Location of Annie Millwood's attack

[JeffHamm]

The difference between the two has a lot to do with the estimation of the "main routes of travel". When Stride is include, both Nichols' and Stride's locations are flagged as being "outliers" - meaning off the major route, and the estimation of the major route passes more vertically (sort of Eddowes to Chapman type thing). A few other calculations change as well, but those changes are more subtle (the output is, after all, influenced by the data you put in). One of the important things to get right is the crime linkage - while one would want an analysis like this to be robust against some erroneous inputs and/or omissions, the better the data the better chance of getting a useful output. At the moment, my analysis program's output is influenced if the map (the image file) doesn't contain a large enough area for the search (like this one; it's not big enough for even the red zones, and the green zones clearly extend off the map; there are two additional zones after that, which have to do with commuters - offenders who live outside the circle (outside the crime region as it's called). The image file really needs to have a large enough "surround" to plot all 4 of those zones, but really big images take a long time to process. Also, I'm still working on the underlying routines (this is all stuff being worked on and researched, and while so far the few tests I've been able to do with solved cases have been encouraging, the sample is too small to say anything other than it's worth continuing with).

For example, here's the BTK (Dennis Rader) output. The blue square near the top (in Park city) marks his residence, which is right in zone 1, and for this the image file is large enough to show all of zones 1-6, so those will be fine, but after that the details from zone 7 onwards will be slightly out, but we're seeing enough to get a good idea.



For the Golden State Killer (Joseph James DeAngelo), again, the map image file I have is undersized, so the zone boundaries are going to be off a bit once the zone extends "off screen". He's right on the border of zone2 and 3, but, provided the areas to the north don't contain part of zone 3 or better, that shouldn't change. At some point I'll need to get the program to deal with the "off image space", but there's a few other things that I need to focus on first. And yes, I fully admit, I'm showing the impressive ones, it's not perfect, and there are cases where it fails miserably. But, while the ideal would be to get something that is always right, for something as complex and individualistic as serial crimes, I would be happy with the more modest goal of getting something that much more often than not provides a better than chance search pattern. And it's only a suggested search pattern, geographical profiling doesn't "solve" crimes, it's just a tool that can be used by investigators as they work on solving the crimes. If they get leads on a suspect that doesn't live in a "suggested zone", then they should ignore the profile. These are useful to suggest where to start to search, particularly if leads are not forthcoming.

[Kattrup]

Thank you, Jeff. It's very interesting.

[JeffHamm]

Thanks Kattrup. It's an interesting area of research, limitations notwithstanding. It's not magic, and it doesn't narrow things down to a specific house or anything like that, rather, it suggests regions to prioritize first for a search, and guides a search pattern. The output of this sort of analysis should always be overridden by actual evidence or leads uncovered during an investigation. If there are multiple "persons of interests", it can be used to help rank them to some extent, as a way of optimizing the limited resources of time and costs. The larger the area over which offenses occur, the larger the zones (each zone has an area equal to 5% of 1/2 the crime region - the 1/2 is because if you search an area randomly, you'll find the location after searching 1/2 of it on average). I'm still doing work on this, and trying to find better and more reliable predictors, and combinations of predictors, so again, I'm not putting these out there as anything other than stuff I find interesting and wanted to share with those who might also enjoy thinking of the locations it highlights and how well they may, or may not, fit in to the bigger picture.

I would really like a decent map that has accurate placements of all the crimes, and all the suspects. I think Kosminski and Tumblety have been suggested to be located north and a bit west, or just a short distance north east, respectively, of Stride, and their locations fall in zones 44 and 31, respectively, which is quite far down (this is including Millwood, Tabram, and all of the C5: I won't go through all the combinations as I'm just presenting this as illustrative and also for that I've used a different map of the area which is larger so the more extreme zones get better calculated, but even in this smaller version those locations are in the green area.

Also, I just double checked my previous post with the "exclude Stride" map, and realized I hadn't all of the settings proper. I've redone that, and it looks like this (so this has Millwood, Tabram, and the C5 except Stride). Notice the two paths of travel are quite different, one capturing the locations running noth-south (Chapman to Eddowes) and the other more east-west, due to Nichol's location. The Goulston Graffito now is in zone 2 (just under the EE in Street), and Hanbury Street is now in zone 4:


And to get a better picture of the output, and to see how having a map that's too small influences the output (has to do with calculating the area a zone spans; even this larger map appears to be too small - the full thing would end up with nothing going off the map. There's an interesting "bump" down in the South East though), I've redone the above analysis on a larger map. I've also put in some letters to mark the location of Stride (the white S), and the K and T mark locations associated with Kosminski and Tumblety (I think K marks his relations' residence, where he was thought to be staying; can't remember where I got the information about Tumblety, but no doubt from a discussion found here). Anyway, neither gets much better if Stride is included. I think there's a suspect, Levy? might be the name, who is very near, and maybe in, zone 1 though. Again, doesn't mean he's the one, just would be a suspect that this analysis would suggest deserves a closer look (I know nothing about him myself, just remember seeing a map with a suspect right in that area, but I can't find it and I'm not sure I'm even remembering the name properly, it is not one of the often discussed suspects though - so not Druit, Kosminsky, etc).



I must really make a huge map that's big enough to show the whole thing. It increases the time for the analysis to run, so for now, while I'm developing and researching, I've been using undersized maps for speed.

- Jeff

[JeffHamm]

Ok, so I've increased the detailed map size from above, and have now come up with the output that presents all the zones fully calculated. The first includes Millwood, Tabram, Nichols, Chapman, Stride, Eddowes, and Kelly, and the 2nd removes Stride. Because the smaller maps are large enough to show all of the top zones, those don't change, it's the zones that go "off the map" that end up being inaccurately presented on the smaller maps. Anyway, I'll present these just so there's a version of the whole analysis available to anyone who wants to have a look.

The eastern small red zone (part of zones 18-20; so low but still "better than chance" if JtR did live there), is interesting in some ways because of the proximity to near direct routes along major roads/streets to both Nichols, Stride, and the main area of the crimes. This would mean JtR is travelling to most of the crimes along Commercial Road. My problem with that area, though, is that if we include Stride and Eddowes, as this analysis does, JtR's known movements are entirely of someone not going to that location. While it might be tempting to try and aruge "ahh, clever boy, trying to throw off the scent", that entirely runs up to the problem of, 1) if JtR were that clever, he's clever enough to know that the longer he's out in public with a human kidney on him, the great and greater chances of getting caught. His primary concern would be to get out of sight. It's important to view these only for what they are, the result of a pattern analysis of locations and how crime locations have related to other offender's anchor points (residences), and to then see what parts may fit best with regards to the specific series under investigation.

And now if we remove Stride from consideration:


And there we have it. Enjoy, I've certainly had fun. This is the first time I've done the full analysis for the JtR crimes so it's interesting to see the whole thing.

- Jeff

[Abby Normal]

Ok, so I've increased the detailed map size from above, and have now come up with the output that presents all the zones fully calculated. The first includes Millwood, Tabram, Nichols, Chapman, Stride, Eddowes, and Kelly, and the 2nd removes Stride. Because the smaller maps are large enough to show all of the top zones, those don't change, it's the zones that go "off the map" that end up being inaccurately presented on the smaller maps. Anyway, I'll present these just so there's a version of the whole analysis available to anyone who wants to have a look.

The eastern small red zone (part of zones 18-20; so low but still "better than chance" if JtR did live there), is interesting in some ways because of the proximity to near direct routes along major roads/streets to both Nichols, Stride, and the main area of the crimes. This would mean JtR is travelling to most of the crimes along Commercial Road. My problem with that area, though, is that if we include Stride and Eddowes, as this analysis does, JtR's known movements are entirely of someone not going to that location. While it might be tempting to try and aruge "ahh, clever boy, trying to throw off the scent", that entirely runs up to the problem of, 1) if JtR were that clever, he's clever enough to know that the longer he's out in public with a human kidney on him, the great and greater chances of getting caught. His primary concern would be to get out of sight. It's important to view these only for what they are, the result of a pattern analysis of locations and how crime locations have related to other offender's anchor points (residences), and to then see what parts may fit best with regards to the specific series under investigation.

And now if we remove Stride from consideration:


And there we have it. Enjoy, I've certainly had fun. This is the first time I've done the full analysis for the JtR crimes so it's interesting to see the whole thing.

- Jeff

Hi jeff
interestingly i just saw a bit on geoprofiling to ID the ripper on the show what on earth? On the science channel. It ids nathan kaminsky of black lion yard as being smack dab in the middle of zone one. According to there analysis.

i beleive they only used the c5 victims.

it would be interesting to see what you come up with using c5 and match it to known suspects to see if any come up in zone one and or to see if black lion yard also comes up as in the middle of zone 1 according to your analysis.

[JeffHamm]

Hi jeff
interestingly i just saw a bit on geoprofiling to ID the ripper on the show what on earth? On the science channel. It ids nathan kaminsky of black lion yard as being smack dab in the middle of zone one. According to there analysis.

i beleive they only used the c5 victims.

it would be interesting to see what you come up with using c5 and match it to known suspects to see if any come up in zone one and or to see if black lion yard also comes up as in the middle of zone 1 according to your analysis.

I've done the analysis below, but I'm not sure where Black Lion Yard is located? Kaminsky, in the A-Z map, would not be in zone 1 though. While it might seem intuitive that zone 1 should be very central, offenders tend to live closer to the edge of the crime zone than the centre (not always, and there are exceptions, but this sort of analysis is based upon probability distributions so it tends to weight very central locations as low probability areas). As can be seen, once Tabram and Millwood, are removed, the pattern of offences no longer has a North-South tendency on the west side. The data set of solved crimes that I have to work from, which is limited - and the crimes are arson, which is a different offense type, finds that in these circular type of offense patterns the offender is often located on the same circle, hence the sort of doughnut shape that's forming. I'm not sure how much I would trust the potential routes (red and green lines) in this case, as really, there's no strong pattern of clustering along the routes (as happens once the Millwood and Tabram locations are included; a N-S type corridor starts to form in the west). But, for the sake of keeping all things the same, I've left them turned on. The area around Tabram has diminished, but tow areas close to Nichols (now the first in the series) start to be suggested. Also, the Goulston Street area remains fairly high (it's now zone 3, but that's still quite high). I've been working on a different approach than the routines that are out there, and while similar regions are sometimes suggested, it is not uncommon for entirely different outputs to result. Again, given that there will always be some in the "low probability zones", it may be that where we differ is one set that are in my low zones are in their high zones, but another set that are in their low zones end up in my high zones. Also, and I can't stress this enough, my stuff is still very experimental and under development. It also needs a lot more testing, and verification. So far, based upon the solved cases I've been able to track down online or in academic papers, it's done pretty good with serial murders despite the data I use to work out the algorithms are arson cases so either I've been lucky so far, or there is something to it.

One thing I would really like to be able to work out is some indication as to whether the offender is likely to be a maurader (meaning lives within the crime zone) or a commuter (travels to the crime zone from outside of it). I believe it's approx 20% of solved cases for serial rape and murder are commuters, and geoprofiling tends to assume a maurader (since that's the majority). That basically means it's going to be wrong 20% of the time because the offender doesn't live anywhere near the zone being suggested.

[JeffHamm]

If I turn off the routes, the number of predictors being used is reduced, and that starts to have a negative impact on it's accuracy. At the moment, this is only based upon 3 predictors as it is (the routes being one of them). There are some other potentially useful predictors that can be added, but so far, based upon my testing, adding them in tends to have a greater detrimental effect when it shifts the zone away from the offenders residence than it helps when it shifts toward. I'm hoping to find at least 1 more well behaved parameter that generally tends to improve the solutions, but still working on that. Unfortunately, for the data set I'm working with, I don't have the sequence order, nor time of day or dates, so temporal aspects of a series I can't build in.

[JeffHamm]

Just for fun, I've turned on all of the predictors I have, not all of which have been tested "in use", but all have some preliminary analyses to suggest they could be useful. The output is fairly similar to the above, but not quite. Now, I'm posting this because some will find this to their liking. The section of zone one in the lower left contains Dr. Thyne's surgery, which some have argued might have been Druitt's "bolt hole" (I believe that idea may have been refuted as he may not have had his surgery there in 1888 if I recall correctly). Also, for those of the Doctor school, or Donston, yes, the hospital is in zone one. What I find interesting is the part of zone 1 that is in the north east, it's close to Nichol's (first of a series is often, though not always, a bit closer to the offender's anchor point), and it's on easy access routes to Chapman (who, being likely killed at dawn, I tend to think lends support to the idea the killer was highly familiar with that area and lived nearby; but again, that's just my own speculation).

As a final warning, however, geographical profiling is based upon the idea of finding a stable anchor point for the offender. Many people in Whitechappel were highly mobile, and might move from doss house to doss house. There is every reason to be cautious as Jack may have been transient, and so the anchor point could be moving. If so, this series is highly unsuited for attempting to locate him - it's not great with moving targets (unless they just shift a bit, there is some tolerance, but if he's starting out from a different location each time, well, the analysis won't apply. It's just math not magic so will always produce something, but if it is garbage in it's garbage out.

[Robert]

Hi Jeff

https://wiki.casebook.org/black_lion_yard.html

[Abby Normal]

I've done the analysis below, but I'm not sure where Black Lion Yard is located? Kaminsky, in the A-Z map, would not be in zone 1 though. While it might seem intuitive that zone 1 should be very central, offenders tend to live closer to the edge of the crime zone than the centre (not always, and there are exceptions, but this sort of analysis is based upon probability distributions so it tends to weight very central locations as low probability areas). As can be seen, once Tabram and Millwood, are removed, the pattern of offences no longer has a North-South tendency on the west side. The data set of solved crimes that I have to work from, which is limited - and the crimes are arson, which is a different offense type, finds that in these circular type of offense patterns the offender is often located on the same circle, hence the sort of doughnut shape that's forming. I'm not sure how much I would trust the potential routes (red and green lines) in this case, as really, there's no strong pattern of clustering along the routes (as happens once the Millwood and Tabram locations are included; a N-S type corridor starts to form in the west). But, for the sake of keeping all things the same, I've left them turned on. The area around Tabram has diminished, but tow areas close to Nichols (now the first in the series) start to be suggested. Also, the Goulston Street area remains fairly high (it's now zone 3, but that's still quite high). I've been working on a different approach than the routines that are out there, and while similar regions are sometimes suggested, it is not uncommon for entirely different outputs to result. Again, given that there will always be some in the "low probability zones", it may be that where we differ is one set that are in my low zones are in their high zones, but another set that are in their low zones end up in my high zones. Also, and I can't stress this enough, my stuff is still very experimental and under development. It also needs a lot more testing, and verification. So far, based upon the solved cases I've been able to track down online or in academic papers, it's done pretty good with serial murders despite the data I use to work out the algorithms are arson cases so either I've been lucky so far, or there is something to it.

One thing I would really like to be able to work out is some indication as to whether the offender is likely to be a maurader (meaning lives within the crime zone) or a commuter (travels to the crime zone from outside of it). I believe it's approx 20% of solved cases for serial rape and murder are commuters, and geoprofiling tends to assume a maurader (since that's the majority). That basically means it's going to be wrong 20% of the time because the offender doesn't live anywhere near the zone being suggested.

thanks jeff
whats zone one on this? the pink areas? also can you lable the colors as to there zones?

[Abby Normal]

I think it would also be interesting, once the zones are plotted, say for just the C5 to keep it as conservative as possible, to also plot the locations of known suspects to see what zones they show up on. For example we know the addresses of the following suspects: Kaminsky, Hutchinson, Lechmere, bury, kozminsky, etc. and any others where we know there addresses in fall of 88.

[Sam Flynn]

I think it would also be interesting... to also plot the locations of known suspects to see what zones they show up on

I've made a start:

[JeffHamm]

thanks jeff
whats zone one on this? the pink areas? also can you label the colors as to there zones?

Thanks AbbyNormal, Yes, pink is zone 1. I've put the key on the output map below (using the last output map posted above), and I've tried to locate as best I can some of the suspects based on the map in A-Z. Each zone is the same size in area (same # of pixels). Take the two most distant crimes (Eddowes and Nichols) and extend beyond each one by 12.5% of that distance, then use that to draw a circle. That's the area of the crime zone where Mauraders are found. If you randomly searched, you would have to search 50% of that on average to find a maurader, so take 1/2 that area. That's the average area you would have to search randomly to find the offender if they were a maurader (and roughly 80% are). Take 5% of the chance area, and that defines the area of a zone. So, if an offender lived in zones 1-20, it's doing better than random searching. By zone 40, you've searched the equivalent area of the crime zone. So, if the offender is not there, you might search a ring around the crime zone of equal area. And so, cyan and blue extend the search similarly; if the offender is a communter and lives outside the crime zone, but in the cyan zone, you're doing better than chance (although you had to exhaust red/green because we start with the higher probability of a maurader).

Anyway, I'm not sure where on Commercial Street Hutchinson was; but if people know, then I'm more than happy for people copy this down to make additions and updates. Feel free to place suspects, or other important locations, on any of the maps I've posted.

- Jeff

[Abby Normal]

I've made a start:

hahahahahahaha!

[Abby Normal]

Thanks AbbyNormal, Yes, pink is zone 1. I've put the key on the output map below (using the last output map posted above), and I've tried to locate as best I can some of the suspects based on the map in A-Z. Each zone is the same size in area (same # of pixels). Take the two most distant crimes (Eddowes and Nichols) and extend beyond each one by 12.5% of that distance, then use that to draw a circle. That's the area of the crime zone where Mauraders are found. If you randomly searched, you would have to search 50% of that on average to find a maurader, so take 1/2 that area. That's the average area you would have to search randomly to find the offender if they were a maurader (and roughly 80% are). Take 5% of the chance area, and that defines the area of a zone. So, if an offender lived in zones 1-20, it's doing better than random searching. By zone 40, you've searched the equivalent area of the crime zone. So, if the offender is not there, you might search a ring around the crime zone of equal area. And so, cyan and blue extend the search similarly; if the offender is a communter and lives outside the crime zone, but in the cyan zone, you're doing better than chance (although you had to exhaust red/green because we start with the higher probability of a maurader).

Anyway, I'm not sure where on Commercial Street Hutchinson was; but if people know, then I'm more than happy for people copy this down to make additions and updates. Feel free to place suspects, or other important locations, on any of the maps I've posted.

- Jeff

thanks Jeff
very interesting! fyi-I don't know hutchs exact address but he was at the Victoria home.