Roughstuff

One reason why (as a statistician) I always howl with laughter about safety 'studies' (about the 3 foot rule, bike lanes, helmets, etc) is they are so childish and infantile. If ya look into them with any detail...this is what ya often see. This is from a report about what happened after traffic cameras were installed:

But at the three city intersections from Feb. 21, when fines began, through May 1, The Post found:

--Rear-end collisions increased to five from two. Rear-end accidents sometimes go up with cameras because anxious drivers are more likely to stop abruptly.

--Overall accidents increased to seven from six.

--The only injury in either period came under cameras, in a rear-end crash in March 2010.



So the study is based on a whopping 10 weeks, a total of 7 collisions for comparison, one ADDITIONAL accident, and only ONE crash with injuries.

The global warmingists and peakoilers would be proud.


roughstuff

Metzinger

I couldn't agree more.
These statistics are just as anecdotal as the temperature tracking of the unusually cold winter we've just had.
Fortunately we've got thinkers like Al Gore and Roughstuff looking at the big picture. Thanks.

Roughstuff

WOW! ME up there with the Goricle! Nunc dimittus!

roughstuff

Seattle Forrest

Almost as proud as those fools who think the moon exists.

The only thing more fun than a conspiracy theory, is a global conspiracy theory.

Roody

You call yourself a statistician. What basis do you have for that claim?

I've taken 4 undergraduate and 5 or 6 graduate classes in statistics. I do NOT call myself a statistician--far from it.)

The Human Car

These same statistical "standards" are used in the medical industry. Be afraid, very afraid.

Roughstuff

I have had a comparable set, but they would be different from a pure 'statistics' course....as ya know, many areas take stats and 'adapt it' to their own field......biostats, etc. I took a zillion econometrics courses. My calling myself a statistician is pushing the envelope, but,

(1) I don't care, I'm self employed so i'll call myself whatever I want. I had a similar argument with some ex-military guy when I called myself a military analyst.....

(2) ya don't need [B] 4 undergraduate and 5 or 6 graduate classes in statistics [ /B] to understand the need for proper sample sizes, control groups, and analysis of within-variance. This 'study' had none of 'em.

roughstuff

genec

I fully agree... it used to drive me completely nuts when we would produce one "golden unit" in early prototyping and sales would then take that unit as the spec unit.

There are variations in life and in engineering... one has to have a sample size greater than one to see those variances.

Of course on the flip side, are safety stats that compare apples with oranges... such as using childrens' scraped knees to cite that MUPs are more "dangerous" than city streets.

JRA

If they say the moon is blue
We must believe that it is true.

Many bicycling crash studies suffer from a glaring lack of adequate sample size. Many also lack adequate knowledge of exposure.
And many suffer from having too many variables that may affect the outcome.

What happens is that the ideologues on all sides interpret the data in a way that suits their bias. And then they call their conclusions scientific. It really is quite amusing.

From the thread title, I thought this would be about cycling statistics but it seems to be just as much about a totally unrelated political debate-- about a topic which suffers less from a lack of data than from the problem of too many variables. Ain't the internet grand?

I-Like-To-Bike

For a whole barrel full of flip side statistical laffs, take a gander at the statistical charades that John Forester has manipulated and conjured to support his "scientifically based" conclusions about bicycling "crashes" (his terminology) and alleged effectiveness of his proprietary educational materials in reducing cycling risk.

Roughstuff

I do want to push it a bit more toward cycling stats, but in and of itself, a small sample size can be overcome. Statistical tests (t, F, etc) are all adjusted for the degrees of freedom and the smaller the sample size, the higher the bar you have to jump over to reach a conclusion. My suspicion however is all that was looked at here was a before/after compare of accidents and injuries.

I LOVED your comment about variables. I was in a seminar once and a guy had, maybe, 3 variable in his model. But he also kept saying 'well of course you need to take that into account as well,' and ever time he did so i put a tick mark on my notepaper, since each that is a variable. By the time I was done his degrees of freedom were negative (speaking to you stat geeks).

roughstuff

boneshake

The only thing worse than bad numbers are random accusations based on NO numbers.

The Human Car

Do you have any support for that statement?


Just joking.