Lies, damn lies and statistics

[MrsMerryHenry]

I've been thinking about how powerful statistics can be since there are so many stats flying around about this, that and the other to do with the dreaded Swine Flu. So here's my 2 pence' worth. I call it When Good Stats Go Bad:

1 (a) If you are pregnant, you are more likely to have mild SF symptoms (NHS website) than severe ones.
1 (b) If you are pregnant, you are 10 times more likely to have complications (BBC news last week).

• Am I wrong in thinking that the NHS and BBC are contradicting each other? Actually I've just double-checked the NHS page - read under the first sub-heading and it appears to be sort of saying both!

• 200 people in the UK have died from Swine Flu. (But how many have not died...? Oh, look - they don't know ).

• 4 pregnant women in my local hospital are currently hospitalised with Swine Flu (but how many have had it with no complications? Oh, look - they don't know ).

It's not that I'm disputing the figures given, what I want is a BALANCED PERSPECTIVE (or a clear one in the case of the NHS website). Which clearly is not on the cards.

Anyway, I'm primed now to jump on any stats given about any subject under the sun - anyone have any other goodies, whether SF-related or not?

Thankfully I changed it last Xmas - rather clairvoyant of me, don't you think?

I've moved on from Thierry, he's gorgeous but he's just so young (tugs cheek skin behing ears).

Lenin - at first I thought you meant 'PR' as in 'Public Relations'!!! Rofl! I was trying to work out how that would be, for the electorate to have its own PR agency!

I get it now. But still - how would that change the short-termism?

Why don't we get taught practical statistics in school? Its not just the general public and the media who have a poor grip on statistics but professionals who really need to understand probability and risks - doctors and the legal profession in particular.

The trouble is, correct interpretation of statistics is often not intuitively obvious. There was an interesting piece in New Scientist recently.

One of the questions posed was this, which you might like to ponder before looking at the correct analysis:

You've just been diagnosed with a rare condition which afflicts 1 in 10,000. The test is 99% certain. Hope or despair?

Umm... I'd think... that if you've already been diagnosed you ought to ignore the first figure and pay attention to the reliability of the test. And so despair - unless the 'rare condition' is one of eternal youth/ bigger brain/ something equally plausible!

I did a whole year on stats at uni. It was the only exam I ever failed. But then that's because I found the teacher nice but dull and incomprehensible, and then applied the tried-and-tested ostrich method to dealing with the problem. Ohhh I learned a lesson of life the day I saw my D grade...

Anyway I agree - it's one of the essential life subjects and you'd probably only need to cover it over a period of a term or so, you wouldn't need to do a whole qualification to cover the necessaries.

Will have a look at that New Sci article though - and possibly weep!

Have you had the jab yet Lenin?

Can anyone help me with these stats?

"Asthma was the most common underlying medical condition that we saw," says Seema Jain, M.D., a medical epidemiologist at the Centers for Disease Control and Prevention. Jain's team analyzed 272 people who were hospitalized with swine flu for more than 24 hours between April and June.
Twenty-five percent were admitted to the intensive care unit and 7 percent died. Overall, 29 percent of children and 27 percent of in adults in the study had asthma, while only about 8 percent of the U.S. population has asthma, according to Jain.
Richard Gower, M.D., the president of the American College of Allergy, Asthma and Immunology, says about 25 million people in the United States have the chronic lung condition, including 8 million children.''

I can't do that one Grimma! It can't be despair or you wouldn't be asking, but why?

Grimma, I don't understand what your question is asking: 1:10 000 (?incidence or prevalence ) ie likelyhood of illness occuring in population and how reliable the test is, really have nothing to do with each other??

Go on, give us the answer, go on, go on, go on, you know you want to .

pofaced, you stats mean that if you had asthma you were more likely to be seriously ill with Swine flu than if you did not have asthma as an underlying chronic health condition (29 percent of those who were v ill with swine flu in the study had asthma whereas only 8 percent of the general population have it).

This things make my brain hurt though...

Pofaced, I think your stats are incomplete - they need to tell us the following:

"7% died" - is this 7% of the ICU patients or (as I assume) of the overall sample of 272?

29% of children and 27% of adults had asthma - it would be useful to know how many children and adults there were in the whole sample. However without this information it's clear that they're saying asthmatics are more vulnerable to SF than non-asthmatics.

While we're on the subject of SF, has any of you been offered a seasonal flu jab while preg before 2009? I was preg in winter 2006, and at the time was classed as asthmatic, so apparently doubly vulnerable. Not one individual gave a whiff of a hint of a suggestion that I might want to consider a jab. Sorry, had to get that off my chest, this SF stuff is getting right up my nose.

Lenin - how's your arm?

Pacific - my brain's throbbing too, but ohhhhhhhhhh how good it will feel if we get full marks!

That New Scientist article is sobering, with the mention of poor Sally Clark at the end. Killed by the judge and jury's inability to understand statistics indeed (and a swollen-headed 'expert' who got away with just making a sum up off the top of his head without bothering to check with any statistician).

Journalists often do make a hash of maths (I am one) because, amongst other reasons

A. We live in a society full of people who don't do maths or lack confidence in maths, and journalists are not exempt. It's fine to 'confess' in a slightly embarrassed manner that you are rubbish at maths in a way that simply would not be acceptable if you were talking about reading or writing.

B. They rely on interviewees who should know what they are talking about and don't themselves have the mathematical ability to spot when AN Other expert who has a paper published in a peer reviewed journal or who is a spokesperson for a leading charity is not giving the full context.

C. They are pushed for time. Really really pushed for time. Try writing 1,500 words on a subject you have never come across before, in an entertaining and interesting manner, in just a couple of hours, including finding and tracking down interviewees, while doing a dozen other things - that's what life's like for general reporters on a national. (Specialist correspondents should have some idea of the context so do rather better.)

D. News editors want stories that are attention-grabbing. And if you hum and haw and don't sell the most attention-grabbing angle, even if it's not quite right, your news editor may well not be a happy bunny. Especially if your rival titles go with a stronger angle.

E. Sources have their own agenda - government wants to get out its healthy eating message or whatever and will tie up all the info that suggests something IS an huge issue.

Or the House of Commons Health Select Committee spends months taking evidence on obesity and decides the angle most likely to attract notice from the media is a 3yo who died from obesity because they 'choked on their own fat'. The media believes what this authoritative source says and repeats it.

Sadly they got it wrong and the poor kid had a very real medical condition - MPs apologise but by then the damage is done and in the public mind, a child died of choking on their own fat. This myth is still occasionally quoted on MN.

The answer is in that link - with graphics, which often help get your head rounds stats.

In short, the answer is 'hope', because with those figures there will be a lot more false positives than actually ill people.

But how do you get PR and maintain the link between the MP and their constituency? That's vital. The party list system we have for Europe is far worse in terms of ensuring you get party lickspittles.

Yes, but the constituency link is vital, or the MP has no reason to give a toss about the voters.

And the party list is horrible - you don't vote for a person, you vote for a party, so any old creep who always obeys the party line whatever the issue gets in ahead of someone with an independent mind.

Grimma you just had me seriously concerned as, after 3 years and a first class honours in statistics, I couldn't follow the logic of your NS example unless it was a screening test offered to everybody, not just people who say have some other symptoms which mean it is more likely that they've got the rare condition. But have now read the extract and am happy again. Phew!

I didn't realise rare condition = lots of false positives. I'd always find out the rate of false positives/negatives when taking a test anyway.

Grimma, still do not understand your example question. And I have now read the New Scientist artikle, still clueless !

Incidence is a population based number, accuracy of test applies to you as an individual. It still does not mean that this is a common illness of course. Or are you asking how likely you are to have the illness before being tested??

Sorry to be a bit tedious, but now I really really want to know

And yes, take party politics out of healthcare planning altogether . Like that's ever going to happen...

...accuracy of test applies to you as an individual

I think thats the fallacy that gets us despairing. Its partly the fault of statistical language too, and our egocentricity. 99% certain doesn't mean its 99% certain you have the disease. It means that if they test 100 people, they'll get the right answer for 99% of them - you're just one of that herd. With the low incidence of the disease, if you happen to be the 1% in error, its much more likely false positive than not. (I think thats how it goes ... I'm not a statistician, just a scientist who sometimes has to try to get a grip on stats).

Anyhow, I think this example served its intended purpose of demonstrating the counter-intuitivity of statistics!

Just to go off at a tangent, I'm sure you'll all be glad to know the government hasn't forgotten avian flu amid the swine. There is a govt funded program to check for incidence in the wild bird population. This is rather charmingly accomplished with the help of a foxy looking dog called Tolly and 19th century duck decoy to catch samples, and some nice WWT gentlemen who swab the wildfowl fore and aft. Highly recommended entertainment for a wet weekend in Gloucestershire . All the birds so far have been flu-free.

thanks Grimma but I now have visions of the government missing an avian flu outbreak because hits Scotland first, not Gloucester!

The way I find it easiest to make sense of Grimma's NS example is to imagine you used this 99% accurate test to screen 10,000 average people for the rare illness. The 1 in 10,000 tells us that there should be only 1 person with the disease. But as the test will give a false positive for every 100 people tested, in our group of 10,000 people we'll get 100 false positives. So in all, you should get 101 positives, and only 1 of them will be a true positive. I.e. less than 1% of the people with a positive test result actually have the condition.

Does that help (and please don't tell me if I've got the stats fundamentally wrong!)?

How's about this then for statistical incomprehensibility:

(from the Health Protection Agency SF report Nov 20th 2009)

"Asymptomatic infection is a well-recognised feature of seasonal influenza. However, the
proportion of those infected with the pandemic (H1N1) virus who have a mild illness or
are asymptomatic has not been well-characterised. Based on serological studies of the
boarding school outbreak, subclinical infection occurred in about one third of those
without symptoms (HPA, unpublished data)."

Am I wrong, or is there some important figure missing from here (as far as I can see it's not mentioned anywhere else in the report)?