The standard of proof and crime prevention: A theoretical and empirical analysis

Introduction

In a perfect world, it would be easy for the criminal justice system (hereinafter, CJS) to distinguish between the innocent and the guilty. But our world is far from perfect. In a world of limited resources and epistemic uncertainty, two types of erroneous outcomes are bound to occur: some innocent people will be falsely convicted (false positives) and some guilty people will be falsely acquitted (false negatives). The CJS of an imperfect world such as ours must accept the inevitability of fallibility. Since error is unavoidable, when it comes to managing erroneous outcomes, the best we can aspire to is to reduce or distribute them.

The total number of erroneous outcomes within the system depends on its level of accuracy. The more accuracy, while keeping the other variables constant (mainly, the standard of proof, hereinafter SoP), the fewer the erroneous outcomes: i.e., fewer guilty people will be falsely acquitted and fewer innocent people will be falsely convicted. Accuracy can be achieved by improving both the amount and quality of the data fed into the system (increased technology, etc.) as well as the quality of data evaluators (better training and education of those who process that data) (Kaplow, 2012: 825; Lilquist, 2002: 99). Of course, accuracy has its price and a central matter for the CJS is how much additional resources it is wise to invest in order to achieve that added level of accuracy (Kaplow, 2012: 828; Kaplow and Shavell, 2002: 248). Furthermore, a society that does not conceive of itself in exclusively cost-efficient terms, nor of its CJS as a mere ‘epistemic engine’ (Laudan, 2006: 2), may have other reasons to maintain rules that are not only cost-inefficient but are also anti-epistemic. Modern CJS often have rules that undermine accuracy (e.g., the prohibition against using illegally obtained evidence; the defendant's right to silence; etc.) in favour of other interests they deem more valuable (e.g., the need to constrain the use of state power; the morality of the criminal procedure; the protection of individual's privacy, etc.). ¹ Accuracy is, however, an important value, not only because reducing both types of erroneous outcomes results in fewer unfair verdicts, but also because it is essential for crime prevention. Thus, rules that thwart epistemic ends in order to secure other values must be well-founded. ² System accuracy raises epistemological and normative questions. While from the former point of view the main question is how to design our criminal procedures if we want to reduce erroneous outcomes, from the latter perspective it is what level of accuracy is desirable. If we agree that people are not entitled to the most accurate procedures, hang the cost (Dworkin, 1985: 72), then the question is how many erroneous outcomes we’re willing to tolerate.

Erroneous outcome distribution also involves normative and epistemological matters. The fundamental question is how we should distribute erroneous outcomes, i.e., what type of erroneous outcome is preferable and to what extent. Even though this is the central issue vis-à-vis distribution, it also raises significant technical and epistemological questions, such as how to implement rules that, on the one hand, accurately reflect the normative judgement concerning the relative weight of the different types of erroneous outcomes and, on the other hand, can be easily understood and followed by triers of fact; what procedural rules distribute erroneous outcomes and what is the best mechanism for that distribution. Even though the most straightforward way to distribute erroneous outcomes is through a SoP (i.e., a rule that establishes the evidentiary threshold for asserting as proven some hypothesis), other rules of procedure or judicial organisation also impact error distribution. For example, if appeals are not allowed when there's an acquittal, then it's harder to convict the innocent, but it's easier for offenders to avoid sanction. Similarly, if a guilty verdict requires the unanimity (as opposed to the majority) of a bench of judges, there will not only be fewer convictions of the innocent, but also of the guilty. If, in addition to unanimity, we increased the number of judges on the bench, then, again, we could decrease the number of false convictions, but also increase the number of false acquittals. Choosing the best mechanism for distributing erroneous outcomes and discovering and quantifying the distributive effects resulting from the system's rules is essential for achieving the normatively desired distribution.

Policymakers inescapably need to address both error reduction and error distribution matters. On the one hand, even a system that sets out to achieve the highest level of accuracy will inevitably incur error; and when that happens, it is necessary to decide how to distribute such erroneous outcomes. On the other hand, a system that concentrates exclusively on erroneous outcome distribution without minding accuracy will achieve unacceptable results. Even though much research is still needed as to the relation between error reduction and error distribution, existing research indicates that changes in one affect the other. For example, changes in distribution mechanisms, such as replacing the SoP for a stricter one, may affect the total number of erroneous outcomes (Kaplow, 2012: 814, 825; Lilquist, 2002: 96). Although reform of the law of evidence must address both matters, here we will solely focus on normative aspects of erroneous outcome distribution, i.e., whether crime prevention should be a factor when stipulating the level of the criminal SoP, and the weight this factor should be given.

This article is structured as follows. The next section describes the traditional point of view with regards to how erroneous outcomes should be distributed in the CJS. It postulates that this point of view defends a markedly asymmetrical distribution of erroneous outcomes to considerably minimise the extent to which the innocent is convicted because it neglects the preventive cost of acquitting the guilty. The following section develops this idea by showing, through strictly theoretical arguments, that there are different ways in which the SoP could be relevant for crime prevention. These arguments suggest that lowering the SoP can result in preventive benefits. This part is followed by a section that examines three possible reasons why preventive arguments are not appropriately considered when establishing the SoP. It concludes that ignoring or underestimating the consequences of the level of the SoP on crime cannot be justified by the distinction between statistical and individualised victims, or by that between distribution and concentration of harm or risk of harm, or ultimately by that between action and omission. The next section examines whether it is possible to empirically prove that a lower SoP increases prevention. It concludes that available empirical evidence on the incapacitation effects of incarceration lends support to that hypothesis. The final section provides a conclusion.

Error distribution in criminal prosecutions

There is a strongly rooted belief in criminal law and philosophical literature that, when judging criminal cases, erroneous outcomes must be distributed asymmetrically to avoid, to the highest possible extent, convicting the innocent. ³ William Blackstone's famous quote best encapsulates this view: ‘it is better that ten guilty persons escape, than that one innocent suffer’ (2016). Of course, this phrase is not to be taken literally as a mathematical formula for tolerable false convictions and acquittals (Wilkinson, 2014: 1109). No one believes that a fair CJS is one in which for every innocent person who is convicted ten guilty ones are acquitted. This is not only because a statistical erroneous outcome ratio (be it that or any other) cannot be safely implemented as a SoP, ⁴ but mainly because any ratio (regardless of what that ratio is) cannot guarantee a fair result. ⁵ Instead, the phrase is a slogan (Reiman and van der Haag, 1990: 227) for a comparative value judgment about the relative costs of different kinds of erroneous outcomes: that a false conviction is morally far worse than a false acquittal. This not only means that erroneous outcomes in the CJS are not to be distributed symmetrically, as in civil proceedings, but also that they are not to be distributed in accordance with a weak asymmetry principle. Instead, according to this view, the CJS requires a strong principle of asymmetry so that the risk of false convictions is minimised to the greatest extent possible. Thus, for the adjudicator to consider the accusation validly proven, a very high evidentiary threshold must be met. ⁶ Justice Harlan's concurring opinion in In re Winship clearly describes this view:

Because the standard of proof affects the comparative frequency of these two types of erroneous outcomes, the choice of the standard to be applied in a particular kind of litigation should, in a rational world, reflect an assessment of the comparative social disutility of each … In a criminal case […] we do not view the social disutility of convicting an innocent man as equivalent to the disutility of acquitting someone who is guilty […]. In this context, I view the requirement of proof beyond a reasonable doubt in a criminal case as bottomed on a fundamental value determination of our society that it is far worse to convict an innocent man than to let a guilty man go free (1970: 371).

SoP and crime prevention: A theoretical analysis

The SoP is relevant to crime prevention in several ways. First, because it may deter potential criminals from committing crimes. Second, because it affects the number of incarcerated individuals who are, therefore, unable to commit crimes while serving their sentences. Third, because it can undermine the public's perception of the CJS and, therefore, result in private vendettas or noncompliance with the law. These arguments are briefly outlined in (a)–(c) below. It will later be argued, on the one hand, that if the system is at least minimally accurate, a low SoP is considerably beneficial to crime prevention (d); and, on the other, that increased criminality resulting from adopting a more stringent SoP is not likely countered by a proportional increase in the number of convictions which keeps the deterrence level constant (e).

Deterrence. The deterrence theory embraces a rationalist account of the human person and causes of criminality. According to that point of view, an individual's decision to commit a crime depends on the expected costs and benefits of that crime. An individual will be deterred when the expected punishment outweighs the benefit of the crime. The expected punishment depends mainly on the severity of that punishment and likelihood of being sanctioned. In turn, the likelihood of being sanctioned depends on the chances of the crime being investigated and tried and the defendant being convicted and sentenced. Therefore, the more severe the punishment and more likely its enforcement, the lower the chances that a person will engage in criminal behaviour. ⁸ A strict SoP makes it harder to convict than a weaker standard. Therefore, all else being equal, the stricter the SoP, the lower the likelihood of being sanctioned and hence the lower its deterrent effect and the higher the crime rate and victim tolls. ⁹

Three possible explanations for neglecting prevention

Below, we’ll look at three possible explanations for the traditional point of view's failure to appropriately consider prevention when establishing the criminal SoP. We’ll argue that this view is the consequence of biases or misunderstanding of deontological thinking.

(a) Statistical v. individualised victims. Cognitive bias may be one possible explanation for the traditional point of view's failure to appropriately consider indirect costs of erroneous outcomes when establishing the stringency of the SoP. Because the SoP operates at trial, it seems natural to focus on the fact that the judge is deciding over the fate of a human being who may be guilty or innocent. But, by focusing the attention on trial, this view distorts the problem, as it either completely obscures the cost of false acquittals on crime prevention or plays it down as an indirect cost merely concerning potential victims. ¹⁴

When the second distortion occurs, the crime prevention costs of acquitting the guilty are certainly recognised, but they are underestimated or dismissed as merely statistical. This distortion consists of incorrectly supposing that falsely convicted innocents are actual (individualised) victims of the CJS, while failure to prevent crime simply involves potential (statistical) victims. But this is a mistake resulting from a focused attention on the operation of the SoP at the moment of trial. Whether a policy results in individualised rather than statistical victims should be irrelevant for our ethical evaluation of that policy (provided that the likelihood of the outcome is not affected). ¹⁵ However, in the case at hand, we’re always dealing with statistical, non-identified victims. In fact, even though the function of a SoP is to establish the necessary degree of evidence required for a conviction in a criminal trial, SoP policies should be evaluated at the legislative level. And, at that level, all victims are statistical. The legislative decision to establish a certain level of the SoP is a decision about what type of error is preferable and to what extent. Whatever the decision would be, there will always be errors and victims of those errors. A high SoP typically results in fewer victims of the CJS (wrongly convicted persons), but in more victims of crime. ¹⁶ A low SoP pushes the scale in the opposite direction. At that moment, lawmakers certainly know that there will be victims of these both types. But they do not know who those victims will be. This explains why the distinction between direct costs (based on the application of a SoP in the criminal trial) and indirect costs (based on the effects of that SoP on prevention) is misguided. From a legislator's perspective, both cases involve statistically predictable costs of a rule establishing a certain level of SoP.

(b) Concentration of harm (or risk of harm) v. distribution of harm (or risk of harm). It has been postulated that the cost of acquitting the guilty is distributed throughout the population as a whole, while the cost of convicting an innocent person affects only one individual. Lord Devlin has advanced an argument along these lines: ‘When a criminal goes free, it is as much a failure of abstract justice as when an innocent man is convicted. Each is a deviation on one side or the other, but an injustice on the one side is spread over the whole of society and an injustice on the other is concentrated in the suffering of one man’ (1958: 135). However, the argument does not hold. The injustice of not punishing the guilty is ‘spread over the whole of society’ if it is conceived of only as a failure to do what retributive justice requires. However, when one considers the effects of false acquittals on crime, the argument fails. The victim of a crime that would not have been committed if the SoP were less strict suffers the full weight of the crime, just like a falsely convicted person suffers the full weight of the punishment (Epps, 2015: 1131; Walen, 2015: 413).

In addition, the reference group must not necessarily be delimited that way. If instead of the falsely accused defendant, the reference group consisted of actual innocent people, that proportion would be significantly altered because the risk of an individual being convicted of a crime they did not commit is much lower than the risk of an individual, who has already been falsely accused, being convicted of that crime. Therefore, depending on how we define this group, we’ll be more inclined to view one circumstance as more serious than the other. We think that the risk of an innocent person being convicted is concentrated because we normally define the reference group within a specific situation (the accused makes up the entire reference group). But we think that the risk of an innocent person falling victim to crime is distributed throughout the population as a whole because we normally define the reference group in general terms. Because we frame the problem differently, our perception of it varies.

(c) Action v. omission: Another likely reason why the traditional point of view undervalues the preventive cost of acquitting the guilty is that it views this matter merely as an omission by the state (who simply allows it to happen). Meanwhile, a false conviction is seen as an active interference in the life of the falsely convicted. According to this view, the moral relevance of the distinction between action and omission is that it justifies a strict SoP (Hosein, 2014: 239, 240). ¹⁸

Once this point is clarified, it is possible to see why the distinction between action and omission is irrelevant in this context. Applied to the state’s regulatory function, this distinction is both conceptually unintelligible and morally irrelevant.

First, it is impossible to speak coherently about actions and omissions (or doing and allowing) in the context of government regulation. If the state sets out to regulate the SoP in criminal trials, it will have to choose among different options that lead to different outcomes. It is not clear how a phenomenological difference is possible when the state adopts a weak standard (which results in more false convictions and fewer false acquittals) than when it adopts a stringent one (which results in fewer false convictions and more false acquittals). Moreover, each of these standards will produce the same type of outcomes, albeit in different proportions. With a strict SoP there will also be false convictions (although less than with a weaker standard) and with a weak SoP there will also be false acquittals and failure to prevent crimes (although less than with a stricter standard). Therefore, when it comes to state regulation, the distinction between action and omission is unintelligible (Sunstein and Vermeule, 2005: 719).

Second, it is hard to see how this distinction could be morally relevant in the assessment of criminal policies. Whether a moral distinction exists between action and omission depends on whether or not there were preexisting duties. If an individual has a duty to protect someone else's legal interests, the question as to whether those interests have been harmed through an action or omission is not morally significant. In a similar vein, if the state has a duty to take effective measures to protect individuals from the unlawful actions of others, ¹⁹ it is not unreasonable to argue that it must consider the effects of its rules on crime, regardless of whether the rule in question benefits crime by merely not hindering it. More importantly, as a life-long committed deontologist like Michael Moore has finally acknowledged, when it comes to matters of institutional design and resource allocation, categorical obligations are out of place. In fact, the justification of our CJS's institutions rests on consequentialist reasoning, because those who design legal institutions ‘necessarily adopt an agent-neutral view’ and, therefore, ‘cannot justly regard their agency as the special source of their (“agent-relative”) obligations’ (Moore, 2007: 1551, 1552). ²⁰

SoP and crime prevention: An empirical analysis

It has been argued thus far that a more lenient SoP benefits crime prevention. Though theoretically possible, the validity of that claim ultimately hinges on empirical evidence. This section analyses relevant empirical research for assessing whether a weaker SoP can indeed benefit crime prevention. First, we’ll examine the results of studies on criminal deterrence. Then, we’ll inspect criminological studies on the preventive effects of incapacitating offenders. Lastly, we will look at econometric studies on the effect of incarceration on crime.

(a) Deterrence. Empirical research on deterrence has shown not only that increasing the likelihood of punishment has a deterrent effect, but also that people are influenced more by the certainty of punishment than its severity. Both conclusions result from solid research in criminology. ²¹ This seems to suggest that changes in the SoP may have a deterrent effect. However, this conclusion may be hasty. None of those studies specifically analyse the effects of changes to the SoP on deterrence. Existing empirical research on deterrence mainly measures the impact on crime of changes in the size of the police force, policing strategies and other policies aimed at improving crime detection and arrest rates. This is a problem because, if we wish to improve deterrence through changes in the SoP, we need a plausible estimation as to if and to what extent the expectation of punishment varies with those changes. Existing studies on SoP are not of great help in that respect because they seek to determine how triers of fact will adjudicate an identical case using a different SoP. These studies may demonstrate how conviction rate varies, when the SoP varies, ²² but provide no insights as to how such outcomes are perceived by potential offenders or how that perception impacts their behaviour. Of course, it does not follow from the fact that there are no existing studies on how variations in the SoP impacts deterrence, that such variations are ineffective for prevention. Absence of evidence is not evidence of absence. Confirming or refuting the theoretical arguments set forth herein requires specific studies measuring, on the one hand, how changes to the SoP impact the perception of potential offenders on the likelihood of punishment and, on the other, how perception shapes their behavioural choices. Without such studies, we can reach no definitive conclusions.

First and foremost, because only a very small fraction of crimes are actually tried (most are not detected, not reported or are closed in the pre-trial phase) (Gramlich, 2019; Robinson and Darley, 2004: 184), it is possible to assume that a lower SoP will not significantly increase the likelihood of punishment. Even when a decrease in the SoP results in a significant increase in conviction rates in tried cases, if the number of tried cases is small compared to the number of crimes committed, the likelihood of punishment (linking the number of convictions to the number of crimes) is not likely to increase significantly. It is true that this per se says little about criminal behaviour because it does not depend on the objective likelihood of punishment, but on how that likelihood is subjectively perceived. And reality and perception don’t always match. Potential offenders may perceive an increased risk even when there is none, which is more likely to happen when public policy changes are widely publicised. But it is difficult to see how a change in the likelihood of punishment as a result of a variation in the SoP could be so widely (and exaggeratedly) communicated to potential offenders to get them to believe that their chances to escape punishment are low. More importantly, it is plausible to assume that potential offenders (at least many of them) will ultimately experience that the policy change did not have the anticipated impact. This is especially true if we consider that most crimes are committed by a small number of offenders and this group is the most knowledgeable as to how the CJS really works. This is not only because they have more incentives to know how the system operates, but also because they have more direct and indirect experience with the system on the count of their environment (Robinson and Darley, 2004: 180, 184). Thus, potential offenders may ultimately learn that if they make it to trial, their chances of acquittal are not high; ²³ but they can also learn that their chances of making it to trial are very low and, therefore, that their chances of getting away with the crime are also high (despite a lower SoP). Of course, because trial rates vary significantly according to crime types, the higher the trial rate of a particular crime, the lesser the strength of the above argument. ²⁴

Second, studies on perceptive deterrence have provided overwhelming evidence that the fear of informal sanctions (i.e., reproval of family, friends, neighbours, etc.) can have a strong or even greater deterrent effect than the fear of criminal punishment (Braithwaite, 1989: 69, 70; Nagin, 1998: 19; 2013: 85, 99; Nagin and Paternoster, 1993: 489; Paternoster and Simpson, 1996: 571; Paternoster et al., 1983: 295). This is not only because of the importance assigned to social standing, employment and financial wellbeing, but also the fact that such adverse consequences take place long before sentencing, typically upon arrest, especially when the crime has received media attention. Because the time that elapses between the commission of the crime and the informal sanction is substantially shorter than that between the commission of the crime and sentencing, the deterrent potential of informal sanctions is not as time-sensitive as formal sanctions. In economic terms, the risk of suffering an informal sanction has a much lower discount rate than the risk of suffering an official sanction (Mamayek et al., 2017; Nagin and Pogarsky, 2004; Polinsky and Shavell, 1999); thus, its deterrent potential is higher. This raises the question of how much the SoP should be lowered to achieve extra deterrence. In fact, if informal sanctions are as effective or more so than formal sanctions, and if proximity in time makes them less time-sensitive, and if all it takes to trigger them is the initiation of proceedings that require a low SoP, then it seems reasonable to assume that only a drastic reduction in the SoP for conviction stands a chance of resulting in extra deterrence.

Third, even though very little is known about how well potential offenders perceive criminal policies and, in particular, how well they perceive that a new policy carries an additional risk of sanction, it is estimated that potential offenders have broad and aggregate perceptions of risk (high and low) rather than specific ones (Kennedy, 2009: 30, 31). If so, then a decrease in the SoP for conviction can have no effect if it does not reverse the standing perception (Apel, 2013: 72). If, as already mentioned, potential offenders are likely more afraid of informal sanctions than formal ones, then changes in the likelihood of conviction will have to be very deep to reverse the perceived risk of sanction that is to a great extent associated with informal sanctions. If, also as stated above, the number of tried cases is meaningless compared to the number of crimes committed, it is unlikely that a less than drastic decrease in the SoP can affect the perceived likelihood of sanction to the point to which it modifies the originally perceived level of risk.

A very drastic decrease in the SoP for conviction could achieve that outcome especially if it is accompanied by a lowering of the necessary SoP for initiating an investigation. If the state decided to drastically lower the SoP for conviction, then, to keep logical consistency, it should also lower the SoP for initiating an investigation. The rationale is that, for it to make sense to have a trial, the SoP for initiating an investigation must necessarily be lower (probably substantially lower) than the SoP for conviction. If both standards were similar, a decision to open an investigation would practically mean a decision to convict and thus to carry out a trial would be nonsensical. Thus, if the SoP for conviction drastically decreases, the SoP for initiating an investigation should logically also decrease (relatively proportionally). Because a lower SoP for investigation will probably result in a greater influx of cases entering the system and making it to trial, such a decrease in the SoP for conviction could be deterrent even when the initial trial rates are very low. Thus, given the system's possible dynamic effects, a drastic decrease in the SoP for conviction would not only increase the conviction rate in tried cases (direct effect), but increase the number of cases tried (indirect effect).

In light of these arguments, only a drastic decrease in the SoP for conviction could stand a chance of resulting in extra deterrence. However, this is not an option for an advocate of deterrence because an extremely low SoP, such as a SoP for conviction that required only a minimal suspicion, could be counterproductive. In fact, a very low SoP would result in the perception that it makes no difference what you do because you could ultimately be convicted anyway. If refraining from crime carries the same cost as not refraining, then there is no reason to refrain. Hence, any incentive to obey the law is eliminated (Png, 1986). ²⁵

This backs the deterrence argument into a corner: a moderate decrease in the SoP would probably not generate additional deterrent effects, but a drastic decrease wouldn’t either. The above arguments indicate that the hypothesis that a decrease in the SoP for conviction deters criminal behaviour (deterrence hypothesis) is less likely than initially suggested by a theoretical model like the one in the first part of this article. However, they are not strong enough to refute that hypothesis. Refutation is only possible by way of evidence; and, as we have seen, existing empirical studies enable no conclusion to be drawn as to the veracity or falseness of this hypothesis.

But even if we assume that a decrease in the SoP will not improve deterrence, we would still not be able to successfully refute the central hypothesis that a decrease in the SoP prevents crime (prevention hypothesis). Next, we will analyse whether existing empirical evidence can support the hypothesis that a lower SoP prevents crime by incapacitating potential offenders (incapacitation hypothesis).

(b) Incapacitation. Criminological studies that set out to calculate the preventive effect of incapacitation part from a model created by Avi-Itzhak and Shinnar (1973). Simply put, the model assumes that active criminals, when free, commit a certain number of crimes per year. Thus, the longer the offender is incarcerated, the fewer crimes they will commit. The number of crimes prevented by the incapacitating effect of incarceration depends, mainly, on the average annual frequency with which offenders commit crimes and the average time offenders spend in prison. In turn, this last variable depends on the severity of the punishment and the number of incarcerated individuals. Thus, the likelihood of the offender being convicted and sentenced to prison is relevant in this calculation.

First, an adequate calculation of the preventive potential of incapacitation must take the replacement phenomenon into account, i.e., that certain crimes don’t stop being committed because the person who would have committed them is in prison. The effect of incapacitation is null or insignificant if those incarcerated are rapidly replaced within a criminal organisation by others who are willing to take their place. Because the replacement phenomenon differs by crime type (Miles and Ludwig, 2007: 291), it would be advisable from a prevention perspective to vary the SoP according to crime type.

Second, the preventive effect of incapacitation depends significantly on the frequency with which offenders commit crimes. Because criminal frequency varies significantly according to the type of crime, the higher the frequency with which a type of crime is committed, the higher the preventive benefits of incapacitating the offenders who commit those crimes (Piquero and Blumstein, 2007: 272). This again raises the issue of establishing differentiated SoP.

Third, because people who commit crimes more frequently are at an increased risk of being caught, the system will naturally select frequent offenders. This is known as stochastic selectivity; and it suggests that the preventive effect of incapacitation may be significant. However, stochastic selectivity also suggests that the preventive benefits of incapacitation decrease to scale, meaning that the more people are incarcerated, even if the system continues to filter out those offenders who are more likely to commit crimes at high rates, the offending frequency of new offenders who get caught will be lower because offenders with higher crime rates will have already been caught before. This, in turn, means that the average crime rate of incarcerated offenders will be higher than that of non-incarcerated offenders. Incapacitation estimates that fail to adjust for this phenomenon overestimate the added preventive effect of increased incarceration because they use data on offending frequency of incarcerated individuals (who have higher frequency rates) to predict how much crime can be prevented by incarcerating those who are currently free (Canela-Cacho et al., 1997: 1567).

Fourth, the preventive effect of incapacitating offenders also depends, to a great extent, on the relationship between the duration of the offender's criminal career and the duration of the punishment (Piquero et al., 2003). Punishments that are longer than the duration of the offender's criminal career are useless in terms of prevention. Therefore, an adequate estimate of the incapacitation effect of incarceration requires detailed knowledge of the factors associated with termination of criminal careers (Farrall, 2000; Farrall and Calverley, 2006; Shapland and Bottoms, 2017; Shapland et al., 2016). Because criminality dramatically decreases with age, the crimes that are prevented while the offender is incarcerated will decrease as the offender ages. This suggests that the preventive benefit of incapacitation decreases with regards to the duration of the punishment as a natural result of aging (Farrington, 1986; Hirschi and Gottfredson, 1983; Ulmer and Steffensmeier, 2014).

Fifth, if incarceration were an effective way to rehabilitate or deter prisoners, the benefits of post-release prevention should be added to the benefits of incarceration. If, instead, incarceration has criminogenic effects, the post-release increase in crime should be subtracted from the preventive benefits of incarceration. Most studies conclude that the impact of incarceration on post-release criminal activity is null or that incarceration has a mild crime-increasing effect (Nagin et al., 2009).

Sixth, and last, incarceration may prevent crime outside the confines of prison, but not within its walls. Thus, when calculating the incapacitating effects of imprisonment, one must also bear in mind the crimes that happen while imprisoned. ²⁶

Studies that take these matters into consideration offer more dependable results on the preventive effects of incapacitating offenders. Most of these studies report their results as an elasticity. The elasticity of crime with respect to incapacitation measures the percentual change in crime vis. a 1% increase in the number of incapacitated individuals. All these studies confirm that incapacitation prevents crime. On average, the estimated elasticity is −0.2%; thus, a 5% increase in incarcerations is needed to achieve a 1% decrease in crime. ²⁷ Therefore, empirical research on the incapacitating effects of incarceration seems to confirm the hypothesis that a lower SoP could possibly reduce crime. ²⁸

(c) Crime and incarceration. Econometric studies on the relationship between incarceration rates and crime rates are also a good source to learn about the potential preventive effect of lowering the SoP. These studies are aimed at determining the total effect of incarceration on crime using aggregated data of prison population and criminal activity.

Empirical research consistently shows that an increase in incarceration rates strongly correlates with a decline in crime rates, and this is the conclusion reached in studies that have taken simultaneity bias into account by using instrumental variables and in those that have not and which tend to underestimate the preventive effects of incarceration. The conclusion about a negative correlation between incarceration and crime is so firmly established among empirical economists that the question today is not whether an increase in incarceration prevents crime, but to what extent (Levitt and Miles, 2007; Spelman, 2000).

Crime-incarceration elasticity estimates vary significantly, not just from one study to another, but from one type of crime to another. After a thorough review of the main studies, William Spelman concluded that crime-incarceration elasticity estimates reach an average of −0.30, but any number between −0.20 and −0.40 is possible, and we should not even be surprised with a result of −0.10 and −0.50 (Spelman, 2000: 484). These wide margins are confirmed by other recent studies. ³⁰ Despite variations in the estimates, empirical studies on the total preventive effect of incarceration also confirm the hypothesis that a decrease in the SoP prevents crime.

Conclusion

Empirical studies on the incapacitating effects of incarceration indirectly support the hypothesis that a lowering of the SoP can have preventive benefits. Econometric studies on the relationship between incarceration rates and crime rates further strengthen that conclusion. Based on those studies, we can conclude that a lower SoP can be beneficial to crime prevention.

However, where exactly the criminal SoP should be set is less clear because no reliable conclusions can be drawn with respect to which SoP maximises prevention. Although empirical studies agree that incarceration prevents crime, there is not much consensus as to the elasticity of crime with respect to incapacitation/incarceration. In addition, it is not clear the extent to which a change in the SoP will impact conviction rates. Therefore, not only is there no certainty with respect to the exact extent to which an increase in the prison population results in prevention, but also with respect to the extent to which a lower SoP increases that population as well. Without this information, it is not possible to propose, with any level of certainty, a specific level of SoP.

However, even if we could determine with relative certainty what standard maximises prevention, that should not necessarily lead us to lower the SoP, because that lower standard may not be desirable all things considered. The fact that a decrease in the SoP benefits prevention merely gives us pro tanto reasons to lower that standard; reasons that, though having genuine weight, may be nonetheless outweighed by other considerations (Kagan, 1989: 17).

There are several reasons for this. First, crime prevention is not the only relevant consideration when designing criminal policies. Because a lower SoP means an increase in the number of incarcerated individuals and incarceration is costly, we need to assess whether the preventive benefits outweigh the costs.

Second, increasing the prison population is not the only way to prevent crime. Therefore, an adequate cost-benefit assessment must not only show that the benefit of a lower SoP outweighs its cost, but also that it has a more positive cost-benefit ratio than other alternative measures (such as an increased police force, social programmes, etc. ³¹ ).

Third, a lower SoP is not the only way to increase prison population. Even when an increase in incarceration is an efficient means of prevention, it has yet to be demonstrated that a more lenient SoP is (under a broad cost assessment) a more cost-effective means for increasing incarceration than other available alternatives. Improving the accuracy of the CJS can result in an increase in the number of convictions without lowering the SoP. Although that strategy is costly, it has the advantage of reducing the risk of convicting the innocent and acquitting the guilty.

An adequate assessment of all these issues requires a comprehensive cost-benefit analysis that exceeds the scope of this article.

In this article, we are satisfied with three more modest conclusions. First, that, all else being equal, a lower SoP implies preventive benefits. Second, that there are no compelling reasons to dismiss crime prevention as a factor to consider when determining the SoP. Third, that appropriately considering crime prevention when establishing the level of the SoP give us pro tanto reasons to adopt a weaker standard than that normally deemed appropriate in criminal cases.