Efficient way to fight crime: Just go where the crashes are
»Play Video
CLARK COUNTY, Wash. – Police officers and sheriff's deputies are always looking for new ways to fight crime, and the Clark County Sheriff's Office has come across some interesting new data that helps it do just that.
The method deals with statistics every law enforcement agency keeps on two issues: Where and how often property crimes, like burglary, occur, and also the same information for car accidents.
They're normally examined separately, but then Clark County Sheriff's analyst Brian Salsig crunched the numbers together.
Salsig – a facts and figures guy – is a math-oriented crime fighter. So when he recently learned about new research connecting car crashes and property crime, he hurried to crunch the numbers.
"I immediately came back to the office and pulled up the graphs and overlaid them, and what do you know, they're absolutely right," he says. "Where we're having our property crime issue is where we're having our traffic accident issue."
Analysts took a map of unincorporated areas of Clark County that showed traffic collision hot spots and compared it to a map of the exact same area that showed property crime hotspots. When they combined the two together, they noticed obvious overlaps and one hotspot stood out: The Interstate 5, Highway 99 corridor.
Deputies plan to use the information to step up their presence in that area. It’s an easy and effective way, they say, to discourage criminals.
"I wish we would have made this correlation years ago," Salsig says. "What's the most efficient way to patrol? Well, obviously, we want to reduce the number of collisions – that's going to make this world a safer place and, if at the same time we reduce the number of burglaries – Holy cow! That makes life really good, right?"
The big question remains as to why this connection exists. There's no solid answer, but one theory is that criminals tend to have a risky lifestyle and that carries over into their driving habits.
Of course, not everyone in a crash is a criminal so the connection may need to be examined a bit more.
Given the current reduction in police budgets, it is difficult enough to respond to most crime scenes to take a report, much less to be proactive in being visible as a deterrent. Â Â
I wonder if they have figured out why there is very little crime in unpopulated areas? The correlation is obvious and if you overlay a population map you will get the same thing. Crime happens where people are and so do traffic accidents, they do not happen where there are no people, not exactly rocket science
Soon we will be able to predict when someone is going to commit a crime before they actually commit the crime. They will send police to arrest them before they have even commit the crime unless of course there is a "Minority Report" then Tom Cruise will step in and save the day.
Wonder if this happened because of budget cuts were there use to be two graph people one working crime division and one working traffic division. When they got only one person working both departments it all fell into place with an overlay.
Hard to drive maybe when on Meth, stoned, or drunk?
 @FreedomRocks most people arrested for burglary are first stopped by police for traffic infraction and the cop sees all the neat stuff you have in your car with the cables cut off and makes them wonder about your form of moving as you told the officer you were doing.
Makes sense, actually. They can also see what kind of population - dense, high turnover, whatever - is in those areas, and also show how economics plays an important factor in these crimes.Â
Sounds interesting, at the very least. I like the idea; hope it works out well.
From story:
' "I wish we would have made this correlation years ago," Salsig says '
Â
Basic axiom #1 from introductory statistics: Association [correlation] does not imply Causation.
Just because there is an association, that does not imply that one necessarily causes the other.
There probably is an even better predictor that hasn't been yet identified; or maybe more accurate to say, 'uncovered' using current methods / approaches.
Â
This line is rather interesting too:Â
'Analysts took a map of unincorporated areas of Clark County that showed traffic collision hot spots and compared it to a map of the exact same area that showed property crime hotspots'
Â
Crimes typically follow and occur in areas where there are denser populations. Â Why not include incorporated areas too? Â At the very least, two different analyses could be done: 1 one for the unincorporated areas, and 2. one for the incorporated areas; and from those, the results could be compared.
Â
Most local municipalities are pretty bogged down just keeping their databases current, fighting their infrastructure, etc. that they really don't get much of a chance to sit down and focus very well on doing 'pure' spatial analysis of phenomena like this. Â Unfortunately, most advanced degree programs in this science tend to focus in areas outside of these techniques (especially grid/raster-based modeling of data variables mapped to a 'surface') -- doing so can yield a range of possible unique results with varying degrees of rigorousness and/or utility.
 @ThePosterFormerlyKnownAsPhredE These data were examined by an employee for the Clark County Sheriff's Office, which patrols unincorporated Clark County, which explains his focus on the unincorporated areas. The data are kinda interesting though, so the analyst would be doing his city police colleagues a favor if he crunched a few numbers for them too..
 @StealthActivist Right. I kinda figured that what was going on.  If the Clark CSO can get help from the county GIS people, they should slap each variable of interest onto a grid and then analyze them in various combinations. For example, in grid format it is relatively easy to apply mathematical operators to find the overall [combined] min/max/mean/etc -- that's mostly a data handling issue.  But, if one wanted to create a full- blown predictive statistical model (as in: yCap = b0 + b1(Var1) + b2(Var2) + ...).
Several commercial GIS packages have this functionality built in, in some way.Â
Â
Where yCap = Probability of crime, Var1 = Distance from previous crime, Var2 = Distance from previous collision, Var3 = Age of Resident [as an example], Var4.. etc.)
Use whatever and how many variables as imagined  I guess.. Then basically, you'd get one data layer (grid) having cell values over the entire area containing a probability given the model inputs.  There are rather large questions of serial correlation/autocorrelation in applying such a model, but I think it would be useful overall to have *some* baseline to compare against.  Also, I'd guess that a predicted map layer created using data for the 11am hour might be very different from one using data at 11pm.  Further, it is possible to weight each variable to assign relative priority / importance and doing such thing might help account for factors like the time difference too.
Â
Â
Just like Numb3rs TV show :-)
Â
 @Oh My Exactly my thought!
American society is imploding and all the reactionary crime predictors will never get to the core of the problem.  Our culture is rotting. Our families are broken, our cities and infrastructure crumbling, our government corrupted and nonfunctional and our economy is on the verge of meltdown. This has not happened overnight, but has been a steady, year by year, decade by decade process.
Â
I know I will tick many people off by saying this, but until we, as a nation, return to our Judeo/Christian morality we will continue down the road to complete and total ruin. Until we realize and believe that the strength of our nation starts with the strength of our families we are doomed. We cannot depend on our stupid, immoral and impotent government to raise our children. Too many of us have become spectators in life and have exchanged the God of the Bible for the god of entertainment and laziness.
Â
I know, I know, many of you will say that Christianity is corrupt and that the Bible is a fairy tale. While I will agree that corrupt people have corrupted the Bible for their own profit, it is in general a huge reason why our nation has been so successful.
Â
Okay, soap box removed and flak jacket put on. I await your slings and arrows.
 @Scotty9 Well put Scotty. I agree wholeheartedly that we need to get to the root of why these crimes are happening instead of just reacting to them!
Â
My family wasn't religious per say, but we lived by the 10 commandments! We also sat down and ate dinner together, no matter how busy anyone was, we spent an hour together to talk about everyone's day.
 @Scotty9Â
No slings and arrows from me Scotty.
Â
We have gone from a self sacrificing society to a self serving society. It is part of the decline of all once great civilizations.
Â
If one looks to Rome they decayed from a strong republic to a brutal dictatorship and of course fell apart. They had their revolutions and got back to a democracy but then deteriorated once again.
Â
All governments have deteriorate to selfishness and the US is no different. Those that don't think this will happen to us are part of the problem. The arrogance in thinking that we can't fail is exactly the reason we will fail.
Â
"For over a thousand years Roman conquerors returning from the wars enjoyed the honor of triumph, a tumultuous parade. In the procession came trumpeteers, musicians and strange animals from conquered territories, together with carts laden with treasure and captured armaments. The conquerors rode in a triumphal chariot, the dazed prisoners walking in chains before him. Sometimes his children robed in white stood with him in the chariot or rode the trace horses. A slave stood behind the conqueror holding a golden crown and whispering in his ear a warning: that all glory is fleeting." - Gen. George C. Patton
Â
Predictive Analysis. Just like what they use in Afghanistan to find IED's before they go boom. This crash vs burglary map overlay is a little more simplistic, but it does work.
This thing is a big joke. Clearly these guys need to look at other factors. Crime is highest along well-travelled areas. And so are car crashes.  Also, both crime and car crashes are higher where young people live.
 @Bill Rogers Any predictive tool for crime is going to be helpful. It may be a big joke in the short term- obviously you don't send extra patrol cars to an area whenever their is a fender bender, but longer term, its valuable.
Â
If you are responsible for planning manpower forecasts, allocating deputies between various areas of the county, siting future law enforcement facilities or planning support functions (vehicles, communications, etc), then having a good forecast of crime is invaluable.
Â
Surprisingly, one other predictor that seems to be true in Clark County is retail activity. Crime definitely follows concentrations of stores.
That reminds me of some old fishing maps that we would use from the Fish and Game Department showing where most fish were caught. Many of the hot spots indicated old ship wrecks or natural geologic uplift zones.Â
Â
This map is for police like those old maps made for fishermen that lead many to fish in the most popular zones that were readily identified by the game department...leading to more fish being caught there simply because there were more hooks in the water.Â
Â
If I were a fish looking for food I would avoid those areas where there were more hooks and seek out the areas were there were fewer fish caught.
 @Icarus Yeah, but you, as a human being, are assumed to be able to reason and utilize logic in order to deduce the most logical outcome of a set of apparently unrelated statistics. Somehow, I don't hold out much hope that you, as a fish, would retain that ability.
Okay I don't see how this is overly helpful, statically interesting yes but shouldn't an area that is already known for high crime be patrolled anyways?
 @BodagrisÂ
Â
Exactly. While it may be interesting, how does it help you? Instead of using traffic crashes to tell you where the crimes are occurring, why not just look at where you police reports are coming from?  Perhaps you can use your crash reports to focus on traffic related issues?
Â
Here are better predictors of where your high crime areas will be:Â high density neighborhoods (translates apartments). Low income, broken families, easily accessible drugs etc
 @Scotty9  @Bodagris Indeed. Anything can be modeled statistically.  It is possible to model the occurrence of crime based on factors such as the ones you mention, plus others could be added as well: time of day, proportion of area visible to street, age of resident, residence of convicted criminal (vs not), etc. Basically, a multiple regression model having an array of independent variables as predictors. Some of the variables need not be continuous, could be ordinal (just ranked data), or even binary (on/off, 0/1) variables.  This really isn't anything new, it's been done for quite a long time. The relatively 'new' aspect of the application of the statistically models is developing the spatial framework or making the analysis operation.  But many larger municipalities do exactly this type thing using 'canned' or developed analysis modules in their GIS.
Believe me, crime analysis is already being done, but not to the level where the information gleaned can possibly be used proactively; except if a patrol officer on patrol happens to spot something criminal happening. No matter how you slice the cake, it still comes to the fact that police cannot prevent someone from committing a crime except by "presence". Law enforcement is a reactive tool, and can never arrest a person for what hasn't yet happened.
 @Scotty9  @Bodagris Here, let me offer a footnote to my own comment: Â
Â
The title reads: 'Efficient way to fight crime: Just go where the crashes are'
What that really means to local PDs is 'how to allocate/schedule resources more efficiently'. Â Of course, out here in the general public we'd sort of hoped that the local PDs that are performing to the max of their ability would be doing this sort of thing already.