Lies, damn lies, misrepresentation, misinterpretation and STATISTICS

[StealthPolarBear]

A thread for anyone who is infuriated by and/or enjoys reading about this type of stuff. I love it, but annoyingly at the moment I can't think of any examples, other than dull work-related ones. For example, if you're looking at something as a proportion of the whole, then you can't consider one thing in isolation. For example, as Trills mentioned on another thread, let's assume heart disease is the biggest killer of adults (which I believe it is). Let's assume 10% of deaths to adults in 1950 were of heart disease, compared to 42% now. Shocking rise? Probably not.

Plus I will attempt to explain the Monty Hall (think that's the name) problem to anyone who is interested and who doesn't already know it.

the deaths to heart disease percentages were made up btw (a "damn lie") :o

I can't be the only one who enjoys this stuff
I estimate 13.2% of the rest of you do

Ooh yes please to the 'Monty Hall problem'.

(I always shout at the tv when stats for the effectiveness of grooming products are given - eg, 87% of women agreed that their skin was smoother after using 'Magocreme' for two weeks. "Smoother than what?" I shout, "Smoother than when you put nothing at all on your skin - therefore 13% of women find that Magocreme either makes no difference or makes their skin rougher".

And how can anyone's hair be 3 times shinier?

Also, "up to 100% x/y/whatever" means nothing does it? The difference will be between nothing and twice what it was before. Pah. Just Pah.

As for heart disease I'm pretty sure that if you add up the people who reportedly die of that and those that die of smoking, overeating and of not exercising then it comes to many more than the total number who died. So unless some people died twice they really mean "contributed to"

"8 out of 10 owners said their cat prefered it" prefered it to what?

oh yes very good point about the effects being compared to doing nothing.
I heard on the radio today that you're more likely to die in hospital if you have surgery near the end of the week (I think, didn't really tune in till half way through). Flippantly, near the end, he said "and this figure is up to 86% on a Friday". I assume that is 86% more likely to die, rather than an 86% chance of death when you go in to have that ingrowing toenail removed!

I assume you're aware of the problem? Gameshow - three doors. Behind one door a flash new car. Behind other two doors, goats (and not even the posh Oxfam ones!). Pick a door, 1, 2 or 3. So you have a 1 in 3 chance of picking the car. A 2 in 3 chance of picking a goat. You pick door 1. OK...now I, the omniscient gamesmaster am going to open a door - here we go, door 2. A goat. You now get the option to change your selection (so from door 1 to door 3, as this is the only one left). Should you swap?

And the answer is yes, as by swapping you increase your odds by 100%

Not sure what you mean, Back. Deaths will be counted as those which have one of the "heart disease" codes in the cause field. The other deaths (obesity, smoking etc) will have whatever caused their death. So in the case of smoking, it may have been lung cancer.
Ah yes, I think I see. The way it works with alcohol admissions is that they assign a rough % of admissions for (eg heart disease) that will have been caused by alcohol. So it may be that research has shown that 40% of heart disease admissions are "caused by" alcohol. I suspect there will be something similar for obesity, smokng etc, and you're right, they will then be double counted in the separate reports (although they're actually reporting different things). Interesting, and something I should know.

"8 out of 10 owners said their cat prefered it" prefered it to what?
Preferred it to a kicking round the sofa and some dog food :o

The issue with the heart disease is that it kills older people - people who, in the past may have died of other causes, much earlier, such as infection or industrial accidents. It's not enough to know the proportion of people dying from X has gone up unless you know whether the rate is also going up and what the rest comprises, and what is happening with those.

That reminds me, when I used to ask what I was getting for Christmas I used to be told "a walk round the table" or " a sack of cinders". I didn't even know what a cinder was! (still not really sure)

Another one - I believe the numbers of children with complex disabilities is rising. This is because premature infants are being kept alive at younger ages because of improvements in healthcare and technology. While it doesn't seem like it at first glance this is a good thing - these are children who would have previously died.

87% of 74 women sampled would recommend it to a friend.

So 10 (9.something to slightly less than exact) women would tell their friends not to buy it, it's shit. Out of 74.

That reminds me, I was recently asked to complete a survey (read out, my answers spoken) in the doorway of my local co-op, within earshot of the tills (it wasn't busy). Luckily I had nothing but genuinely nice things to say about them but I can imagine if I'd wanted to rate them badly I'd have been a bit nervous next time I popped in for a pint of milk!

And I bet the options to select were:

Definitely
Probably
Maybe
Definitely not absolutely not, I'd rather stick pins in my eyes

With the top 3 counting as a "yes" - sneaky!

The skin cream and shampoo ads do my head in! How can they justify the claims based on so few people, who were probably paid to take part in the research?! 'Our advertising and marketing budget for this year is 10 million quid - let's survey 70 women!'

Anyone know what would be a fair number of people to test on?

Great thread!

Oooh, this bugs me too - especially the shampoo ads. My biggest annoyance though is people's lack of understanding of risk which the media usually play up no end e.g "Eating bacon increases the chance of dying from a horrible gruesome death by 25%", which of course means absolutely nothing without the baseline and even then doesn't really tell you anything unless you know your own personal risk factors.

There was a great article on the BBC website a couple of years ago which explained it really well - I'll see if I can find it.

Well, the reason they don't do better research into cosmetics and hair products is because if they were actually to prove some benefit about them, the product may have to be classified as a drug, and they wouldn't be able to sell it in the same way.

DD1, who is 14, often tells me that 84% of statistics are made up.

People in general dont understand risk. You see it all the time, especially in the discussions about breastfeeding and formula. Formila feeding increases the risks of gastroenteritis. The fact your formula fed baby didnt get it doesnt negate that statement. Ditto if your breastfed baby did get it. (Not turning this into a breast/formuka debate btw!). Also dismissing research with "but I bet those mums were the middle class ones". Proper research should control for these factors but people always assume that if a randomised controlled trial hasnt been carried out, then the results are crap. Yet somehow no one doubts the ljnk between smoking and lung cancer. I wonder how they got ethics approval for that rct.

Is that reallh the case sanity? If a cosmetic product has an effect it gets reclassified? How interesting.

news.bbc.co.uk/1/hi/magazine/7937382.stm

A lot of his articles were very good and are available in the drop down list at the side.

Ooh claire thanks fo that. I should be working...must savethem to read tomorrow

Stealth, can I ask for this to be moved to Science and Nature Club? Statistics being 79% more important in reaching scientific conclusions than anything else, after all .

I don't know, shall we survey a sample of posters on the thread and see what they think? We'd need to weight it to give more weight to more recent posters.
(fine by me )

It is amazing the amount of "proper" scientific research that gets written when the researchers don't appear to understand statistics and risk. And then you get the books that are absolutely based on actual scientific evidence and not at all biased towards the opinion of the author.

Take 3 in a bed for example which makes my blood boil every time the author merrily says something along the lines of "30% of SIDS cases happen in co-sleeping babies which means that 70% happen in cots so actually MORE babies die in cots so it is more dangerous." NO YOU MORON! IT MEANS NOTHING UNLESS YOU ARE LOOKING AT IT AS A PERCENTAGE OF THE NUMBER OF BABIES WHO USUALLY CO SLEPT OR WHO USUALLY SLEPT IN A COT! And that i before you even get around to bf vs ff, usual place of sleep vs place of sleep whenthe death occured and all the rest of the variables. The Politics of breastfeeding is guilty of similar offences too.

Exactly, Indith, exactly