it is just that there are a range of acceptable methods of statistical analysis.

some methods aren't considered acceptable or aren't considered acceptable to use on certain kinds of data. if you do that then people typically jump on you for bad statistical analysis.

but there are still a range of acceptable statistical methods.

so what tends to happen in practice is you pick the statistical method that most shows what you want to show.

at a very simple level... one has to pick the scale of the graph. one wants the scale to be small if the differences one has found are only just statistically significant (aka small differences). by having a small scale the visual impression is that the differences are large. that is just one simple way in which the methods you use affect the information you convey which influences the take home message.

if you look at individual statistics... i've heard the best predictor of violence is history of violence. so if you take one individual and track them through time then pick a time slice of the individual (say age 20) and ask 'is this person likely to commit a violent crime?' then if he has committed a violent crime in the past it is more likely he will in the future.

but another way to look at it is in terms of a sample size of the population. take a time slice of people (say age 20) and then take the numver of offenses over the next 10 years and then ask 'how many of those offenses were committed by people with a past history of offending?'.

subtly different questions...
subtly different research methods...
subtly different statistical analyses...

different answers.

science is like that... that is why 'facts' are challenged. that is why 'facts' change. while you can't quibble about what the particular study found you can certainly quibble over whether the sample size was adequate, whether they were testing what they thought they were testing, whether their statistical analyses were legitimate, whether other statistical analyses would lend themselves to very different conclusions, whether the findings generalise back to the general population, whether the findings generalise back to interesting information about the hypothesis they began with.

different studies are often interpreted in contradictory ways... science is really a rather messy business when it comes down to it...

and in all this... i worry about individuals getting lost in statistics. i worry about MY chances of improvement given MY dx. I don't care if 99.99999% of people with my dx get worse - I could be that 0.000001 person who improves. And you know what... The liklihood of my improving given that I improve is 100% so there!

I kinda like the TAB motto:

You know the stats... Now beat 'em!!!!!

But sometimes the gamble is my life...
Others lives...

Stats change depending on the information that is considered relevant to be plugged in.

For example... What is the average age that people 'like me' live to?

I'm female.
I'm white.
I have brown hair
I'm a smoker
I have 10 fingers

How do you decide which information is and is not relevant? What information you choose to use affects the stats you get.

So... IMO it is wise to be wary of stats... Because... If they really pissed you off it is only too easy to construct your own set (by legitimate research methods) to show... Whatever you want!!!!!

(Though that is probably overstating the point)

Crazy scientists huh

:-)