Stats Chat

This one is from Young Labour, and as with the previous examples it was retweeted  into my Twitter feed. It’s a useful contrast to the previous, since it’s misleading only in the standard ways you expect from political graphs; it’s not an outright distortion.

Good points: there is a reasonable data range, which is important for time-series comparisons. The vertical axis is labelled, and the labels match the quantities they are supposed to match.

From an economic point of view, there’s a problem.  The increase in debt under John Key’s government was largely due to the Great Recession.  Labour, one would hope, would also have run deficits during a major recession — that’s what you do — and there’s no way that the Great Recession can be blamed on the NZ National Party.  So, as is often the case, there are important confounding factors in the time series.  The numbers are right, but they don’t really support the obvious causal conclusion.

The graphical criticism: the vertical axis begins at 5% rather than at 0.  While for line graphs there are a range of  views about when zero needs to be included and how, there really aren’t for bar graphs. The area of the bar is supposed to convey information.

If you look up the data source, you find that it truncates the vertical axis in exactly the same way, which is presumably the reason (though hardly an excuse)

Redrawing, we can see what the graph should have looked like; I think the distorting effect is real but relatively small, especially for the last decade (which, again, not an excuse)

And, in fact, using the wonders of PowerPoint and image transparency, I found that the Young Labour graphic and the one from the TradingEconomics.com can be exactly superimposed — which I suppose argues against the vertical distortion being deliberate.

There’s just one more thing.

You will have noticed, perhaps, that my redrawn graphic has two red bars on the right-hand end, but Young Labour’s has three, with the last two at about the same height. If I hadn’t gone to the effort of superimposing the two versions I would have thought it was probably due to different divisions of the time axis or something.  But it’s not. An extra bar has been added. That’s a bit suspicious.

Or, it would be if it weren’t about the same height as the previous bar.  It turns out, if you go to the Treasury’s  Interim Financial Statements for the four months ending 2019-10-31, you find a figure of 20.2%, which seems to be the height for the last bar. On the one hand, that’s not precisely comparable to the rest of the graph; on the other hand, it’s probably the best that could be done, and shows a reasonable degree of care.

• Bloomberg News writes about drug prices.  Their point is expensive cancer drugs, but their graph also shows how drugs for heart disease have become much, much less expensive as they go off patent (the light-grey area)
  
• Census 2018 small-area data now available.  These go down to SA1, the smallest area unit for published results now.  About two-thirds of SA1s are a single meshblock, but some are multiple meshblocks to get the population size higher. Read the footnotes (always, but especially with the 2018 census)
• Jamie Whyte of Open Data Manchester has extended his inequality graphs to NZ (click to embiggen). Each blob shows a general electorate, and the width of the blob shows the proportion of people at various levels of the Index of Multiple Deprivation. So, for example, Manurewa is mostly high-deprivation, Selwyn (bottom right) is low. In the middle of the graphic, Hutt South and New Plymouth have opposite shapes: one has most people in the middle, the other has more people at the bottom and the top. There are two important caveats here compared to the UK version: this is based on everyone living in a general electorate regardless of which roll they are on, so it misses the Māori electoral roll and it misses the impact of List MPs.
  
• Twitter thread from Peter Aldhous (of Buzzfeed), on DNA testing and media coverage over the past decade
• NZ is still under-vaccinated for measles. At the moment there’s still a shortage of vaccine, but at some convenient point in the future you should make sure you’ve had your two shots.
• Figure.NZ is seven years old — 44,000 graphs of NZ data
• ‘Even if Kāinga Ora’s intentions were good, she was still concerned about “a creeping, increasing type of surveillance capacity being built … without mechanisms that allow communities to be really engaged in the process”‘.  Donna Cormack on the plans to put lots of sensors in state houses. There are obvious good uses of this sort of data, and bad uses, and the project would be better off with more information about how the data would be restricted to good uses (if that’s the plan).

Alex Brae, who writes The Spinoff’s  daily news summary  (which you should sign up for), is having “Say something nice about a journalist” week.

As a site that mostly specialises in not saying nice things about journalists, we should probably do the same. Here’s a selection: I’ve probably forgotten some.

• Kirsty Johnston has had a wide range of important stories this year, and is a good reason for subscribing to the Herald.
• The Herald data journalism team: Chris Knox and Keith Ng (Keith also did some non-data-journalism in Hong Kong)
• Jamie Morton, the Herald science reporter, has reported a lot of interesting and important science.
• Farah Hancock has written some very good environmental and health stories for newsroom:  most recently, on the claims that a secret lab had found huge amounts of 1080 in some dead rats when Landcare didn’t, but also on pharmacies pushing homeopathic non-remedies, and on the measles outbreak
• Eloise Gibson, also of newsroom, for her coverage of Sir Ray Avery and how some of his inventions are progressing and being evaluated, and for stories about radiata pine (big carbon sink) and about water-quality modelling
• Joel MacManus had a very good piece at Stuff about algorithmic inputs to parole and sentencing

There’s another badly-distorted bar chart being circulated, approved by Simon Bridges. It purports to show median rents in September 2017 and now.  I’m not actually going to reproduce the chart. Instead, look at this chart from interest.co.nz for the period covering the time since the 2011 election (the red line).

Over this time, rents have increased.  The trend is reasonably consistent, not far from 5.5%/year; it might be accelerating a bit.

The purple line shows the trend from September 2017 to now; you can see for yourself how it compares to the previous trend, and decide whether that’s good or bad, and who is to blame.  Presenting just the purple line, without the context, would clearly be a bit misleading.

The green line shows the trend from September 2017 to now more as the latest National Party bar chart depicts it: the bar corresponding to current rents is 39% longer than the bar corresponding to 2017 rents, which would be correct if the current median rent were $556.  Presenting the green line as the trend is more than a bit misleading.

(PS: I don’t want to get heavily into the political fact-checking business, but if the other parties are circulating graphs like these ones I’d be happy annoyed enough to write about them, too)

Multiple people have pointed me to this picture, authorised by Simon Bridges.

It’s designed to look like a barchart, but if it were a barchart the dimensions of the colored bits would have a closer quantitative relation to the numbers being displayed. I’m going to assume the numbers are correct, because that’s the sort of thing people are more careful about. The bars aren’t.

To start with, the red bar is wider than the blue bar, which is a well-known graphical exaggeration technique. But that’s not the real problem. The real problem is the heights.

On my laptop screen, I measured the red bar as 61% higher than the blue bar. But $2.23/L is only 16% higher than $1.91/L: that’s nearly a four-fold exaggeration of the difference

The dark red ‘Tax’ section of the red bar is 92% higher than the dark blue ‘Tax’ section of the blue bar, but $1.12 is only 29% higher than $0.87: more than a three-fold exaggeration.

If the heights were proportional to the numbers, as in a real barchart, it would look like this:

But wait, there’s more!  The blue bar is averaged over nine years of National government; the red bar is from last week.  That means the difference between the blue and red bars is partly inflation.  Over the time since September 2008, the RBNZ calculator says there has been 18% inflation.  We could say, as a rough approximation, that the data spread over nine years should get half the inflation applied.  Doing this would wipe out about half the difference between the blue and red bars, to give a comparison like this one:

There has been an increase in petrol prices due to tax increases under Labour. It’s not anywhere near as big as National’s graphic implies.

Q: Did you see that a harrowing study of 46,000 women shows hair dyes are heavily associated with cancer?

A: Harrowing?

Q: According to FastCompany.

A: That’s just the headline, though.

Q: “You know how you don’t see very old people with dyed hair? There may be a reason for that: Hair dye is heavily associated with cancer.”

A: That would be a heavy association.

Q: So is it true? There’s a link, even.

A: “We observed a 9% higher breast cancer risk for permanent dye use in all women but little to no associated risk for semipermanent or temporary dye use.”

Q: That’s… not as harrowing as I expected. Would that kill off a big fraction of hair-dye users?

A: Well, not very many men. For women the lifetime risk, which is the cumulative risk by the time you’re ‘very old’, is about 1 in 8. Increasing that by 9% would be about one extra breast cancer case per hundred hair-dye users.

Q:  So that’s not why very old people don’t use hair dye

A: There are probably other factors in play, yes.

Q: It’s just breast cancer, though. What about the effect on other cancers?

A: They didn’t look.  They mostly expected a risk on breast cancer, because of theories about ‘estrogen disruption’.

Q:  The story says the effects are bigger in Black women.

A: It looks like they probably are, though there’s a lot of uncertainty because Black women were only 10% of the study.

Q: Nothing about hair dye in men?

A: It was a study of sisters of people with breast cancer, so no men.

Q: Also too, no generalisability?

A: It’s not quite that bad, but, yes, it’s not clear how well this generalises. It could be that these women are more susceptible to hair dyes. Or not.

Q: What did previous research find?

A: It’s mixed. A big study in Black women in the 1990s didn’t find an association, but another current study did.

Q: So do we believe it?

A: We aren’t Black women who use hair dye. On several counts.

Q:

A: It certainly could be true, but the evidence isn’t overwhelming.

Team Ratings for Round 8

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, 38 of which were correctly predicted, a success rate of 77.6%.
Here are the predictions for last week’s games.

Predictions for Round 8

Here are the predictions for Round 8. 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 30 matches played, 23 of which were correctly predicted, a success rate of 76.7%.
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.

Newshub (and a lot of the British press) reported a couple of weeks ago “New blood test could detect breast cancer five years before symptoms“.

There’s a problem. Well, more than one problem.

First, the accuracy of the test is terrible.  It missed the majority of cancers and falsely diagnoses about twenty percent of the normal samples as having cancer.  There’s no way anyone would use a test like that.

Second, the story says “They estimate that, with a fully-funded development programme, the test might become available in the clinic in about four-to-five years.” If they had a working test, that might be true. But they don’t. So it isn’t.

And finally, all the breast cancer samples were taken from people who had already been diagnosed, so the idea that you’ll get early diagnosis this way is, at best, hopeful.