon, there is a wide range of products I can get, from cheap and low-quality to expensive and high-quality. For products like food, there is a greater guarantee of quality: I can feel confident that most food I can legally buy is not going to make me sick. This creates some market restriction: I, as an enlightened consumer, simply cannot buy food below a certain quality, even if I would rather pay less money in exchange for an increased chance of disease.

Healthcare is even more restricted: medical associations control the content of medical education and the processes of licensure and certification, which produces a floor on the type of care you can get. This isn’t necessarily a bad thing —I’m not saying I want a pre-Flexner Report world where anybody can hang a shingle and say they’re a doctor— but it does show that healthcare isn’t a free market. Indeed, we don’t want it to be a free market, precisely because we’d rather have all our physicians meet a certain standard of training. (We have gotten clever in more recent years about how to redistribute the workload of care between physicians, nurses, physician assistants, and so forth, but it’s important to remember that all these other people practice under a physician. The only potentially interesting example are chiropracters, or maybe physical therapists, although even they have their own methods of certification.)

What this means for healthcare policy

Even in Arrow’s day, there was concern about growing healthcare costs:

Insruance removes the incentive on the part of individuals, patients, and physicians to shop around for better prices for hospitalization and surgical care. The market forces, therefore, tend to be replaced by direct institutional control.

In other words, health insurance (in the sense of protection against costs, not a guarantee of care) improves social welfare, but we can’t expect insurance to function well enough on its own, in a free market.

Arrow also notes that there are apparently “enormous economies of scale in the provision of insurance”, based on data about how much insurance companies have to spend to administer individual versus group healthcare plans. In other words, the simple administrative costs of selling an insurance policy are so much higher when selling to an individual compared to selling to a group. Arrow notes that this makes “a very strong argument for widespread plans, including, in particular, compulsory ones.”

Git \‘er done

2021-03-13T00:00:00-05:00

As a sort of side hobby, I’ve started thinking about how political speeches could have been delivered better, or just differently. So I imagined how Biden’s speech about vaccination in spring and summer 2021 would have sounded like, if written by Larry the Cable Guy:

My fellow Americans: We have got to crush this virus. The only way we will get everyone back to work, the only way we can make sure no one dies from just being with other people, is to crush this virus. And it is my pleasure to unveal my three part plan, to crush coronavirus. First, you have to get your vaccine. Second, rrrbody needs to social distance, at least until July 4. Finally, you have to don a mask, at least for the next few months.

That’s right, America: git ‘er done!

The Supreme Court is funny

2021-03-09T00:00:00-05:00

I’m reading Legislation and Regulation, a law textbook about, among other things, how judges interpret written law (or “statutes”). I’ve been surprised by (1) how many important cases hinge on the reading of a single word in the text of a law, (2) how devoted the Supreme Court is in to trying to understand that single word, and (3) how funny they can be about it.

Here’s an example. Smith v. United States was a Supreme Court case in 1993. The facts of the case are these: A certain John Angus Smith traveled from Tennessee to Florida to buy cocaine, planning to return to Tenneesee and re-sell at a profit. While in Florida, Smith made an offer to someone who was, unbeknowst to him, an undercover cop. The offer was that Smith would trade his MAC-10 submachine gun for two ounces of cocaine.

By the time this case reached the Supreme Court, it was already decided that Smith was guilty of drug trafficking. The question was whether he would get extra jail time, potentially another 30 years, because he, in the words of the relevant law, “during and in relation to any crime of violence or drug trafficking[,] use[d] or carrie[d] a firearm.” The operative question was: did Smith’s offer to trade a gun for drugs mean that he “used a firearm” in relation to a drug trafficking crime?

Smith argued that the phrase “use a firearm” in the law means “use a firearm as a weapon”. Because he did not fire the gun or threaten anyone with his MAC-10, he should not be subject to the extra jail time.

Unfortunately for Smith, the Court’s majority decided that, yes, trading a gun for drugs qualified as “using” a firearm. They said, although this use of a firearm is not the first thing that comes to mind when someone says “use a firearm”, it nevertheless is a “use” of a firearm. As an analogy, they say, consider a cane. If you ask, “Do you use a cane?” the first thing that comes to mind is whether you use a cane as an aid for walking. But, the Court argues, this does not mean that having a decorative cane in your hallway is not a “use” of a cane. They then point out that the most famous “use” of a cane in American history is when Senator Preston Brooks nearly killed Senator Charles Sumner on the floor of the Senate chamber by beating him with his cane in 1856. Indeed, they observe: using a cane as an instrument of punishment was once so common that “to cane” meant “to beat with a cane”.

If this diversion into the history of the cane was not amusing enough, the Court then noted that there are indeed “uses” of a firearm that would not be covered by the text of the law. For example, if a drug trafficker used a gun to scratch his head, this would not constitute a “use” of a firearm in terms of the law because, although scratching one’s head is certainly a “use” of a firearm, such uses do not count “unless [the gun] facilitates or furthers the drug crime […] [T]hat the firearm served to relieve an itch is not enough…” The Court then opines whether, in their reading, pistol-whipping, an offensive use of a firearm that does not involve firing it, would qualify as “use” under the law.

The dissenting Justices start by noting that “use” is such an “elastic” verb that it could be employed in so many ways, noting that “use” in “he uses tobacco” and “he used to smoke tobacco” mean different things. They then addressed the questions of the decorative cane and the head scratching:

When someone asks, “do you use a cane?,” he is not inquiring whether you have your grandfather’s silver-handled walking cane on display in the hall; he wants to know whether you walk with a cane. Similarly, to speak of “using a firearm” is to speak of using it for its distinctive purpose, i.e., as a weapon. [… However] It would also be reasonable and normal to say that [someone] “used” [a MAC-10] to scratch his head. When one wishes to describe the action of employing the instrument of a firearm for such unusual purposes, “use” is assuredly the verb one could select. [… However] It is unquestionably not reasonable and normal, I think, to say simply “do not use firearms” when on means to prohibit selling or scratching with them…

In a footnote, the dissent wrote that it is clear that the ordinary use of “use a firearm” does not include selling it because a witness who “sold his grandfather’s Enfield rifle to a collector” would be justified in answering “no” if asked in court whether he had “ever ‘used a firearm’”. Another reason to think that the law didn’t mean “use” to include “trading” comes from yet another example in the dissent:

[T]he statute provides that its prohibition on certain transactions in firearms “shall not apply to the loan or rental of a firearm to any person for temporary use for lawful sporting purposes,” I have no doubt that the “use” referred to is only use as a sporting weapon, and not the use of pawning the firearm to pay for a ski trip.

It’s interesting to me that, while I’m sure the Court was not intending to be funny per se, that humor arises from the particular examples that they found and used to explore the boundaries of the term “use”. Does it follow that an exploration of the boundaries of a term must inevitably lead to funny examples? Would a non-native English speaker find any of this funny?

What's wrong with Ubuntu on my laptop

2020-12-16T00:00:00-05:00

I’ve been a many-years user of Macs, and so I was a little deflated when my job sent me a Windows machine, a Lenovo T490S. So I immediately installed Ubuntu as a dual-boot, hoping that I could live forever in Linux-land and closely approximately my Mac experience. Here was the result:

Pro’s of Linux-land:

Boot-up is fast. Once I enter my login credentials, the desktop is up and ready to go in a second. This is a big contrast from the slow, churning startup of both Mac and Windows.
There’s no bloat. There’s nothing I didn’t ask for here.
For a modeler/data scientist, there’s no better environment than being able to fire up the terminal and be ready to compute. Conda helped a lot on Mac to get the OS environment (with its Python 2) separated from my work environment, and WSL gave me Linux within Windows. But nothing beats Linux for this.

Con’s of Linux-land:

A few things didn’t work on my laptop. It’s amazing how these small things can become such a part of your routine that they seem almost essential.
- Swipe gestures to go forward/back in my web browser. Apparently there’s a Chrome extension that kind of does this, but not well enough to be seamless.
- Windows and ChromeOS use the same keystroke (Ctrl+Tab) to move between browser tabs and terminal tabs. Weirdly, in Ubuntu, the terminal tabs need Ctrl+PageUp, which I find tremendously unnatural. For reasons I had to dig to find and then didn’t understand, you can’t remap the terminal to accept Ctrl+Tab instead.
There’s no Word or Excel on Linux. I know, there are ways around it, but it’s going to be that much of a pain, I’d rather just fire up Windows instead. (Google Drive has gotten really close, so that I know edit even my .docx resume in Drive. But it’s not sufficiently seamless to collaborate.)
~~There’s no Google Drive backup program.~~ OK, although this is true, I’ve found that you can mount your files locally using gnome-online-accounts (link).

I recognize that these are such small things (aside from Office software!) that I might be able to hunt around among other distros to get a better experience.

August 2021 update: I recently tried elementary OS, which impressed me both by its beauty and niceness, but also by the fact that:

My most needed swipe gesture, two fingers to go back on a webpage, worked right out of the box on my Lenovo ThinkPad X1.
The keyboard shortcuts were very customizable, and Ctrl+Tab did the sensible thing of changing tabs in the terminal.

So Linux can’t easily get arond the Word/Excel problems, but I think that’s what we’re down to nowadays.