Article: Fingerprint evidence to harden up at last

[Boyd Baumgartner]

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

Fingerprint evidence to harden up at last
22 March 2010 by Linda Geddes, Seattle
Magazine issue 2752.


HOW courts use fingerprints, the icons of forensic science, is on the cusp of a much-needed revolution. In response to claims that their use is often unscientific, statistical tools are now being designed to determine the probability that "matches" between a suspect and prints from a crime scene are spurious.

If used in court, these tools could vastly reduce the number of people convicted of crimes they didn't commit. But their use may also face resistance from some fingerprint analysts who are reluctant to give up the long-held, but incorrect, notion that crime-scene fingerprints could only have originated from one person.

Since their first use in 1892 (see timeline), fingerprints have formed the heart of many criminal trials. But recently the realisation has dawned that two prints deemed a match by an expert may in fact come from two separate people, due to human error, coincidence, low-quality prints or a mixture of all three.

In 2005, Itiel Dror and Ailsa Peron of the University of Southampton, UK, showed that fingerprint examiners could come to different conclusions about whether two fingerprints matched when presented with the same prints under different circumstances. Fingerprints have also led to some high-profile mistakes. In 2004, they led the FBI to incorrectly identify Oregon lawyer Brandon Mayfield as a suspect in the terrorist bombings in Madrid, Spain (try our fingerprints quiz to see how well you do).

But fingerprint analysts have been slow to acknowledge the problem, and still present matches to courts with no accompanying error rate - a state of affairs that shocked the US National Academy of Sciences (NAS), which presented a report on US forensic science last year. "The NAS committee was horrified by the testimony of the fingerprint analysts," says Thomas Bohan, who stepped down as president of the American Academy of Forensic Sciences (AAFS) last month. "[One] guy came in and said: 'It's infallible. It has a zero error rate. If we say it's a match, there's no one else in the universe it could be.'"

The NAS report, which called for new research to assess fingerprint reliability, shook forensic scientists. A year on, steps are now being taken to make the use of such evidence more scientific.

Last month the US National Institute of Justice commissioned the School of Law at the University of California, Los Angeles, to investigate error rates associated with fingerprint analysis. "Even though we have been using fingerprint evidence in court for almost a hundred years, not nearly enough is known about how often fingerprint examiners might make mistakes, or in what circumstances," says study leader Jennifer Mnookin. The plan is for fingerprint analysts and defence lawyers to use the results of this study to give judges and juries a clearer idea of the accuracy of fingerprint evidence.

The attitudes of forensic scientists are also changing. In a 2009 survey, 75 per cent of US forensic scientists believed that innocent people may be in jail or on death row because of fingerprint errors. That's up from 56 per cent in 2007, says Samantha Neal of West Virginia University Forensic Science Initiative in Morgantown, who presented the results at a recent AAFS meeting in Seattle. "I think the NAS report has had an impact on the perception of fingerprint analysis," she says.

Some analysts are now taking matters into their own hands. Cedric Neumann at the UK's Forensic Science Service in Birmingham and his colleagues have created a program that calculates the statistical weight that should be attached to a given fingerprint match. Such statistics are already used when a DNA match is presented in court.

Computer programs, such as those used by US immigration, can already match a scanned fingerprint against a database to confirm people are who they say they are. But these require good-quality prints, whereas crime-scene prints are frequently smudged or only exist in partial form. Other programs are used to pull up possible matches from a database of criminal suspects, but a human analyst ultimately decides whether there is a match.

Saying a fingerprint is a match or not doesn't represent the actual weight of the evidence
This is where problems can creep in. Analysts will pick a number of features from a fingerprint and compare these with the second print, declaring it either a match, an exclusion (a non-match) or inconclusive. But not all prints are of equal quality.

"Saying something is excluded or included doesn't represent the actual weight of the evidence," says Glenn Langenburg, a certified fingerprint analyst with the Minnesota Bureau of Criminal Apprehension, who has been training US analysts to use Neumann's program.

Analysts are also under no obligation to document how they reached a conclusion, so the level of accompanying information varies between labs from a little to extensive, Langenburg says.

To solve these problems, Neumann's team asked 40 people to print their fingers 704 times. Then they created a program that uses this data, together with fingerprints taken from 12,000 people, to come up with a probability that a crime scene print matches a given suspect's.

"It asks 'what proportion of [features] from the suspect look like the crime scene mark, and what proportion from fingers in the general population look like the crime scene mark?'" says Neumann, who presented the program at the AAFS meeting. "Instead of going to court and saying: 'It's him because I say so, and I have 25 years' experience and I've never been wrong', I can say, 'It's him because the data supports it and here's how I came to that conclusion'," he says.

Justice Joseph Maltese of the New York Supreme Court agrees that if validated, such a model would be a welcome addition to the courtroom. "'I know it's a match because you say it's a match' isn't good enough today," he says. "If you could buttress that with some statistical data that would be very useful."

There are still some kinks to iron out. One is that the probability given by the program depends on the number of features an analyst highlights as "of interest", which can vary widely between analysts. "You could have very confused juries. We need a system to help us select features reliably, routinely and reproducibly," says Langenburg.

One such tool is being developed by Virginia-based company, Noblis, in consultation with the FBI. Although the exact details are not yet public, the tool maps fingerprints for regions of high and low reliability, flagging up the areas that analysts should focus on. "It's very promising," says Christophe Champod of the University of Lausanne in Switzerland, who is working on a program similar to Neumann's.

Even if fingerprint analysis can be made more consistent, there's also the issue of what happens in court. "It will be interesting to see what jurors make of it," says Christine Funk of the State of Minnesota Board of Public Defense. "For my whole career, it has always been definitive that a fingerprint either belongs to an individual or it doesn't. For me as a defence lawyer it says: 'this isn't an absolute certainty'."

Langenburg agrees that a statistical model may make it easier for defence lawyers to undermine fingerprint evidence, a big fear for some analysts. "They have been taught not to open the door for the defence," he says. But if analysts are serious about transforming their field into a scientific discipline, that's a risk they may have to take.

[sharon cook]

[kevin]

the realisation has dawned that two prints deemed a match by an expert may in fact come from two separate people, due to human error, coincidence, low-quality prints or a mixture of all three...also face resistance from some fingerprint analysts who are reluctant to give up the long-held, but incorrect, notion that crime-scene fingerprints could only have originated from one person

The article kind of contrdicts itself there and I wouldn't expect a reporter to fathom the arguement in its entirety but it brings up a good point as to the direction the research is taking.

I actually like the fact that statistical probabilities can be used to support, confirm, or play a QC role in FR analysis and I hope it takes a direction that is constructive and not an attempt at proving 'the doppleganger' exists There was some good discussion on that very topic a few threads back and its worth arguing over but there are definitely a variety of viewpoints on the subject nowadays.

Contemporary research that further validates what the science has displayed for the past 100+ years is needed and welcomed and I'm looking forward to seeing the research come out, but I don't think it will turn the fingerprint science upside down or on its ear nor do I believe the NAS report shook forensic science to its core like it is portrayed in news articles.

[atorres]

Sharon

Coming from someone who has taken Glenn and Cedric’s class where they discuss these issues that you bring up, plus a lot more it was hard reading your reply and not responding. A lot of what you bring up is not at all what is being proposed, at least by Glenn and Cedric (who the article references quite a bit).

First thing:

“Let's see, considering that DNA analysis is absolutely nothing like fingerprint analysis...SURE! Let's use the same statistical model! I guess using a statistical model that has nothing to do with fingerprint analysis...”

The model/system that is used for fingerprints is not the same model that is used for DNA. The concept is the same and the way the statistical information is presented is along the same lines, but the models are not the same. The statistical model that is used has everything to do with fingerprints.

“Bean-counters rejoice! Soon fingerprint analysis will be totally untouched by human hands. Only machines that have a verifiable error rate because they can be calibrated will be doing fingerprint analysis. I'm looking forward to the ridiculous statistical modelling that's on its way...and, of course, the debunking of said modelling.”

Even before taking their class, I knew that it would never come down to machines running the show, so to speak and after taking their class they will tell you that that is not at all what they are promoting or saying will happen. It’s not going to come down to being totally untouched by human hands, look at DNA analysis if you will… an examiner still has to look at the data/information given and report their findings in court. They use statistics and yet an examiner is still involved. As I have heard many times, this will just be another tool in your tool shed/chest! As to the “ridiculous statistical modeling” statement… I have to ask: have you taken their class or done any research into what is being proposed? And if not, then how do you know it’s ridiculous? You can’t make a statement like that without looking at all the facts and at least trying to understand it and be open-minded about it.

“and it's not based on the probability of the co-efficient of a ridge ending upwards multiplied by the number of ridge endings upward in a standard ulnar loop, surrounded by at least two, but not more than four ridge endings downwards within one, but not more than four, standard ridges, which for the purpose of this model are defined as...blah, blah, bla-de-blah...”

This is not exactly how the probability is calculated!

I highly recommend taking Glenn and Cedric’s class (for you and anyone else), if anything just to get a better idea of what is actually being proposed and talked about. You may not like it or agree with it but at least you may having a better understanding of it at the end. I thoroughly enjoyed the class! It’s not something that you will be able to gestate right away because I know I haven’t, but at least I have a better understanding and see things in a different light. Anyway, I definitely know what your opinion is on the subject and now you have mine, but I do hope that for you and for other’s that if you maintain this strong opinion that you don’t discourage others who might want to learn and understand more about this.

Anne

[Charles Parker]

I have to agree with Kevin and Anne. I would like to see it first before I am ready to dismiss it.

Also I have seen in some cases where LPE have explained how they got from A to Z and they did not convince me. The days of "I am an expert, this is what I see, take it or leave it" (IPSE DIXIT) I believe are coming to an end.

[Ann Horsman]

Right from the beginning I have attended several of these workshops and discussions put out by Glenn, Cedric, Christophe and most recently Paul Chamberlain at the fall 2009 NEDIAI Conference. Yep, was skeptical at first but as time goes on I see potential as an addition to our 'tool chest'. I wish the researchers all the best in devleopment, just make the result user friendly!

[sharon cook]

Note to self: don't use sarcasm anymore.

[printlady]

I agree that statistics and probablity models will help clarify the expert’s conclusions in court. But studies also need be conducted to develop statistics quantifying examiner error, preferrably before critics accuse us of ignoring it.

[kevin]

Note to self: don't use sarcasm anymore

haha-I do feel you on some of that Sharon.

I don't disagree that there are some people who would love to see a statistical model published just so they could use it to attack fingerprints (The word Haber comes to mind....)

I think it will all come out in the wash and the current stat research can be a useful tool without being to over-reaching or burdensome but we'll all have to wait and see. It might not be something the FP community can implement or it might be a great tool we can add or modify to fit the needs of the CJ system.

The article by Linda Geddes(?) defintely doesn't paint the FP community in the best of lights though and I imagine it is by design.

[JohnT]

My concern with this computer generated statistical model - shared with Glenn & Co at a course in 2007 - is this...

We are looking for 'agreement in sequence, having sufficient uniqueness (detail) to individualise'. Where errors such as McKie, Madrid etc. seem to mostly occur is in the 'agreement' of the detail - not in the 'sufficency' of the detail. (Not that I agree with the constant suggestions that we have a large number of errors lurking in the closet.)

As much as the research being done by the likes of Glenn, Cedric, Christophe & Paul is interesting, to me it does not address the real concern of being confident that we do have agreement when we say that two prints have a common source.

Therefore there seems to be the potential for prints to be plotted on this system and a high likelihood of a match is then generated, yet a competent latent print examiner could clearly find that the two do not have a common source.

The additional suggestion that this system can give us confidence in those marginal matches that have agreement but lack that sufficiency for us to be completely confident is also of concern. Firstly I don't find this to be a regular occurance, in fact it would be extremely infrequently and secondly this model does not (unless things have changed) take into account third level detail.

I do find the research interesting, yet I still cringe at the very poorly written and researched New Scientist article.

[Neville]

Hi Sharon
sarcasm is the lowest form of wit.

Being in a county so low on earth I recognised it as such, I thought it was very funny, well done, Bridgette would be proud of you.

JohnT maybe the name of the publication is a clue (New Scientist )

[raybie]

The article is all about edutainment/infotainment.

Those who sensationalize (as pseudosciences do) tend to do it through the media rather than through scientific discourse or scientific peer review, anecdotal evidence is used to justify claims.

[Gerald Clough]

I think there is both a misapprehension of the nature of conclusions and a decided desire to maintain the misapprehension. It works for the defense bar, and it more readily produces research money. Consider how the results of such studies could be applied. If I follow the potential development of statistical studies based on the nature of features and their interrelationships to it's ultimate potential, I find that it really ends up with something we already know, but it cannot produce very precise numbers that relate to the real world of examinations. From such studies, I can only get a somewhat better feel for what I mean when I say it's possible for two portions of skin to produce indiscernable impressions. I get a rough "index of caution," where before, I only admitted the possibility. I am still far from any metric that relates to the cause of nearly all identification errors that don't involve plain fraud, mislabeling, or gross incompetence. That common cause of error is misinterpretation of impressed features, erroneously characterizing the skin features that made the impression.

The misapprehension is the position that fingerprint conclusions are not subject to differences of opinion. We're very fond of pretending there are no differences of opinion, only errors and correct conclusions. But the nature of expert evidence is that there can indeed be two opposing opinions. Since the ground truth can only rarely be known, it is purely a matter of who a finder of fact takes as more or sufficiently believable.

Now, given the poor representation of fingerprint evidence today in court, it is decidedly to the advantage of whoever wants to discredit a conclusion to pretend that it is either an unsupportable statement of scientifically certain physical reality (which should never be our position) or that it is the product of an inbred community of examiners who have talked themselves into unjustified confidence to the point that "resistance is futile." If they can maintain that view, then any statistical studies are useful in wrongly casting the conclusion as a statement of scientific certainty or in disparaging all conclusions. They can only mount that attack if they can characterize the target as too unreliable to be heard.

The proper approach, sponsoring opposing opinion, is more effort, and they believe it has the lamentable tendency (from the defense view) to substantially confirm the first conclusion. In this opposition, the results of statistical studies do have value. They help demonstrate the reason a different opinion on conclusion or on confidence in a conclusion should be considered. This is strictly a strategic issue. It does not go to revealing potential error. From our experience with errors, that is most often revealed by reexamining the interpretations of specific features and relationships. Statistical studies of frequency of occurrence of characterized features don't help there. In fact, nothing helps, except actually opposing opinion on the way impressed features are interpreted and demonstrations. And since plain errors are not often going to turn up, maintaining the misrepresentations of the nature of conclusions is to their advantage.

[fingerprint]

The article is all about edutainment/infotainment.

Those who sensationalize (as pseudosciences do) tend to do it through the media rather than through scientific discourse or scientific peer review, anecdotal evidence is used to justify claims.

I agree with you one hundred percent, there are so many different forms of PR and twisting everything around. When I first became a private investigator I was really looking forward to joining a community of collaboration, innovation, and kindness (and don't get me wrong I have made some great great friends), but i have been amazed at some of the unethical things that happen in our business. Just the other day I was doing some research on my competition, and it really seems like that's what this is all about. I mean, have you all seen any of the reviews and listings on angies list for private investigators?