Stats Chat

Team Ratings for Week 9

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 49 matches played, 36 of which were correctly predicted, a success rate of 73.5%.
Here are the predictions for last week’s games.

Predictions for Week 9

Here are the predictions for Week 9. 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 Round 14

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 87 matches played, 66 of which were correctly predicted, a success rate of 75.9%.
Here are the predictions for last week’s games.

Predictions for Round 14

Here are the predictions for Round 14. 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 Round 12

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 65 matches played, 32 of which were correctly predicted, a success rate of 49.2%.
Here are the predictions for last week’s games.

Predictions for Round 12

Here are the predictions for Round 12. 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.

• Arithmetic fail by Bloomberg Australia: 126 cases over two days is not ‘more than double’ 62 cases in one day.
  
  They’ve since edited to “N.Z. Adds 126 Cases in 2 Days…New Zealand’s daily toll of new local infections has risen from the 62 reported on Dec. 24.”, which is at least arithmetically correct, since 2×62 is 124, and 126 is larger than 124.  Interestingly, the story now says “(Updates with state case numbers in third paragraph)” but doesn’t mention the correction to the maths

• The Washington Post reports on using convalescent plasma — antibodies from people who’ve recovered — to treat Covid. A new trial has positive results, but the story seriously underplays the previous trials with negative results. The story emphasizes that we’re short of treatments for Omicron so a new treatment would be more valuable, but that’s only true if it works, which is what’s in doubt
• Cruise ships are getting Covid outbreaks. Ok, yes, I’m shocked too.  More seriously, the problem is numbers. There were 3500 people on the ship. If each one is 99.95% sure to be Covid-free, that still comes to more than one expected case. At the sort of Covid prevalences the US has now, symptom screens and pre-departure tests aren’t good enough to get a high probability of a safe cruise.

There are two main sorts of Covid tests available: PCR tests and rapid antigen tests.¹

PCR tests have essentially no false positives and ok-ish false negative rates.  They’re processed in lab; overnight if you’re lucky and the lab isn’t overworked. Rapid antigen tests are less accurate: they have rare but non-negligible false positives and they turn positive later in the infection process than PCR tests.  As it says on the tin, they’re rapid: you don’t need a lab and can get results in minutes. The two test types are useful for different purposes.

In New Zealand until recently, Covid was rare and occurred in limited clusters. In this setting, when the public health system is doing zero-tolerance control, there’s not much use for rapid antigen tests².  If you are symptomatic or a contact you should isolate and get PCR tests; if you are asymptomatic and not a contact then you’re very unlikely to have Covid and any positive test is likely to be a false positive.  Anonymous at-home testing allows someone who has symptoms or is a contact to test positive and keep it quiet — maybe stay away from Nana’s 90th birthday party or drinks at the pub, but still go to work and go to the supermarket. The public health system would like this not to happen.

When Covid is everywhere, as in the US and the UK, quite a lot of people will be exposed and not know about it, so asymptomatic community testing is no longer useless. The problem with PCR is that it’s slow, especially when the labs get overworked: you won’t get the result until at least tomorrow and perhaps a few days later.  Rapid antigen tests are very valuable because they are rapid — they tell you, with imperfect but useful accuracy, whether you are infectious right now.  If you’re planning an in-person party or meeting or movie or flight or date, that’s what you want to know.  In the UK and Europe, the ‘lateral flow test’ type of rapid antigen test have been very useful; they miss some infections, but they catch quite a lot.  In the US they would have been useful, but they’re expensive and availability has been limited. The US government has just moved on making the tests a little more available — they’ve bought half a billion tests, nearly two for each US adult.

In New Zealand at the moment the situation is more complicated.  We’re on the boundary between zero Covid and low-level suppression.  The government is still trying to keep control of test results, which has clear benefits in contact tracing and elimination, but removes the ability for everyone to use rapid tests to reduce their individual risk of spreading Covid.  Whether you think the government is making the right decision here depends a lot on how much you trust the public health system, and on how much you trust other people.

¹ yes, and a whole bunch of other minor options
² except perhaps for border workers or customer service people during an outbreak

There’s a piece in Stuff about a startup using NFTs to do carbon capture.  I’m not going to get into a general discussion of NFTs here¹.  What’s StatsChat-relevant about the story is the general principle that when you have two numbers you should do something with them.

The company in question has a current product and a planned product.  The current product is a subscription that offsets more carbon than you use on Instagram, for $5/month.  The planned product is a set of $1000 NFTs that will offset 1 tonne more carbon than they cost to produce, and will potentially generate royalties on future resales to offset more emissions.

Looking at the Instagram offset subscription, you’re getting about 6kg of offset per dollar, so about $160/ tonne.  The NZ emissions trading scheme price is about $70/tonne. You might be worried about the quality and reliability of the NZ offsets — I haven’t looked into this in any detail, and you probably haven’t either — so you might be willing to pay more for offsets you trusted more and which had the potential to develop new technologies.   Or you could buy for US$15/tonne directly from Tradewater, one of the companies used by Cool Points Club, whose approach is to prevent emissions of used refrigerant gasses by incinerating them. The monthly price does include GST and the cost of running the system; I don’t think it’s great value, but it’s priced transparently and it’s probably capturing money that wouldn’t otherwise be spent on carbon offsets.

The NFTs initially cost about $1000 for one tonne of offset. That’s very expensive.  You can buy 14 tonnes of generic NZ offset for that much, or 66 tonnes directly from Tradewater (give or take any GST liability).  If these are going to be worthwhile, nearly all of the value will have to be in the NFT, not in the offset.

¹ and since I moderate the comments, neither are you

• Pfizer’s vaccine trial in kids 2-5 years old wasn’t successful: the dose (1/10th adult dose, 30% of 5-12yr dose) seems to be too low. They will try a three-dose series.  It did work in 6-24 months kidlets.  Pfizer is also testing third-dose boosters for all child age ranges
• “Why trust and transparency are vital in a pandemic” from the UK Office of Statistics Regulation. They note “It will not always be possible to publish information before it is used publicly. In these cases, it is important that data are published in an accessible form as soon as possible after they have been used, with the context provided and strengths and limitations made clear. “
• Lee Wilkinson, a pioneer in statistical computing and graphics, passed away on December 10.  Among other influential contributions, he developed the statistical package ‘Systat’ and wrote ‘The Grammar of Graphics’
• Florida is a counterexample to correlations between Covid vaccination and politics in the US. That seems to be partly because their data are wrong. “People age 18 and over in the 33122 area code had more than a 2,700% vaccination rate, according to the data….’That’s the airport,’ Gelber said.”
• Via Jenny Nicholls on Twitter, a Washington Post story on injuries from bouncy castles. The story quotes the number of injuries as 82,203 from 2008 to 2013 and as one every 46 minutes in 2013.  You might think about whether these are compatible and which sounds bigger.  It also works out as about 0.2% of the roughly 8 million unintentional injuries in kids leading to emergency department visits. Which I think is more than I’d expect

There’s still discussion about whether the Omicron covid variant is milder than Delta. We don’t really know yet, but this post is about why that’s not even the question.

First, Omicron is still scary: people do end up in hospital; people do die; even a ‘mild’ case can still really suck; and we have literally no idea what proportion of people will get Long Covid. If it’s milder, it’s still very much in the Do Not Want category.

Second, we do know that the proportion of people who get hospitalised will probably be lower than with Delta, and that isn’t the answer to ‘mild or not?’ The primary facts about Omicron are that (a) the vaccine is definitely much less effective at preventing infection, but (b) the vaccine is probably still somewhat effective at preventing severe disease.   Suppose, to give us something to work with, an Omicron infection was exactly as likely to cause hospitalisation for an infection in vaccinated individual as Delta, and was exactly as likely to cause hospitalisation for an infection in an unvaccinated individual as Delta.

If you (Dear Reader) are vaccinated or otherwise immune, you’re more likely to be hospitalised by Omicron because you’re more likely to be infected. The vaccine protection is less, even with a third dose, and the prevalence will be higher so you’re more likely to be exposed. If you aren’t vaccinated, you’re more likely to be hospitalised by Omicron because you’re more likely to be infected: the communal vaccine protection is  less so the prevalence will be higher and you’re more likely to be exposed.  So, in that sense Omicron is worse: you are more likely to get sick, more likely to be hospitalised, probably more likely to die than if Omicron hadn’t come along.

On the other hand, the fraction of cases who end up in hospital is likely to be lower than we were seeing with Delta.  That’s because we will have a larger fraction of cases in vaccinated people, and these are less likely to end up in hospital.  The number in hospital will go up, but by a smaller multiple than the total number of infections.

So, if the question about a milder variant is “will the fraction of people with serious disease go down?” the answer is probably “yes”. If the question is “will the number of people with serious disease go down?” the answer is probably “no”.  If the question is “should I relax because it’s not serious?”, the answer is “holy fuck no”.

• From Ars Technica, Report reveals which sealed NES games are the rarest of the rare. This is relevant because most of the story is about selection bias “Wata’s sealed-NES report, for instance, only shows one graded, sealed copy of Jeopardy!, a game that most collector’s regard as pretty common.
  This disparity could be because sealed copies of Jeopardy! happen to be much rarer than open boxes or loose carts. Or it could simply be that almost no one has bothered going through the time, expense, and hassle of going to Wata for a professional grade on a relatively ignorable game like Jeopardy!.“
• Phillip Bump, of the Washington Post, is starting a newsletter “How to read this chart“
• NZ police release an independent report on facial recognition technology
• The police, and various other agencies, have asked the Ministry of Health for data from Covid contact tracing. They were (correctly) turned down.
• According to UK supermarket chain Tesco, via Wales Online,  33% of people in London and 39% of 18-24 year olds in the UK celebrate Thanksgiving. I’m reasonably sure this isn’t true, but it doesn’t seem possible to find out any more about where they got the numbers.
• NZ Herald, Nov 22 “Auckland CBD sinking into anarchy and resembling 1980s New York, city leaders told. Newsroom, Dec 6, “yeah nah”
  

This isn’t precisely statistics in the media, but it’s research about the sort of stories we discuss a lot.  A new research paper in Nature looks at three estimates over time of the proportion of people vaccinated in the US. Two of these were based on large self-selected sets of respondents, the third was much smaller but had an attempt at random sampling.

What’s interesting about this is that we know the truth, pretty well.  US States kept track of vaccinations and the CDC collated the data.  There aren’t many examples where we have that sort of ground truth — the closest we come is elections, and even then we only get the truth for one point in time.

Here’s a graph from the research paper:

The two ‘big data’ estimates were much more precise than the smaller survey, but also much more biased: they were confidently wrong, where the small survey was pretty much right.  For some reason (and it’s not hard to think of possibilities) people who were vaccinated were more likely to respond in the big unselected data sets.

This is a general ‘big data’ phenomenon: when you get more data it tends to be of lower quality.  It’s very hard to overcome the data quality problem, so you will often get worse answers, but your estimation procedure will tell you they are much better. The ‘margin of error’ on the 75,000-person Census Household Pulse is much smaller than on the Axios-Ipsos survey, but the actual error is much larger. If you’ve seen lots of 1000-person surveys reported in the media and wondered why they aren’t bigger, this is the reason.  It’s not that you can’t do a 10,000-person survey; it’s that it needs to have much higher data quality than a 1000-person survey to be worth doing.

Now, ‘big data’ isn’t useless. It can be possible, with detailed enough data on a large number of people, to get around the data quality problems.  The polling company YouGov has had some success with large unselected samples and reweighting them to match the population. But that’s only possible where you have good data for the sample and the  population — the Nature paper hypothesises that collecting political affiliation and rurality might have helped, but the ‘big data’ surveys didn’t.

I didn’t have anything to do with this research, but one of my research areas is combining big databases and small samples in medical research: in the small sample you can afford to get accurate data and then you can use the big database to get extra precision.