Archive - Short essays, research and random thoughts

Residual confounding in Environmental Health - November 2019

I have written up a document that goes into detail on the subject of residual confounding in environmental health. The crux of the analysis is that when scientists improve the measurement of exposure to hazards without also improving the measurement of confounders, it leads to biased results. The document incldues R code that you can use to see the results yourself.

On Losers - October 2019

What is a loser?

In the traditional sense of the word, a loser is simply someone that has not won. This is a descriptive and sometimes useful definition, but of course ‘loser’ is often used to imply a pathology of failure–someone who never wins, who can’t succeed at anything, and that we as a society don’t value.

I am not satisfied with this definition, so I offer an alternative:

Loser: a person for whom success is uncorrelated or negatively correlated with demonstrable merit.

Using this definition, a person can be a loser in two ways:

A person with great potential that is wasted.
A person with great success due to something other than demonstrated merit.

Both 1 and 2 require a little explanation. First, consider the concept of wasted potential. To waste potential means failing to live up to what one could have done had they tried. Trying and failing doesn’t make someone a loser. A loser is anyone who doesn’t make use of the talents they have due to things like fear of failure, sloth, or sense of entitlement.

Second, consider the concept of merit. Merit is ability, skill or talent (inherited or not) that is relevant to success. A great hockey player that records a best selling album of mediocre country music is a loser in the country music domain. His success in hockey doesn’t demonstrate his merit in other areas. So the hockey player is a loser in one domain and not a loser in the other. For merit to be meaningful, it has to be linked to success.

Why should you accept my definition?

Calling someone a ‘loser’ in the traditional sense is to ignore important many uncontrollable factors that contribute to failure and lack of success–like bad-luck. People should not be held responsible for bad luck. Bad luck doesn’t make a person a loser.
Undeserved success is economically inefficient. Merit-less success rewards people based on attributes that are not relevant to the creation of value. Labelling people with unmerited success as ‘losers’ is a way of knocking them off their roost (at least verbally), and perhaps making way for those more deserving of praise.
Fear of being thought of as a ‘loser’ in the traditional sense may discourage risk-taking. The world benefits from some risk-taking; it doesn’t make sense to condemn people for trying and failing. Trying and failing is taking one for the team. By my definition, trying and failing doesn’t make someone a loser.
If you accept my definition, you won’t know whether a person is a loser or not unless you get to know them, and compare their success against their merit. Nobody is a loser by default.

Here are some examples to help illustrate the meaning and usefulness of this term.

At least three of Donald Drumpf’s children are clearly losers (Jr. and Ivanka and the other one). As far as I am aware, they have not been tested in this world, yet they seem to wield great power and reap the rewards of ancestral financial success. They did not ‘make it’ in business or politics or anything in a way that demonstrates a competence commensurate with their positions. Losers.

Members of royal familes are pretty well all losers. Inherited wealth and power? Meh.

Dictators are always losers; their power and wealth is not a reflection of demonstrated ability to governing. They have skills best suited for other kinds of occupations--like bullying, stealing and murder.

Big lottery winners are all losers. Nobody that wins the lottery wins because of merit, just dumb luck. Lottery losers are losers too (in the traditional sense), so it would seem that playing the lottery is for losers.

Athletes are pretty well never losers; sports are great at demonstrating merit. People don’t win at sports by luck alone; any sport that you can win based on luck alone isn’t a sport. People who try and lose at sports aren’t losers either. To paraphrase 90% of the high school gym teachers who ever lived: the only losers are those who don’t try!

On the other hand, super famous actors and musicians are all kind of losers. We may love them, but demonstrating merit in the acting/music industries is pretty tricky. I firmly believe that any rock ‘n’ roll orchestra that is filthy, insanely successful in money and fame should be viewed with suspicion. Sure, Nickelback has sold millions of records, but they are clearly losers.

Mark Zuckerberg, Bill Gates, Steve Jobs, Elon Musk, Jeff Bezos: mostly losers. They all have demonstrated some merit, but it is pretty weakly correlated with their success. Check out a recent computer simulation that makes this point beautifully. If they were multi-millionaires, they may not have been losers. But as billionaires, losers they be.

Doctors make a lot of money, and are highly regarded culturally, but most aren’t losers. Doctors are under close professional scrutiny, and it takes a lot of work to enter the profession–most of them have to work their butts off to get into and out of medical school, especially these days. People who are bad at these jobs don’t keep them for very long. Doctors are not losers. Of course, most people in health care aren’t losers. They have tough jobs and neither their pay or cultural recognition is excessive. The same can be said about school teachers who care and are good at their jobs.

However, many University professors are kind of losers. Universities are not good at getting rid of under-performers, and success (in terms of salary and reputation) usually goes up over time irrespective of merit. There is evidence that the research system as a whole is not always great at allocating success (see the same simulation above). I’ll fully acknowledge that I am probably at least 37.5% loser (+/- 20% 19 times out of 20).

Conclusion: we’re mostly not losers

Most regular folks with jobs, and/or families and/or that contribute to their communities are not losers. In fact, 80-90% of the human population are probably not losers. Your neighbour who keeps you up at night with his banjo music and pool parties? Sorry, he’s probably not a loser, just annoying.

Uncertainty, coordination and climate change- June 2019

Today I spent a couple hours reading two papers authored by Scott Barrett, an environmental economist. One paper is Climate treaties and approaching catastrophes and the other (co-authored with Astrid Dannenberg) is "Sensitivity of collective action to uncertainty about climate tipping points". The first paper is purely theoretical, and the latter has theory and an economic experiment.

Barrett’s focus in these papers is on climate change treaties. Climate change treaties are a challenge because costs of CO2 pollution are externalized, but countries are inclined to ‘free ride’ since there is no global government to punish non-signatories, or countries that fail to meet their commitments. For this reason, many argue that global-scale climate treaties are doomed to fail.

These papers show that coordination can work to address this climate treaty problem. Coordination occurs when actors work in their self-interest towards a goal that is collectively beneficial. When coordination is possible, there is no need for an enforcement mechanism; the treaties work because adhering to commitments are in the private interests of actors.

A number of conditions must exist to see effective coordination in abating climate change, but the one of Barrett’s focus is the reduction of uncertainty at the threshold which a climate catastrophe would occur. In short, as uncertainty in the threshold gets smaller, the more likely that an actor will follow through with abatement measures–like reducing CO2 emissions. Provided the uncertainty is small enough (in proportion to the damages resulting from climate change, and inversely proportional to the difference between costly abatement and the benefits of climate change) then actors will choose to take measures to avoid catastrophe.

Lucky for you, I wrote some R code that you can use to experiment with the model he proposes!

In addition to demonstrating this theoretically, Barret and his co-author use some hypothetical choice experiments to test this theory in the lab. The experimental results are fairly consistent with their theoretical findings; the authors found that most people would choose to abate when the catastrophic tipping point is certain, and nobody would abate when it is very uncertain.

What this research shows is the importance of reducing scientific uncertainty, and specifically, that reducing uncertainty about thresholds of climate catastrophes may be key to getting useful and effective climate treaties.

Parity in hockey: a simulation - October 2019

I wrote a bit of R code to explore the ‘parity question’ in the NHL; specifically, if all teams were more or less equal, what would we expect to see in terms of regular season point totals? You can find the code for generating a hockey schedule here and the simulation code here.

The scheduler is just a dirty optimizing algorithm that I came up with that is probably inferior to what the NHL uses, but it seems to work. You can set the schedule parameters and it seems to work fairly well provided the inputs result in a feasible solution; for example, don’t lower the max games too much or it won’t be able to solve. I had to create the scheduler to generate a realistic play schedule that the hockey simulator could use.

For a given schedule, the hockey simulator plays a season of hockey and sums up point totals. For the purpose of this experiment, I assume that all teams have an equal chance of winning. If a team loses, they have a x% chance at getting a loser point (by default, this is set at 11%).

I then run the simulation many times. One way to investigate the results is to plot out the distribution of top regular season team (the team with the most points). Here is a histogram of the maximum point totals over 1000 simulations:

How does this compare to real data? Well, it depends on the season, but generally speaking, the best teams in the real world do better than the best teams in the simulation. This suggests that the NHL has not yet reached the point of real parity. The far left values on the figure below tell us the mean, range (95%) and maximum (of the maximum points) over 1000 simulations. Tampa Bay outperformed the parity simulation even after 1000 simulations–suggesting that their regular season performance in 2018/2019 was far from a matter of luck.

Nevertheless, imagine if 2018/2019 Tampa Bay was an anomaly, and the trend of real NHL point maximums continues downward–where the best regular season team gets point totals that approach 107 or so in a few years. At that point, it might suggest that the NHL has reached some level of true parity, where the vagaries of year-to-year luck (like injuries) will play an increasingly large role in determining the success of a team, rather than true superiority of skill and tactics on the ice.

Data, infant mortality and anti-abortion activism - May 2019

The following question crossed my mind recently: how many lives are lost from excess infant mortality?

I asked this question because I wondered if anti-abortion activists couldn’t better spend their time saving the lives of children that died in their first year of life, rather than protesting at abortion clinics. Saving the life of a child who has been born seems like an easier political task, and will almost certainly be more effective.

We have a pretty good sense that this is theoretically possible because of the considerable variation in infant mortality that exists worldwide. Most poor countries have high infant mortality rates (IMR), and most rich countries have low infant mortality rates. There are a number of countries in Africa with IMR above 50 per 1000 live births (meaning 50 children die in their first year for every 1000 live births). There are rich countries with IMR below 5. The difference between these numbers–roughly 45 per 1000 births–tells us that preventing infant death could be very impactful.

In the US, there is variation in IMR across ethnic communities. The IMR among African Americans is 11.4, and 4.9 among whites. All of these babies are being born in the US, but white babies are much more likely to survive past 1 year of age. Preventing infant deaths in African American and other marginalised communities would save lives without the controversy.

In Canada, the IMR for the Inuit population is a shocking 17.7, three times the national average. Improvements to maternal care and access to quality health care in remote rural areas would save the lives of many living infants. Given that there is little popular or political appetite for outlawing abortion in Canada, reducing IMR among the Inuit seems infinitely more productive way to spend time for people who are concerned with preserving lives of children.

Globally, the potential numbers of lives saved would be much greater. I used World Bank data to estimate the number of infant lives that could be saved worldwide if we lowered IMR to 5 per 1000 for in every country in the world that currently has an IMR greater than 5. You can download the R code here. According to these calculations, we could save around 3.6 million children every year by lowering IMR to 5.

Practical solutions for lowering infant deaths in poor countries are fairly well known. Worldwide, many infant deaths occur because of preventable infection. Most of these infections could probably be treated with investments in medical infrastructure and better access to antibiotics. Other deaths are caused by an absence of simple medical interventions and lack of training. It seems that anti-abortion activists could put their time into lobbying pharmaceutical companies to increase affordable access to antibiotics in developing countries, or improving access to maternal care in the developing world.

Closer to home, we could do more to increase access to health care and support mothers living in poverty, particularly in rural and remote areas. There are many lives to save, and best of all, none of them come with legal wrangling or political and social controversy. It seems that anyone who genuinely cares about saving lives for their own sake could do a lot about it without entering the quagmire of anti-abortion activism.

No room for human error - March 2019

Jaskirat Singh Sidhu caused the bus crash in Saskatchewan that killed over a dozen people. It’s a tragic event that received lots of media attention, in Canada, and elsewhere. The judge of the case recently handed Sidhu an 8-year prison term as punishment for causing the crash. A public transcript of the judgement can be found here.

Here is a list of important facts concerning this case.

Sidhu pleaded guilty to causing a collision that killed 16 people and injured 13. Specifically, he pleaded guilty to 16 counts of dangerous driving causing death, and 13 counts of dangerous driving causing bodily harm. These are criminal code offences in Canada, with a maximum sentence of 10 years in jail.
Sidhu expressed remorse, and there is no evidence of any intent to have caused this collision.
There were multiple stop sign warnings indicating that Sidhu was required to stop at the intersection.
The visibility was not an issue (due to weather or time of day)
Sidhu was not speeding at the time of the collision.

Here is some other information that is not specific to this case, but is important for context, and available to anyone who cares to look.

Many drivers have driven through stop signs and other traffic controls. If your personal experiences don’t convince you of this, the empirical evidence is abundant. In a U.S. study a few years ago, 13% of drivers at a stop sign controlled intersection did not stop at all and 52% came to only a rolling stop [1]. Older data from Canada showed similar numbers [2]. In a 2009 study of older drivers in the US, almost 16% of drivers over 66 years of age failed to stop at a stop sign at least once over a 5 day period [3]. Stop sign violations are second only to speeding violations on a per-kilometre-travelled basis [4], and are responsible for between 5 and 10% of collisions involving casualties [5].
Stop signs are not as effective as common sense might dictate. Multiple signs might even be counter productive, and changing controls from yield to stop may even increase risk of motor vehicle collision [6]. Some have argued that signs are a distraction, that we may become desensitised to them the more of them that there are, and we’re better off reducing the number and size of traffic controls [7,8].

On balance, this information shows that humans make mistakes–both in missing or ignoring traffic controls, and in assuming that traffic controls necessarily increase safety. Sidhu’s responsibility for the collision is not in question; what is in question is what punishment is appropriate when someone makes a fairly common mistake that leads to an incredibly unlikely and tragic outcome?

The judge’s view is to punish him with a jail sentence of 8 years. This is below the maximum. My gut tells me this is unfair, but I am not a judge, so I’m not sure my gut matters.

However, I do take exception with some of the reasoning offered by the judge. In her sentence ruling, the judge argued that it was “baffling and incomprehensible” that Sidhu missed the stop signs that were posted. As noted above, it should be neither baffling nor incomprehensible to anyone who cares to consult research on the efficacy of stop signs, or reads local newspapers [10].

Humans make mistakes. Sidhu’s mistake was not unique, not uncommon, and not malicious. What was uncommon was the convergence of other facts surrounding this event–that a bus full of teenagers happened to be in the wrong place at the wrong time.

The fact that the judge was motivated to sentence him to 8-years at least in part because she can’t imagine how a person could not have seen the stop signs is simply ignorant. It is clearly possible for a person to make this kind of mistake, as well as a variety of other perceptual and judgement errors that leads a person to run through a rural intersection without stopping. It happens all the time.

Another problem I have with the reasoning of the sentencing seems to have been the magnitude of harm. In her decision, the judge writes “a sentence of more than six years is mandated due to the horrific consequences of his actions.”

On the one hand, the judge acknowledges that Sidhu had no intent to harm, but was simply very negligent and inattentive. On the other hand, she is holding him responsible for the gravity of the outcome; had he been equally inattentive but killed one person, it seems that he would have received a shorter sentence.

In legal circles this may make sense, but to me it seems illogical. Sidhu’s crime was dangerous driving, and his dangerous driving directly increased the probability of causing a collision. He is accountable for that. However, this increase in probability had no corresponding impact on the consequences once a collision occurs; the bus full of young hockey players would still have been at that intersection whether or not he stepped on the brakes in time to avoid the collision.

Obviously the bus driver and victims of the collision were not at fault here. But assigning fault to Sidhu that is proportional to the loss of life suggests that he was responsible for the severity of the consequences, in spite of him having no control over them. I’d understand this conclusion if the judge ruled that he intended to cause the collision. However, she accepts that Sidhu “did not deliberately drive through the intersection.”

Conclusion

Something in me is unsettled by the fact that Sidhu is being punished harshly partly because of the number of lives lost in the this tragic event. He was negligent, and punishment may be required, but was he really so much more negligent than the average truck driver, or even the average person?

I’d like to think all of us can relate to making mistakes. I guess we should just feel lucky that most of our mistakes are very unlikely to harm anyone, as it seems that the courts have little tolerance for human error.

References

https://ageconsearch.umn.edu/record/207330/files/2012v51n3_07_StopControlledIntersections.pdf
https://www.researchgate.net/publication/232585442_An_opinion_survey_and_longitudinal_study_of_driver_behavior_at_stop_signs
https://www.sciencedirect.com/science/article/abs/pii/S0001457509001195
https://www.sciencedirect.com/science/article/abs/pii/0001457570900084
https://www.transportation.alberta.ca/Content/docType47/Production/AR2015.pdf
https://www.sciencedirect.com/science/article/abs/pii/0001457585900053
https://www.psychologytoday.com/ca/blog/adaptive-behavior/201605/death-stop-sign
https://www.theatlantic.com/magazine/archive/2008/07/distracting-miss-daisy/306873/
https://thestarphoenix.com/opinion/columnists/gormley-when-did-stop-signs-become-optional