The GAMSAT is an entrance test that all prospective students to medical schools in Australia must take.
I want to use a hypothetical scenario about this test to understand how it might be possible to determine whether it constrains the rate at which new doctors are trained.
The hypothetical
Some people say that this test affects the total stock of doctors and hence the price of medical services.
You have the following information and are asked whether this is a potentially important concern.
In addition, you know that
Only those with a bachelors degree are eligible to take the test.
The number of people graduating with bachelors degrees each year is nearly a consistent 20,000 per year, adding to a large pool of candidate test takers.
Those who do not pass the GAMSAT can re-sit the test as many times as they like in subsequent years.
Those who pass have the option, but no obligation, to attend medical school.
You must re-sit the test if you do not go on to medical school within three years.
100% of those that decide to attend medical school complete it and become practising doctors.
You are asked to advise whether the pass rate contains information about the degree to which the entrance test determines the stock of practising doctors. Some say the high pass rate and ability to re-sit the test shows that the GAMSAT test is not a constraint on the supply of doctors.
Let us think this through.
The system perspective
The first thing to do is get a good understanding of the system with the numbers involved. The below diagram shows how the stock of potential candidates flows through the testing system to become doctors. There are three decision points.
The choice to take the GAMSAT test
The pass/fail choice
The choice to proceed to study after a pass
I draw these choices as taps that control the flow of “water” into the “buckets” (stocks of people at each stage). Notice that two of the choices return the people back to the pool of candidates—the pass/fail, and the study/delay choices.
Quite clearly the most important choice in getting water from the stock of potentials to the stock of doctors is the choice to sit the GAMSAT test in the first place.
This choice has by far the biggest effect on the outcome, with its variability accounting for the variation of flows through the system by a magnitude of 16x. One year 50 people took the test. One year it was 800.
None of this variation appears due to the GAMSAT test as the pass rate is unchanged and the choice to proceed is unchanged (we will return to this assumption).
By looking at the system in this way we can see that the maximum amount of additional doctors getting through the process by removing the GAMSAT test is 6%. It is likely to be less than this because those who fail often repeat the test.
You conclude that the GAMSAT test is at most an extremely minor factor influencing the rate of supply of new doctors.
A new argument
However, some argue that there is no evidence in the 94% pass rate that the GAMSAT is not a major constraint.
The argument is that the existence of the test reduces the number of applicants. Those who are likely to fail will know in advance and choose not to take the test. Therefore, even if the pass rate was 100% the GAMSAT could still be a major restriction on the flow of new doctors. It might be a plausible assumption that the variation in the choice to sit the test is explained by the number of people who believe they will pass it.
So we have two potential mechanisms of actions of the GAMSAT test.
A direct effect due to the pass rate
An indirect effect due to reducing the number who choose to take the test
How could we tell if the second mechanism was important?
We could look further up the system and see if the variability of the choice to take the test is related to factors regarding the test stringency, or other factors. But how would you measure test stringency if not for the pass rate?
You would need a third variable that measures test stringency that is unrelated to the pass rate, and that correlates closely with the number of test-takers. Possible? I’m not sure.
The problem is that if the second indirect effect dominates, then what are we to make of variation in the pass rate? What if a 10% pass rate is the norm, and that falls to 5% when the number of test-takers is high? This would surely indicate that the indirect effect is minimal and that people do not have a good idea of whether they will pass in advance. Or that they are willing to take the chance even if they have a good idea in advances.
Whichever way you cut it, the presence of an indirect effect surely must show up in the pass rate to some degree.
What have we learned?
It seems logical that there is information in the pass rate about the degree to which the GAMSAT test can reduce the flow of new doctors compared to if the test did not exist.
In the real world, and not the hypothetical I described, the pass rate for the GAMSAT is about 20-25%. In fact, the pass rate is itself determined by a quota on new university places. The test doesn’t constrain new doctors because the university quotas do it, and that quota determines the passing grade and hence the pass rate.
The reason for explaining this is because this post is not about medical school. It is about town planning. The “entrance test” in the planning system is a planning application, which is required to (re)develop a property.
Many argue (e.g. point 4 here) that just because 90%+ of the planning applications are approved that this doesn’t indicate there is at most a small effect on the rate of new housing supply. They argue that the indirect effect dominates and that’s why the approval rate is high.
But this leaves us with a conundrum. We know that a property with a planning approval is worth a lot more than one without. Therefore there is a large payoff to getting an approval. Just like there is a large payoff to becoming a doctor.
Yet candidate medical students are willing to sit a test with a near 80% failure rate, often repeatedly, to get that payoff. However, property owners are not, even though the payoffs can be worth tens of millions of dollars or more.
While an indirect effect surely exists in both medical school entrance tests and town planning applications, the pass rate also contains information about the existence of this effect.
“There are two schools of thought… Science stands for healthy scepticism… asking for better evidence… Then you have a second school of thought that is public health… it has the stance that we have a crisis, we are like an army, the platoon must do this or that. Anyone who leaves the platoon must be shot down.”
I like the way John Ioannidis has characterised the COVID public health response. The science and scepticism approach has been overridden by the public health army approach, which has little need for evidence. I recommend his presentation in this video, which is the source of the above quote.
The two schools of thought might explain the “you are better than this” responses I sometimes get on Twitter when I raise concerns that the public health policy seems detached from the scientific and logical reality. I hope it’s because they want me in their army. I hope it’s not because they have given up seeking the truth.
For some reason, I have a brain that can’t stop trying to seek out contradictions and the underlying logic that makes sense of the world. Scientific scepticism seems hard-wired. For example, when I look at Australia’s superannuation system, logic forces me to conclude that the system as a whole makes funding retirement harder, not easier. So I say we should dismantle it altogether.
I predicted that house prices in Australia would rise in May last year and people scoffed. Someone told me I should hand back my degrees. But the underlying logic I saw was correct (or at least useful for prediction).
Being right when the mob is wrong is, unfortunately, never popular.
In fact, a good rule of thumb is that there is no new information when someone says something popular. There is a huge amount of information when someone risks their reputation to say something. This is why John Ioannidis remains one of the few experts whose words contain actual information. He risks his reputation to say them.
This blog post is about the scientific and sceptical school of thought on COVID policy. It provides a glimpse of the contradictions and the underlying logic I see at play. Some of my previous thoughts and comments on COVID policy can be found here.
Seeking logic and evidence
Vaccines are the path
The current marching song is that vaccines are the path to freedom. Recently promoted by the Grattan Institute, an 80% vaccination target gets discussed as the key to returning to normal life.
There are two problems with this. First, getting 80% of adults vaccinated is quite difficult and has only been achieved in a small number of places.
Second, highly vaccinated places are getting COVID, some more than at any time before (e.g., Iceland, Israel, San Francisco, with surely more to come). While vaccines appear to be reducing mortality rates from COVID, differentiating the effect of the vaccination from the effect of previous population exposure is quite a challenge.
It seems to only make sense to vaccinate the elderly given the risk relativities and the limited effect on transmission.
Blaming one side of politics or the other for the “botched vaccine rollout” looks like nonsense to me when the experience elsewhere is that the level of vaccination is not having a major effect on subsequent virus waves.
Masks work
The only problem with this idea is that you cannot see it in the population-level data. In fact, you cannot even see much evidence that masks work in surgical theatres. Here is a thread containing many studies showing they don't.
Masks have become political symbols. And people love it.
The seasonal resurgence of COVID across the US despite vaccination and masking is quickly turning into a political problem to be manoeuvred around, not a health issue. Yet more signs that COVID policy is not health-driven.
Vaccine passports
Another chant I hear is that vaccine passports are necessary. But if vaccines work, we do not need a vaccine passport. The vaccinated are not at risk and their presence in the population reduces virus spread regardless. If vaccines do not work, then a vaccine passport is not stopping the virus from circulating and spreading amongst the vaccinated.
Notably, the recent data shows that vaccines wear off and vaccinated people get COVID at quite a high rate compared to unvaccinated (perhaps as much as 80% after six months) and are likely to transmit at a similar rate. This is why vaccine boosters are being planned.
I cannot see how the current crop of vaccines gets anywhere near a reasonable benchmark for restricting movement. Most people when pushed seem happy that vaccine passports would be used as an incentive to get vaccinated rather than as a direct health measure.
R0 talk and “exponential” threats
Despite a high reproduction rate and infectiousness, many COVID Delta waves have fallen off dramatically with relatively low infections (e.g. India). R0 does not seem to give any indication of the final size of an outbreak.
Another big unknown in the modelling is the degree of prior immunity in terms of the variation in COVID waves over time and between regions (such as if previous local viruses conferred some protection in the population), and in terms of potential for reinfection.
Lockdown cost-benefit
The lack of discussion about costs and benefits from masks and lockdowns is mass willful blindness. When an attempt is made, or some concession is made that the approach of evaluating costs and benefits is sound, usually another panicked argument is substituted instead.
Way back in the early days of COVID we saw some appalling attempts at cost-benefit analysis. One was out by a factor of 1,000! You could not be more wrong if you tried. Despite this, these same people are pretending to have been right all along and are still being taken seriously by the media.
That basket case Sweden
Sweden had a roughly 5% increase in total deaths in 2020 with no vaccine and no lockdowns (98,000 against ~93,000 expected deaths). For context, total deaths increased 5% in Australia from 2015 to 2017 (144,000 compared to 137,000 due to a 1.5% increase followed by a 3.5% increase).
Sweden saw no increase in deaths in any age group under 50 years.
When faced with these facts some say Sweden did reduce mobility voluntarily and that made the difference. But this merely implies that compulsory masks and mass lockdowns are not necessary and do not make a difference. You cannot have it both ways.
Kids and vaccines
Plenty of medical experts and ethicists warn about the risks of vaccinating children. They are rightly cautious. If one death per million from the AZ vaccine applied to Australian children that would kill 7 kids if they were all vaccinated. How many would it save? Given the low risk of COVID in children that number seems to be roughly 14 to 20. Are you happy with that trade-off?
This paper estimates the likely range of vaccine-related deaths if 80% of 18-59 year olds are vaccinated at 17 to 153. Given how little vaccines seem to stop virus transmission these risks need to be carefully assessed.
Suicides
A concern of mine has been that lockdowns would result in a rise in suicides. Thankfully that has not happened, but that does not mean there is no harm from lockdowns. U.S. data is showing a 50% rise in emergency department visits by teenage girls involving suspected suicide attempts.
The material prepared for quarantining households is predicated on the fact that forcing people to stay in their homes for weeks on end will lead to people bashing each other. Recent surveys of domestic violence care agencies suggest this has been the case.
Surveys show huge increases in depressive symptoms during lockdowns. These human well-being costs are real.
Media reporting for the army
I want to also demonstrate how easily the media falls into line with the public health army.
You may have seen the below chart showing that death from the AZ vaccine is less likely than being killed by a lightning strike in any year. Did you ever question it?
I did not at first either. But once your logical mind is brought into action you have to ask some questions. This is a classic example of the media picking and choosing “facts” and repeating them until they become the truth. Here’s the AMA President repeating it. Expect to hear it in casual conversation.
But there are two problems with this “fact”. First, the AZ vaccine has seen 7 deaths in Australia from roughly 7 million doses, so that risk is closer to double the 0.5 per million presented. Much more for all side-effects. Second, a risk of 0.4 per million for lightning strikes implies about 10 lightning strike deaths in Australia per year. But in reality, it is usually less than 2 (average of 1.9 for the past decade). So this is overestimated by a factor of five.
These two corrections mean that the AZ vaccine is ten times more likely to kill you than lightning. This “fact” is off by a factor of ten. The vaccine risk is still low. But this is hugely misleading and certainly is not going to promote trust in authorities when the error becomes more widely known.
Do you think the author of the original article presenting this “fact”, or the editors at The Conversation, actually care? Nope.
I have been reliably informed that someone with a keen eye for statistics approached the author to request they update the chart with more accurate statistics (their original lightning strike stats were simply lifted from here). But no. No action. The editors prefer to keep the wrong statistic on this hugely important topic rather than issue a small correction. Off by a factor of ten is totally acceptable as long as you are marching with the public health army.
And what of the risk of people dying with COVID? Why not put that on the chart and make a decent comparison. There has been almost no attempt at putting COVID risk in context in the media.
Perhaps the reason is that the data doesn’t sing to the public health army marching song. Take the Swedish data again. For ages 0-19 the risk of dying from COVID after 18 months of community transmission including two waves of infections, mostly with no vaccine, masks or lockdown, was 3.7 per million (9 deaths out of 2.4 million population). On an annual basis that is 2.5 per million. If we partition the data to account for co-morbidities, a healthy young person’s risk of COVID death gets much lower. Lower than the one in a million risk from the AZ vaccine? Probably not. But not a big difference, and certainly not enough difference to warrant the calls for rushing to vaccinate children.
Another place the media seems to be wrong is the story that vaccines produce better immunity than recovering from COVID. You might have heard this or seen a tweet like this.
So let us check the source of this claim. Nope. The study has no comparison between recovery-induced immunity and vaccine-induced immunity. It does show that some well-known immune responses do wane over time after infection. But this natural immune response may still be more persistent than the response from vaccines. However different evidence would be needed to establish the relativities. That doesn't stop the authors from making this claim, which is strange considering that one of the findings is that there is a subpopulation of people with a super strong and persistent immune response. Could they be simply chanting the public health army marching song?
Predictions
All of this has been a long-winded way of saying that a lot of what you hear about COVID and vaccines and the effect of our policy choices is incomplete, misleading, or plain old wrong. The one part that does make sense is quickly getting vaccines to the elderly—the overwhelming evidence for this conclusion is why every place is doing it regardless of differing views on masks, lockdowns, vaccine passports or border controls. In my view, vaccinating the elderly is one of the few policy actions the evidence favours.
The rest of the actions only make sense if you are in the business of marching a public health army and don’t care where that army is going or how many of its own it loses along the way. Lockdowns cost a huge amount of lives, masks don't do anything at a population level, and vaccine passports make no sense given the type of vaccinations available.
If the underlying logic of COVID I have identified is roughly true, then I should be able to make some predictions. Here are some.
There will be a time in the next two years when Australia has a much bigger COVID outbreak than any yet despite being hugely vaccinated. Depending on the political fallout from 2021 we may even collectively take no action. No masks. No lockdowns. No border closures.
Australia will see a year with a 7% increase in all-cause deaths (about 10,000) in the next decade and no one will notice. Given the ageing population and the normal variation in deaths each year, this makes sense. I’m actually being intentionally bold on this prediction. Realistically a 5% increase (7,000 extra deaths), or 134 extra deaths per week, is more likely to be observed.
Vaccine passports of some sort will be enacted against all evidence. They will be cheered by the mainstream media as they justify all the terrible policies the public health army has forced onto us so far. No one will care that the vaccines wear off or that the vaccinated transmit the virus to a similar extent after six months or so. The public health army will march on from the vaccine race song to the vaccine passport song, to whatever else keeps the marching going.
How the analysis looks to me
There is a spoof viking show called Norsemen on Netflix. In it, the characters talk about customs of life and death in a hilarious matter-of-fact way. I feel like I am living in a spoof Netflix show. The wonks are arguing the finer points of how to skin a virgin alive to please the gods while I stand by looking at the evidence that suggests rejecting the premise altogether. If only our policy choices today were a laughing matter.
I have a good track record of making “mad” predictions about Australia’s housing market that turn out to be correct.
In May 2013 I wrote that "if you have been holding off purchasing a home because of the risk of capital losses, then these risks are probably lower now than at any time in the past decade."
In May 2020 I argued that prices were more likely to rise than fall.
Here is another prediction. In the next fortnight, all Australian households will complete the census, as they do every five years. When the data comes out, it will show a slight uptick in the homeownership rate compared to the 65.4% from the 2016 census.
Why am I predicting this?
First, I want to look at the long term trends. Australia’s homeownership rate (share of households who own the home they live in, including if they have a mortgage) peaked in 1966 at 71.3%, having risen from 52.6% just two decades prior.
That boom in homeownership was brought about by heavy-handed government intervention in the housing market, including
rent controls that persisted post-war and incentivised landlords to sell,
public finance for first home buyers building new homes,
large scale public housing with tenant purchase programs,
along with many other interventions.
The market era from the 1980s onwards has seen homeownership rates fall from 70.4 % in 1986 to 65.4% in 2016.
However, there are ups and downs within the slight downwards trend over the last 30 years.
In a housing market, an imbalance of home buying between landlords and first home buyers leads to changes in overall homeownership. More landlords selling and more first home buyers buying is the only mechanism that changes the overall distribution.
Let’s look at those two elements of the market.
Since there are no available direct records of whether a property buyer is a landlord or first home buyer in any transaction, we can look at the patterns of mortgage finance.
The plot below shows the share of housing finance going to first home buyers and investors (sorry for no pre-2002 investor data).
The green shading shows the census periods with rising homeownership, and the red shading is falling homeownership, with the percentage point change in homeownership marked.
Since the last census in 2016, first home buyers have been large a proportion of all new mortgage lending, consistent with the 1991-2006 period of slightly rising homeownership (up from 68.8% to 69.8%).
But the stark change recently is the decline in investor lending in the market since 2016. In the 2011-16 period, investor lending was 39.4%, but since 2016 it has been 30.8% on average. Not buying and selling are somewhat equivalent asset investment options for investors. They both reduce the allocation to housing (just like not selling and buying keep more of an investment portfolio in property). We can then infer that the decline in investor buying is likely related to more investors selling.
This might not be the case. A confounding factor is the change in risk-weighting by banks for investor lending due to the fallout of the 2017-18 royal commission. Perhaps this means relatively fewer investors are selling compared to buying, but the overall level of activity fell substantially.
Another factor worth keeping in mind is the make-up of total households. The recent period of declining homeownership coincided with a period of rapid migration. These new households were more likely to be renters, at least for a short period. Most of this sub-set of renting households, including foreign students and temporary workers, have left Australia since COVID.
This combination of factors is why I am predicting a bounce in homeownership in the 2021 census.
