+44 (0)20 7799 6677

An author responds to Richard Garside’s Police, Prisons and Crime Rates

nick cowen, 13 July 2012

Guest post by Dr Marianna Koli

Department of Economics, University of Birmingham

We add to Nick Cowen’s previous response to Richard Garside’s Police, Prisons and Crime Rates. We thank Richard Garside for his comments. However, from a methodological standpoint we would like to respond to some of his concerns.

While Garside rightly draws attention to some of the one-sided press coverage, his own comments contain the same kind of incompleteness. Similar to some of the press coverage on our study, he fails to report the significance of our finding regarding short custodial sentences, and writes: ‘[a researcher] in search of prison as the answer to crime will surely find it’. However, we write in our paper: ‘Particularly for robberies, it seems short sentences are counterproductive… For those offenders who are candidates for short sentences, non-custodial alternatives could be considered.’ In the case of robbery, we find 36 months the minimum period where a custodial sentence has a statistical crime-reducing impact. We state a number of reasons for this, including poor access to rehabilitation programmes for those serving short sentences. We do not find prison is the answer to all crime. If anything, we are arguing for less prison for some criminals. We do believe that our results suggest longer prison sentences are appropriate for some very serious offenders.

Garside states that of all the variables used in our study, the presence of young people ‘has the strongest impact on recorded crime rates’. We can only assume this comment is based on the youth coefficients being the highest of the variables studied. However, there is no statistical meaning that can be attached to the relative sizes of regression coefficients between different types of variables. In other words, detection rates and sentence length are just as significant as the presence of young people in determining crime rates. We do find a larger proportion of 15-24 –year-olds in an area correlates with lower crime rates, and this is consistent with the basic labour economics theory that individuals implicitly assess both current and likely future income when deciding on present actions. Future income is adversely affected by engaging in crime today, and young people stand to lose the most.

Regarding Garside’s comments about the value of quantitative studies, we disagree. Such studies are sorely needed and a valuable contribution in such a contentious and politically and emotionally charged field of study. Unlike Garside in his past work (linked to within his commentary regarding our work), we do not engage in judgment on whether any particular offence, such as speeding or possession of cannabis, should land one in the Offenders Index, or indeed make any other comment regarding crime-recording practices. Making such value judgments is the domain of democratically elected politicians and lobbyists, not of academics. We analyse offences recorded in the past 20 years with the various crime-recording systems that have been in place in that period, and find that there is a benefit to incapacitating would-be serious re-offenders through custodial sentences. Regarding criticism of using recorded crime data, as Nick Cowen ably demonstrates in his response, the crime rate trends displayed by our data are likely to reflect reality.

The classical economists that devised the idea of rational human behaviour fortunately recognised that rational decision-making does not have to entail a conscious and formal list of pros and cons, as alluded to in Garside’s comment: ‘Rational calculation might inform consumers’ decisions to switch electricity suppliers. Applying the same logic to the complex social relationship that is crime is a mystifying simplification.’ Perhaps Garside believes criminals do not respond to incentives. It is, however, well documented in a number of criminal biographies and testimonials (e.g. Leech, 1993) as well as qualitative academic studies of crime (e.g. Shaftoe, 2004:69; Hallsworth, 2005:109-141) that criminals do make rational lifestyle choices. Richard Garside’s implicit suggestion that criminals are mentally and emotionally a different species from the rest of society leads to myopic, misleading and dangerous conclusions. We believe in a criminal’s ability to think just as logically – not more, not less – as a law-abiding citizen.


Hallsworth, S. (2005) Street Crime. Willan Publishing, Cullompton, Devon.

Leech, M. (1993) A Product of the System: My Life In and Out of Prison. Victor Gollancz Ltd., London.

Shaftoe, H. (2004) Crime and Prevention: Facts, Fallacies and the Future. Palgrave Macmillan, Basingstoke, Hampshire.

11 comments on “An author responds to Richard Garside’s Police, Prisons and Crime Rates”

  1. To those support less prison sentences. I’ll inject with the usual “thickies” comments here: I don’t much care as to whether the criminal is rehabilitated or not. The reason why is because it is so self-evidently a clear choice: Either support longer prison sentences, or tolerate innocents experiencing violence and fear, thereby granting tacit approval for more of the same. Theft is a nasty personal attack made by someone who evidently has callous disregard for others sufficient to make them a danger. The fact there is less a correlation between this act receiving a sentence and reoffending is irrelevant. I find it neither clever nor civilised to risk it. In fact, I rather think it immoral to put others in danger, while also tolerating the wider societal effects of crime.

    My argument for long prison sentences doesn’t rest, therefore, on the myopic perspective of a criminal’s wafer-slim chances of redemption. Nor, that said, the desire for vindictive revenge. Simply, it is to note that there is nothing at all kind or civilised in an act of knowingly encouraging and accommodating incivility or violence.

    Instead of this myopic, ‘far end of a f***’ debate, therefore, why not study the more important eco-system of events that follow a criminal act. The impact on the victim. The impact of their subsequent behaviour on others. The costs (e.g. NHS, lost income, insurances). The opportunity costs. Look at the message sent to others who might consider similar acts; the effects on their victims et al. The effects on a locale. Then, look at the expansive potential of long prison sentences: My bet is that a three year sentence for theft (et al) would set a deterrent far more effective than any half-baked early intervention programme. Such is the deterrent, it also, I’d bet, steers people away from the serial decisions that typically line a path to criminality; the wilful abdication of responsibility to work and pay one’s way; casual vandalism and violence, drug abuse, and all other such choices.

    Work out the costs of crime (and insurances don’t forget) and police and hospital bills that could be then saved by long sentences; and then compare with the benefits if diverted into meaningful, positive programmes to care for the vulnerable and/or break down socio-economic barriers for the countless majority of poorer young people who would benefit. Count up what happens when comunity areas get regenerated, that are presently left to rot (with their inhabitants) as hopeless causes.

    Work out this equation – that of whether locking up, for far far longer, the micro segment amongst us, of thugs and thieves, makes society better or worse. My bet is its a “no-brainer” (apt, you say). To many who read this, its a thickie’s argument: So, perhaps answer it then if it so simplistic. What are you seeking to achieve as your objective? To me, I admit my own sentimentality in that I find one cracked skull, one blighted life of an innocent person is too much to tolerate, but then you evidently don’t think so? So to me, much of this focus on the criminal and their “rehabilitation” comes across as nothing more than self-indulgent, sickly-sentimental displays of “cleverness” that I can barely conceive are even sincere.

  2. We appreciate your comments on the methods we employ in our study but I would point out that Fixed Effects IV is a standard and widely accepted method for such analysis (see, for example, the celebrated paper by Machin and Meghir, 2004 we cite).
    If you take a look at how Fixed effects IV estimation is carried out you would no doubt realize that your concerns are unfounded and the diagnostics we have carried out and the tests we have reported are appropriate. There are several good books (e.g. Wooldridge) which you can refer to if you wish to familiarize yourself with the methods.
    The paper is online and if you have further questions, please feel free to email us.
    In any case, we thank you again for pointing out a transcription error when reporting the findings in our tables.
    The paper has been peer reviewed.

  3. With the greatest of respect, Siddhartha it isn’t for me to respond to Marianna’s comments. I’m not making any claims about the effectiveness of prison sentences or otherwise and I’m not the co-author of a report that has been picked up by the media as evidence for more a punitive sentencing policy. I don’t have to answer the questions because I’m not making any claims. I am, however, asking questions about the methods employed in your study which in some cases I believe are questionable.

    I notice my point about peer-review was ignored. So I’ll ask again, was your original paper put out for a blinded peer-review process so that it could be scrutinised by independent and suitably qualified academics prior to publication? It is boldly asserted in the introduction to the report that “there are few rigorous studies which look at the impact of sentencing on crime” – This really isn’t my area at all (I’m not even a quantitative researcher really), but I’d be interested to hear what Professor Julian Roberts at Oxford University has to say about that. I’m pretty certain his work on sentencing is rigorous. It also happens to have been published in peer-review journals.

    And did you or did you not run collinearity diagnostics on the predictor variables that you entered into your model? If not, how do you know that the regression coefficients on which you have based subsequent predictions about crime rates are accurate?

    In any case, there is a fundamental flaw in any predictions that might be drawn about the impact of longer prison sentences. You acknowledge elsewhere in the report that most determinate prisoners serve 50% of the sentence in custody. The rest of the sentence is spent under licence supervision. Some prisoners are even released early on home detention curfew (tagging). So when you have used ‘sentence length’ or variations of this as a predictor variable for subsequent crime rates, how can you be sure that any reduction in crime is an outcome of the custodial element of the sentence? Could it not be the case that a longer sentence might, in fact, be efficacious not due to the greater exposure to custody, but to the greater length of supervision in the community? i.e. you can’t tell if it is the custody or licence supervision that is leading to the negative regression coefficients.

    A complex picture indeed.

  4. Nowhere have we talked about prison sentences as a solution to theft. Nor do I see that picked up by the press. Theft and sentencing appears nowhere in the press release either. Further note, notwithstanding the typo I had said:
    ‘Our analysis finds a strong and negative relationship between detection rates and crime and a similar negative relationship between sentence and crime for three of the four crime categories (with it being statistically significant for two out of the three).’ (p.3)
    We sincerely apologize for the typographical error in transcribing a coefficient from the table into the discussion in the text (which we have now changed in response to your comment). However, we emphasize that our discussion about the efficacy of sentencing was almost entirely on the other three categories (burglary, fraud where we introduce a non linear specification in addition to the linear to consider the possibility that short sentences may be counterproductive-see Tables 1 and 2). I am therefore surprised that you continue to ignore the substantive message of the paper and choose not to reply to Marianna’s earlier comment.
    We can certainly disagree about whether you believe criminals respond to incentives and on the efficacy of prisons but I hope we can have a substantive debate about that. Regards our position on prison, It clearly states (p.12):
    ‘Our interpretation of this is that prison can have very different effects on different offenders. This suggests that an optimal policy will target repeat and serious offenders for long sentences while using alternatives to custody for other offenders. This complex picture offers some support to the view that ‘prison works’ as an important way of reducing crime while warning that prison sentences can also be misapplied.’

  5. Yes, Marianna, but it’s a very obvious error – and based on it you’ve made a claim in your paper which patently is not true – “Sentencing is negative and significant for all crime categories except robbery which is positive but insignificant”. This is evidently not the case, theft was negative and non-significant (rather than ‘insignificant’).

    Now I don’t know a huge amount about statistics, but even I picked up on this and it leaves me wondering how many other errors there are in the paper and why weren’t they picked up on during peer review? I’m assuming that this paper was subjected to a process of peer review? After all, if it’s the basis of a Civitas report which has made national headlines surely it’s very important that the research has been subject to a high degree of academic scrutiny.

    Whilst we’re on the subject of non-significance, the p-values for each of the linear (table 2) and quadratic (table 3) regression models all appear to be non-significant. And the R-squared values for these models are hardly anything to get excited about. For instance, the R-squared for your quadratic model for burglary (which is the highest of both the tables) is 0.26 – so it accounts for only 26% of the variance in your outcome variable (crime rate)… and that’s the best one.

    And why are there no collinearity diagnostics reported? How is anyone reading the paper supposed to know that there isn’t a problem in your models with correlation between the predictor variables? If you don’t report the tolerance values or Variance Inflation Factors, people will just assume that they weren’t done and therefore the reliability of your regression coefficients becomes very questionable.

  6. So why report this as a main effect? There is a danger that people won’t read as far as page 8 of the report.

    If as you say the statistical work is rigorous, why on page 8 of the original paper (Section 3: Results – A: Linear model) is there a different mistake with the reporting of the same statistic? Here it is reported that “For theft the corresponding figure is 0.05%”.

    One typo is a simple mistake, two different typos with the same statistic looks like carelessness.

  7. Elsewhere in the report (page 8) we state the theft coefficient is not statistically significant, and the sentence you refer to has now been clarified in the paper. The statistical work in the paper is rigorous and the commentary reflects our results accurately.

    We would be interested in comments regarding our finding of very short prison sentences potentially being counterproductive to crime reduction.

  8. One has to be very dubious of a report which claims as a ‘main result’ at the bottom of page 3, “For sentences, a 1% increase in sentence decreases…theft by 0.16%”.

    This is wrong on two levels.

    Firstly, the reported regression co-efficient in Table 1 is -0.0156 (so it should be 0.016% and not 0.16%).

    Secondly, the standard error reported in the table is big enough for this co-efficient to be discounted as not significant.

    You have to ask the question, if very basic errors like this appear in report – has the correct degree of rigour been applied to the rest of it? Problem is, journalists pick up on these ‘main results’ and take them as gospel.


Keep up-to-date with all of our latest publications

Sign Up Here