I dissagree with your policies on DNA

[Panzer]

In your manifesto I agree with most of the things you put in, the only thing I really dissagree with is the part about DNA. I think once you have a sample it should stay in the records as it might be able to help in future investigations on any future crimes.

My view is that everyone DNA should be recorded and taken a birth. I can see why this is aganist peoples privacy but to me it make it easier for criminals to be caught and narrow down the people who are resposable for murders.

The advantages out way the disadvantages in my opinion.

[cc]

Dear panzer,

We just analysed the DNA data you provided at birth, and we have determined that you have a high risk of cancer and are incredibly stupid. Because you are likely to die young and are an inherent moron, we will be unable to accept your application for a job/loan/unemployment benefit/scholarship/insurance/whatever.

Yours sincerely,
The UK government/your future employer/...

[rancidpunk]

Dear panzer,

We just analysed the DNA data you provided at birth, and we have determined that you have a high risk of cancer and are incredibly stupid. Because you are likely to die young and are an inherent moron, we will be unable to accept your application for a job/loan/unemployment benefit/scholarship/insurance/whatever.

Yours sincerely,
The UK government/your future employer/...

[Finlay_A]

^^This,

Also, it does away with the whole presumtion of innocence thing, which I'm quite fond of. It is another of many ways in which every citizen is treated as a suspect. See also excessive CCTV, monitoring internet traffic, etc.

Not to mention it makes it A LOT easier to frame innocent people.

[azrael]

In your manifesto I agree with most of the things you put in, the only thing I really disagree with is the part about DNA. I think once you have a sample it should stay in the records as it might be able to help in future investigations on any future crimes.

My view is that everyone DNA should be recorded and taken a birth. I can see why this is aganist peoples privacy but to me it make it easier for criminals to be caught and narrow down the people who are resposable for murders.

The advantages out way the disadvantages in my opinion.

The problem, panzer, is that the innocent shouldn't need to prove their innocence, the legal system needs to prove someone's guilt. I really don't support the notion that the innocent have nothing to hide and therefore should have no privacy. The innocent do have things to hide without being criminals. I might want to hide the fact that I am genetically predisposed to cancer. I might want to hide the fact that I borrow teen vampire romance novels from the library (I don't, I swear I don't!).

On DNA samples specifically, what is the rate of error? I'm not sure off the top of my head, but let's assume it is incredibly low. No matter how low the error rate is, surely the chance of a mistake increases with the more samples on record. Let's look at the situation where DNA records are kept from anyone every arrested, even if they were later found innocent. A DNA sample from a crime gets matched against a database off 1000 samples, or 1000000 samples, and this particular DNA is not on record, there is more chance of a false positive when there is a bigger database matched against. When in a court of law talking about error rates, the jury will think "well there might be a mistake, but the DNA was on record, so the person is a criminal anyway, so they probably did this too". So by keeping innocent people's DNA on record you increase the chances of mistakes, and you prejudice jury's against possibly innocent people.

Your answer might be to put everyone on the database, to reduce the assumption that being on the DB means you are a criminal. What happens when we share that DB with the rest of the world and they do likewise? You then have billions of samples to match against - what's the chance of a false positive then? When searching for needles in haystacks, we do not want bigger haystacks!

[apricot]

A lot of the problems suggested take common sense, I mean if you DNA matches someone in brazil it's unlikley the brazillian would be accused of the crime as it would be impossibe for him to commit a crime.

The amount of DNA analysed by forensics is actually only a small fraction of the total human genome do it would be difficult and expensive for everyones DNA to be properly analysed. The means specific markers must be searched for. Even with such a limited scope I think the odds of a false positive are 2 billion to 1 , plus any identical twins you have.

So there is a small chance of being falsley accused if anyone has studied psychology you would know how inaccurate eye witness testimony is. DNA databases have been used in Greenland (or it nay be Iceland) with great results in terms of health benefits... I'll have a look tommorow for some more specifics as it would be an interesting case study.

However I do agree DNA evidence shouldn't be the be all and end all of criminal conviction and circumstantial DNA matches do not equal a guilty criminal

[azrael]

It doesn't matter how long the odds are, when the false positive happens it happens. http://en.wikipedia.org/wiki/Sally_Clar ... l_evidence

[Panzer]

Nothing is going to be 100% perfect, people have been accused of crimes they have not done without DNA evidence. And about the jury been prejudice is down to there judgement not because the samples are on record

[apricot]

It doesn't matter how long the odds are, when the false positive happens it happens. http://en.wikipedia.org/wiki/Sally_Clar ... l_evidence

under the same reasoning eyewitness testimony shouldn't be used especially as how inaccurate it is.

[cc]

If we don't have a national fingerprint database, why should we have a national DNA database? Has the right to privacy suddenly gone out of fashion? Why did the people who wrote the Human Rights declaration take the effort to put in the privacy clauses? Why did they think privacy is important?

Here's something different: What if some day 1984 becomes reality and we need to turn to resistance/guerrilla warfare to save ourselves from our own government? If there is a DNA database, full CCTV coverage, full communications monitoring and guards with guns patrolling the streets, the fight is already lost! Sure, that eventuality seems particularly distant right now... but Hitler was massacring people just a measly 65 years ago. Who knows what the world will be like 65 years from now?

[Finlay_A]

If there is a DNA database, full CCTV coverage, full communications monitoring and guards with guns patrolling the streets, the fight is already lost!

We'll find a way around that I'm sure

[JohnB]

On DNA samples specifically, what is the rate of error? I'm not sure off the top of my head, but let's assume it is incredibly low. No matter how low the error rate is, surely the chance of a mistake increases with the more samples on record. Let's look at the situation where DNA records are kept from anyone every arrested, even if they were later found innocent. A DNA sample from a crime gets matched against a database off 1000 samples, or 1000000 samples, and this particular DNA is not on record, there is more chance of a false positive when there is a bigger database matched against. When in a court of law talking about error rates, the jury will think "well there might be a mistake, but the DNA was on record, so the person is a criminal anyway, so they probably did this too". So by keeping innocent people's DNA on record you increase the chances of mistakes, and you prejudice jury's against possibly innocent people.

Your answer might be to put everyone on the database, to reduce the assumption that being on the DB means you are a criminal. What happens when we share that DB with the rest of the world and they do likewise? You then have billions of samples to match against - what's the chance of a false positive then? When searching for needles in haystacks, we do not want bigger haystacks!

Something to bear in mind... uniqueness is often assumed for records in DNA databases (rather like uniqueness is assumed for fingerprints), however real life is a bit more messy than that, and all biometrics recorded and matched have some inexactness and there is a possibility of identifying a "match" that actually is no such thing. DNA "profiles" are a selection of markers from the genome, and are reduced to a set of about 10 numbers or pairs of numbers, rather like the way auto finger print matching is done by reducing the full print down to a selection of points and therefore to a set of numbers.

What are the chances of having an unrelated hit / false match like that? It turns out to be potentially rather higher than might be imagined, it's worth looking up the birthday problem and the prosecutor's fallacy. When it's something like matching a biometric to a database, however, where there's a very small chance of a random match, but a very large database, we can very easily calculate approximately how often to expect duplicates in the database (you can do an exact calculation, but it's more fiddly, doesn't matter for this purpose).

Let's assume the odds against a random match are 1 billion to 1 against (the actual odds claimed for DNA testing vary slightly, but it's around that order of magnitude). As we'll see, it doesn't really make much difference even if it's quite a bit more precise than that. Let's also assume that there are very few in the database that actually are exact matches of another profile, so few that we can discount the possibility of triples or more (we can get away with that assumption until the database is large enough to have significant numbers of duplicates).

So what's the number of pairs of profiles we expect to see, if we had a database including everyone in the UK, about 60 million? Let the number of such pairs be n. Now, if we pick two profiles at random, for them to be the same, we must first pick one of those which is not unique in the database, there are 2n of those, so the chance of that is 2n / 60 million. The second one we pick, however, is almost certainly not going to match, unless we pick that 1 other profile out there that matches; so the chance of that is 1 in 60 million. We multiply those, the chance that we pick two people at random and find they have the same profile is therefore 2n / (60 million)^2, or n in 1,800,000,000,000,000. However, this must also be the same as what we already know the chance of random match to be (how do we know that? see below), 1 in 1 billion.

As a result, if the test has an accuracy level of only a billion to one, we can expect to see approximately 3.6 MILLION people in a full UK database who exactly match someone else in the population! Actually, not quite, because once the database is that large we can't quite make the assumptions I did, and should do the full calculation; it doesn't matter, though, the simplistic calculation is enough to see that there's an issue.

Maybe the test could be made more precise? It turns out that also doesn't work. Even if the chance of a random match is 10 or 100 billion to one, it's still enough to be an issue in a very large database (such as an entire population), and the resolution needed to achieve near-uniqueness turns out to be related to the square of the population size; which means whatever you do doesn't work as you scale up (e.g., think about doing this across the EU population, rather than just the UK...). Also, there's an issue with making the test more exact, because then you're more likely to fail to match when it is actually the same person - which is a trade-off AIUI to be expected with any kind of biometric.

Now, as for the claimed accuracy of the test, the way this is calculated is by looking at each of the 10 or so elements in the DNA profile, each of which show up in the population with varying frequency, which has been measured for representative samples (no issue with this, standard stuff), and so we can say for each one, this one is 1 / 10th of the population, this one is 1 / 17th of the population, etc. All those are then multiplied together, and that's what gets quoted (1 billion to one, or whatever) as the "headline stat". That's not necessarily wrong, either, but... I have a couple of potential issues with the way it gets bandied around. Firstly, that's a lot of multiplications, and there's an assumption in there about independence, and secondly the end result is very small / very large depending which way you look at.

So... I'd really like to know, from a large database like the UK NDNAD, what level of duplicates is actually turning up; because the numbers say there should be some, as the database grows, and unless the test is vastly more accurate than is claimed, the present database is getting large enough to expect some. Count those, the ones that really are not the same person under an alias etc, and the math I started with can be worked backwards, we'd actually have a direct experimental measure of how precise the biometric was, rather than the guess based on multiplying all those probabilities mentioned in the previous paragraph. One other thing to bear in mind, the math above is a bit idealised, assuming that each person is equally likely to be unique or not in the database. That's not necessarily the case, as it could be worse than that, it's conceivable that particular family or racial groupings are less unique in the database than others (I don't know; but it's possible).

Anyway, this post is probably a tl;dr thank you to anyone who's had the patience to read through this far. DNA can be a hugely useful tool for helping determine guilt, or innocence, however having thought through how this pans out with very large databases, I myself am deeply sceptical of the idea of having a very large (or universal) database. If you have a smaller database, if there's reason to suspect someone AND there's a hit on the database, then it becomes compelling evidence; even doing a DNA sweep e.g. of a town or village could be very effective (the numbers work then; if you find a match it becomes very likely you've got the right person, and can look for direct evidence as well).

What you shouldn't do, IMHO, however, is to store all of those (most of them are, and will remain, innocent of any wrong doing), and keep adding to the database/keep trying for "cold hits", that will lead to problems, and like wanting to extend CCTV everywhere - there comes a point where it just isn't bringing in enough results compared to other things we could put resource into (e.g., old-fashioned police and detective work, not instead of, but as well as, all the fancy forensics).

[azrael]

Here's another one for maths fans...

If DNA profile samples from all unsolved crimes are stored in one database, and every person in the UK has their DNA profile stored in a second database, and every couple of days the databases are matched off against each other, what are the chances of someone being accused of committing a crime that occurred before they were born?

[Finlay_A]

I'll get to work on azrael's maths homework later tonight I'm off uni for summer but might as well have some fun

It doesn't matter how long the odds are, when the false positive happens it happens. http://en.wikipedia.org/wiki/Sally_Clar ... l_evidence

under the same reasoning eyewitness testimony shouldn't be used especially as how inaccurate it is.

The thing about eyewitness testimony, as opposed to DNA, is the perception of accuracy. Most people, from watching thhings like CSI, assume that DNA evidence is completely infallible. When I say "most people" this could mean the politicians that write the laws as well as the jurors that pass verdicts. Given the last government's record on science, and I have no reason to believe the next will be any better, this is quite worrying.

With witness testimony however, the perception matches the truth (or close enough), probably because most people can understand it. In the case of a false positive: with witness testimony this probably wouldn't convict an innocent person (definitely not by itself); however in the eyes of the jury a DNA match might be convincing enough.

[Pete]

The problem is that Jurors "understand" DNA (and fingerprints)

However the other problem is (and this is John's post in tl;dr) that p(Positive Match) is not equal to p(Positive Match|Innocent)

That is to say that where the probability where there is a random chance of a match of 1 in 3 million (0.0000003%), this is not the chance that the crime was committed by someone other than the accused. Rather we must divide the into that the population. Given the UK population of approx. 60million, this gives 20 matches. This means that the chance of someone other than the defendent committing the crime (given no other evidence) is not 1 in 3 million, but 19 in 20(95%), or just a 5% chance of guilt.

For anyone interested in this and more statistical joy (Such as how a possitive HIV test is only right about 50% of the time) I highly recommend Reckoning with Risk by Gerd Gigerenzer.

[JohnB]

The problem is that Jurors "understand" DNA (and fingerprints)

However the other problem is (and this is John's post in tl;dr) that p(Positive Match) is not equal to p(Positive Match|Innocent)

Yes... right.

That is to say that where the probability where there is a random chance of a match of 1 in 3 million (0.0000003%), this is not the chance that the crime was committed by someone other than the accused. Rather we must divide the into that the population. Given the UK population of approx. 60million, this gives 20 matches. This means that the chance of someone other than the defendent committing the crime (given no other evidence) is not 1 in 3 million, but 19 in 20(95%), or just a 5% chance of guilt.

But this is not. Not at all, it's necessary to consider the population size, but it's not a simple division, it's much more counter-intuitive than that. If it was that simple, then a billion to one accuracy would pretty much be definitive, but it's not. With a random match frequency of 1 in 3 million, there will be a large number of other people in the population who also match an individual biometric; but not necessarily 20, there will be a distribution, some could be much more rare, and some might match far more than 20.

Back in the tl;dr post, I did some math, and showed that if the random match frequency is 1 in one billion, you'll have ~5% who have one or more matches in the UK population (can't be bothered to work out the exact distribution right now, if I get round to it I'll post it), and such matches are common enough that there'll be a few where there could be two or three or four such. So a test that accurate is certainly strong evidence to investigate further, but in the absence of any other evidence the question mark over it is easily enough to establish reasonable doubt, a small but significant possibility of convicting an innocent. So it's compelling, but not enough by itself.

If the population size can be reduced (e.g., two-thirds of the full database can be excluded entirely on other grounds) or if other direct evidence is discovered, then that becomes sufficient to establish the case beyond reasonable doubt. But... note that establishing that the DNA sample is in fact from the crime scene is also not automatically equivalent to establishing guilt (someone can have been present, and even have committed crimes, but not be guilty of the offence in question), and it's also possible that their DNA may have become associated with a crime scene without their presence, so a prosecution needs to establish the provenance of how the sample was collected etc.

One of the arguments against having a full DNA database which I myself find particularly telling... If we made such a thing, and routinely matched every crime scene sample against the full UK database to generate leads, then it seems to me there would be a big incentive for actual criminals to drop false DNA at crime scenes. It wouldn't surprise me, even, in that circumstance, if a "cottage industry" of criminals supplying such, were to develop, and criminals can use amplification techniques as well as anyone else. Even if it was only a temporary respite, knowing that there was a fair chance someone innocent somewhere out there would be investigated instead of pursuing the actual criminal, it would become worth the criminal's effort, especially for serious offences where the potential penalty is very great.

Conversely, if the database is limited to only those previously charged or convicted of a crime, then that particular criminal tactic won't work; dropping random DNA collected from someone else has no real effect if the chances are they aren't on the database, so it's unlikely to become the "criminal fashion", and it can be argued that having already been charged/convicted of an offence that automatically checking them is something they've brought on themselves, and that it's going to produce results much more often than checking the whole population all the time. The only caveat I'd put to that, is that rather like the way the Rehabilitation of Offenders Act works that a former offender should have some route back to "normality" and eventual removal from the database, especially if their offence is minor (non-violent; not attracting a custodial sentence, for example), and that only the most serious violent offences should result in permanent retention on the database.

[azrael]

And you all thought this thread was dead, but couldn't help posting this when I read it a moment ago

http://www.newscientist.com/article/mg2 ... ?full=true

[cabalamat]

That is to say that where the probability where there is a random chance of a match of 1 in 3 million (0.0000003%), this is not the chance that the crime was committed by someone other than the accused. Rather we must divide the into that the population. Given the UK population of approx. 60million, this gives 20 matches. This means that the chance of someone other than the defendent committing the crime (given no other evidence) is not 1 in 3 million, but 19 in 20(95%), or just a 5% chance of guilt.

You're right. Personally I think Bayes' Theorem should be standard part of the maths curriculum.

[Finlay_A]

That is to say that where the probability where there is a random chance of a match of 1 in 3 million (0.0000003%), this is not the chance that the crime was committed by someone other than the accused. Rather we must divide the into that the population. Given the UK population of approx. 60million, this gives 20 matches. This means that the chance of someone other than the defendent committing the crime (given no other evidence) is not 1 in 3 million, but 19 in 20(95%), or just a 5% chance of guilt.

You're right. Personally I think Bayes' Theorem should be standard part of the maths curriculum.

It is. The problem is that maths is taughts in school as per "memorize and regurgitate this formula" method, so the pupils don't really learn how counter-intuitive maths can be until they get to uni

[cabalamat]

That is to say that where the probability where there is a random chance of a match of 1 in 3 million (0.0000003%), this is not the chance that the crime was committed by someone other than the accused. Rather we must divide the into that the population. Given the UK population of approx. 60million, this gives 20 matches. This means that the chance of someone other than the defendent committing the crime (given no other evidence) is not 1 in 3 million, but 19 in 20(95%), or just a 5% chance of guilt.

You're right. Personally I think Bayes' Theorem should be standard part of the maths curriculum.

It is. The problem is that maths is taughts in school as per "memorize and regurgitate this formula" method, so the pupils don't really learn how counter-intuitive maths can be until they get to uni

Sadly, a lot of subjects are taught like that in schools. The reason is a combination of (i) pressure to achieve exam results and (ii) exams that can be passed by rote memorizing rather than actually having to understand stuff.

For this reason, exams should concentrate on measuring people's understanding rather than their ability to memorize. This is particularly important when people are growing up into a society where any fact is only a Google search away.