# Study: French three-strikes law did not deter or reduce piracy

**doctorow**#1

**euansmith**#2

2,000 doesn't seem to be a very large sample group.

What happens if a user is "cut off"? Are they prevented from going to another provider? What happens if you move into an address that has been cut off?

**duznanski**#3

2,000 is actually a pretty good-sized sample group. The accuracy of a poll increases not with the ratio of sample/population, but with the the size of the sample alone, at least until you get to significant percentages of the population... but asking 7 million people for a poll is a little bit out of the question.

Using the number quoted in the paper from IFPI of "27% of internet users visited an unlicensed music site in any given month", a sample size of 2,000 gives a 95% margin of error of about 2% -- the true value has a 95% chance of being between 25% and 29%.

**euansmith**#5

Thanks I never did stats at school and the ones we use at work are concocted by people who can't even work out 23 as a percentage of 100.

n00b question: Why would the 95% margin of error change based on the sample size?

**duznanski**#6

The Central Limit Theorem is the official mathematical explanation, but consider for a moment dice. If you rolled one six-sided die, you could get numbers from 1 to 6; and they're all equally likely. Roll two and average them, and you're most likely to get 3.5 -- 1 and 6 are still plausible though. roll three and average, 1 and 6 each only show up 1/216 of the time, under 0.5%. Roll a thousand dice, you're likely to get around 3.5 still, though it's very unlikely to hit it exactly-- but if everyone in the world does this, maybe 15 people will get outside of the range 3.2 to 3.8, and they won't do that by much. Pick up a basic statistics test and look through the chapters about sampling distributions and hypothesis testing to get more information.

**euansmith**#7

Thank you When I'm trying to work out things like; what are the odds of me rolling a higher number on 5D20 than on 3D20, I dig out Excel and go all Monte Carlo; making 10,000 rolls I'm sure I could work it out with exponential or something if I understood how to do it. I think I'm more a hit it with a hammer kind of bloke.

**GilbertWham**#10

Well, clearly, if what they're doing with HADOPI isn't working, they just need to *do it harder*, and it'll fix itself. That always works.

**RichP**#11

2000 is sometimes a good sampling size, but it often isn't, and pollsters often overstate the accuracy of their polls. In polls of this nature there are at least 4 major potential problems. The first is that the sample population is usually not fully randomized, and bias may be introduced by the selection method. The second is that those who respond are not necessarily random. The third is that people make errors in their responses, either deliberately or otherwise. The fourth is that some phenomena do not fit the normal distribution well, especially when it comes to uncommon events.

**Pradaldi**#14

They don't cut the internet anymore since Jully 2013, you receive a brunch of emails and letters and a 1500 Euro bill.

Anyway, I think they cut the internet to someone only one time since 2010, it always was a really controversial penalty.

**Immutable_Mike**#15

Let's test that argument on unfettered capitalism first.

If it works there, we'll try it somewhere else.