Posts from September 2016 (33)

If you’d like to comment on or debate any of this week’s Stat of the Week nominations, please do so below!

From the Guardian

The first international beauty contest judged by “machines” was supposed to use objective factors such as facial symmetry and wrinkles to identify the most attractive contestants. … But when the results came in, the creators were dismayed to see that there was a glaring factor linking the winners: the robots did not like people with dark skin.

The way statistical `supervised learning’ algorithms work is that you give the algorithm a lot of measurements, and a training set of ‘correctly’ classified examples. It tries to find the best way to get the ‘correct’ answers from the measurements, in an objective, unprejudiced way.

The algorithm doesn’t know or care about race. There’s nothing in the maths about skin colour or eye shape. But if it never saw a ‘beautiful’ label on anyone who looked like Zoë Kravitz, it’s not going to spontaneously generalise its idea of beauty.

We’re lucky people learn in a much more sophisticated way than computers, aren’t we?

• From Vox: Amazon is a giant purveyor of medical quackery

• From Joel Eastwood on Twitter: different approaches to balancing geographic accuracy and number of voters in maps of the US (the ‘one cow, one vote’ problem, as I like to call it)

• The American Statistical Association is running a competition for students on predicting the US election results

• From Matt Levine at Bloomberg View: “Measurement is sort of an evil genie: It grants your wishes, but it takes them just a bit too literally. “

• From 538: Is A 50-State Poll As Good As 50 State Polls?

• See evolution in action: antibiotic resistance bred in the lab, and filmed on an agar sheet

Q: Did you see “women are infinitely more likely than men to suffer from virtually every form of headache there is”?

A: What would that even mean?

Q: Well, it says “statistically’, so you should know.

A: No.

Q: Could it be like ‘literally’ and it just means ‘a lot’. Like ‘five times as much’ or something?

A: I suppose, but it still isn’t true.

Q: What does the story say?

A: “a review of 24 worldwide studies published in the journal Headache in 2011 found that while more than half of women polled (52 per cent) reported having a problem with headaches at the time of the research, only 37 per cent of men did.”

Q: And is that right?

A: My English teachers would say they’re missing commas after “studies” and “2011”, but I might not have the high ground in that kind of debate.

Q: Restricting to statistical pedantry, is it right?

A: Well, the first step would be to find the paper, which they don’t make all that easy.

Q: You mean they don’t link?

A: That’s part of it, as Mark Hanna eloquently complained on Twitter. But they also don’t give the author names or the month.

Q: Did you find it?

A: Yes.

Q: And?

A: Here’s the relevant graph

Q: Wait, what? More than 80% of men and women in North America have headaches right now? But only about 50% of people in total?

A: Looking up the reference that little ¹ points to, it seems ‘current’ headache really means ‘in the past three months’, or even ‘in the past year’, depending on the study.

Q: And the total being much less than either men or women separately?

A: ¯\_(ツ)_/¯   If I had to guess, maybe the studies that separated out men and women used a longer time period

Q: And is the time period why there are more headaches in North America than Europe?

A: Could be. Or the quality of the cheese. Or the fact that they’re in an election campaign, like, half the time.

Q: I’m sensing you don’t like this graph.

A: There are others

Q:  Right. 7.4% in women is less than 6.4% in men. ಠ_ಠ

A: Actually, the lines are ok, it’s just that the numbers are in the wrong places. If they’d written the numbers on the y-axis like we’ve been doing for centuries, they’d be ok.

Q: Ok. But we nearly digress. You’re saying ‘infinitely more” means something in the range 1.5 times to three or four times more?

A: And that some of the comparisons are a bit dodgy.

Q: It seems to have taken more work than necessary to establish that.

A: It sure has.

Q: Would you like to quote some of Mark Hanna’s tweets on linking?

A: “Listen up, reporters & editors. Every day I see you publish articles using “studies say” that make it hard or impossible to find the studies. You should be improving public understanding of science, but instead you are training your audience to believe in “studies say”. 

From the Herald (from the Daily Telegraph)

A revolutionary blood test, which acts like a smoke detector to spot cancer up to 10 years before symptoms appear, could be available within five years.

It looks like this is genuinely impressive research, and deserves its spot at the British Science Festival, but it’s harder to assess the realism of the claims. What do we actually know now? Well, less than we should, because the claim is based on a press release and interviews about unpublished research. However, earlier research by the same group is available, with a bit of detective work.

In a conference abstract published in February, they report what they were trying to do: measure mutations in a specific gene in red blood cells. As the Herald story says:

Scientists at Swansea University have discovered that mutations occur in red blood cells way before any signs of cancer are evident.

But it’s more than that. Mutations in red blood cells occur before cancer even exists — another reason this test is potentially useful is for studying low levels of mutations that would have a very low chance of leading to cancer, so that the risk of realistic doses of potential carcinogens can be assessed. Since the test picks up mutations in the absence of cancer, there’s justification for worrying about false positives.

In the February abstract they had used the test on 121 people, and were claiming five-times-higher mutation rates in people with cancer than healthy people. Now they have 300 people and are claiming ten-times-higher rates — one possible explanation is that they’ve made the test more selective somehow and so are picking up fewer uninteresting mutations.  In any case, progress. The earlier data didn’t look as if it could support a useful test; the new data might be able to.

We still don’t know about the false-positive rate — with 300 people tested, it’s too early to say.  The false-positive rate is important for another reason, though.  The Independent has another story, quoting the lead researcher

Professor Jenkins said they needed to find evidence that it would work for other cancers, but added it would be hard to imagine that it would not.

“It would be really difficult to think why it would only affect oesophageal cancer,” he said.

As he says, it’s hard to think why oesophageal cancer would be unique — though you might expect some cancers to be different. For example, in cervical cancer, the mutations are caused by a virus that only infects certain cell types, so it might not cause mutations that show up in red blood cells.  But if we assume many cancers show the same pattern of red blood cell mutations, assessing the usefulness of the test gets more difficult. Suppose a positive result means you’re going to get some type of cancer over the next ten years, but it could be almost any type. What would the next step be?

There’s another important point in the first sentence of the Herald story. It contains two numbers. One is bigger than the other.  As far as I can tell, this test is done on freshly-collected blood, and hasn’t been done on large numbers of healthy people yet. If the test is available within five years, it will, at best only come with reliable information for five years after testing.

Team Ratings for Finals Week 1

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 192 matches played, 123 of which were correctly predicted, a success rate of 64.1%.
Here are the predictions for last week’s games.

Predictions for Finals Week 1

Here are the predictions for Finals Week 1. 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 4

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 22 matches played, 19 of which were correctly predicted, a success rate of 86.4%.
Here are the predictions for last week’s games.

Predictions for Round 4

Here are the predictions for Round 4. 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 6

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 19 matches played, 12 of which were correctly predicted, a success rate of 63.2%.
Here are the predictions for last week’s games.

Predictions for Round 6

Here are the predictions for Round 6. 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.

• Facebook’s step towards predictive models instead of people for its news editing didn’t work very well last week

• Why more interesting local restaurants tend not to show up on internet maps.

• A preview of Cathy O’Neil’s book about data science and its potential dangers, coming out tomorrow.

• From Flowing Data, how the frequency of different household structures has changed in the US since 1970

• A map of the world’s languages — showing the difficulties in definition, since all the Chinese languages are lumped in together when probably equally distinctive languages from different countries are given separately.
  

Each week, we would like to invite readers of Stats Chat to submit nominations for our Stat of the Week competition and be in with the chance to win an iTunes voucher.

Here’s how it works:

• Anyone may add a comment on this post to nominate their Stat of the Week candidate before midday Friday September 9 2016.
• Statistics can be bad, exemplary or fascinating.
• The statistic must be in the NZ media during the period of September 3 – 9 2016 inclusive.
• Quote the statistic, when and where it was published and tell us why it should be our Stat of the Week.

Next Monday at midday we’ll announce the winner of this week’s Stat of the Week competition, and start a new one.

(more…)