Posts written by Thomas Lumley (2645)

Ben Goldacre on Twitter, about a “heart age test” that “has so far told eight in ten people that they are at higher risk of serious illness because their heart is prematurely aged.”

“that’s either a fumbled implementation turning coefficients into automated patient advice; or it’s a radical new NHS screening programme announced, oddly, only to individual members of the public, one by one, through an app”

Screening is always easy to sell, so it’s useful to remember a sound-bite: “Screening is the opposite of treatment: you come in healthy and go out sick”

From the New York Times

For the first time, we’ll publish our poll results and display them in real time, from start to finish, respondent by respondent. No media organization has ever tried something like this, and we hope to set a new standard of transparency. You’ll see the poll results at the same time we do. You’ll see our exact assumptions about who will turn out, where we’re calling and whether someone is picking up. You’ll see what the results might have been had we made different choices.

Business confidence surveys are in the news recently.  I thought it might be useful to look at why anyone even does them.

Obviously, there’s no great social interest in making businesses happy and confident. Businesses are people only as a legal fiction. And even the business-related confidence of business owners and executives isn’t an obvious target — if it impacts something that matters, such as GDP or employment, we’ll see it there, and if it doesn’t, well then.

What makes countries around the world measure business confidence is that it’s thought to be a leading indicator.  That’s a slightly unfortunate phrase; we don’t mean it in the “very good indicator, we have the best indicators” sense. “Leading indicator” is less metaphorical than that: it means something that goes up and down earlier than the real economy does, and so helps with economic forecasting.

It’s plausible that business confidence could be useful this way.  Your boss knows if you’re going to be laid off before you do. A business owner has more idea than anyone else whether she’ll respond to a drop in sales by cutting production or expanding marketing.  If we could extract this information from a representative sample of businesses, it should help predict the economic future. Business confidence figures also have the advantage of timeliness: it takes a lot longer to get unemployment or GDP data collected and published.

On the other hand, business owners and managers have opinions as well as knowledge. Part of ‘business confidence’ will reflect the general political reckons of business owners and managers, both their overall feeling about the party in power and their views on bike lanes or the Americas Cup. That part of business opinion isn’t likely to help with forecasting.

Whether a business confidence survey will be economically useful is an empirical question, to be answered by data. It’s likely to help if the survey has questions designed to measure plans and intentions rather than feelings, and it’s likely to help if there’s a well-defined business population and a representative sample of it. The United Nations Statistics Division has written a book on how and why to conduct business confidence surveys and other ‘economic tendency surveys’, with chapters on questionnaire design, sampling frame, weighting, and other nerdy issues. If you feel like running one, you should maybe read it.

• There’s been a lot of coverage of a recent paper claiming ‘no safe level’ of drinking.  The best rebuttal is probably this one from David Spiegelhalter.
• Harvard Business Review published “Want Less Biased Decisions? Use Algorithms”. There’s a rebuttal at fast.ai pointing out that the initial argument misses a lot of the scaling-up harms of algorithms.
• Mark Hanna looks at Official Information Act reporting delays

When government policies change in different ways in different parts of the country, you can try to estimate the impact of the changes by comparing trends in affected and unaffected areas.  This week I’ve seen two examples, one in the NZ media and one from a Twitter request.

Chris Knox, in the Herald, has both an interactive graph and a story about the changes in petrol prices, using data from Gaspy. There was a clear jump of 11.5c/litre when the 11.5c/litre tax was increased.  What’s not so clear is whether petrol companies have subsequently tried to spread the price rise to other parts of the country.  Averaged over the whole rest of the country the gap has narrowed from the 11.5c start, but that’s driven by the South Island.  In the Waikato and Bay of Plenty regions, which were previously the most similar to Auckland, there’s no sign of price spreading.

So, right now, it’s unclear whether Auckland is bearing the whole cost (as it should) or just most of it.  The Herald story makes this fairly clear, quoting the economist Sam Warburton, who has been looking at the data — but also quoting someone who doesn’t need data to be sure that the costs are being borne by the regions. A similar analysis was done by Peter Ellis, using Sam Warburton’s data from pricewatch.co.nz, with basically similar conclusions.

The other example from this week is a research paper looking at how fatal workplace injuries changed in US states that introduced medical cannabis laws. They found a decrease starting at the time of legislation and slowly increasing to about a 1/3rd reduction after five years, compared to states without legal medical cannabis.  The analysis looks sensible and the data on fatal workplace injuries are of high quality, but there’s still a bit of potential concern about publication bias — the researchers might have published a negative result, but I wouldn’t have heard about it and told you.  I’ve written before about an analysis that suggested banning single-use plastic bags caused deaths from food poisoning —  and while the analysis looked reasonable it was clear based on additional data that the conclusion wasn’t plausible.

On the cannabis issue what’s most unclear is the contribution of various possible reasons to the reduction (if it is real):

• medical cannabis works to treat, say, pain or nausea, so people are healthier and better rested and don’t get injured
• medical cannabis substitutes for legal or illegal opioid use, and while it might increase injury risk compared to nothing, it’s better than opioids
• medical cannabis ends up in recreational use and substitutes for alcohol, and while it might increase injury risk compared to nothing, it’s better than alcohol
• medical cannabis ends up in recreational use and substitutes for alcohol, and people are less likely to use at work.
• it actually makes you safer

To the extent that the second reason is important in the US, where there’s an opioid use epidemic, it won’t generalise to New Zealand. The other reasons would probably generalise.

This year’s census already been the subject of some news reports, some of which have quoted me.  I’ve now been appointed to an external data quality panel advising StatsNZ,  and I will be much more limited in what I can say in the future here or to journalists.  That’s partly just ordinary confidentiality to make the panel discussions work effectively, and partly because we’ll be seeing unreleased data whose disclosure is prohibited by the Statistics Act.

• From the Guardian: Mapping the world’s cities where you can live comfortably without heating or air conditioning reveals how few boast such ideal climates.  Among the cities that apparently don’t need heating or AC are Auckland and Melbourne.  Not entirely convincing.
• From the Herald:  Scientists accidentally discover pill which could stop weight gain. What they actually discovered was a way to genetically engineer mice not to gain weight.  There’s a drug used to treat glaucoma that affects the same biochemical mechanism that the genetic engineering did — but it’s used as eyedrops to treat glaucoma, so that’s not great evidence it would be safe and effective as a pill.
• Peter Ellis has been analysing petrol-price data after the Auckland tax was imposed: “after the spike caused by the tax, fuel prices in Auckland and in the rest of the country are converging somewhat (although much less than the full cost of the tax), and plausibly this is because of companies’ price adjustments down in Auckland and up elsewhere to spread the cost of the tax over a broader base.”
• I’ve written a program to produce a map of Wellington-area buses showing how many are late.  It’s not quite real-time; it shows roughly the past hour. The map is here (you can click on the markers); the (more-technical) blog post explaining how it works is here.

• Navigating word relationships from Wikipedia, spaceship-style:
• But just because social problems aren’t surprising doesn’t mean they aren’t worth mapping. Just because maps are similar doesn’t mean they’re actually the same. From Pro Publica, on maps of inequality in Chicago and similar topics
• Privacy issues with deep neural networks: they remember a lot more than they are asked to. 
• The Royal Statistical Society announces its Statistical Excellence in Journalism awards. Well deserved.

Q: Did you see that smartphones make you go blind?

A: Doesn’t it depend on what you do while you’re using them?

Q: No, the headline says Blue light from phone screens accelerates blindness, study finds.  And it goes on Light from digital devices triggers creation of toxic molecule in the retina that can cause macular degeneration

A: Yeah nah

Q: They didn’t study phone screens?

A: No

Q: Macular degeneration?

A: No

Q: Retinas?

A: Not as such, no.

Q: Ok, so was it mice? It’s always mice, isn’t it.

A: No, this was cells grown in a lab from standard cell lines then genetically engineered to produce the chemicals the eye uses to see blue light. Some of them were originally derived from mouse cells, and some were originally derived human cells — like the famous HeLa cell line.

Q:

A: You were going to mention that Thor movie, weren’t you?

Q: No, I’ve read The Immortal Life of Henrietta Lacks. Everyone should. But we nearly digress. If they didn’t use digital screens, what did they use? Sharks with lasers on their heads?

A: Close. No sharks. LED lasers.

Q: So why does this show phones make you go blind?

A: That isn’t what they were trying to do. They already believed blue light caused macular degeneration, and they were trying to find out how that works, on a molecular level. It’s clearer from their press release, though that still talks a lot about phones — the newspaper didn’t make this one up.

Q: Is it in the original research paper?

A: No, that’s written in High Biochemist. It’s got subheadings likeBLE-retinal induced PIP2 distortion is independent of GPCR-G protein activation

Q: How do phones even compare as a source of blue light, compared to other sources? Police car lights? University-themed webpages? The sky?

A: Even though your eyes squinch up in bright sunlight, the sun and the sky are going to be the big contributor

Q: Especially if your phone and computer switch to a tasteful sepia colour scheme at night, like they tend to nowadays.

A: So, maybe sunglasses.

• From the NY Times Upshot blog: a randomised trial finds little or no effect of providing a workplace wellness program — but within the trial, the people who ended up using the program were healthier. It would have looked effective without randomisation
• The US National Academy of Science joins the groups saying it’s a bad idea to add a last-minute citizenship question to the US census.
• “Raising the Bar” is a set of 20 talks in bar by Auckland academics, held on Tuesday 28th August. Some of them are sold out already, but the remaining ones include Andrew Chen on  privacy implications of modern  surveillance systems and Cather Simpson on useful and fun things she does with lasers.
• This graph appeared at vox.com,
  As Kieran Healy tweeted “that 1-year, ~15lb-per-person jump in vegetable fat consumption c. 2000 is weird, and a candidate for the rule of thumb that sudden jumps in a time series are often due to changes in measurement criteria”.  And so it was.  Official statistics agencies try not to change their definitions without a good reason, and put this sort of thing in footnotes. Which you need to check.