Thursday, October 12, 2017

Plumbers, police, and hotspots

Now Chris Blattman has new paper on policing which is, willy-nilly, being cast as raising doubts on "hotspot" policing - use of digital technologies for intensive policing of areas where crime is concentrated. 

The paper finds that while such policing "deters crime and violence" in those areas and "reduced the most serious violent crimes" (rape and murder) in the aggregate, it had limited impact on the "number of total crimes deterred". But it found "spillovers", "pushed property crime around the corner" etc. The tone of the paper (and you can just browse to feel what I mean) unmistakably gives the impression that the paper evaluated "hotspot" policing and found limited benefits. 

Given the framing, it is only natural that people will raise doubts on "hotspot" policing. I can also imagine this paper triggering off more research on distant concerns like spillovers. While nothing as damaging as this can happen with this paper, the costs of such digressions are non-trivial.  

This is all very unfortunate and a testament to the state of development economics research. I actually think that the paper's primary endeavour itself is questionable. Why do we need to test the efficacy of "hotspot" policing?

To answer this, we need to deconstruct "hotspot" policing. There are two elements here. One, prioritising and targeting work. Two, using digital technologies (data analytics and visualisation dashboards) to help with decision-support on the prioritisation. 

Do we need evidence about these two elements? In more simple terms, in a system with scarce resources and several competing needs, isn't it a natural order of things to prioritise them using the most effective reporting and monitoring mechanism possible? 

Have we ever sought evidence on whether great Dashboards are a step in the right direction in the corporate world? Did the first set of corporate users demand evidence from IBM before placing orders for Cognos?

I see Dashboards as a logical progress in the transition of data management from the far less user-friendly stages of massive paper Registers and the confusing array of rows and columns of Excel sheets. Registers and Excel sheets contain data. Dashboards contain information. 

There are two logical criticisms to "hotspot" policing. One, given that actual decisions get taken at smaller jurisdictions, the human agents may have the bandwidth to be able to have a good real-time mental picture of the hotspots. A Station House Officer or Police Inspector, especially in urban areas, who has served for a couple of months, should have a fairly good idea of the high crime incidence areas, people, and categories. So why need a "hotspot" map? Two, even if the police manager is able to prioritise, he cannot, for contextual reasons (say, political complusions or there are too many high crime areas and too few constables to spare), deploy staff as he/she wants. What is the need of something that cannot always be acted on?

The responses are also two-fold. One, it may so be that police managers already have a graphic mental knowledge of all this stuff. But a mental map is a private information. Instead, information captured in this manner is amenable to effective monitoring across the organisation and is therefore a public information. This is true of any organised hierarchical system. Two, the objective should not and cannot be to have precise prioritisation and its implementation by all police managers. Even if a small share of them start using it, which is most likely in any large system, in a decidedly second-best manner, that itself would be a big progress. We would have set the stage for diffusion of an undoubtedly more efficient approach to policing. Given the weak state capacity in police systems, any improvements in monitoring capacity enhances administrative efficiency in the aggregate. 

Or the researchers could claim that they have generated evidence which can help limit the damage with fancy ideas which lead to massive wasteful spending. But how costly is a freeware application like this, this, and this that police jurisdictions can use to translate their data to actionable information? 

I see several pathways to change. One, some of the more enthusiastic police managers in a system find great value in using such technologies to prioritise and monitor their beats. Positive deviances emerge. Two, as Justice Brandies said, sunlight is the best disinfectant. Hotspot applications can shine light, bright and deep, into the crime data collected by police systems. Actionable information emerges. At least some insiders and outsiders can learn and act. 

I am unambiguous and unashamed in support of "hotspot" policing. All policing jurisdictions should, among other things, strive for spatial (and on temporal and human dimensions too) prioritisation and targeting of their policing efforts and they should use technologies like "hotspots" and other types of Dashboards! Of course, needless to talk about all caveats of data privacy and data biases, and the need for their mitigation.

This is not all that is bad with the paper. Spending too much research effort dwelling on displacement is a psychologically misleading illusion. Talk of constructing a fictitious straw-man and then refuting it! Why did we ever think that there would not be any displacement? Isn't it natural for thieves to move to the margins if an area suddenly becomes more intensively policed? Who said crime fighting is a binary game? It is far from the case that if we focus efforts in an area, crime will be eliminated.  

Policing is a repeat game. Actually, police officials, unlike economists, deeply internalise this plumbing reality. Focus on the existing hotspots and reduce incidence there. Respond to the emergent trends and redeploy. Keep iterating and aggregate crime will reduce gradually. If available, even use machine learning applications that have predictive analytics to help double down on the displacement locations. And alongside, also take all the other complementary steps required, not just including those by the police, to address crime. At least some police jurisdictions will have the good fortune of confluence of positive factors that contribute to the emergence of positive deviances in a few years. 

And in several policing concerns like traffic safety, where accidents happen more because of location characteristics and are less likely to be "displaced", just "hotspots" visualisation can be useful as decision-support. 

The most disappointing part comes somewhere near the end (italics mine),
But if crime is easily displaced, then targeting, coordinating, and concentrating resources in high-crime places may not be the right approach after all. Rather, it might be wiser to target the specific people who commit crimes or particular behaviors. Displacement may be inherently less likely than in place-based approaches. This is the spirit of focussed deterrence, which identifies the small group of people who commit serious crimes and use threats and incentives to keep them from offending (Kennedy, 2011). This is also the spirit of cognitive behavioral therapy, which fosters skills and norms of non-violent behavior in high-risk young adults (Heller et al., 2017; Blattman et al., 2017).
In an ideal world, we need to address the underlying socio-economic reasons for crime and several other dimensions, including focusing on the most riskiest people. Many of these dimensions are beyond the control of police departments. And in the real world, we all know the difficulty, even impossibility, of breaking down silos and addressing problems in a comprehensive manner. Should we wait for that ideal world to arrive before we try this out?

Who says "hotspot" targeting trades-off police system resources with "person" targeting? How can we say that it should at all be one or the other? Wouldn't measures to target specific people become likely more effective when the deterrent effect on places is higher and vice-versa? 

This is pulling stray research strands, the only intersection being that they were all done by the same researcher, and generating a grand policy narrative about addressing crime and violence. Frankly I just don't have the energy to go on. 

Talk of writing a paper on the obvious! And even making a splash out of it! Why is academic research so plumbing-free?

No comments: