# Where's my traffic singularity?January 11, 2008 8:36 AM   Subscribe

If I wanted to extrapolate how many years it's going to take before traffic fatalities were reduced to a statistically insignificant occurrence, how would I do that? I would use the data from the Fatality Analysis Reporting System.
posted by bigmusic to Travel & Transportation (14 answers total)

It appears that the number of traffic fatalities are only increasing over time, so...you can't?
posted by kittens for breakfast at 8:50 AM on January 11, 2008

kittens for breakfast, if the population in general was growing at a faster rate than the total number of traffic fatalities, theoretically you could still get there.
posted by jrishel at 8:58 AM on January 11, 2008

Response by poster: Look at the bottom of the page, those numbers are declining. And if I do an auto calculate in excel with those number only. The first year I get negative numbers is 2051, using the fatalities per mile numbers. With the fatalities per vehicle, it goes below zero on 2060, with fatalities per per driver - 2121 and per population 2115. I somehow doubt that excel filled in the right numbers doing the auto calculate though.
posted by bigmusic at 8:58 AM on January 11, 2008

@kittens: Depends if the fatalities are increasing slower or faster than the amount of traffic or journeys. If it's slower, then in theory it could still become statistically insignificant at some point, (although there's clearly a limit to how many cars or journeys we can have).
posted by wackybrit at 8:59 AM on January 11, 2008

Have you read this wikipedia article.

This is not answer to your question but the amount traffic deaths are not going in any directions where they reach insignificance. Sweden has a zero-traffic-death vision written down in law and we overshot our target for 2007 by almost a 100 %

Traffic is what kills most people under 40 in the western world. The only way to make it statistically insignificant the way is to start killing lots of young people in other ways than traffic and I don't think that is a very good solution to the problem.
posted by uandt at 9:03 AM on January 11, 2008 [1 favorite]

@kittens: Depends if the fatalities are increasing slower or faster than the amount of traffic or journeys. If it's slower, then in theory it could still become statistically insignificant at some point, (although there's clearly a limit to how many cars or journeys we can have).

Yeah, either the statistical number would go way down because overpopulation forces a much greater reliance on public transportation, or it could possibly remain static (or increase) as cars out of necessity carry more and more people (and a single car accident kills more people than it does today). I think there are too many unknowns to reach a figure that's meaningful in a predictive way, but if we don't care about that, then yeah, a figure could be arrived at.
posted by kittens for breakfast at 9:30 AM on January 11, 2008

You might also want to take a look at the "extrapolation error" section of this wikipedia article on extrapolation.
posted by shothotbot at 9:30 AM on January 11, 2008

Your question is ill-formed, and I think comes from a misunderstanding of what statistical significance is. You need to identify a null hypothesis H0 to test against an alternate hypothesis HA. Here's what I imagine you are trying to test:
H0: the mean number of fatalities per population in the current year = 0
HA: the mean number of fatalities per population in the current year > 0
If there were a very small number of reported fatalities and a significant false reporting error rate, it's possible that the true number of fatalities is actually 0. If this were more than 5 percent likely, then you might not be able to reject the null hypothesis. The number of deaths could be called statistically insignificant.

But that is not going to happen anytime soon. I imagine the false report rate for that system is very low. Just as long as there are even a few deaths, the number will still be statistically significant when compared against a null hypothesis of 0, no matter how big the population gets. You might decide that it is insignificant for other reasons, but you will have to define those criteriaâ€”it isn't what statistical significance means.

Also, if your model predicts that there will be a negative number of fatalities, it's obviously the wrong model to use around that time period.
posted by grouse at 9:38 AM on January 11, 2008

What exactly are you trying to get at with this figure?

Short of some kind of paradigm shift there is going to come a point where nothing you do is going to make an improvement in the system. Making a crappy system better is easy. Improving an OK system takes work. At some point you get to where the difficulty involved in improving the system so outweighs the benefits that no one even tries.
posted by Kid Charlemagne at 10:44 AM on January 11, 2008

You have no model for this data. Fatalities/driver were increasing, then started decreasing. Do you think there is some factor which explains this? Is there some plausible relationship to time? First you need to think about model discovery.

Second, a method which is frequently used for rates (which are greater than zero and less than one by definition and so can not possibly have a linear relationship to independent variables which can get really big) is logistic regression.
posted by a robot made out of meat at 10:44 AM on January 11, 2008

Also, you could say that traffic death is already statistically insignificant. ~ 40 K traffic death vs total 2.4 million dead each year in US. If traffic death went down to zero that is just a 1.7 % decrease and I dare say that this would make no significant impact on total us death toll in a statistical sense. Anyone with access to all the data can probably point out how wrong i am but the fact still stands that traffic is and will be a major killer in the age group people care about. Young people.
posted by uandt at 6:00 PM on January 11, 2008

uandt, you also misunderstand what "statistically significant" means. Please read my previous comment and the linked Wikipedia entry.

you could say that traffic death is already statistically insignificant

No.
posted by grouse at 6:26 PM on January 11, 2008

grouse, Of course you are right but my point was that in any phb-presentation I would present any number within the standard deviation of a set (or just just too small too look nice on a pie chart) as "insignificant" and these people would probably believe me to.
posted by uandt at 3:22 AM on January 12, 2008

Yes, there are other kinds of significance, other than statistical significance. Unfortunately, you'll have to decide on your own what significance means in other senses.

While I agree you could convince PHBs that something less than two percent is insignificant (in a non-statistical sense), I think they are perhaps making a mistake. First, when you are dealing with really large numbers, percentages that may have been insignificant at lower levels become much more significant. And as you point out, motor vehicle deaths are inarguably significant for certain age groups.

Also, in the first instance here, the CDC divides deaths into broad categories. The fifth-ranked category, with 4.4 percent of deaths is accidents (unintentional injuries). This will show up in your pie chart presumably. And the number one cause of accidents is motor vehicles.
posted by grouse at 4:05 AM on January 12, 2008