Posts from November 2013 (53)

I’ve written before about the Seralini research that involved feeding glyphosate and GM corn to rats. Now, Retraction Watch is reporting that the paper will be retracted.

This is a slightly unusual retraction: typically either the scientist has a horrible realisation that something went wrong (maybe their filters were affecting composition of their media) or the journal has a horrible realisation that something went wrong (maybe the images were Photoshopped or the patients didn’t actually exist).

The Seralini paper, though, is being retracted for being kinda pointless. The editors emphasise that they are not suggesting fraud, and write

A more in-depth look at the raw data revealed that no definitive conclusions can be reached with this small sample size regarding the role of either NK603 or glyphosate in regards to overall mortality or tumor incidence. Given the known high incidence of tumors in the Sprague-Dawley rat, normal variability cannot be excluded as the cause of the higher mortality and incidence observed in the treated groups.

Ultimately, the results presented (while not incorrect) are inconclusive, and therefore do not reach the threshold of publication for Food and Chemical Toxicology. 

They’re certainly right about that, but this is hardly a new finding. I’m not really happy about retraction of papers when it isn’t based on new information that wasn’t easily available at the time of review. Too many pointless and likely wrong papers are published, but this one is being retracted for being pointless, likely wrong, and controversial.

[Update: mass enthusiasm for the retraction is summarised by Peter Griffin]

The Mayor of London, Boris Johnson, on equality:

‘It is surely relevant to a conversation about equality that as many as 16 per cent of our species have an IQ below 85,’ before facetiously asking the audience if anyone had a low IQ.

Now, there are lots of problems with the over-intepretation of IQ as a predictor of economic success.  There are problems with the definition of IQ from factor analysis: positive correlations between test results will lead to an apparent single factor explaining them, even when this about as untrue as possible.

But that’s not the most egregious statistical point. IQ, because it’s a hypothetical reconstruction, not a directly measured thing, doesn’t have any intrinsic values.  When you design tests to estimate intelligence, you can choose to scale the mean and spread of the distribution to be whatever you like. The people who designed IQ chose to set the mean at 100 and the standard deviation at 15.  That doesn’t tell you anything about how variable intelligence is between people, it’s just a choice of scaling, part of the definition of IQ.

Boris’s observation that 16% of people have an IQ below 85, then, is absolutely content-free. It’s like saying our schools are bad because 50% are below the median, or that NZ rugby teams aren’t very good because only one of them holds the Ranfurly Shield. The fact that more than 1 in 7 people is below the 16th percentile is, actually, not at all relevant to a conversation about equality.

The Daily Mail strikes again, in today’s Herald story headlined “Gene explains why girls like sweet stuff more than boys”

The actual research did an experiment to look at differences in what snacks children chose. The difference they found was in fat consumption, not sugar, and, overall, girls actually ate a bit less than boys, not more. Here’s the graph from the paper

The potentially interesting finding was that the difference between kids with two versions of the gene was larger for girls than boys, suggesting that the gene may act differently in boys and girls.

From David Spiegelhalter, William Sutherland, and Mark Burgman, twenty (mostly statistical) tips for interpreting scientific findings

To this end, we suggest 20 concepts that should be part of the education of civil servants, politicians, policy advisers and journalists — and anyone else who may have to interact with science or scientists. Politicians with a healthy scepticism of scientific advocates might simply prefer to arm themselves with this critical set of knowledge.

A few of the tips, without their detailed explication:

• Differences and chance cause variation
• No measurement is exact
• Bigger is usually better for sample size
• Controls are important
• Beware the base-rate fallacy
• Feelings influence risk perception

From the Herald, under the headline “Could you have the binge drinking gene?”

It’s long been known that a penchant for alcohol may be in the genes, and scientists say they may now be a step closer to understanding why.

They have found that a faulty gene may cause binge drinking – and that mice with the mutation overwhelmingly prefer the taste of alcohol to water.

This story is from the Daily Mail again; the Herald’s own reporters write better medical science stories.

In fact, the research is looking at mice to see the effect of mutating one of the genes encoding something called the GABA-A receptor. There’s some genetic evidence that differences in this gene are related to alcohol dependence (not binge drinking, which isn’t the same thing) in humans, and the researchers are interested in how the effect might work. They say

Our understanding of the genetic and molecular basis of alcohol dependence is incomplete. Alcohol abuse has long been associated with facilitation of neurotransmission mediated by the brain’s major inhibitory transmitter, GABA, acting via GABAA receptors (GABAARs). Recently, a locus within human chromosome 4, containing GABAAR subunit genes… associated with alcohol dependence in humans. … However, the neurobiological basis by which genetic variation translates into alcohol abuse is largely unknown.

This research into how the genetic differences might work is interesting and has potential applications in treatments for addiction, but we know that variants in this particular gene predict almost nothing about alcohol dependence in humans. That’s typical in modern large-scale genetics: genetic variants common enough to study in large numbers of people usually have very small effects, and they are important because they provide tiny points of light illuminating the complex biological mechanisms of health and disease.

You can get another useful bit of context by searching on “binge drinking gene”

• Daily Mail, March 2011: It has long been believed that alcoholism runs in the family – now scientists have pinpointed why. They have identified a binge-drinking gene, offering new hope in combating the growing social problem, it was revealed today.
• Daily Mail, December 2012:  A newly discovered addiction gene could be fuelling teenage binge-drinking, research suggests. The mutant version of the RASGRF2 gene makes the brain more sensitive to habit-forming rewards such as alcohol, studies have shown.

We were clearly due for identifying the binge drinking gene again about now. But if you want to know if you are at risk of binge drinking, counting your drinks will be much more informative than measuring your genes.

Well, one of the ways.

Professor Terry Speed, an Australian statistician and geneticist, who gave public lectures in New Zealand this year as the Royal Society of New Zealand’s Distinguished Speaker, recently won the (Australian) Prime Minister’s Science Prize.

BBC News Magazine has an interview with him.

The US Food and Drug Administration has sent a letter to 23andme, one of the companies that will genotype you and provide lots of information from the sample, telling them to stop. It’s a tricky situation.

This product is a device within the meaning of section 201(h) of the FD&C Act, 21 U.S.C. 321(h), because it is intended for use in the diagnosis of disease or other conditions or in the cure, mitigation, treatment, or prevention of disease, or is intended to affect the structure or function of the body. For example, your company’s website at www.23andme.com/health  (most recently viewed on November 6, 2013) markets the PGS for providing “health reports on 254 diseases and conditions,” including categories such as “carrier status,” “health risks,” and “drug response,” and specifically as a “first step in prevention” that enables users to “take steps toward mitigating serious diseases” such as diabetes, coronary heart disease, and breast cancer. Most of the intended uses for PGS listed on your website, a list that has grown over time, are medical device uses under section 201(h) of the FD&C Act. Most of these uses have not been classified and thus require premarket approval or de novo classification, as FDA has explained to you on numerous occasions.

On the one hand, I can’t see any valid social interest in stopping people from knowing their genotypes if they want to. On the other hand, the FDA has a point about marketing.

They raise two isssues.  The first is that 23andme make lots of (fairly weakly supported) claims about the usefulness of the results in disease prevention. The second is that some of the genotype information is actually clinically relevant and that they have not demonstrated sufficient accuracy in their results. The first issue is essentially a misleading advertising problem; the second is a quality assurance problem.

There are two things that can go wrong with the clinically useful results. The first is simple error: the genotype assay could give the wrong result, or you could be given results from someone else’s sample. This should be low probability, but it’s important to know how low — 1 in a 1000 would definitely be too high.

The second issue is interpretation. Suppose you have a lot of family members with breast cancer, and you suspect a BRCA1 mutation is responsible. You might be relieved if you test negative, and think your risk isn’t especially high, but that’s only a reliable conclusion if your family’s cancer risk was actually due to a BRCA1 mutation, not to some other genetic risk factor.

Update: I should probably note that 23andme could fix what I think are the actual problems, but this wouldn’t necessarily satisify the FDA.  The FDA aren’t currently being unreasonable oppressive Luddite statist bureaucrats, but they’re probably happy to be to if that’s the option on offer.

• UK government “Behavioural Insights Team” wins a government innovation award for using randomisation to test more effective tax reminder letters.
• David Spiegelhalter is Unhappy with the UK Press Complaints Commission
• ‘Crowdsourcing‘ gives fast, detailed interpretation of satellite images from Philippines, to provide information about typhoon damage.
• XKCD
• 

(not that bloggers are necessarily any better)

Showing what can be done straightforwardly with online data, the site Fbomb.co (possibly NSFW) is a live map of tweets containing what the Broadcasting Standards Authority tells us is the 8th most unacceptable word for NZ.  Surprisingly, it was written by a Canadian.

Spotted this when reading Stuff’s homepage on my phone over the weekend:

This is an obvious case where rewording a statistic makes it incorrect. When you read the full article, it’s a lot clearer:

“The Department of Internal Affairs has already blocked 34 million attempts within New Zealand to access at least one of 582 child sex abuse sites blocked by government filters since 2010.

And the number is rising, with roughly a million requests a month within New Zealand being blocked in the first half of this year.

However, while the stream of blocked requests was huge, the actual number of people consuming child sex abuse material in New Zealand was small and most of them were being caught, he said.”

According to Statistics New Zealand, 2.8 million New Zealanders were connected to the internet in 2012.

Update: David Farrar and Whale Oil blogged about this too and the headline has since been changed.