Hold me now, I'm hoping that you can explain

[venta]

Yesterday, I pottered along to the Barbican to see what was going on at their Brainwaves Weekender. Among other things, I parked myself on the floor to watch Helen Arney present the unappealingly-titled I'm a Neuroscientist, Get Me Out Of Here!.

I'd have titled it "Listen to five smart people, who are neuroscientists, talk entertainingly and informatively about their subject". Which isn't as snappy, but there you go. Anyway, they were entertaining and informative.

Annoyingly, the bit that has really stuck with me is a silly little niggle. At the beginning of the 45 minute session, each of the tame neuroscientists was invited to present two "facts" for the audience to guess which was true and which was false. One of the neuroscientists, whose subject was research into Alzeimher's disease, stated "You are more likely to get Alzheimer's if you have a relative who suffers from it".

This was revealed to be her untrue "fact". She said that actually, the biggest risk factor for Alzheimer's was getting old. Most cases of Alzheimer's just occur spontaneously. Only 1 in 20 cases are hereditary, so people who worry a lot about developing it because they have relatives with dementia are worrying unnecessarily.

Now, I'm pretty sure that we're all at equal risk of getting old and developing a spontaneously-occuring disease. And for those who have a relative with Alzheimer's, there's that extra chance of it being a hereditary trait. So actually, by my reasoning, you genuinely are more likely to get Alzheimer's if you have a relative who suffers from it. Possibly not massively more likely, but more likely. Her untrue "fact" was actually a true fact.

Scientists who don't seem to understand stats and probability scare me :(

Comments

[bateleur.livejournal.com]

Scientists who don't seem to understand stats and probability scare me :(

I advise you not to investigate the field too closely. I wasn't even good at stats at uni and I have so far encountered so few scientists who understand stats properly that I can literally count them on the fingers of one hand. (Admittedly accompanied by many more for whom I don't know either way.)

The big problem is that scientists fall into two categories:

1) Those who don't work with stats. It's not too surprising they don't have much grasp of the finer points.
2) Those who do work with stats. And the trouble here is that they have been given rote methods to use, which they don't understand, but are prepared to argue for.

Result: When actual understanding is needed, they all come unstuck.

This is something of a rant topic for me, since I get very tired of explaining correlation vs causation to people who ought to know better. (Bonus points if they respond "Yes, of course I know the difference, but what are the chances that we have all this evidence when there's no causal link?") And don't even get me started on scientists saying "No, don't worry, we adjusted for that" when all they mean is that they did a linear regression. (Bonus points in this case for "No, it's not linear, it's multivariate!")

[venta.livejournal.com]

Ditto, I'm no statistician myself. And I can understand how scientists can get bogged down in the advanced stuff (although that doesn't mean it doesn't matter that they do!)

Just in this case, I didn't even feel it was a finer point!

[bateleur.livejournal.com]

Yeah, I see what you mean. I suppose it depends a bit on whether you think the problem is a sloppy use of English or an inability to recognise that 1 x 1.05 > 1.

[venta.livejournal.com]

Yes... it's possible that when she meant "having a relative with Alzheimer's means you are significantly more likely to develop it" as her untrue fact. Which is just sloppy English, but I do kind of expect scientists to care about that sort of detail!

(Actually, in fairness, I guess it's also possible that she was speaking in an unfamiliar environment, and delivered her sentence badly due to nerves.)

[sushidog.livejournal.com]

The thing is, when talking to non-scientists, scientists tend to have to dumb down a bit; the average person in the street doesn't use the word "significantly" to refer to statistical significance, after all.
But also, it's quite possible that, statistically speaking, what she said was true. If one in twenty cases of Alzheimers is hereditary, and having a relative with hereditary Alzheimers increases your risk by, say, 15%, then within the general population (including the other 19 in 20 whose Alzheimers is not hereditary), having a relative with Alzheimer's may not result in a significant increase in risk. For the relatives of the 1 in 20, it will, but you'd have to know the root cause of your relative's Alzheimers to increase your statistical (but not actual) risk.

Erm, I think. I've only had one coffee today though.

[venta.livejournal.com]

I'm quite happy with the explanation "I think you'll find it's a bit more complicated than that", and if someone had directly asked her about the risks I'd have let her off. As it was, having been warned to bring two "facts" with her, she brought one that seemed quite unsuited to its job.

For contrast, the two "facts" offered by the next neuroscientist were:

1. Rats laugh if you tickle their tummies (but at such a high pitch that you need special machinery to hear it)
2. In my lab, I have created synaesthetic mice.

... which seemed much better choices :)

[sushidog.livejournal.com]

I wasn't even good at stats at uni and I have so far encountered so few scientists who understand stats properly that I can literally count them on the fingers of one hand. (Admittedly accompanied by many more for whom I don't know either way.)
But what's your sample size and your p value?
:-P

[bateleur.livejournal.com]

:-D

Actually... hard to say in both cases, because there's a very large unbiassed population from which a biassed sample is being selected by virtue of their claims being drawn to my attention.