杏吧原创

IN 1995, Steve Fortin pleaded guilty to a savage attack on a woman in Maine, in which he broke her nose, bit her, strangled her into unconsciousness and raped her. So when it was discovered that Melissa Padilla, a mother of four from New Jersey, had been found murdered with similar injuries years before, police linked Fortin to the case.

He was convicted, but the New Jersey Supreme Court later overturned the conviction on the grounds that linkage analysis 鈥� the technique of connecting serial crimes by similarities in the criminal鈥檚 behaviour 鈥� lacked sufficient scientific reliability and should not have been used in court testimony. Fortin was later retried without evidence from the linkage analysis expert and is serving a 30-year sentence for Padilla鈥檚 murder. The case confirmed what many had suspected: despite being used to generate leads for sex attacks and murder cases, linkage analysis actually had little evidence to back up its claims.

That could be changing. Several groups have developed linkage analysis software and are now carrying out the research needed to establish the technique as a more rigorous tool. The work challenges some of the common assumptions about patterns of criminal behaviour 鈥� that serial murderers will mutilate their victims鈥� bodies in the same way, for example. It also suggests that linkage analysis could be applied to a much broader range of crimes than at present.

Linkage analysis differs from the 鈥渟ignature analysis鈥� used by specialist police forces such as the FBI. 鈥淭hey鈥檙e looking for very, very specific behaviours, or combinations of specific behaviours, many of which are ritualistic and some of which are very bizarre,鈥� says of Carleton University in Ottawa, Canada. These are quite rare, and tend to be confined to only the most disturbing cases. 鈥淲e鈥檇 love it if an offender left the ace of spades after every rape or murder, but it鈥檚 just not like that,鈥� says Adam Gregory, behavioural investigative advisor for the UK鈥檚 Serious Crime Analysis Squad (SCAS), which helps police to link seemingly motiveless murders, sexually motivated murders or rapes by strangers.

Linkage analysis works on similar principles, but can be applied to a wider variety of crimes. 鈥淭he assumption is that a given offender will exhibit similar behaviours across their crimes, and that these will be relatively distinct from those of other offenders committing similar types of crimes,鈥� says Bennell.

However, the success of the technique depends on the ability of police officers to recognise these similarities. And this is where the problem lies, because individual officers will form their own subjective impressions of crime situations. What鈥檚 needed is a consistent way of categorising and analysing behaviours, and weighing up the probability that they can be used to link crimes.

Bennell has been developing just such a system. It is a computer program that can read a database of crimes in which behavioural domains 鈥� such as the method of entering a victim鈥檚 property, or items stolen 鈥� have been detailed by the police. The program then comes up with a set of behavioural similarity scores for any two crimes, with one rating in each domain. 鈥淚f the crimes have been committed by the same person you鈥檒l often find higher levels of similarity in certain behavioural domains,鈥� he explains.

The trick is identifying the behaviours to rely on. Some behavioural domains may not be particularly useful for linking purposes, because they are not exhibited in a consistent fashion across crimes. In other cases, consistency will exist, but the behaviours will not be distinct 鈥� for example, different burglars may steal very similar items and so property stolen cannot usually tell you if two crimes are linked.

However, after applying the system to a database of solved crimes, regression analysis can provide a weighting to similarity scores, according to how predictive of a link they actually are. This enabled Bennell鈥檚 team to work out which behaviours were most useful in linking serial burglaries. By combining the most predictive behavioural domains, the program produces a probability that two crimes were committed by the same offender. In the case of burglaries, for example, Bennell鈥檚 team found that a short distance between offences was far and away the most accurate predictor of a connection.

鈥淭he program produces a probability that two crimes were by the same person鈥�

Several groups have since used the technique to evaluate the use of linkage analysis in car theft, commercial robbery and sex offences. A different statistical technique has also been used to look at behavioural patterns in serial murders. These studies are turning up a few surprises (see 鈥淐riminal behaviour patterns鈥�).

In serial murders, for example, it seems that only the domain of planning and control 鈥� activities such as destroying evidence, and bringing gloves or rope to tie up the victim 鈥� could potentially be used to link serial crimes, but the method of body disposal, type of mutilation, sexual acts, theft and weapon use are less consistent between these crimes (Behavioral Sciences and the Law, vol 25, p 527).

In serial sex offences, the distance between attacks, methods of controlling the victim and methods of evading capture tend to be consistent between crimes by the same offender, but behaviours reflecting the attacker鈥檚 personality are less consistent. 鈥淭he domains that are coming out as the best predictors are those that depend less on the situation,鈥� says Jessica Woodhams at the University of Gloucestershire in Cheltenham, UK, who led the study and presented the results at a forensic psychology meeting at the University of Leicester, UK, last month. Details like bringing twine to tie up a victim are decided in advance, while what the criminal actually does to them may be influenced by how much the victim struggles, for example.

Bennell hopes the research will result in improved police databases capable of supporting this kind of analysis. 鈥淲e envision a computer program that provides the police officer with a means for uploading data about specific crime-scene behaviours鈥� likely right at the scene using a hand-held device,鈥� says Bennell. The computer program would then combine and weight the behaviours, and suggest to the officer whether to treat the crime as a linked one or not.

He has some preliminary evidence that this could work. In a recent study, he asked a group of 40 university students and another group of 31 police professionals to look at the crime records for 38 burglaries and decide which were linked. Before taking part, half of the volunteers in both groups were told that the likelihood of two offences being committed by the same offender increases as the distance between the crimes decreases. While this 鈥渢raining鈥� improved the accuracy of their decisions, none of the groups performed as well as the program. 鈥淓ven when people are told to use information that is useful to them, sometimes they don鈥檛 have the ability to identify the correct links,鈥� says Bennell.

Professional crime analysts are not completely convinced, however. Previous research has shown offenders only become consistent over time, says Gregory. 鈥淢ost of the time we鈥檙e trying to link offenders鈥� first and second, or first and third, offences 鈥� a time when offenders will be at their least consistent.鈥�

Bennell stresses that the system would simply be an additional tool in the crime-analyst鈥檚 kit. They will still 鈥渘eed to read the victim鈥檚 statement to understand the context鈥� in which the crimes were committed, he says.