This would work if the cameras took pictures of every car, but in practice, if any car going too slow probably (95%?) doesn’t trigger the camera, then you’d only have, say, 5% of your 990, meaning that 99 out of 2000 drivers would be wrongly ticketed, compared to 20 speeders, of whom 2 are incorrectly done.

Of course, if the camera has an even lower error rate for below it’s set limit, then the discrepancy is less bad. But there really would be two error rates here.

]]>I don’t claim to know the actual percentage, though I’d be surprised if the average number of driving minutes that are spent traveling above the speed limit is more than a few percent of the total number of driving minutes.

]]>self driving cars will never happen… it would be too much of a loss of revenue to the states, unless they tax the hell out of them to make up for all the moving violation fines they would be missing out on (as properly programmed and operating self driving vehicles would not commit moving violations without an outside cause beyond their programming).

]]>Interesting math, but if you think only 1% of drivers speed, you’re out of your mind.

]]>WHO do you sue??

Who thinks the STATE should be doing the STATES OWN JOB??

This is a policing job, and the Police should be doing it, NOT A CORP..

You mean make the city government look like they’re intentionally giving out tickets illegally

]]>Instead of investing in red light cameras invest in /traffic planning/. Reward drivers with green lights for doing the speed limit (once they’ve synced to the waveform) and set actual /speed guides/ that post the speed to go at to hit the green light. Do that and you won’t even need to have ‘speed limits’ because drivers will want to just keep getting green light after green light.

Even better, once self driving cars (please arrive soon) hit the market they can time when to leave and how fast to go so that you’d /never/ hit a red light.

]]>*The city issued roughly 700,000 speed camera tickets at $40 each in fiscal year 2012. If 10 percent were wrong, 70,000 would have wrongly been charged $2.8 million.*

It’s actually potentially much, much worse than that.

Let’s suppose you have a population of 1000 drivers, of whom 1%, so 10 of them, speed. The camera is wrong 10% of the time, so of the 10 speeders one doesn’t get ticketed and of the 990 non-speeders 99 *do*.

That’s 108 tickets issued, of which only nine are legitimate.

That’s 11 out of every 12 tickets, not just one out of every ten, wrongly issued. If the real percent of speeders is close to 1%, the real number of wrong tickets may well be a lot more than 70,000. It could be as many as 640,000.

The higher the percentage of cars actually speeding at the camera sites, the lower this is, but it would have to approach 50% speeders for the number of wrongly-issued tickets to be as *low* as 70,000; or else the camera would have to be biased to falsely clear speeders more frequently than it falsely tickets non-speeders.

Unfortunately, the actual, separate false-positive and false-negative rates aren’t mentioned in this article. Someone who knows them could plug them into Bayes’s theorem to figure out the actual percentage of tickets that were wrongly issued. The math would be:

p(non-speeder|ticketed) = p(ticketed|non-speeder)p(non-speeder)/p(ticketed)

with p(ticketed|non-speeder) being the false-positive rate, p(non-speeder) being the proportion of cars passing cameras that are truly within the speed limit (calibrate using a more reliable data source, perhaps a trustworthy cop with a radar gun if you can find one), and p(ticketed) being tickets issued (700,000) divided by total traffic volume past the cameras during the relevant time period. All the p() numbers should be between 0 and 1 (multiply the output by 100 to get a percentage).

(More information on this math is at http://yudkowsky.net/rational/bayes)

]]>