Posts filed under General (3156)

On Ezra Klein’s blog at the Washington Post, Dylan Matthews has published one of the least convincing scary climate change graphs I’ve even seen.  The graph purports to show that weather fatalities are increasing in the USA, and this is  attributed to climate change.

The most obvious feature of the graph is the decrease in deaths due to lightning, which is partly because people spend less time outside and partly because of better forecasting.   This is also an excellent illustration of the fact that stacked plots don’t work very well for unordered categories.

There’s an increase in the light blue section in recent years, but it’s obvious that the data weren’t recorded far back into the past. Matthews says “Other” includes cold, winter/snow, rip current and miscellaneous weather events.  Part of the increase in that category reflects better NWS record-keeping, but the spike is striking, nonetheless.”  Deaths from cold and snow are certainly not a new phenomenon in the US, and on average are probably decreasing with climate change.

The 1995 Chicago heat wave is visible.  This genuinely was an event of the sort that is going to be more common with climate change, but again, the deaths from heat weren’t recorded back into the past.  Just off the graph to the left is 1936, when more than 5000 people died in the worst heat wave in US history, a number that would swamp the weather events of recent years.

Even the hurricane trend is misleading: most hurricanes now kill relatively few people because of better warnings and better buildings.  There’s a big spike for Katrina, again a major disaster, but only the third worst US hurricane in terms of deaths. The Galveston hurricane of 1900 killed more than four times as many people.

And even taking the numbers at face value, the population of the US now is 2.5 times what it was in 1940, on the left-hand side of the graph. The death rate from weather in the US is lower than it was in the past.  It’s also lower than it’s going to be in the future, a conclusion which, unfortunately, is grounded in a lot more than this graph.

The Guardian has a map (also with animated thingies) of all the hurricanes since 1851.

d

One of the most dramatic features is the semicircle of hurricanes in the Southern Hemisphere.  That’s because the ocean isn’t warm enough in the other half: see this graph from NOAA

Since the oceans are getting warmer, you might expect hurricanes to be getting worse.  It’s a bit more complicated than that, since climate change is warming the upper atmosphere more than the surface, which would tend to decrease hurricanes.  Still, NOAA expects more really intense hurricances and greater total rainfall from them, and they think it’s more likely than not that the overall frequency will go up.

After teaching an online course “Computing for Data Analysis” to 50,000 people through Coursera, Roger Peng has decided to do it all again on his blog next month.

Think of it as an Advent calendar for aspiring DIY statisticians.

A list of 20 from NetMagazine, via BetterPosters.

Yes, I did check that R was on the list before passing it on.

The Herald is enthusing about the new worlds of chocolate, and the boom in consumption in various parts of the world.

Japan aside, chocolate sales are also growing 30 per cent year on year in China, while in India – even though one in two Indians have never tasted chocolate – growth is 20 per cent, for a treat taken ultra sweet and milky.

The Quarterly Bulletin of Cocoa Statistics, as you would expect from the name, takes a less effervescent attitude. Their data show a steady increase of about 3% per year for the past decade, with modest year-to-year fluctuations in production smoothed out by stockpiles.

That’s Britomart and Newmarket. I just thought it was an interesting statistic, in the Herald’s coverage of the new “integrated” ticketing system (it works on trains, and also on trains.)

Stuff, this time:

Auckland teenagers had the highest number of convictions for drink-driving, with 1553 in the past two years.

Christchurch teenagers weren’t far behind, with 1383, while 793 Hamilton teenagers were convicted and 728 Wellingtonians.

That’s out of a national total of about 9700 per two years (so this could also have been in our ‘think of a number and double it’ series).  Based on total population it looks as though the drink-driving rate in Auckland is about half that in the rest of the country — and any journalist should know that Auckland is roughly 1/3 of the country and Wellington and ChCh are each very roughly 1/3 of Auckland.

With a few minutes at Stats New Zealand we can check the denominators and see if teenagers are distributed differently than people.  At the 2006 Census, the number of 15-19 year olds was 99,000 for the Auckland region, 33,000 for the Wellington region, and 36000 for the Canterbury region (ChCh is about 60% of Canterbury), out of 300,000 for the whole country.  These follow the overall population pretty accurately

So, the numbers are showing a news story that’s the opposite of what was actually reported: Auckland has dramatically lower rates of conviction for drink-driving among teenagers than the country as a whole and than the other major cities.

A standard criticism of science, especially in areas you don’t agree with, is to quote the ‘obvious’ findings that scientists have wasted time and money checking.

Quite often, though, the obvious doesn’t happen.  Stuff has a story about a randomized trial of intensive weight-loss propaganda in people with Type II diabetes: the trial is being stopped early, after ten years, because it isn’t working.  People in the intervention group are losing weight, and improving their control of blood sugar, but they are still getting heart attacks and strokes at the same rate as the control group.

An earlier trial had given a lot of hope by showing that quite small increases in exercise and reductions in weight could stop pre-diabetic levels of insulin resistance progressing to full-blown diabetes.

The current trial is led by Professor Rena Wing, of Brown University, who is hardly a heretic on the general weight-loss message.  Her research focuses on ways to maintain weight loss, and when the trial started she said

“Getting people to lose weight is talked about as the best treatment for both preventing diabetes in people who are overweight and for treating individuals who already have the disease,” Wing said. “If the long-term data from the study support these approaches, it will make it easier to go to patients, insurance companies and public-policy makers and say, ‘Yes, weight loss works over the long run.’”

We don’t have the full details of results, which are still be analysed. However, the trial succeeded in getting people to about the same amount of weight as in the earlier diabetes prevention trial, and to keep the weight off.  There were also improvement in glucose control, in sleep apnea, and in general quality of life, which might be sufficient motivation for weight loss, but not in heart disease and stroke.   This was a large, well-funded, well-designed study, and I would have bet on the results being the other way.

There won’t be any shortage of people who don’t want to believe the results, and it will be interesting to see if any of them can come up with some useful explanation for why the other differences between treatment and control groups don’t translate to reductions in heart disease or stroke.  [There will also be plenty who are happy to just say the results are wrong without needing anything as difficult as an explanation]

Unless there is something the researchers have completely overlooked, this trial adds to the long list of things that improve blood test results but don’t provide the expected health benefits in Type II diabetes.

An analysis reported in JAMA, and more accessibly by the LA Times shows that really dramatic beneficial effects of new treatments, even in good-quality clinical trials, are usually over-estimates.

They looked at 85,000 treatments for which there were multiple trials and binary (yes/no) outcomes measured.  In the 8000 treatments where the first trial showed a very large benefit, 90% of the time the subsequent trials showed a smaller benefit.

Also, when  large benefits were seen in well-replicated studies these were not typically for life or death issues: the researchers say

These include clinical benefits such as control of nocturnal enuresis with alarms in children, or symptomatic improvement with 5-aminosalicylic acid in ulcerative colitis; mostly mild or modest harms such as burning with capsaicin or local tenderness with the influenza vaccine; laboratory-determined response such as induction of hepatitis B surface antigen with hepatitis B vaccination; and control of acute pain with analgesics such as etoricoxib or diclofenac. As shown, all of these effects corresponded to very large absolute risk differences.

These findings aren’t actually surprising: if a treatment does very well in its first clinical trial it was probably lucky as well as effective.  Very large benefits in survival are very rare, so if you think you’re seeing one, you’re probably wrong .  It’s still useful for someone to put together all the data confirming what we expect, and it’s useful to remember when you see a headline about miracle cures.

The Herald has a headline “Young Kiwis among most savvy web users”, which would be justified if the phrase ‘most savvy’ was omitted.

The story comes from AVG, who sell internet security software, and so want you to be nervous online.  They have been conducting a set of surveys starting with kids and are now on to 18-25 year olds.  All we’re told about the survey methodology (by AVG, not by the Herald) is that the surveys were conducted by Research Now, so it’s not completely bogus but might be more or less representative or, um, ‘targeted’.

However, we do get told that there were only 4400 respondents in 11 countries.  Even with a completely random sample of 400 in each country, that gives a margin of error for comparing two countries of about 7 percentage points.   You can see a set of infographics from the survey here, and it’s pretty clear that a lot of the differences are compatible with random sampling error.

There certainly isn’t a general tendency for Kiwis to be more ‘web savvy’: while NZ has one of the lowest rates of admitting to accessing banned sites while at work, we have one of the highest rates of allowing colleagues to see everything that friends can see on Facebook.  The proportion for other questions, such as ‘friend’ing your boss, was intermediate.