Geographic Profiling

[Scott Nelson]

It does appear to get a little better as the numbers go up, and in the vicinity of 10 is probably where it asymptotes, but performance with as few as 4 cases is still very good (as in overall 50% fall in the hot spot area; that's the pink, yellow, and orange areas; and 75% when the light red area is included. That corresponds to roughly 10% (8.89%) and 20% of the total search space; (I think I said 7.5% before, it's 8.89% for my own routines, which I round up to 10% to be conservative) for 50% or 75% return. If I limit it to only cases with 10 or more, 50% fall in 7% of the search area, and 75% in 15%.

It's a huge regional area of dense population, Jeff. Fifty to one hundred confirmed data points may do it, but not 5, or even 10. I worked with this sort of stuff for years as an engineer (not criminal profiling, but for chemical tracking, fate and transport analysis, etc.). The more confirmed data points, the better the definition of trends.

[DJA]

Asymptote - Wikipedia

[JeffHamm]

It's a huge regional area of dense population, Jeff. Fifty to one hundred confirmed data points may do it, but not 5, or even 10. I worked with this sort of stuff for years as an engineer (not criminal profiling, but for chemical tracking, fate and transport analysis, etc.). The more confirmed data points, the better the definition of trends.

Hi Scott,

The spatial analysis isn't about pinpointing an individual, though that is the impression one can get from the public press or entertainment. It's about prioritizing areas, and those areas will indeed contain a huge number of people. There have been a couple of instances where a profile's peak is impressively close to the offender's actual residence, but those are just the most dramatic ones to advertise and certainly not the norm. The typical outcome is far less likely to make sales, and there are those who are selling products for doing this. That's one of the reasons I mention that my data set has some selection issues, many of the cases I have come from books published on this topic, and the ones chosen tend to be the impressive ones. When I compare, for example, Rigel vs Dragnet, Rigel does better for the cases from books by Rossmo, and Dragnet does better for the cases by Canter, and my routines do better for the data set I use to develop it. No surprise there. But of course, that tells me there's a bias in what cases are being illustrated. If I showed only results from the cases I use to develop the routines, then it shouldn't be surprising if those routines work particularly well. A fair test requires a set of data that isn't used in the development of any of the algorithms.

And it is also the case that the more spread out the offense locations, the larger the search area, so the larger the area suggested to be prioritized. If the cases create a search area circle of say 10km radius, then the search area is 314.15 square km. Reduce that to 10%, and you're still dealing with an area of 31.415 sq km, and that has a 50% chance of containing the offender. Expand it out to around 62.83 sq km, and you now have a 75% chance of the offender being in there. That's not locating an individual, that's reducing the search area, but it's still a big area. Having a quick look at some numbers, if we consider just the pink area (that marks off 1% of the search area, so I'll call that pinpointing), out of 100 test cases that I have, Rigel puts 4 of them in that zone, and Dragnet got 10, and my routines got 7. So "pinpointing" (and that's being generous, as again, that would be an area over 3 sqr km) is not the norm, but those are the cases that most people hear about because it makes for a sexy story.

And as I mentioned, more offenses does result in some improvement, it's just doesn't appear to be as unstable as implied when dealing with fewer than 10 cases. One thing, though, is that as the number of cases goes down, one does have to increase the estimated search area. If one doesn't do that, which I have in these cases, it's possible the loss of accuracy is greater than I've found. That will take me a bit of time to explore though. I know, however, that it does not require 50-100 locations to get a stable pattern that serves the purpose these types of analysis are providing. Again, it's not about pinpointing, but improving the efficiency of the search pattern. It's a very broad brush, which I would suspect is unlike the types of solutions for situations you're working with, where the precision would have to be so much higher. I have a suspicion, in fact, that if one were to have 50-100 locations, the routines currently in use might fare quite poorly as they are derived from much more sparse data. Once an offender is up to that number of offenses, I think the analysis routines would in all likelihood have to be modified to deal with that. I don't know that for a fact, but there would be an increase in information that is normally not present in a lower number of offenses and that's going to start influencing things negatively as the routines will effectively misinterpret it (again, that's my gut instinct, not one based upon any data; but one I would like to be able to actually test so that I knew).

Like I say, a lot of what is presented in the public domain is hype and advertising and sexy stories. The reality is, these routines do extract information from the offense locations, but the information that is there is a relatively low spatial frequency type of information, so the result is of low resolution. Reducing the search area to 10% or even 20%, can still mean you're left with a very very large area, it's just less than when you've got no analysis.

What I would like people to understand is that these types of routines do provide some information worth considering, but they are not solutions nor are they anywhere near as precise as the impression one gets.

- Jeff

[rjpalmer]

It's a huge regional area of dense population, Jeff. Fifty to one hundred confirmed data points may do it, but not 5, or even 10. I worked with this sort of stuff for years as an engineer (not criminal profiling, but for chemical tracking, fate and transport analysis, etc.). The more confirmed data points, the better the definition of trends.

How can one hope to factor in the fact that potential victims are more likely to congregate in certain areas? It's not like the 'targets' are spread evenly over the entire East End. If unfortunates are known to frequent say, the lighted sidewalk by London Hospital, or Aldgate tube station, or a street known for pubs, this would have an enormous influence on the 'data points,' would it not? Even if he fails to find a victim by the train station, he is now on High Street, and the flow of foot traffic might be pushing him along specific lines of egress, etc. etc.

As I see it, even if the murderer is to some degree a 'local,' he would still be 'commuting' to areas where he was likely to find what he was looking for.

Thus, I'm assuming that an analysis of this sort would be more relevant to arson or burglary, where the potential 'data points' are more or less ubiquitous, but less relevant to street prostitution or bank robbery, where they wouldn't be.

[Scott Nelson]

Yes, agree RJ. It wouldn't account very well for moving/commuting "data points".

[JeffHamm]

How can one hope to factor in the fact that potential victims are more likely to congregate in certain areas? It's not like the 'targets' are spread evenly over the entire East End. If unfortunates are known to frequent say, the lighted sidewalk by London Hospital, or Aldgate tube station, or a street known for pubs, this would have an enormous influence on the 'data points,' would it not? Even if he fails to find a victim by the train station, he is now on High Street, and the flow of foot traffic might be pushing him along specific lines of egress, etc. etc.

As I see it, even if the murderer is to some degree a 'local,' he would still be 'commuting' to areas where he was likely to find what he was looking for.

Thus, I'm assuming that an analysis of this sort would be more relevant to arson or burglary, where the potential 'data points' are more or less ubiquitous, but less relevant to street prostitution or bank robbery, where they wouldn't be.

Hi rj,

Those concerns are all part of the reasons why this sort of analysis is not about pinpointing. There are those who are looking into trying to factor just that sort of information. I've not visited that literature for awhile so I can't say how much difference it makes, but my recollection is that it was not all that much. It might, however, play a part in separating out commuters from marauders. If that "pre-screening" could be done, that would be an important first step in determining whether or not these sorts of analysis are suitable (as they tend to focus on marauders). Anyway, it is for reasons like the ones you've mentioned that this sort of analysis does not produce exact locations, but rather is about narrowing down the search space. Much like determining the known associates of a victim (partner, family members, friends, colleagues, associates, etc), is a good first step in deciding what people in a city to interview first - the offender is often in those groups, but not always. This does the same thing for space, the offender is likely to have ties to the indicated location, so start looking there. In that respect, the patterns are helpful. But when it gets put out to the public that it "points to the doorstep of the offender", that's incorrect. That rarely happens, and while it would be cool if that outcome was very common, it's only verily commonly advertised, not the common outcome. As I say, no suspect should be considered cleared simply because they don't fall in the hotspot, and no suspect is considered "proven guilty" just because they do. But, all other things being equal, a suspect in the hotspot would be favored over one who was not, but only in terms of ranking their probabilities - it would still take real evidence to determine which, if either, was the offender.

- Jeff

[JeffHamm]

Here's an example of the kind of thing that often is what people hear about.

Below is the output from the Hillside Strangler case from back in the 70s. The red locations mark out the encounter locations, the blue is one of the offender residences (there were two of them), and the cross is the peak location of the profile. It's almost right on top of the residence. If I'm trying to convince police to buy my routines, I'm going to show them this, or if the press want to write a story, they are going to select this one even if I showed them a bunch of other examples, none of which were so close because it adds to the "wow" factor of the story. Does this example reflect the norm? No, if it did, the profiles would just be a single location. They map out regions for a reason. Most offenders are not at that location per se, but within vicinity of it (and by vicinity, that can still be quite a large distance if the offender is committing offenses over a wide range). It's the selective exposure to this sort of result that creates either overconfidence or heightened skepticism in what this can do. I think the latter is more justified though, because in the end, it is this sort of bias in the results that get exposure that tends to be associated with claims that really are unwarranted. It cannot do what is publicly claimed (or at least implied), so being skeptical of those claims is a good thing. On the other hand, it does provide some useful information, so it is worth taking note of, in the same way one might take note of a personal relationship between a victim and a suspect. It's not proof, it's not evidence, but it is something not to be ignored completely. My own interest in this area of research, in fact, started off because I thought it was complete rubbish and I expected to find there were some major flaws in some of the underlying statitics. One thing that does get highly reported is a very strong relationship between the diameter of the crime range (smallest circle that encloses all the offenses) and the maximum distance an offender travels to commit an offense. There's a correlation of over 0.9 between those, and this gets reported in the literature all the time and the slope from that relationship is used to suggest the offender lives near the edge. But it's absolutely bogus, and simply reflects the fact that if the crimes are more spread out, the maximum distance will have to increase as well. And a careful examination of where in that circle an offender lives by dividing the circle up into a disk a cocentric rings of equal area, show no tendancy for an offender to be more towards the edge than towards the centre. But, as I further explored other aspects, it appears there are some regularities that can be found, which genuinely surprised me.

Anyway, you are correct about there being all sorts of things that produce noise, but that doesn't negate the fact that on a broader level the analyses do much better than chance of defining a region that contains the offender. And that can be useful, even though it doesn't tell you who in that region the offender is.

[DJA]

How can one hope to factor in the fact that potential victims are more likely to congregate in certain areas? It's not like the 'targets' are spread evenly over the entire East End. If unfortunates are known to frequent say, the lighted sidewalk by London Hospital, or Aldgate tube station, or a street known for pubs, this would have an enormous influence on the 'data points,' would it not? Even if he fails to find a victim by the train station, he is now on High Street, and the flow of foot traffic might be pushing him along specific lines of egress, etc. etc.

As I see it, even if the murderer is to some degree a 'local,' he would still be 'commuting' to areas where he was likely to find what he was looking for.

Thus, I'm assuming that an analysis of this sort would be more relevant to arson or burglary, where the potential 'data points' are more or less ubiquitous, but less relevant to street prostitution or bank robbery, where they wouldn't be.

Ironically,the first murder was fairly close to the London Hospital.
Second and third a bit further away which makes sense.
Fourth roughly the same and likely very close to his home away from home.
Mary Kelly seems to have been his ultimate target.

Even when all the evidence is ignored,any heat map radiates out from the London Hospital.

[caz]

How can one hope to factor in the fact that potential victims are more likely to congregate in certain areas? It's not like the 'targets' are spread evenly over the entire East End. If unfortunates are known to frequent say, the lighted sidewalk by London Hospital, or Aldgate tube station, or a street known for pubs, this would have an enormous influence on the 'data points,' would it not? Even if he fails to find a victim by the train station, he is now on High Street, and the flow of foot traffic might be pushing him along specific lines of egress, etc. etc.

As I see it, even if the murderer is to some degree a 'local,' he would still be 'commuting' to areas where he was likely to find what he was looking for.

Thus, I'm assuming that an analysis of this sort would be more relevant to arson or burglary, where the potential 'data points' are more or less ubiquitous, but less relevant to street prostitution or bank robbery, where they wouldn't be.

I had been thinking along similar lines, RJ, and was reminded of the case of Colin Ireland, the Gay Slayer, who lived in Southend on the Essex coast, but picked up all five of his male victims in the same gay pub in south west London in 1993, and murdered them when they took him back to their homes.

Here's an eye-opening link to the basics of the series:

https://en.wikipedia.org/wiki/Colin_Ireland

A rare commuter series, granted, but when all the victims are in the same small area, geographically, when they encounter their killer, and the murders remain unsolved, there is simply no way of knowing all the factors which combined to determine the killer's choice of comfort zone, and his own close proximity to it may or may not come into it. It's not as if London didn't have a good transport network in 1888, to allow for anyone with murder and mutilation in mind to travel to where he could make it his territory and feel on familiar ground each time.

The argument often goes that if JtR came into the killing zone from anywhere outside, he could have found equally suitable victims in many other parts of London and would not have been restricted to that small area. But this misses the point because it takes no account of a repeat offender's character and foibles, which by definition will include repetitious behaviour when he finds something that has worked for him once, and the risks are worth taking to try it again.

[Trevor Marriott]

Even when all the evidence is ignored,any heat map radiates out from the London Hospital.

As it does from the St katherines and The London Docks 5 mins walk from Whitechapel!

https://www.trevormarriott.co.uk

[Abby Normal]

I had been thinking along similar lines, RJ, and was reminded of the case of Colin Ireland, the Gay Slayer, who lived in Southend on the Essex coast, but picked up all five of his male victims in the same gay pub in south west London in 1993, and murdered them when they took him back to their homes.

Here's an eye-opening link to the basics of the series:

https://en.wikipedia.org/wiki/Colin_Ireland

A rare commuter series, granted, but when all the victims are in the same small area, geographically, when they encounter their killer, and the murders remain unsolved, there is simply no way of knowing all the factors which combined to determine the killer's choice of comfort zone, and his own close proximity to it may or may not come into it. It's not as if London didn't have a good transport network in 1888, to allow for anyone with murder and mutilation in mind to travel to where he could make it his territory and feel on familiar ground each time.

The argument often goes that if JtR came into the killing zone from anywhere outside, he could have found equally suitable victims in many other parts of London and would not have been restricted to that small area. But this misses the point because it takes no account of a repeat offender's character and foibles, which by definition will include repetitious behaviour when he finds something that has worked for him once, and the risks are worth taking to try it again.

youre not maybrick signalling again are you caz? ; )

well the druittists will agree with you anyway. No, the ripper was more than likely a local man who knew the area like the back of his hand. he knew the local peoples behaviors (including the prostitutes), the alleyways and courts and shortcuts, maybe even the police beats. No outsider could have gotten away with what happened on the night of the double event.
And the shout of lipski was an insult that another local like abberline even understood.

[rjpalmer]

he knew the local peoples behaviors (including the prostitutes), the alleyways and courts and shortcuts, maybe even the police beats. No outsider could have gotten away with what happened on the night of the double event.

You're confusing the punter and the prostitute.

[Abby Normal]

You're confusing the punter and the prostitute.

no. im not

[rjpalmer]

Sure you are, Abby. It's not how street prostitution works.

'Unfortunates' hang out on street corners for a reason: so the punters, who are very often from outside the area, can find them.

Once found, the women do the rest. THEY are the experts when it comes to back alleys, police beats, etc. The punter? Not so much.

The cry of 'Lipsky' is a more sophisticated argument.

[rjpalmer]

What does 'local man' even mean in an area like East London? A large slice of the population is from somewhere else. Rent is cheap, so people flop their temporarily, and once you're there, you are a 'local.'

Violina was from Manchester, but was temporarily living near Hanbury Street. He is now a 'local chap.'