"Tony Raven" <
[email protected]> wrote in message
news:
[email protected]...
>
> So why do you think the graphs for head injuries in adults and children
> (Figs 1 & 2) only move within normal statistical fluctuations, with
> significant changes in helmet wearing levels. Indeed in the first year of
> Fig 1, helmet wearing quadruples but the head injuries go up, not down,
> although I doubt the increase in injuries is more than statistical
> fluctuations in the small sample. Meanwhile in Fig 2 helmet wearing goes
> from 0 to 40% in 3 years yet head injuries stay within the sample natural
> noise variations of 25 +/- SQRT25.
>
I'm going to pull a bit of rank here. I actually AM a statistician.
Figures 1 and 2 are basically **** from a causal standpoint. As the authors
note their model didn't fit well. Even HAD the model fit well, you couldn't
rule other other problems (such as, for example, the amount of cycling
changing over the time of the study). If figure 1 and 2 looked the same,
but were based on a much larger sample, they still would be basically ****
from a causal standpoint. Fundamental flaws in a logical argument aren't
helped much by a larger sample.
It's also relatively poor statistical hygiene to compare normalized rates
(e.g. percentage of helmet use) with unnormalized rates (number of head
injuries). For example, if the population doubles in San Diego but the
behavior stays the same, the percent of helmet use wouldn't change but head
injuries would be expected to double. Lots of people make this error,
though, and the population isn't growing that fast in San Diego over 5
years, so this is minor.
I haven't read the whole study thoroughly enough to have an opinion on the
rest of it, nor do I really want to get involved in the helmet wars.