Okay, so those of you who have read my blogs will be familiar with my views on numerical targets in policing – in short I argue that they should all be abolished, without exception. This is because they generate perverse incentives and behaviours, drive waste into the system, skew operational activity, encourage the practice of ‘gaming’, and result in a less efficient service to the public.
But let’s put all these terrible consequences to one side for a moment, and look at numerical targets from a different perspective…
Being an optimist, I don’t actually believe that those who advocate or impose targets are necessarily bad people. Command and control through targets has been the norm for years, and at first glance some targets may even seem appropriate. Politicians and public service managers install them in the hope of improving performance, increasing efficiency and accountability, reducing costs, and so on. Let’s face it – it would be a bold move for such a person to genuinely discard management through targets if it was all they had ever known and trusted in. It could be quite an uncomfortable Epiphany for some.
So, for the purposes of this experiment, let’s explore the possibility that it could be feasible to set a target that is appropriate within a public services setting, and which comes with a warranty never to cause any of the horrors described above. A target that can only do good. A target that is absolutely proper, and completely defensible against any statistical or scientific argument; the perfect target.
Right, how do we set it?
I suppose the first logical step is to choose an area that is considered to be important enough to measure. This makes sense to me (genuinely), as I believe that we need appropriate indicators which feed into a proportionate performance measurement system. After all, if we don’t measure anything, we don’t know how we are doing, but worse still, we will miss important changes in the data which act as a signal (see my blog ‘Stay Calm and Don’t Knee Jerk’ for more on this, and Statistical Process Control charts in general).
The other important ingredient in effective performance measurement and operational deployment is a set of clearly defined priorities. So, for the purposes of illustration, a sensible and measurable policing priority could be to strive to reduce crime. No one is going to argue that the police shouldn’t try to reduce crime or arrest criminals, after all. (I’m keeping it simple here for the sake of the next bit - I appreciate external factors such as social deprivation, substance abuse, economic drivers, and other determinants affect crime rates, and it would be arrogant to suggest that crime reduction is totally within the gift of the police to control).
So now we have a clearly defined and measurable objective against which to set a target and measure subsequent performance. Let’s give it a try using widely-adopted techniques.
This chart represents the crime rate over a period of twelve months (or it could be twelve weeks, twelve years etc – it doesn’t matter). The actual figures that would normally appear on the vertical axis aren’t important, but for this demonstration let’s assume that the mid-point (horizontal blue line) is 1,000 crimes. The variation in the data ranges from 850 crimes (points 4 and 9 on the horizontal axis) to 1,100 crimes (point 2 on the horizontal axis).
For what it’s worth, the chart displays a good degree of control, with limited variation, suggesting that the system is stable and setting a target won’t actually affect the data. But that shouldn’t stop us trying. Here are some common methods of setting targets:
Method 1 is the tried-and-tested “add or subtract a few percent from last year’s figure”.
Method 2 is called “let’s use last year’s figure as a benchmark”.
Method 3 is simply a variant of Methods 1 or 2, which is to choose any of the data points (or just pick a number out of the air) and designate that figure as the target.
In the example of the crime rate chart above, this would mean that the crime rate at point 12 could be interpreted completely differently depending how the target had been set:
1. If the target was to reduce crime by 5% compared to this time last year (point 1 on the chart), then it was right on target, coming in at 950 crimes (point 12).
2. If the target was set using point 2 on the chart (1,100 crimes) then we see a huge reduction of 150 crimes at point 12 (almost 14% reduction). Well done!
3. If the target was set using point 9 (850 crimes) then it’s bad luck for you, as crime at point 12 has gone up 11.8% compared to this point.
The fact is that as the systemic variation in this chart is already stable and within the limits (horizontal red lines), setting a target within these limits will have no effect. (Other than to cause the type of unpalatable consequences mentioned briefly at the beginning of the article, and I did say I wasn’t going to go there, sorry). The reason it will have no effect is because the normal variation will be unaffected, meaning sometimes the target will be met and sometimes it won’t.
If a target is set above the upper limit it will be unattainable, as it is outside of the capabilities of the system. If it is set below the lower limit, there is no incentive to maintain current output and there is a risk that performance will deteriorate to meet the target.
So, where (and how) can you scientifically set the target?
The problem with all of these approaches is that they are entirely arbitrary. Each of the data points is subject of normal variation, so to designate one of them as a target is exactly the same as saying ‘that figure is normal and should be aspired towards’. Why?
How can it be ‘normal’, immune from natural variation or external influences? Why is it ’normal’ compared to the week before, or six months previous? How can it be set in stone as the benchmark, or more bizarrely, be amended by a couple of percent in one direction or the other to generate a target? Where do these percentage adjustments come from? How are they calculated? Even if a comparison is made against a long-term average, guess what – about half of the time the figure will be above average, and about half of the time it will be below the average! Target-driven performance management operates in a binary world of either ‘everything is doing just great’, or ‘things are getting worse’. There is no other position. Reality is not like that.
Even if you want to try and set a numerical target for all the right reasons, it just doesn’t make sense. Listen out for hollow proclamations of success that begin with statements such as “…compared with this time last year…” or “…the target for ‘X’ has been exceeded in 95% of cases…” etc. (This latter example is even more extreme as it is a target on a target! 95% of what?) Credible? You decide.
Another major problem with attempting to set numerical targets is that (as Deming pointed out) the system itself is responsible for around 94% of performance. Setting a target does not affect the system, as the system does not understand targets. Simply exhorting the workers to work harder or to try and achieve an aspirational numerical goal does not work. Even if it did, this tactic could only ever affect 6% of performance. Why would an organisation put so much effort into an area of such minimal leverage, when there is that big 94% just waiting, begging to be improved so that it can fulfil its potential?
Furthermore, to base targets on the current output of the system is to admit defeat. It is like accepting that we can’t do any better; that the system must be at capacity. The fact is that the system is capable of a lot more but is clogged up with the wrong stuff.
The failure to understand that all public sector numerical targets are a) completely arbitrary and b) scientifically impossible to establish in the first place, is the first mistake of those who promote their application. You wouldn’t set a target for the amount of hours the sun shines in a day would you? Why not – what’s the difference?
The second mistake is to react to whether normal statistical variation meets these targets or not, as it is exactly the same as reacting to something that isn’t there. The worst case scenario is the emergence of those hideous consequences I said I wouldn’t talk about; the best case scenario is that you just waste your time and effort, cause valuable resources to be diverted from elsewhere, and have absolutely no impact on what you are trying to achieve.
As the greatest opportunity for performance improvement lies within the system, this is where effort should be focused. The first step is to reduce waste. Waste is the activity within a system that does not provide value to the service user, such as unnecessary internal reporting requirements, or time spent reworking what wasn’t done properly in the first place. If waste can be reduced (or ideally, eliminated), this generates capacity that results in a more effective system and improved service delivery. These improvements outstrip anything that even the most ambitious numerical target could aspire to reach.
I argue that the only goal worth striving for is 100%, all of the time. Surely this can be the only perfect target. Realistically, we are never going to eliminate all crime or catch 100% of car thieves for a variety of reasons, but that should not stop us aiming to reach far beyond the artificial constraints of the arbitrary numerical targets we are subject to. ‘Aim for the stars and you might hit the moon’, as they say. A numerical detection target of 10% is like aiming for the top of a bungalow.
Finally, aside from the rights and wrongs, whys and wherefores of numerical policing targets, if anyone out there can present a sound, statistically robust, scientifically rigorous theory for determining them in the first place, give me a shout. I contend that it is impossible.