Statistics and Misidentifications - The weeks Detail

[Michele]

Steve,

I agree that Bayes Theorem could be useful to our industry but I don’t agree that the application of it is as straight forward as you suggest. One of the main problems people have with applying Bayes Theorem is unintentionally using irrelevant information and presuming it’s relevant information. To prevent this from happening and to objectively peer review how it’s being used, everything needs to be clearly stated at the beginning.

For example, Bayes Theorem is looking for the probability of an event happening (I’ll call this ‘ a ’ to keep in line with your website) given that we have additional information (I’ll call this ‘ x ‘ since you have) and it’s written p( a / x ), which equals the probability of ‘ a ‘ happening given that we know that ‘x’ has occurred.

Here’s an instance that shows how simple and useful this can be.
Suppose a latent is found at my desk. The probability of if being mine is very high (given that we know I sit here) and the probability of it being Steve Horn’s is very low (given that we don’t think he’s ever been at my desk). Should a solid conclusion be based on this information? If that’s all the information we have, then forming a conclusion by mathematical probability may be the best conclusion we can arrive at. But if we have additional data (like physical characteristics in the print that we could use to do a comparison) then forming a conclusion based on limited data isn’t a scientifically sound practice, even if statistical probability is a scientific tool. Using mathematical chance and ignoring other relevant information is never a valid means of coming to a conclusion.

In this situation, I’m a little confused about what probability we are trying to find (a). I’m also confused about the information we already know (x). What are we looking for and what is the condition??

The probability of an error given an ID was made?
The probability of an error given an ID was made to the victim or other elimination prints?
The probability of an error given the ID is the only evidence against this person?
The probability of an error given she said it wasn’t hers?
The probability that Shirley is telling the truth given an ID was made (regardless of if it’s erroneous or not)?
The probability that Shirley is telling the truth given a valid ID was made?
The probability that Shirley is telling the truth given an erroneous ID was made?
The probability that Shirley wasn’t in the house?

On a superficial level, these may seem like the same thing but exactly how ‘a and x’ are defined is crucial to looking at the probability of it happening.

Some of the conversations in this thread are interesting but I don’t see the relevance (such as if a subject was wearing gloves). I’ve been involved in cases where the subjects wore gloves but still left a print (without removing his gloves). This may seem unlikely to those outside our discipline but once shown how this works it’s easy to understand. My point is that the inferences we come to will change based on how we understand the initial assumptions.

[Outsider]

Daktari,

I'll try and explain things in a similar way to your balls example.

There are 100 balls in a bowl, 90 white, 9 blue and 1 green. Let’s say we get a prize for picking a blue ball and a pie in the face for picking the green ball (and we get ignored for choosing a white ball). A blind person will have a 9 in 100 chance of getting a prize and a 1 in 100 chance of getting a pie in the face because they are picking at random (and a 90 in 100 chance of getting nothing). A colour-blind person can tell if a ball is white or coloured so they will have a 9 in 10 (90%) chance of getting a prize and a 1 in 10 (10%) chance of getting a pie in the face because they will chose a coloured ball. A person with good eyesight will have a 100% chance of getting a prize because they can see the balls clearly. It is interesting to note that a colour-blind person has a greater chance of getting a pie in the face than either a blind person picking at random, or a fully sighted person who never makes an error.

The police can only choose between fingerprint-identified people who agree that they deposited the print and those who deny having deposited the print. They cannot distinguish between the people who are denying because they have been correctly identified and are lying, and those who were misidentified and are telling the truth. They are like the colour-blind people choosing a ball.

The way to get colour-blind people to get closer to a 100% chance of getting a prize and closer to 0% chance of getting a pie in the face is to increase the ratio of blue balls to green balls in the bowl. In an ideal world we would make sure there are no green balls at all, but if we cannot do that we should make sure there is a plentiful supply of blue balls. Similarly if we cannot ensure that the error rate of fingerprinting is zero, we should only be doing fingerprint work in places that the chances of finding wrongdoers lying is relatively high. The way to do this is to wait till we KNOW that a crime has occurred, then go to the crime scene without delay and secure it.

The trick is to realise that Marion Ross’s house was not the scene of an offence of entering a crime scene without permission. There was not a jot of evidence before the Shirley McKie identification to suggest that anybody did this and no evidence has subsequently emerged. As I have said many times before, the Shirley McKie case was started from an identification rather than from a crime so the case (or more correctly “A” Shirley McKie case) could have happened during any fingerprint inquiry at any time anywhere so there is an unlimited opportunity for a misidentification to be at the heart of the case. In a normal crime-led case the opportunity for misidentification is limited because the crime limits the number of latents in the inquiry. Now, it would be reasonable for Daktari to say that in such an unlimited area and time period the opportunity for a police officer to enter a crime scene without permission is similarly unlimited. He or she would be right but we are in the situation of having to try to predict what is the ratio of these two chances. That is not good enough to make an accusation that will wreck someone’s life. Compare that with a crime scene where we KNOW that the criminal was present and the chances of him touching something and leaving an identifiable print are overwhelmingly higher that the chances of a misidentification in the few hundred latents that have a connection with the crime.

[Michele]

Steve,

Like I said before, lots of interesting scenario’s but it’s irrelevant information without knowing what problem we’re trying to solve. I don’t know if I made it clear or not but my question of “What is the probability we’re trying to find? And what is the condition we’re assuming to be true (what do we already know)?” was a question for you. I gave 8 possibilities, are one of these the problem that you’re solving for on this website
http://www.stevehornsc.pwp.blueyonder.co.uk/bayes.htm ??

[Daktari]

Thanks Q It's not often someone agrees with me, even if it's in part! But watch your back if McKie see it.
One slight quibble you say

Let's say the EVENT is a person entering my restaurant. You may find the probability of the event is higher around Noon to 1PM (people on their lunch hour) vs. 10 AM when everybody is working. The event doesn't change, but the probability changes over time to reflect the conditions.

The probabilty of a person entering a restaurant at a specific time is a discrete event with a discrete probalilty based on all sorts of factors.

[g.]

Daktari

Well I definitely agree in this instance it is a discrete value. The rate of value changed with the conditions of the scenario. For example, we might discuss 150 patrons per hour (lunch time) vs. 50 patrons per hour (morning tea time).

We can then ask questions, such as:

1) What's the probability I will have at least 10 patrons in the next minutes (during a lunch rush)

or what's the probability I will have exactly 30 patrons in the a.m.

2) or With what certainty can I say we will have 140 patrons over the lunch hour...

etc. So while the rate/event can be discrete, the probability density functions (pdf) associated can be continuous functions (such as in the case of a Normal distribution). You are correct though, that with respect to Poisson (as in my example) the pdf is discrete. But depending on the questions asked some discrete events (e.g. how many people will see Spiderman 3 at a certain movie theater) can have a continuous area associated with the distribution that needs to be integrated and solved using calculus to determine the probability (hence why people use the standard tables, or just normalize everything into easily tabulated Normalized scores (z scores).

Finally even a discrete value (such as the estimator in the example) can be often described as a random variable, with a mean value and variance associated with it. So as in the example of the rate of 150 people in the noon hour, there is variance and randomness associtted with it, since it is an estimator. The true value may vary daily, but over long periods of time the average is 150. And it's still a useful estimator. We see these types of random fluctuations in stocks, market value, etc. where those estimators definitely fluctuate with market bumps, but yet remain fairly constant when averaged.

As with anything, the rabbit hole can go as deep and as complicated as people want to make it...

So yes Daktari, "quibble seen, and raised". =) But point taken.

oh and it's g. (as in Glenn) not Q, as in 007)

Oh and yeah, I have to agree fully with Michele's points about stating what these priors are in Steve's assumptions. The concept of Bayes can be useful here, I just, like Michele, had problems with how it's set up. Her analogy of the latents at the desk is perfect. What is missing from the equation is the likelihood ratio of the evidence itself (minutiae in agreement, or other metric to calculate the Bayes' posterior). Sorry Steve, not getting the set up of these problems. I think you might be overcomplicating the situation. But I don't know, I am not a statitician, so if you can show me the math (I looked at the website, but it's just a bunch of assumed frequencies, without actually a likelihood ratio with respect to the evidence, the sensitivity/selectivity of the tests, etc. The meat of the equation is missing, as far as I can tell...but again, it's not my forte'.

Cheers all,

g.

[Outsider]

Hi Michele,

Sorry it took so long to reply to your points, life (work) got in the way.

lots of interesting scenario’s but it’s irrelevant information without knowing what problem we’re trying to solve

I agree we need to get this clear. The only thing that matters is this: During a court case which includes a fingerprint identification, or at the point of accusing somebody of lying as a result of a fingerprint identification (it might not become a court case), how sure are we that the identification is correct?

The only identifications that reach this point are those where the identified person cannot give an innocent explanation for being at the location, so it is after a non-random selection process. Only a small proportion of good identifications will be selected and these are the criminals but nearly all misidentifications will be selected because the misidentified person will almost certainly deny having deposited the print. So even if the identification error rate is tiny, that does not necessarily mean that the error rate at the point of making an accusation is tiny. At this point a denied identification can only come from two sources, a good ID and the person is lying or a misidentification. So the certainty is the ratio of these two probabilities.

The example I gave of criminals using gloves was just to illustrate a point that if there is next to no reasonable chance that the criminal has left a print, the frequency of prosecution will be low yet the (small) frequency of misidentification will be the same as any other case with a similar number of latets. At the point of an accusation of lying, the certainty or confidence level of the accusation is the ratio of these two frequencies (good ID and lying, to misID). So even though we make fewer accusations of lying if we know that there is a very small chance that the criminal left a latent, the chances of accusations being wrong will be higher than in cases where we know that there is a good chance that the criminal left a latent (this is the prior probability in Bayes Theorem). I assumed that leaving a fingerprint with gloves on was impossible, if that is not the case then maybe it is not such a good example to illustrate the point

Keep in mind that in both the gloved and non-gloved robberies there was at least a prior crime to limit the number of latent prints in every inquiry. In the Shirley McKie case there was no prior crime so there was an unlimited opportunity for the case to be started off from a misidentification.

Using Bayes Theorem to look at this is, in some ways, taking a sledgehammer to crack a nut (some people like sledgehammers). All you need to ask is this: Given that there was an unlimited opportunity for the case to be started from a misID and an unlimited opportunity for a police officer to enter a crime scene without permission and lie about it, what is the ratio of these two probabilities? To be confident enough to make an accusation that will wreck someone’s life and career, we would need to be sure that for every one misidentification that might happen, thousands of police officers must be entering crime scenes without permission who would lie about it if identified from a fingerprint. This is the sort of ratio that happens in normal crime scenes, thousands of criminals are depositing identifiable fingerprints for every one misID.

[Daktari]

Sorry Glenn, for misreading your name. But worse was to come I thought for a second the others were discussing Bayle's Theorem. You must remember Allan Bayle.

[adroitcaledonian]

Bayles theorum... Christ.

You don't mean this Allan Bayle do you?

26 August 2001 - The Telegraphhttp://www.telegraph.co.uk/news/main.jh ... fing26.xml
Britain's leading fingerprint scientist has resigned from the Metropolitan Police in order to testify in court against what he believes is flimsy forensic evidence that has led to unsafe convictions.

In an exclusive interview with The Telegraph, Allan Bayle, a former detective whose evidence helped to convict the Lockerbie bomber, has alleged that dozens of innocent men are in jail because of poorly recorded fingerprints.

Mr Bayle, 55, left the police in June after a career spanning 25 years. He is now preparing to give evidence on behalf of a businessman who is currently serving 30 months for burglary.

From his home in north London, he said last week: "I wanted to end my days in the police force but I have to speak out. Officers are not being properly trained and are not recording the evidence properly. There are innocent people in jail now because of these mistakes."

Mr Bayle will give evidence in the appeal case later this year of Alan McNamara, who was imprisoned on the evidence of a fingerprint. "I was not allowed to take the stand for the defence when I was in the force, and the only way I could give evidence was to leave," he said.

...

He provided evidence during the Lockerbie trial which helped to convict the Libyan bomber Abdelbaset al Megrahi of the murder of 271 people in December 1988.

Sun 19 Feb 2006 Scotland on Sundayhttp://news.scotsman.com/topics.cfm?tid ... =258742006

...

A second theory brings in the shadow of the Lockerbie bombing. Mackay's explosive report into the McKie case that August came three months after Boyd began the prosecution of Libyan suspects Abdelbaset Al Megrahi and Al Amin Khalifa Fhimah. The eyes of the world were focused on Scottish justice. What would it have said of that system if - just as the Crown was trying to convict the bombers - it emerged that fingerprint officials had been involved in "criminality and cover-up"?

....

TWO American fingerprint experts were warned by the FBI to back off from the Shirley McKie case for fear it would scupper the trial of the Lockerbie bombers.

David Grieve, the senior fingerprint expert at Illinois State police, said that FBI agents pleaded with him to stay silent, fearing the case "would taint the people involved in Lockerbie".

Campaigners for the McKie family last night claimed that the plea to "let everything drop" shed new light on why the former policewoman was denied justice. They believe that the Crown was determined to protect the reputation of the Scottish justice system at a time when it was coming under international scrutiny.

...

Wertheim and Grieve, both internationally respected fingerprint experts, were central in clearing McKie in 1999 when she was accused of having left her fingerprint at a crime scene. The case left the Scottish justice system open to claims its fingerprint evidence was unsafe. FBI officers took both aside before the Lockerbie trial in the Hague began in February 2000.

Grieve, the senior fingerprint expert at Illinois State Police, said: "I was asked not to mention anything about the case and not to publicise it because we had to think about the higher goal, which was Lockerbie."

He also claims that the FBI had been visited weeks earlier by an official from the SCRO.

"I was pulled aside and given a lecture on the importance of not embarrassing a 'sister agency' which had 'very important and high profile' cases pending of an international significance. I knew the reference was to the Pan-Am bombing," he said.

Wertheim, a fingerprint expert of 20 years' experience, added: "I was at the FBI for a meeting and one of their people approached me and made the suggestion that I let everything drop."

Iain McKie, Shirley McKie's father, said yesterday that he believed Lockerbie provided a motive for the 'cover up' over his daughter's case.

He said: "I have always suspected the Lockerbie connection, but when I put it to the Lord Advocate, I got nothing from them. I could never understand why they treated my daughter like that. Lockerbie would give them that motivation."

...

"Fingerprinting, properly administered, is an exact science. Ask any of the genuine experts, such as Allan Bayle, formerly of Scotland Yard, or Pat Wertheim, the American expert who testified so brilliantly at McKie's trial in 1999. But even common sense should tell us, given the fact that people have been executed - and still are in some parts of the world - on the strength of a fingerprint, that it has to be precise. "




Surely a contradictory state of affairs?
I'm surprised that the journalists who wrote these articles weren't aware of Bayles Good Cop, Bad Cop involvement.

I'm sure you all are too when you consider that the co-author of both articles was one Marcello Mega, a McKie friendly journalist at the Scotland on Sunday.

Iain mate, if you are reading, I would stop the sensational Lockerbie nonsense, you are only going to embarrass your erstwhile houseguest Big Al' Bayle.

[adroitcaledonian]

From the Brazil trial at Leeds last year:

Mr. Wright : So you take the view that British Experts are not up to it is that what...?
Mr. Baille (sic): No they are not and ACPO know that and it is time somebody did something about it.


http://seattletimes.nwsource.com/html/l ... nt07m.html

Only a single piece of evidence linked Portland lawyer Brandon Mayfield to terrorist train bombings in Spain. Yet it was powerful evidence.
Not just one but three FBI examiners concluded that a fingerprint left on a bag of detonators linked to the March 11 attack that left 191 dead came from Mayfield. As always, they were so sure of their finding that they called it a "100 percent identification."

...

The mistake has tainted the FBI's once-vaunted reputation for fingerprint work. Yet some legal and forensic experts say the blunder comes as no surprise because the bureau fails to rigorously train, test and oversee its examiners.

...

Although fingerprint identification is called a forensic science, few examiners at the FBI or elsewhere are trained scientists. There are no national standards for becoming a fingerprint examiner. You don't need a college degree. You don't need a license. You are not required to be certified.

...

However, this seemingly sterling record isn't proof of extraordinary expertise at the FBI because the test is so easy, said Allan Bayle, a top fingerprint expert formerly of Scotland Yard.

"If I gave my experts these tests, they'd fall about laughing," Bayle once testified.

The FBI tests its examiners using fingerprints that are much clearer and more complete than the latent prints typically lifted at crime scenes, Bayle said. Also, the prints used in the test don't change from one year to the next, so examiners could remember the answers from the year before. Finally, the test was watered down in 1998 and no longer includes prints that do not have a match, a way to see if examiners make "false positive" identifications, as happened to Mayfield.

One FBI examiner who missed an identification on the 2000 test was asked to explain himself. He wrote a four-page memo to supervisors, mocking the test for being too easy but also claiming that examiners routinely cheat on the test by discussing their answers with one another, according to sources familiar with the memo.

..

The FBI's methods may now change, something Bayle believes is long overdue.

The question now, in Bayle's view, is: How many innocent people are in prison because of mistakes by the FBI's fingerprint lab?

And there was me so impressed by Pats statement of qualification at the end of McKies book.

Talk about a loose cannon.

[charlton97]

Adroit,

quote, after quote, after quote, after quote. So, I too would find it easy to trawl press clippings and court transcripts to try to score some points for one opinion or another, after all, one universal truth is that press and media will always misquote, miss attribute and misrepresent. The media will often substitute fact with sensationalisms too. So well done for finding the blatently obvious. I think Allen Bayle can sometimes be his own worst enemy in the way in which he handles media and communicates his message, but the message should not be lost just because you, or others want to use such quotes to try to score a point or two.

There are some absolutely brilliant latent print examiners in the US and worldwide. There are some truly awful ones too, as evidenced by the goings on in Florida at this time. In the UK, and this is recognised by Pat Wertheim, in case you want to take this on board too, the UK has some of the finest and well trained officers to be found anywhere. But do not take comfort from that too much. Bayle is right in many ways, how many experts in UK can explain biology, morphology, statistical probability, can demonstrate regular further education, self learning, research, conference attendance. Many, old sweats, as I refer to them qualify as 'experts' and then spend the next 30 years telling everyone to believe me, I am an expert, without ever backing that up with evidence of that expertise.

But I don't lay all blame at the examiners. I also lay blame at the criminal justice system, the lawyers and barristers as wel call them in UK. They rarely call experts to court, rarely cross examine with any enthusiasm, rarely know enough to give the average expert in court a challenge.

Bayle is right to be worried about standards. He is right to be concerned to see uniformity of knowledge and understanding across the global forensic domain. That is his message, and I support it. Do not corrupt this message to suit your own very narrow agenda here. The issue of developing the science is too important for individuals to close down the debate and close down the concepts being put forward here.

My request is for all parties to be very very careful about how we discuss issues and debate them. Top defense lawyers must read forums like this one and see just what a rag tag and unprofessional lot we are sometimes. It seems we cannot even agree to disagree without resort to the sort of pranks and japes that I would ground my children for. Such playground antics do no one credit and does the profession great great harm. I can only assume Adroit that you are not a true fingerprint expert? Otherwise I am sure you would understand the damage you and others do by continually persuing the retort, counter retort strategy. Allan Bayle may be guilty in the past of a certain media naivity. I think even he would concede that. But are you listening to the message Adroit? All experts must up their game. All experts must ensure the ongoing evolution of our science and other forensic sciences. That is the nature of expertise, and I agree with Allan Bayle, many many experts around the World are not up to it. But of course there are many who are. The trick here is to attain uniformity of quality. That will be achieved through understanding and consensus. Not by cheap back biting and inuendo.

[adroitcaledonian]

I think there is more than niavity at play here Dave, when Bayle went to the press and said:

But Allan Bayle, an independent fingerprint expert, who described Scottish fingerprint services in Glasgow as the "worst in the world", said there could be many more.

...

Mr Bayle was asked to look at a palm print left at the scene of a robbery. The SCRO believes the print was left by a man accused of breaking into the premises.

However, Mr Bayle and his colleague, John MacLeod, said the print was falsely identified. The case against the accused man was later dropped, although the Crown Office claims it was not because the fingerprint evidence was contested.

http://news.scotsman.com/topics.cfm?tid ... =783352006

&

"Mr Bayle told the BBC's Newsnight Scotland programme that the SCRO in Glasgow was becoming a "byword for incompetence".

He said: "I was approached by the solicitor dealing with this particular case and he showed me a palm mark and a copy of the accused.

"When you first look at it you think all the lines are in the right place but when you actually look at doing the comparison it is nothing like it and it is very serious.

"It only took me a couple of minutes to see it is not identical."

http://news.bbc.co.uk/1/hi/scotland/5018748.stm

...he knew exactly what he thought he was doing. Think about the timing of it. Designed to do the maximum amount of damage to the SCRO in front of Justice One.

It's a shame that the press bias in this case prevented the same media attention to what we now all know, that these claims were incorrect.

Now the public at large won't read this site, but they will read the Scottish Broadsheets and they do watch Newsnight Scotland and the various panorama/frontline Scotland documentaries.

I am glad you are wise to the ways of many journalists. Unfortunately much of the public are not, and they also tend to trust those who present themselves as experts.

[Outsider]

If I might drag the thread on-topic again:

Prior probability is not such an alien concept. If there was a robbery in a shop and security cameras showed that the criminals did not touch anything other than the things they took away, the police would not even bother to send a fingerprint team to the location because they would know that the prior probability of a latent having been deposited by the criminal is zero. If fingerprint analysis was done (maybe the TV footage was viewed later) then we can assume that the risk of a misidentification in this inquiry is the same as the risk in any similar robbery. So if a suspect was identified from a latent, and the suspect denies having visiting the shop, then we can say that the certainty of an accusation (the ratio of the probability that the latent was deposited by the criminal to the probability that a misID has occurred) is zero after we view the security camera footage (but not before).

I have spent a bit of time playing the devil’s advocate and trying to argue with myself that the effect I am speaking of is maybe not such a big deal after all. I did this by thinking what would be necessary for an accusation such as the one against McKie (where there is no evidence that an act of wrongdoing has occurred other than the fingerprint ID) to have the same certainty as an accusation of carrying out the crime that the fingerprint team went to the location to investigate?

We have to start from assuming that there would have to be on average one police officer disobeying orders and entering the crime scene in each and every criminal inquiry in the land. Honestly, that has to be the starting point. This is because we KNOW that the criminal visited the crime scene, so to be in an equivalent position we would have to KNOW that a police officer entered the crime scene without permission. Let’s look at the uncertainties. The criminal might not leave a useable latent at the scene, but then the disobedient officer might not leave a latent so those are equivalent. The criminal might agree that he deposited the print, but then so might the disobedient officer (in which case there is no dispute). The criminal’s fingerprints might not be on file anywhere, but the prints of a police officer in the investigation team are always available so here is one factor which would increase the probability of finding a latent print of a disobedient officer compared with the person carrying out the crime.

The certainty of an accusation of lying is the probability that the dishonest person’s fingerprints are identified compared with the probability that a misidentification has occurred. In the absence of any data I would assume that it is about as likely that a police officer is misidentified at a crime scene as a suspect (there must be a lot of police officers on elimination lists who do not have permission to enter the crime scenes) so lets assume that this part of the equation is the same.

In reality it is ridiculous to suggest that there is a disobedient police officer visiting every crime scene in the land. The only uncertainty on the side of identifying the criminal is whether or not his prints are on file. If the proportion of crime scenes that have a visit from a disobedient police officer is, say, one in 10,000 and the proportion of crimes that are carried out by people whose fingerprints are available for comparison is one in 10 (9 inquiries out of 10 come up with nothing), then the accusation against Shirley McKie will be about one thousand times less certain than an accusation of carrying out the crime under investigation.

[Outsider]

From Seattle Times 07 June 2004

The FBI, which has resisted outside reviews, does have its own internal proficiency tests that examiners must take each year. In a court review of 447 tests taken from 1995 to 2001, examiners missed only three identifications — less than a 1 percent error rate

I remember after an internet search a while ago I found Kasey Werhiem reporting in New Scientist an error rate of about 1 in 3000 comparisons in an informal study (please correct me if I am wrong).

In the Marion Ross murder there were 400 latents in the murder house each of which would be checked against the 150 people on the elimination list. That is 60,000 comparisons and 150 identifications in a single case. It only needs one comparison to go wrong for a Shirley McKie case to be generated or for the wrong person to be accused of the murder.

Even if 99 out of 100 of these errors are trapped at the verification stage, I don’t think that these “per identification” or “per comparison” error rates would result in a satisfactory “per accusation” error rate.

Calculations: 1 in 3,000 comparisons result in errors. Verification traps 99 out of 100, so error rate of 1 in 300,000 comparisons after verification. 60,000 comparisons in one murder inquiry, so one in every 5 inquiries has in an erroneous accusation.

[mdavis]

Sticking my nose into an old thread, I cannot but help insert a comment concerning potential error rates. Throwing statistics around has, obviously given a generic overview of what a pseudo-mathematician might think when observing a few case or test results for the generic latent print examiner. But I think even the current argument (discussion?) of probabilities demonstrates that these numbers can vary widely.

Speculation of routine numbers of mid-idents stems from the assumption that the fault lies with the inevitability of error with all examiners, when in fact it should be applied to each individually. The stats do not take into account individual levels of conservatism regarding the line between a latent print of value and a latent print of insufficiency -- the independent "A" of ACE-V. As an examiner gains experience, there is a tendency to push the envelope, sometimes a bit too far, in "A". Virtually all the mis-idents on high profile cases have come from "borderline" impressions, yet examiners have often pushed ahead and caused themselves and their departments great harm and embarrassement in making an ident when, perhaps, the latent print should have failed "A".

Other factors, as we all know, include blatent dishonesty, failure to follow the letter and intent of independent "V", potential bias, and political pressure. These other factors are ethical failures, not errors in comparison per se. Whether or not they should be included in "error" studies is controversial.

As we gain experience, it should be obvious that we must continually guard against over-confidence, potential for bias, and complacency. Each examiner has a slightly different threshold in "A". Each ident carries with it the potential for ending a career and/or ruining the life of an innocent victim. We often place too much pressure on failing to make an ident "when it's there." I wonder how many of us out here would be willing to submit our entire life's work to expert re-examination of called idents? How many idents would be "controversial"? How many mis-idents, if any? I would like to think that there are hundreds that would pass muster.

There's an old country and western lyric here in the U.S. that fits rather well: "Know when to hold 'em, know when to fold 'em, know when to walk away, know when to run." Hanging on to a bad ident for dear life when the impression is border-line and the ident is not rock-solid is the cause of all this grief. As someone's appropriate tag line says "you're only as good as your last ident." Arrogance has no place in forensics.

[Michele]

Steve,

With all due respect, I think I’m going to bow out of this conversation. There are just too many issues being discussed at once and several more issues being brought up each time someone tries to resolve one question. It also doesn’t seem wise to try to plug in numbers to a formula before agreeing on the basic assumption that apply to the situation. Not only is the formula and initial assumptions still being discussed but now you’ve added arbitrary numbers which aren’t agreed on either. Additionally, now even more questionable mathematical formulas are being thrown in the mix.

I think you have some valid thoughts but it doesn’t seem like a chat board is the best venue to refine these kinds of ideas.

Sincerely,