60% of drivers give false details when stopped.



Tony Hogarty <[email protected]> wrote:
> On Wed, 22 Nov 2006 13:37:13 +0100, GeoffC wrote:
>
>> What makes you think it will be limited only to car drivers? Anyway
>> it doesn't really appear to be absolutely foolproof.
>>
>> http://www.theregister.co.uk/2002/05/16/gummi_bears_defeat_fingerprint_sensors/'

>
> The article you quote is over five years old - in terms of IT that is
> ancient so I expect that there will be improved technology now.


I wish I could share your confidence in the infallibility of new technology.
Somehow I have a sneaky feeling that however high they build the walls,
there will always be somebody who finds a longer ladder.
If somebody steals your house keys it is relatively easy to change the locks
and buy new keys thus rendering the old keys useless. What if somebody
manages to "steal" a copy of your prints?..........buy a new set of fingers?

--

Geoff
 
On 2006-11-24, GeoffC <[email protected]> wrote:

> I wish I could share your confidence in the infallibility of new technology.
> Somehow I have a sneaky feeling that however high they build the walls,
> there will always be somebody who finds a longer ladder.
> If somebody steals your house keys it is relatively easy to change the locks
> and buy new keys thus rendering the old keys useless. What if somebody
> manages to "steal" a copy of your prints?..........buy a new set of fingers?


Have a longer ladder:

http://www.ccc.de/biometrie/fingerabdruck_kopieren.xml?language=en

There's even the possibility of attacking the database directly by using
fingerprint generators to exhaustively search for fake prints that hash down
to values similar to those of real prints.

Regards,

-david
 
Mark Thompson wrote:
> Andy Leighton wrote in news:[email protected]:
>
> >> An error rate of 1% represents a non error rate of 99%.

> >
> > Sure. It would equate to over 650,000 incorrect profiles if there
> > was a mandatory database. I don't think you can just hold up your
> > hands and say it is right most of the time.

>
> Um, I think I've misunderstood you. Wouldn't a 1% error rate mean it *was*
> right most of the time (about 99% of the time, I s'pose).


No. This is where Bayesian stats comes in.

If a test has a false positive rate of 1%, ie is right 'most of the
time', and 1 in 10 million is the criminal, what is the chance of a
positive test being the criminal?

simple words (I won't even need to quote bayes theorem).

>From that 10 million people , 1% would be falsely identified positive

(-1 for the pedantic - we'll assume the real criminal is always tested
positive). That is 100,000 people.

So 99,999 times out of 100,000 the test is wrong. That is why a test
that is right pretty much all of the time is also invariably
unreliable.

The IT technology is irrelevant. The maths is at least 100 years old
and well understood [1] (Bayes theorem itself is 300 years old). It
does not change. The numpties who try to implement something they do
not understand is a problem.

...d

[1] This means that there exist a finite number of people who
understand it well. It does not mean that in general people will
understand it well, though it shouldn't be beyond the ken of a
reasonable GCSE or Standard grade maths candidate.
 
On 2006-11-23, Sniper8052(L96A1) <[email protected]> wrote:
> David Nutter wrote:
>
>> I think you are confusing two issues: the desire of police officers to do
>> their job without excessive bureaucracy and the desire of society to limit
>> police powers to reduce the impact of abuse, whether by individual
>> corruption or the invidious effects of particular laws (e.g. stop and
>> search).

>
> There is not now, nor has there ever been a general police power to stop
> and search. This power to search MUST be exercised on grounds that
> would be reasonable to an impartial third party.


Like this you mean?

http://gizmonaut.net/bits/suspect.html

This doesn't seem remotely reasonable to me, though maybe I'm not impartial
enough. For a more recent example see the knife scanner trials (Operation
Shield?) in stations - nominally voluntary but with the obvious effect that
anyone refusing to be scanned was likely deemed to be "acting suspiciously" and
searched, yielding plenty of stabbers and other nasty things. This shows two
things, firstly that the Shield trials showed more about the effects of a
decent police presence than the scanner technology and secondly that police
can usually find a reason to pass the time of day with any citizen should
they wish to do so.

While such powers are indeed necessary for efficient (as you put it)
policing, great care should be taken when exercising them and even more when
extending them. 'Sus' and similar practices in NI undoubtedly were a major
factor in the 80s riots and the Troubles respectively. The [ab]use of those
powers and the consequences thereof cannot be laid at the door of the
'civil liberties' brigade, since they argued against them.

*snip*

> The continued use of the term Stop and Search is misleading and
> devisive. It is used by the press and others who wish to raise distrust
> of the service by being deliberatly emotive and provocative.


Yet abuse and to a lesser extent mere use of such powers is divisive and
provocative all by itself. Interesting though it is, this discussion is
beside the main point that I was trying to make: when creating new powers
for the police or any other organ careful attention should be paid to what
happens when that power is abused and to the likelihood of any such abuse.
This is very difficult to do well, but that doesn't mean we shouldn't try.

>
> While these two concerns may overlap in some cases, they are not
>> the same. My facile interpretation: the police want certain laws largely for
>> their convenience (e.g. 90 days detention) whereas wider society wishes to
>> be safe from overbearing policing and both want to control criminality.

>
>
> Wider society?
>
> Do you mean Mr and Mrs Average?


No, I literally mean 'wider society' - that collection of different social
groups often with massively different social mores but with some shared
values. Mr and Mrs Average don't really exist in that context, if at all.
They perhaps typify one of those groups, no more.

>
> I can tell you, in my experience, what Mr+Mrs Average, with their two
> children and dog want. They want to be able to live their lives and
> raise their children in safety, with an efficient police, fire, health
> and social service and without the fear of being subjected to threats
> from groups of any faction. In my experience Mr+Mrs Average don't give
> a fig about how that comes about, just that it does.
>
> I agree that both parties, the public and criminal/suspects, must have
> adequate protection. I do not see that Mr+Mrs Average want the
> laws/powers that could protect the majority to be rendered ineffective
> or impossible to administer by the vocal minority of do-good activists
> who have their own, unelected, knife to grind. If they feel they are so
> right then they should get elected to parliament.


I take your point to a certain extent but find the idea that civic discourse
starts and ends at the ballot box every few years peculiar. Assuming for a
moment that 'Mr and Mrs Average' do exist, they may not care per se how
their lifestyle is kept safe but will certainly care if overzealous policing
either affects them directly or causes such disruption that their way of
life is affected by the fallout.

JOOI, have you taken a flight recently?

> We as a group decry BHit and their ilk as being unrepresentative of the
> majority cycling group. Why should Mr+Mrs Average be subject to a lobby
> group surplanting their elected groups authority to pass and administer
> legislature?


You are correct to point out the similarity between BHIT, who promote an
officious imposition by the state based on highly questionable evidence, and
elements of the police who promote...er...officious impositions by the state
based on highly questionable evidence (ID cards, 90 days and magical
fingerprint readers). Ultimately we're all still more-or-less free to argue
the point according to our personal lights, thank goodness.

>> In this case, the police apparently want a handy way to check drivers'
>> identities to save time and paperwork. Even assuming the fingerprinting
>> technology works properly a significant number of drivers will not be on the
>> fingerprint database (at least at first) and so will still need to be taken
>> to the station or trusted to produce their documents later. So, why not just
>> legislate to require the carrying of licence and insurance documents, as
>> happens in other European countries? That would solve the specific problem
>> here without requiring any gadgetry.

>
> I do not know but I would make a suggestion that once we have a national
> ID scheme the system will be used to help ensure that the person
> presenting the card is the person it proports to represent. Whilst not
> foolproof it makes copying the card more difficult.


That's a whole other argument. Nevertheless, even if the system works as
advertised (doubtful, for a whole host of reasons) it's still a decade away
and consequently doesn't help to solve the immediate problem that these
readers puport to solve (false driver details).

>>
>> No, I think the agenda here is something different. Given the history of the
>> DNA database, I think these devices are to be used as a convenient method of
>> snaffling lots of fingerprints into the database initially without explicit
>> legal authority. Such a collection is undoubtedly considered mighty useful
>> in some quarters but I personally believe the costs to be too high to
>> proceed, especially in such an underhand manner.

>
> I doubt that, a significant amount of information has to be entered into
> the database to have the prints acceptable. Also one fingerprint is not
> a large amount of use in abstract identification of offenders where as
> checking a residual fingerprint against a known set/subset is.


Yet exactly this situation occurred with the DNA database, legalised
retrospectively by the 2001 Criminal Justice and Police Act. Efficacy or
legality of the collection process or resultant data will carry little
weight with those who could sanction such an imposition.

> I have very strict views on policing and how it should be done. I have
> stated my view on many occasions but it boils down to,
>
> 'Keeping a straight bat and playing the game...'
>
> Justice is not always about punishing the offender. Justice, in its
> abstract form, is about the adminisration of the wishes of society to
> live free from fear of whatever criminal activity the offender has
> commited. It is the confusion between protecting society and punishing
> the offender that many do not grasp.


Nothing I disagree with there.

>
>>... our remaining liberties are the only thing we can use to protect
>> ourselves from being tyrannised, whether by criminals or an overbearing
>> state. All other desirable societal structures are built on these, whether
>> governance/policing by consent or any meaningful concept of equitable
>> justice. If you weaken them to facilitate vengeful chastisement of minor
>> wrongdoers you risk the very basis of the society you puport to protect.

>
> I do not advocate vengeful chastisement and I never have. Ever. I
> advocate a strong system of laws, administered fairly, governed by legal
> restraints and passed by our elected representatives.


You spoke of the 'pendulum of justice' which has been 'swung [...] away from
the victim' and I assumed that you were talking about Reid's "rebalancing"
act. This was careless of me and I apologise; your point of view is more
subtle than that and I fear we may actually be in violent agreement here!

> If you weaken the ability of the law to be administered then you harm
> society. In my book "...the needs of the many outweigh the needs of the
> few" but the few must be legaly protected within that system.


Fair enough as a utilitarian philosophy, but surely you agree that this
ability to adminster the law is likewise weakened when excessive and easily
abusable powers are granted, or similarly when the police/security services
are seen to be above the law they profess to uphold?

Policing is all part of governance by consent.

> All to often though, in my opinion, the vocal few seek not to provide a
> legal safety but to destroy the purpose of the law and its
> administration for whatever reason.
>
> If I felt that I could not work freely and fairly I would not be a
> police officer. I did not spend 13 years of my life serving my country
> to supress the freedoms I risked my life to preserve.


Commendable, but there are plenty of stark lessons from history here:
briefly that in the right situation it's very easy for good men to do
extraordinary evil if that is what the system demands. It is easier for such
demands to become commonplace when the organs of the state have great
discretionary powers and little oversight from the governed. We're not there
(yet) but I and it would seem others become concerned everytime we (or our
elected representatives :)) take us a step towards that sort of dystopia.

Regards,

-david
 
On 24 Nov 2006 22:32:39 GMT, David Nutter <[email protected]> wrote:
> On 2006-11-23, Sniper8052(L96A1) <[email protected]> wrote:
>> David Nutter wrote:
>>
>>> I think you are confusing two issues: the desire of police officers to do
>>> their job without excessive bureaucracy and the desire of society to limit
>>> police powers to reduce the impact of abuse, whether by individual
>>> corruption or the invidious effects of particular laws (e.g. stop and
>>> search).

>>
>> There is not now, nor has there ever been a general police power to stop
>> and search. This power to search MUST be exercised on grounds that
>> would be reasonable to an impartial third party.

>
> Like this you mean?
>
> http://gizmonaut.net/bits/suspect.html
>
> This doesn't seem remotely reasonable to me, though maybe I'm not impartial
> enough.


Or even the widespread use and misuse of section 60 in Scotland for the G8.

> For a more recent example see the knife scanner trials (Operation
> Shield?) in stations - nominally voluntary but with the obvious effect that
> anyone refusing to be scanned was likely deemed to be "acting suspiciously"
> and
> searched, yielding plenty of stabbers and other nasty things.


Except it didn't. During the trial period (8 weeks) in London some 1148
people were stopped out of 9728 total. Only 48 were for possession of
offensive weapons. There were 59 arrests for other reasons. In
additional usage it has often had even lower success rates - for example
in Largs 240 people were stopped an no knives found. Interestingly the
modus operandi also included sniffer dogs - in an operation intended
to target knife crime.

It is likely that normal policing practices, stopping and searching those
whom the police have reasonable grounds for suspecting that they are
carrying knives, would have caught similar numbers without searching
as many innocent individuals and would have cost massively less.

Interestingly there was a recent report in the Observer of a man who
refused to be scanned. He was backed up by a few other passengers who
also refused to be scanned. His argument was that the police did not
have reasonable grounds. After a short discussion the police declined
to take any action.

--
Andy Leighton => [email protected]
"The Lord is my shepherd, but we still lost the sheep dog trials"
- Robert Rankin, _They Came And Ate Us_
 
Mark Thompson wrote:
> Andy Leighton wrote in news:[email protected]:
>
>>> An error rate of 1% represents a non error rate of 99%.

>>
>> Sure. It would equate to over 650,000 incorrect profiles if there
>> was a mandatory database. I don't think you can just hold up your
>> hands and say it is right most of the time.

>
> Um, I think I've misunderstood you. Wouldn't a 1% error rate mean it
> *was* right most of the time (about 99% of the time, I s'pose).


A 1% error rate would make it useless as sole eveidence and only of us as
supportive evidence.

See: http://en.wikipedia.org/wiki/Prosecutor's_fallacy


--
Andy Morris

AndyAtJinkasDotFreeserve.Co.UK

Love this:
Put an end to Outlook Express's messy quotes
http://home.in.tum.de/~jain/software/oe-quotefix/



--
Posted via NewsDemon.com - Premium Uncensored Newsgroup Service
------->>>>>>http://www.NewsDemon.com<<<<<<------
Unlimited Access, Anonymous Accounts, Uncensored Broadband Access
 
AndyMorris wrote in news:[email protected]:

> A 1% error rate would make it useless as sole eveidence and only of us
> as supportive evidence.


Yebbut I didn't think anyone was arguing that.
 
On 24 Nov 2006 05:20:53 -0800,
[email protected] <[email protected]> wrote:
>
> Anthony Jones wrote:
>> [email protected] wrote:
>> > If I were advocating DNA as the sole criteria for guilt I would be
>> > forced to agree with you in the intrests of being fair to the argument.
>> > I am not advocating that though. I am advocationg narrowing the
>> > unknown from the population to a number more likely to have commited
>> > the offense and then having a full investigation.

>>
>> Yet we already have the courts convicting solely on the basis of DNA
>> evidence:
>>
>> http://en.wikipedia.org/wiki/Regina_versus_Denis_John_Adams
>>
>> Advocating a system that might work in theory is not the same as arguing
>> that it would work in practise.
>>
>> Anthony

>
> Which could equally be used to illustrate how the jury trial was no
> safeguard of justice and the public. It is not a reason to throw out
> the baby with the bath water.
>
> There have always been miscarriages of justice and there always will be
> no matter what system is used.
>

But a DNA database will just lead to more miscarriages of justice (and
hence the guilty escaping punishment)

If you have someone who was spotted at the scene, victim identifies, has
a motive etc, etc, and then you do a DNA test and find that their DNA
matches then it's icing on the cake. But you don't need a database to do
that.

If instead you trawl the database for a match and then look for other
evidence that support that - someone might have seen someone who looks
like the person who's dna matched, the victim picked them out from an
identity parade where they were only present because their DNA matched
in the first place then you are basically only using "bad" forms of
identification and the fact that the DNA matched is useless.

DNA recovered from a scene is usually of low quality. Let's assume a
false positive rate of 0.01%. Consider the following scenario - a crime
committed - say a rape. The police decide to do a trawl of all 50000
"local" men. Thanks to a crystal ball we happen to know that the perp
was actually not local. What is the probability that _exactly_ one of
those 50000 men match the DNA profile?

It's about 3.5% or just over 1 in 30. For comparison, the probability
that none of them match is 0.7%. 2 is 8.4%, 3 is 14%, 4 and 5 is 18%.

So you are likely to focus your efforts down on maybe up to a dozen
innocent men[1]. One of them is bound to come up matching on some other
"evidence". Indeed, what some bright spark will do is put the 6 matches
in an identity parade with another 6 men - if the victim picks one of
the matches he's "obviously" guilty

Read the story of John Munden (a police officer) who did time because of
odds of around 1/3000 (not dna) even if we assume pure chance and not an
insider job.

Tim.

[1] Actually probably a few less than this.

--
God said, "div D = rho, div B = 0, curl E = - @B/@t, curl H = J + @D/@t,"
and there was light.

http://tjw.hn.org/ http://www.locofungus.btinternet.co.uk/
 
Andy Leighton wrote:
> On 24 Nov 2006 22:32:39 GMT, David Nutter <[email protected]> wrote:
>
>>On 2006-11-23, Sniper8052(L96A1) <[email protected]> wrote:
>>
>>>David Nutter wrote:
>>>
>>>
>>>>I think you are confusing two issues: the desire of police officers to do
>>>>their job without excessive bureaucracy and the desire of society to limit
>>>>police powers to reduce the impact of abuse, whether by individual
>>>>corruption or the invidious effects of particular laws (e.g. stop and
>>>>search).
>>>
>>>There is not now, nor has there ever been a general police power to stop
>>>and search. This power to search MUST be exercised on grounds that
>>>would be reasonable to an impartial third party.

>>
>>Like this you mean?
>>
>>http://gizmonaut.net/bits/suspect.html
>>
>>This doesn't seem remotely reasonable to me, though maybe I'm not impartial
>>enough.

>


I can see why they challenged you in the first place, given the date and
your behaviour at a particularly sensitive time. After that it just
becomes bizarre.

The Terrorism Act 2000 was used in your case. As you know that is not
what is normally referred to as 'stop and search' and gives officers the
power to stop any person for the purposes of a terrorism search.
Grounds for that search are not required but the act must be in force in
the area at the time. This is authorised by the Home Secretary on
application of a chief constable. The only police area that has a
continual authorisation is the City of London Police and they have to
re-submit their application every month to six weeks. They are heavily
monitored to make sure the authority is not abused.

The Terrorism Act 2000 is not a General Power therefore and in the
context of the original approach to you the officers actually extended
more information than they needed to. As for the rest I take you at
your word and leave it at that.

>
> Or even the widespread use and misuse of section 60 in Scotland for the G8.
>


Scottish law is different to UK but as I am aware an exclusion zone was
imposed which led to confrontation when 'activists and protesters' tried
to breach the security. Had they behaved in a less confrontational
manner perhaps things would have been different. I doubt either party
can claim a high ground but in general police are reactive to disorder
or civil disobedience. That those actions were proportionate depends on
the viewpoint and agenda of the participants of either side.

>
>>For a more recent example see the knife scanner trials (Operation
>>Shield?) in stations - nominally voluntary but with the obvious effect that
>>anyone refusing to be scanned was likely deemed to be "acting suspiciously"
>>and
>>searched, yielding plenty of stabbers and other nasty things.


Sticking plaster to a problem which exists but not a real solution in my
opinion. Deployment to nightclub queues might be more effective in
finding weapons and deterring them from being carried. But as a tool
for altering the conception that one will always get away with it there
is some validity to its deployment on a random basis across time and
location.

Sniper8052


All,

Whilst, it may be, if and but are all fine conjectures about a possible
miscarriage of justice the point which all of you are missing is that
there are miscarriages of justice now and that no system can be perfect.
That DNA might lead to a miscarriage of justice on occasion if it were
used as the sole basis of investigation is a concern but it is not a
reason to deny the positive benefits its linkage to a national ID card
scheme could bring to the 'super majority' of the population.

To deny that safeguard to society because of that concern is in my mind
criminal. DNA represents the biggest step forward in the prevention and
detection of crime since fingerprinting and blood typing. If there are
legitimate concerns then those need to be talked about in a facilitating
debate so that the tool may be used within a legal and equitable
framework.

I frankly don't understand the statements about Bayesian stats. Are you
saying that because it fails to find a single match, but may find a
number of false matches within 1%, that it is in fact 99% inaccurate?

Can you explain this to me so that I can understand your argument. Use
simple terms and small numbers please.

Sniper8052
 
On Sat, 25 Nov 2006 19:17:12 GMT,
Sniper8052(L96A1) <[email protected]> wrote:
> Andy Leighton wrote:


>> Or even the widespread use and misuse of section 60 in Scotland for the G8.
>>

>
> Scottish law is different to UK but as I am aware an exclusion zone was
> imposed which led to confrontation when 'activists and protesters' tried
> to breach the security.


But it was English policemen as well as Scottish policemen up there,
however as far as I know that part of the Criminal Justice and Public
Order Act 1994 applies north of the border as well as south.

Section 60 was used as a justification for many searches and not just
in the exclusion zone - there was a section 60 order in place for the
whole of Edinburgh. At the time Jon Senior wrote about when he was
stopped and searched when out on a normal bike ride whereas cars were
not (see almost on topic). As you should know, but some other people
probably do not, section 60 is about searching for weapons and other
dangerous instruments. It gives police the power to do so without any
suspicion at all. As for misuse there were some reported cases of police
arresting people for not giving names and addresses when stopped under
a s60.

I'm not trying to rag on the police or you in particular but these
kind of powers really do annoy a lot of people because they can be
applied on a constable's whim without need for reasonable grounds.
In effect they treat everyone as a potential suspect.

--
Andy Leighton => [email protected]
"The Lord is my shepherd, but we still lost the sheep dog trials"
- Robert Rankin, _They Came And Ate Us_
 
Sniper8052(L96A1) wrote:

>
> I frankly don't understand the statements about Bayesian stats. Are you
> saying that because it fails to find a single match, but may find a
> number of false matches within 1%, that it is in fact 99% inaccurate?


Hi Sniper,

No this does not say that. You can have a test that is pretty accurate
but still makes mistakes. It may be that if you actually catch the
culprit there is a very high chance that the test will 'identify' him.
For example if you have 6 suspects there is probably a good change that
only his DNA will match. And even if you get two matches, good police
work should be able to figure out which is the best bet.

The problem is that if you're just trolling through a database looking
for anybody recorded there, apparently it is likely that you'll get
hits on that 1% just by change. So there are the police saying we know
it's one of these when the real culpit is on the Costa del Sol laughing
his head off and the police are wasting their time trying to 'prove'
one of those 12 or so people are guilty.

In medicine it is called a false positive.

As a somewhat different example, lets say we have an early detection
test for colon cancer that detects 95% of all colon cancers among those
tested who really have colon cancer but which also mistakenly
identifies 25% of those people tested who do not have colon cancer as
having the disease. So almost all the people with colon cancer are
identifed and have an excellent chance of survival. However a number
of healthy people get one big scare.

This means that we have a false positive rate of 25% and a false
negative of 5%. I don't know if a statistician would agree here but
basically I see the rates as independent of each other.


>
> Can you explain this to me so that I can understand your argument. Use
> simple terms and small numbers please.
>
> Sniper8052



Baysian stats are not really simple nor are they too complicated.
Have a look at http://en.wikipedia.org/wiki/Bayesian_inference. It is
'fairly' clear. I just glanced at it but I think the simple examples
may help.
 
Sniper8052(L96A1) wrote:

> I frankly don't understand the statements about Bayesian stats. Are you
> saying that because it fails to find a single match, but may find a
> number of false matches within 1%, that it is in fact 99% inaccurate?
>
> Can you explain this to me so that I can understand your argument. Use
> simple terms and small numbers please.



Let us take the two examples quoted previously. We have a DNA test that
has a uniqueness value of 1 in a million. By that I mean that there is
a 1 in a million chance that any two individuals will have the same DNA
match.

A crime has been committed and evidence is collected.

In Scenario A DNA is collected and processed. A trawl through the DNA
database identifies a single individual with that DNA profile. What is
the probability that the individual committed the crime?

The value we are trying to find is the probability P that the
individual is the criminal, given the DNA match.
This is expressed as P(C|D)

To calculate this we need to know the prior probability that the person
is the criminal, P(C), the prior probability of the person having that
DNA, P(D) and the probability of having the DNA if you are the
criminal, P(D|C).

These are put together as P(C|D) = P(D|C) P(C)/P(D)

In a population of 10 million, you would expect 10 people to have the
DNA match.

P(C|D) = 1 * .0000001 /.000001 or .1

In case B there is evidence to rule out all but 6 suspects (excluding
the DNA evidence cos it isn't yet available)

Now we get a positive DNA match. to one suspect. What is the
probability it is the killer?

Put in the right numbers

P(C|D) = 1*6/6
or a rather strong probability.

It's late and I have made some errors in the calculation of P(D) which
are to do with the distribution of the samples and why the numbers
shouldn't quite equal 1, but I'll leave those as an exercise for the
reader.

In short, if you trawl the DNA database, it will give you an idea of
how many people in the catchment area would have the same DNA
signature. Once you have used it to get that pool of suspects (most of
whom are not identified as they are not on the police DNA database) you
cannot use that information to discriminate any further amongst the
suspects. To give a crude example:

The killer wore size ten shoes.
Here is a man wearing size ten shoes.
Is he the killer?

I would hope you need far mroe information than that. It is also the
case that we do not know what the allele distribution of the markers
used for DNA testing in local populations is. Where communities are
more insular one would expect the stats to be skewed from a broad UK
average. Likewise, no-one knows what the distributions are by race.

Hope this helps.

...d
 
David Martin wrote:
> Sniper8052(L96A1) wrote:
>
>
>>I frankly don't understand the statements about Bayesian stats. Are you
>>saying that because it fails to find a single match, but may find a
>>number of false matches within 1%, that it is in fact 99% inaccurate?
>>
>>Can you explain this to me so that I can understand your argument. Use
>>simple terms and small numbers please.

>
>
>
> Let us take the two examples quoted previously. We have a DNA test that
> has a uniqueness value of 1 in a million. By that I mean that there is
> a 1 in a million chance that any two individuals will have the same DNA
> match.
>
> A crime has been committed and evidence is collected.
>
> In Scenario A DNA is collected and processed. A trawl through the DNA
> database identifies a single individual with that DNA profile. What is
> the probability that the individual committed the crime?
>
> The value we are trying to find is the probability P that the
> individual is the criminal, given the DNA match.
> This is expressed as P(C|D)
>
> To calculate this we need to know the prior probability that the person
> is the criminal, P(C), the prior probability of the person having that
> DNA, P(D) and the probability of having the DNA if you are the
> criminal, P(D|C).
>
> These are put together as P(C|D) = P(D|C) P(C)/P(D)
>
> In a population of 10 million, you would expect 10 people to have the
> DNA match.
>
> P(C|D) = 1 * .0000001 /.000001 or .1
>
> In case B there is evidence to rule out all but 6 suspects (excluding
> the DNA evidence cos it isn't yet available)
>
> Now we get a positive DNA match. to one suspect. What is the
> probability it is the killer?
>
> Put in the right numbers
>
> P(C|D) = 1*6/6
> or a rather strong probability.
>
> It's late and I have made some errors in the calculation of P(D) which
> are to do with the distribution of the samples and why the numbers
> shouldn't quite equal 1, but I'll leave those as an exercise for the
> reader.
>
> In short, if you trawl the DNA database, it will give you an idea of
> how many people in the catchment area would have the same DNA
> signature. Once you have used it to get that pool of suspects (most of
> whom are not identified as they are not on the police DNA database) you
> cannot use that information to discriminate any further amongst the
> suspects. To give a crude example:
>
> The killer wore size ten shoes.
> Here is a man wearing size ten shoes.
> Is he the killer?
>
> I would hope you need far mroe information than that. It is also the
> case that we do not know what the allele distribution of the markers
> used for DNA testing in local populations is. Where communities are
> more insular one would expect the stats to be skewed from a broad UK
> average. Likewise, no-one knows what the distributions are by race.
>
> Hope this helps.
>
> ..d
>



All,


John, David,

Thank you for pointing me toward the information I needed and thank you
for taking the trouble to explain the problem to me.

I can now see the concern that is being expressed. I will explain my
conception of that so that you can gauge that I have understood you
correctly.

Bayes Theorem 1 states that if I use my database to conduct an initial
search for the offender then I lower the reliability of the result being
accurate as the knowledge (prior) of the person investigated being
guilty relates to the 1% false positive I.E.: 1:650,000

So that although I now have a smaller section of the community to
conduct my search for the true offender my initial search states that
all of them could be the offender and DNA can be of no further use to me.

True that I can then use other investigative techniques to narrow the
new sub set,

Sex, age, description, proof of alibi etc but on a broad basis my
liklyhood of selecting the correct offender from this group is

1/650000

assuming they all were the correct sex, age and description.

Bayes Theorem 2 states that if I use my database on a group after I have
selected a number of suspects my result will be more accurate; to the
point that it will almost invariably always be right.


This presents a dilemma. Given that Bayes Theorem is correct then the
outcome of Theorem 2 is that the offender may never be caught but if
they are and we then apply a DNA test then they are guilty - PDQ.

The outcome of Theorem 1 is that there is an initial liklyhood that one
of the 650,000 hits is the correct person but that after this I will
have to rely on proving my case through other means.

If I cannot disprove the alibi, 'he was at home in bed', I have no means
of improving my case and the offender has to be found not guilty.

Given the above I still prefer the use as in Theorem 1. The offender is
in there somewhere but with legal safeguards. I still think that DNA
used in this method would offer a significant safeguard against many
offences.

Unfortunately we do not have a Bayes Theorem 3 to represent the likely
distribution of the offender being guilty under the present system
without DNA. Can that even be calculated?

If Bayes Theorem 1 then represented a significant improvement it would
then show its validity as a basis of qualified investigation; given the
correct legal safeguards.

As I see it we should be discussing legal safeguards rather than not
having a national population database.

Many thanks

Sniper8052

I am going to bed now.........Zzzzzzzzz
 
Sniper8052(L96A1) wrote:
>
> Bayes Theorem 1 states that if I use my database to conduct an initial
> search for the offender


This is the first, and possibly most worrying, error. You are assuming
that because DNA was found, or at least purported to have been found, at
the scene, that it must belong to the offender.

You must first prove that the DNA in question could not have got into
'the frame' by any other means, and that those claiming to have found it
there actually did. Then, having established that the DNA was actually
there, you need to be sure that the only possible way it could have got
there was if its 'owner' was personally present at the scene. Then you
know you have the DNA of a potential witness, and maybe even a suspect.
DNA cannot prove who the 'offender' was though.

--
Matt B
 
Sniper8052(L96A1) wrote on 26/11/2006 11:44 +0100:
>
> If I cannot disprove the alibi, 'he was at home in bed', I have no means
> of improving my case and the offender has to be found not guilty.
>
> Given the above I still prefer the use as in Theorem 1. The offender is
> in there somewhere but with legal safeguards. I still think that DNA
> used in this method would offer a significant safeguard against many
> offences.
>


Wrong. You don't even know if the offender is in your 650,000 your way
and at least one of them is bound to be unable to pass your other
subjective tests even though they are innocent e.g. they have no alibi
for where they were at the time. The beauty of the Bayes 2 scenario is
that you don't need a national database. Once you've identified someone
by the other characteristics you can take their DNA and test it against
the evidence.

Lets take another slightly more controversial proposal. Lets have a
database of everyone by race. Then when a person commits a crime and
the evidence from the CCTV is that he is Chinese, you can search your
database for all the people that are Chinese. Then by your rules, all
you have to do is find a Chinese person who can't show an alibi and
looks like the person on the CCTV. Do you think that will have caught
you your criminal?

The correct way to do it is use your normal investigative skills to
identify a list of suspects from whom you can eliminate those that don't
look like the person on the CCTV. Then you have a small number of
people left for whom you have to show by other means which one it is.
That way your false positive rate is very much lower.

--
Tony

"Anyone who conducts an argument by appealing to authority is not using
his intelligence; he is just using his memory."
- Leonardo da Vinci
 
Executive summary; broadly correct but with wrinkles:

Sniper8052(L96A1) wrote:
> Bayes Theorem 1 states that if I use my database to conduct an initial
> search for the offender then I lower the reliability of the result being
> accurate as the knowledge (prior) of the person investigated being
> guilty relates to the 1% false positive I.E.: 1:650,000


It also a priori assumes that the person who committed the offence is
in the database. Note that I didn't mention false positive or false
negative rates, assuming a perfect test.

> So that although I now have a smaller section of the community to
> conduct my search for the true offender my initial search states that
> all of them could be the offender and DNA can be of no further use to me.


You also have to know that all the potential criminals are in your
database. That means taking a DNA sample of everyone who enters the
country, otherwise you cannot be sure that your reduced search space
includes the target.


> True that I can then use other investigative techniques to narrow the
> new sub set,
>
> Sex, age, description, proof of alibi etc but on a broad basis my
> liklyhood of selecting the correct offender from this group is
>
> 1/650000
>
> assuming they all were the correct sex, age and description.


No. if you identify a group of size n where you know the group must
contain your target, the chance of picking your target from that group
at random is 1/n

> Bayes Theorem 2 states that if I use my database on a group after I have
> selected a number of suspects my result will be more accurate; to the
> point that it will almost invariably always be right.


You don't use the database except to establish the prior and
conditional probabilities. It is the selectivity of the test with
respect to the size of the group that matters.

> This presents a dilemma. Given that Bayes Theorem is correct then the
> outcome of Theorem 2 is that the offender may never be caught but if
> they are and we then apply a DNA test then they are guilty - PDQ.


Precisely. Or one should say, most likely to be guilty if the DNA is
not used to select the suspect in advance.

> The outcome of Theorem 1 is that there is an initial liklyhood that one
> of the 650,000 hits is the correct person but that after this I will
> have to rely on proving my case through other means.
>
> If I cannot disprove the alibi, 'he was at home in bed', I have no means
> of improving my case and the offender has to be found not guilty.
>

Yes. That is correct.

> Given the above I still prefer the use as in Theorem 1. The offender is
> in there somewhere but with legal safeguards. I still think that DNA
> used in this method would offer a significant safeguard against many
> offences.


It has the fallacy that I could commit an offence and you wouldnever
catch me because I am not in your database.

> Unfortunately we do not have a Bayes Theorem 3 to represent the likely
> distribution of the offender being guilty under the present system
> without DNA. Can that even be calculated?


There are models one could use. The 'usual suspects' model is
frequently used. Whether that is because it is easier to identify
repeat offenders rather than new offenders is open to debate.

> If Bayes Theorem 1 then represented a significant improvement it would
> then show its validity as a basis of qualified investigation; given the
> correct legal safeguards.
>
> As I see it we should be discussing legal safeguards rather than not
> having a national population database.


A national DNA database would have missing data, particularly from
those groups least likely to comply with the law. It would have
erroneous data. A lower bound estimate for the quality of the database
could be obtained by looking at the scientific DNA databases. There are
many reasons why motivated scientists should put together a high
quality database compared to relatively poorly paid technicians who
would be employed to meet the processing needs of such a project. From
my experience as a professional DNA database wrangler (yes it is what I
do from an informatics point of view) I am seriously unconvinced that a
national DNA database could provide anything like the level of
confidence I would want. You would have to retest every identified
suspect to verify their database entry as an absolute minimum.

Whilst one can see the desirability in abstract terms for a (what was
that film where he keeps swearing to get enough endorsement dockets to
use as toilet paper) type society where you cannot do wrong, in
practice it has to be run by the civil service and administered by
unmotivated non-specialists.

...d
 
In article <[email protected]>
David Martin <[email protected]> wrote:
<snip>
>
> Whilst one can see the desirability in abstract terms for a (what was
> that film where he keeps swearing to get enough endorsement dockets to
> use as toilet paper)


Demolition Man
:)
 
Sniper8052(L96A1) wrote:
> David Martin wrote:
> > Sniper8052(L96A1) wrote:
> >
> >
> >>I frankly don't understand the statements about Bayesian stats. Are you
> >>saying that because it fails to find a single match, but may find a
> >>number of false matches within 1%, that it is in fact 99% inaccurate?
> >>
> >>Can you explain this to me so that I can understand your argument. Use
> >>simple terms and small numbers please.

> >
> >
> >
> > Let us take the two examples quoted previously. We have a DNA test that
> > has a uniqueness value of 1 in a million. By that I mean that there is
> > a 1 in a million chance that any two individuals will have the same DNA
> > match.
> >
> > A crime has been committed and evidence is collected.
> >
> > In Scenario A DNA is collected and processed. A trawl through the DNA
> > database identifies a single individual with that DNA profile. What is
> > the probability that the individual committed the crime?
> >
> > The value we are trying to find is the probability P that the
> > individual is the criminal, given the DNA match.
> > This is expressed as P(C|D)
> >
> > To calculate this we need to know the prior probability that the person
> > is the criminal, P(C), the prior probability of the person having that
> > DNA, P(D) and the probability of having the DNA if you are the
> > criminal, P(D|C).
> >
> > These are put together as P(C|D) = P(D|C) P(C)/P(D)
> >
> > In a population of 10 million, you would expect 10 people to have the
> > DNA match.
> >
> > P(C|D) = 1 * .0000001 /.000001 or .1
> >
> > In case B there is evidence to rule out all but 6 suspects (excluding
> > the DNA evidence cos it isn't yet available)
> >
> > Now we get a positive DNA match. to one suspect. What is the
> > probability it is the killer?
> >
> > Put in the right numbers
> >
> > P(C|D) = 1*6/6
> > or a rather strong probability.
> >
> > It's late and I have made some errors in the calculation of P(D) which
> > are to do with the distribution of the samples and why the numbers
> > shouldn't quite equal 1, but I'll leave those as an exercise for the
> > reader.
> >
> > In short, if you trawl the DNA database, it will give you an idea of
> > how many people in the catchment area would have the same DNA
> > signature. Once you have used it to get that pool of suspects (most of
> > whom are not identified as they are not on the police DNA database) you
> > cannot use that information to discriminate any further amongst the
> > suspects. To give a crude example:
> >
> > The killer wore size ten shoes.
> > Here is a man wearing size ten shoes.
> > Is he the killer?
> >
> > I would hope you need far mroe information than that. It is also the
> > case that we do not know what the allele distribution of the markers
> > used for DNA testing in local populations is. Where communities are
> > more insular one would expect the stats to be skewed from a broad UK
> > average. Likewise, no-one knows what the distributions are by race.
> >
> > Hope this helps.
> >
> > ..d
> >

>
>
> All,
>
>
> John, David,
>
> Thank you for pointing me toward the information I needed and thank you
> for taking the trouble to explain the problem to me.
>
> I can now see the concern that is being expressed. I will explain my
> conception of that so that you can gauge that I have understood you
> correctly.
>
> Bayes Theorem 1 states that if I use my database to conduct an initial
> search for the offender then I lower the reliability of the result being
> accurate as the knowledge (prior) of the person investigated being
> guilty relates to the 1% false positive I.E.: 1:650,000


There's only one theorem. I think you're just seeing it written in more
than one way.
However I think you have some of it.

However you need to know what the prior value is to use the theorem
which is David's case B Thus you might first want to say that from
experience 90% of the culprits in this type of crime are in my
catchment area. Then maybe make an assumption, again based on past
experience that 50% of the beggers have a prior arrest. Now you have
a good guess about the changes that the culprit is in the database.
>
> So that although I now have a smaller section of the community to
> conduct my search for the true offender my initial search states that
> all of them could be the offender and DNA can be of no further use to me.


Yes but in the DNA problem there is no a priori reason to think that
anyone in the database is guilty since we know we have a false postive
problem. See above about estimating the chances that they are in the
database.

>
> True that I can then use other investigative techniques to narrow the
> new sub set,
>
> Sex, age, description, proof of alibi etc but on a broad basis my
> liklyhood of selecting the correct offender from this group is
>
> 1/650000


I don't think so. We still have no idea if the culprit is in the
database. We are still going to get those 10 or 12 hits by chance.
>
> assuming they all were the correct sex, age and description.


Oh yes assuming that you have something to go on. If all you know is
that a crime was committed you have nothing to go on. If you have any
information to start with like the information that you have a male of
about 6 ft height then you can use the database to reduce the margin of
error. Remember though, there is still no guarantee the culprit is in
the data base but you might be able to elimnate everybody in the
database too.
>
> Bayes Theorem 2 states that if I use my database on a group after I have
> selected a number of suspects my result will be more accurate; to the
> point that it will almost invariably always be right.


Well not invariably right but much much more likely. And if you have
more information you can increase the accuracy. Six foot male in
database, no alibi for time of crime. Good! And so on.
>
> This presents a dilemma. Given that Bayes Theorem is correct then the
> outcome of Theorem 2 is that the offender may never be caught but if
> they are and we then apply a DNA test then they are guilty - PDQ.


Not that they are guilty but the likelihood that they are goes up.

> The outcome of Theorem 1 is that there is an initial liklyhood that one
> of the 650,000 hits is the correct person but that after this I will
> have to rely on proving my case through other means.


Yes but we still don't know the chance that the person is in the
database at all.
>
> If I cannot disprove the alibi, 'he was at home in bed', I have no means
> of improving my case and the offender has to be found not guilty.
>
> Given the above I still prefer the use as in Theorem 1. The offender is
> in there somewhere but with legal safeguards. I still think that DNA
> used in this method would offer a significant safeguard against many
> offences.


No, the problem is with the false positives. We know that there will
be 10-12 people tossed up by mistakes. What we don't know is the
probability that the culprit is in the database at all or if we will
get him in that 10-12 person group.

There is also the thing about a false negative. We might just miss
them even if they are in the database. However if you have good reason
to think that they are in the data base you can keep looking for more
evidence to further refine the estimate.
>
> Unfortunately we do not have a Bayes Theorem 3 to represent the likely
> distribution of the offender being guilty under the present system
> without DNA. Can that even be calculated?
>
> If Bayes Theorem 1 then represented a significant improvement it would
> then show its validity as a basis of qualified investigation; given the
> correct legal safeguards.
>
> As I see it we should be discussing legal safeguards rather than not
> having a national population database.


Possibly but the problem is that there is no guaranttee a) they are in
the database, b) that the database is correct, and c) that you really
will get them if they are in the database.

Using David's case B you have a good chance of identifying the culprit.


Using Case A means that you know you have accidentally identifed some
innocent people and _may_ have identified the culprit but you have no
idea of the odds on it.

John Kane, Kingston ON Canada

John Kane, Kingston ON Canada
 
John Kane wrote:

> > As I see it we should be discussing legal safeguards rather than not
> > having a national population database.

>
> Possibly but the problem is that there is no guaranttee a) they are in
> the database, b) that the database is correct, and c) that you really
> will get them if they are in the database.
>
> Using David's case B you have a good chance of identifying the culprit.
>
> Using Case A means that you know you have accidentally identifed some
> innocent people and _may_ have identified the culprit but you have no
> idea of the odds on it.


And caused resentment amongst those who have been pulled in from the
database trawl for questioning etc.

There are good reasons for a national DNA database. Crime prevention
might be erroneously thought to be one of them.

...d