J
James Annan
Guest
I sent the following to "Injury prevention on-line" over a
week ago, but it shows no sign of being published and my follow-
up email has not been answered. I guess someone might as
well see it, typo and all. Isn't the internet great for
vanity publishing?
Neither Cook nor Sheikh replied to my email, either.
Their original article, to which this refers, can be found
on
http://ip.bmjjournals.com/cgi/content/full/9/3/266
-------------------------------------------------------
----------
"Fundamental error in "Trends in serious head injuries..."
Cook and Sheikh 2003"
-------------------------------------------------------
----------
<!-- article ID: 9/3/266 -->
<P> The main conclusion of Cook and Sheikh (2003), that a
bicycle helmet prevents 60% of head injuries, is
incorrect due to a fundamental error in the way they
have treated their percentages. A correct analysis
demonstrates unequivocally that there must be major
confounding factors in their data set that they have
failed to take into account, and therefore any estimate
of helmet effectiveness is purely speculative.
<Q> Assuming that their basic analysis of the data is
correct (although the numbers they quote in the text do
not actually appear to match the figure plotted), they
arrive at a figure of a 3.6% for the reduction in the
head injury (HI) rate for cyclists, over and above the
"background" reduction that pedestrians have also seen.
They assume that this drop in HI is due to increased
helmet-wearing. However, this reduction is presented in
terms of the number of percentage points, and relative
to the baseline value of 27.9% HI for cyclists in 1995-
6 it actually represents a
3.6/27.9 = 13% drop in the HI rate.
<P> The decrease in the number of helmetless cyclists over
the same interval is 5.8 percentage points from a
baseline of 84% unhelmeted, giving the percentage drop
as 5.8/84 = 7%. Cook and Sheikh calculate helmet
effectiveness to be given by the ratio 3.6/5.8 = 60%.
However the correct expression to use is 13/7 = 186%.
In other words, "helmet effectiveness" is so high that
each helmet does not just save its wearer, but a non-
wearer too. At this rate, head injuries would be
eliminated completely if just a little over half of all
cyclists wore them! This is clearly ludicrous.
<Q> A more reasonable conclusion to draw from this would be
that there are some other factors that are responsible
for the large drop in HI rate, and therefore any
attempt to attribute some part of the total 30%
(8.49/27.9) change to the provably marginal impact of a very
small number of extra helmet wearers is at best
highly speculative and fraught with inaccuracy.
<P> What makes this all the more poignant is the fact that
the authors have recently produced a book entitled
"Basic skills in statistics"!
<Q> James Annan
week ago, but it shows no sign of being published and my follow-
up email has not been answered. I guess someone might as
well see it, typo and all. Isn't the internet great for
vanity publishing?
Neither Cook nor Sheikh replied to my email, either.
Their original article, to which this refers, can be found
on
http://ip.bmjjournals.com/cgi/content/full/9/3/266
-------------------------------------------------------
----------
"Fundamental error in "Trends in serious head injuries..."
Cook and Sheikh 2003"
-------------------------------------------------------
----------
<!-- article ID: 9/3/266 -->
<P> The main conclusion of Cook and Sheikh (2003), that a
bicycle helmet prevents 60% of head injuries, is
incorrect due to a fundamental error in the way they
have treated their percentages. A correct analysis
demonstrates unequivocally that there must be major
confounding factors in their data set that they have
failed to take into account, and therefore any estimate
of helmet effectiveness is purely speculative.
<Q> Assuming that their basic analysis of the data is
correct (although the numbers they quote in the text do
not actually appear to match the figure plotted), they
arrive at a figure of a 3.6% for the reduction in the
head injury (HI) rate for cyclists, over and above the
"background" reduction that pedestrians have also seen.
They assume that this drop in HI is due to increased
helmet-wearing. However, this reduction is presented in
terms of the number of percentage points, and relative
to the baseline value of 27.9% HI for cyclists in 1995-
6 it actually represents a
3.6/27.9 = 13% drop in the HI rate.
<P> The decrease in the number of helmetless cyclists over
the same interval is 5.8 percentage points from a
baseline of 84% unhelmeted, giving the percentage drop
as 5.8/84 = 7%. Cook and Sheikh calculate helmet
effectiveness to be given by the ratio 3.6/5.8 = 60%.
However the correct expression to use is 13/7 = 186%.
In other words, "helmet effectiveness" is so high that
each helmet does not just save its wearer, but a non-
wearer too. At this rate, head injuries would be
eliminated completely if just a little over half of all
cyclists wore them! This is clearly ludicrous.
<Q> A more reasonable conclusion to draw from this would be
that there are some other factors that are responsible
for the large drop in HI rate, and therefore any
attempt to attribute some part of the total 30%
(8.49/27.9) change to the provably marginal impact of a very
small number of extra helmet wearers is at best
highly speculative and fraught with inaccuracy.
<P> What makes this all the more poignant is the fact that
the authors have recently produced a book entitled
"Basic skills in statistics"!
<Q> James Annan
