Stats Chat

Team Ratings for Postponed Matches

The basic method is described on my Department home page.
Here are the team ratings prior to this week’s games, along with the ratings at the start of the season.

Performance So Far

So far there have been 63 matches played, 43 of which were correctly predicted, a success rate of 68.3%.
Here are the predictions for last week’s games.

Predictions for Postponed Matches

Here are the predictions for Postponed Matches. The prediction is my estimated expected points difference with a positive margin being a win to the home team, and a negative margin a win to the away team.

There was a mysterious piece of vaccine news yesterday when German paper Handelsblatt claimed that the Oxford/AstraZeneca vaccine wasn’t going to be approved in Europe for people over 65 because it had only 8% efficacy in that group.  From the start, he claim of 8% efficacy was clearly either misquoted or misleadingly calculated, but it caused quite a bit of angst — and it is quite possible that the vaccine won’t be approved for older people

The problem is that the vaccine basically wasn’t tested in older people.  In contrast to the Moderna and Pfizer trials, the Oxford/AstraZeneca trials mostly recruited people under 55. The set of four trials reported in December had 428 people over 70, and fewer than 600 between 55 and 70.  Exactly none of these people had the low-dose first injection where the vaccine looked more effective. There’s more data from additional trials, eg in India, but still not very many older people recruited. In fact, about 8%^* of all the trial participants were over 65.

Testing in older people is valuable because they’re the single most important group to vaccinate, and because the immune system changes with age.  It’s always possible that a vaccine will be less effective in older people, because they won’t generate as good an antibody response. In normal circumstances, a vaccine with trial data like this would probably be approved with some upper age limit — partly because of real uncertainty and partly because you want to discourage manufacturers from doing trials that don’t include the most important target populations.

Right now, though, we’re seriously short of vaccines. The trials confirm the Oxford/AstraZeneca vaccine does produce good concentrations of antibodies in older people, so there’s reason to expect it will prevent disease. As an additional encouragement, the two mRNA vaccines didn’t show any signs of being importantly less effective with age.  Soon there should be data from a Johnson & Johnson vaccine using the same basic technology as the Oxford vaccine, and it will be even more reassuring if that is effective in older people.  All in all, it wouldn’t be unreasonable to use the Oxford vaccine at least in the short term, but it’s not really surprising that there’s a controversy.

It is more surprising that Handelsblatt would get the facts so badly wrong.

Update:  An actual table with actual data has escaped on to the internet, saying that the point estimate for efficacy in 65+ year olds is 6.3%, with a confidence interval from -1400% to 94%. That is, the vaccine could make you 14 times more likely or 19 times less likely to get Covid, or anything in between. That still doesn’t really explain where the 8% number came from, but it does show the basic problem that there just isn’t direct information on efficacy in older people.  You have to either rely on the antibody levels, or not. 

^* No, that apparently isn’t just a coincidence

Team Ratings for Round 15

The basic method is described on my Department home page.
Here are the team ratings prior to this week’s games, along with the ratings at the start of the season.

Performance So Far

So far there have been 100 matches played, 65 of which were correctly predicted, a success rate of 65%.
Here are the predictions for last week’s games.

Predictions for Round 15

Here are the predictions for Round 15. The prediction is my estimated expected points difference with a positive margin being a win to the home team, and a negative margin a win to the away team.

Team Ratings for Postponed Matches

The basic method is described on my Department home page.
Here are the team ratings prior to this week’s games, along with the ratings at the start of the season.

Performance So Far

So far there have been 62 matches played, 41 of which were correctly predicted, a success rate of 66.1%.
Here are the predictions for last week’s games.

Predictions for Postponed Matches

Here are the predictions for Postponed Matches. The prediction is my estimated expected points difference with a positive margin being a win to the home team, and a negative margin a win to the away team.

Team Ratings for the Currie Cup Final

The basic method is described on my Department home page.
Here are the team ratings prior to this week’s games, along with the ratings at the start of the season.

Note that Cheetahs2 refers to the Cheetahs team when there is a Pro14 match. The assumption is that the team playing in the Pro14 is the top team and the Currie Cup team is essentially a second team. Possibly there will be no such clashes this year

Performance So Far

So far there have been 21 matches played, 16 of which were correctly predicted, a success rate of 76.2%.
Here are the predictions for last week’s games.

Predictions for the Currie Cup Final

Here are the predictions for the Currie Cup Final. The prediction is my estimated expected points difference with a positive margin being a win to the home team, and a negative margin a win to the away team.

DOT loves data has produced an analysis of alcohol consumption in NZ last year, and Justin Lester tweeted a graphic.

It’s interesting to see the impact of the lockdowns, though it’s very limiting that that the graphic refers only to liquor stores (and maybe supermarkets?) and is in terms of cash spent rather than alcohol purchased, and is only for the start of the first lockdown. The ‘increase’ includes stocking up for consumption during lockdown, as well as any real increase in consumption.  Obviously, even if total alcohol consumption had remained the same, expenditures at liquor stores would have increased with pubs and restaurants closed. And even if total consumption had decreased it’s possible that expenditures at liquor stores could have increased.

Judging from last year, we should get StatsNZ’s alcohol sales data for 2020 around the end of February, which will allow a more useful comparison.

One positive point about the graphic, though: the areas of the bottle things on the graphic are actually proportional to the numbers they represent. For example, Tauranga City’s number  went up 115%, and the bottle (on my screen) is 715×274 pixels compared to 490×186 pixels for 2019, a factor of 2.15. You often don’t see that.

Team Ratings for Round 19

The basic method is described on my Department home page.
Here are the team ratings prior to this week’s games, along with the ratings at the start of the season.

Performance So Far

So far there have been 93 matches played, 61 of which were correctly predicted, a success rate of 65.6%.
Here are the predictions for last week’s games.

Predictions for Round 19

Here are the predictions for Round 19. The prediction is my estimated expected points difference with a positive margin being a win to the home team, and a negative margin a win to the away team.

Team Ratings for Postponed Matches

Another update because of a rescheduled match. And what is even weirder, another match listed as a Round 12 match. Can the season get any more messed up?

The basic method is described on my Department home page.
Here are the team ratings prior to this week’s games, along with the ratings at the start of the season.

Performance So Far

So far there have been 57 matches played, 38 of which were correctly predicted, a success rate of 66.7%.
Here are the predictions for last week’s games.

Predictions for Postponed Matches

Here are the predictions for Postponed Matches. The prediction is my estimated expected points difference with a positive margin being a win to the home team, and a negative margin a win to the away team.

Producing plausible but newsworthy predictions can be hard. It’s especially hard in areas where there are constraints that the predictions should really satisfy.

In the NZ$ Herald,

Professor Tim Congdon from the Institute for International Monetary Research said it is extremely unlikely that the excess money will be sucked out of the system again. He warned US inflation could explode to double-digit levels before the end of next year.

“The likelihood of US inflation exceeding 3 per cent is very, very high. In my view, it is more likely to be 5pc to 10pc when it peaks, probably before mid-2022,” he said. Such an outcome would turn the global financial system upside and have dramatic ramifications for asset prices of all kinds.

One constraint here is the that the US Treasury sells both ordinary bonds and inflation-protected bonds. The ordinary bonds pay ordinary interest; the inflation-protected ‘TIPS’ pay out interest plus inflation-adjustment.  So, if you expect inflation to be 2%, you will be willing to accept 2% lower interest on the inflation-protected bonds.   The gap between TIPS and ordinary bonds tells us about the market’s expectations of inflation

Here’s a graph from FRED, the wonderful economic stats website of the Federal Reserve Bank of St Louis, showing the five-year-average inflation rate at which you’d break even by buying TIPS.

The breakeven inflation rate has been rising, but it’s still only about 2%.  If inflation is going to average 2% over five years, it’s not going to hit 5% next year and 10% in 2022.

Now, the US financial markets aren’t infallible. You could tell a coherent story about how the US financial markets are wrong, and how there is some good reason to expect inflation that they don’t know about or don’t believe. Perhaps the M2 money supply indicator is much more important than anyone except this guy realises. I’m not an economist; I wouldn’t know.  But that’s not the story that’s being told.

There’s increasing consensus that the new B 1.1.7 variants^* of the Covid virus really are importantly more transmissible than the previous versions, which then raises the question of whether NZ-style lockdowns will still work.  You’ll see people trying to do simple arithmetic on the basic reproduction number, R, to work this out, but it’s more complicated than that.

In a very simple epidemic model, the number of people infected by a case is the number of people they come into contact with, multiplied by the probability that they infect each one.  You would think of interventions such as lockdown as affecting the first factor and differences in the virus as affecting the second. In this model, if a new variant has R about 1.3 times higher than the existing virus, for the same contact rate, it is increasing the infection probability by 30%.  If lockdown decreases R for the old virus from, say, 2.5 to 0.8, then it would decrease it from 2.5×1.3 to 0.8×1.3 for the new virus. That’s not a completely useless way to think, but it’s nowhere near good enough for government work.

The issue is that interventions don’t affect all transmission equally.  Imagine we had the impossible perfect lockdown, so you only came into contact with people in your bubble — any essential shopping was done by perfect zero-contact delivery or something. Obviously, the virus would not spread between bubbles; that’s exactly what we’re assuming. However, the effective reproduction number would still be greater than 1 initially, because the virus would spread within bubbles.  A combination of testing and isolation and just running out of susceptible people in affected bubbles would get the effective reproduction number down near zero eventually, but there would never have been any between-bubble spread.

In the real world, or even in New Zealand, we won’t have the impossible perfect lockdown. But back in April alert level 4 reduced between-bubble transmission a lot, and it will be combined with now knowing about masks and airborne transmission plus now having much more testing capacity than in April plus now having faster tracing than in April.  Is that enough? Well, you need a more sophisticated model to tell you, one that’s between our two simple extremes of no population structure and perfect lockdown.  We have those models, and what I’m hearing is that level 4 lockdown would work, but it’s not clear that level 3 would work.

There are other related questions where you need a model rather than simple intuition. For example, is the new variant more likely than the old ones to appear with no obvious contact to the border (as in August), making a lockdown more likely? Apparently, the answer is “yes, much more likely”.

It’s amazing how far you can get with unstructured differential-equation models for epidemics, and they are still valuable, but for some questions you need to model something closer to the real population, and it’s harder and takes longer and the results don’t necessarily have a simple intuitive explanation.

^* if you don’t say “China virus”, you probably shouldn’t be saying “UK strain”