Monthly Archives: August 2014

Stay dry

Cross-posted from The Gaping Silence. Long (5000 words).

1. It’s great when you’re straight, yeah

I have a beer most nights – just the one, usually. I have one dry day every week and often two, and tot up my weekly units every so often (the number’s never been at all alarming). But I drink most nights, and certainly most weeks – well, all weeks. I don’t think I’ve gone as long as ten days without a drink since an ill-advised ‘detox’ attempt back in the 90s (three long weeks of muzzy caffeine withdrawal); before that you’d probably have to go back to my teens or early 20s.

I’m not a heavy drinker and never have been; I’ve never worried I was overdoing it, never felt I ought to cut down, never been tempted to go Dry for January or whatever. (I have occasionally worried that I might overdo it; this may be connected to my lack of anxiety about my actual consumption, by a kind of precautionary anxiety principle.) If I did try and abstain for as long as a month, mind you, I think I’d find it tough. Pete has written a couple of times about his own personal ‘dry January’ routine (“I try to go dry for January every year, and have done so for years – since long before it became a piece of nonsense to beat people with”). His account makes it sound – as Guy Debord wrote of his own alcoholism – émouvant mais difficile:

In the first few days, you notice the better sleep, the higher energy, the greater clarity of thought. My blood pressure, which landed me in hospital in October, is now verging on normal. After a couple of weeks, you realise you’re thinking differently. You’re more in the moment, more thoughtful, more connected. This is not always pleasant. But like the physical benefits, it does feel like it’s doing you some good.

By the second week you start to feel like a cultist praising the virtues of abstention. By the third week, you start to notice that everything is bright and shiny and hard. Perhaps a little TOO bright. It’s natural and healthy to sometimes want to fuzz the edges and turn the lights down to mood. I’ve missed that. But I’ve missed the sensory experience of drinking – the aromas and tastes of good beer, cider, wine, sherry and the occasional malt whisky, and the stories that go with them, the associations they have, the connections they make, the contemplations and flights of fancy they inspire – a whole lot more.

Whether that ‘you’ would include me – whether somebody who averages (say) one and a half beers per day is limescaling his sensorium in this way, such that three weeks’ abstention would give his system a hard reset and make the world turn dayglo – is an interesting question; I tend to think the answer’s No, but I’m not in a hurry to find out. That small point aside, it’s clear from Pete’s post that it was quite a long month. Which is no surprise – alcohol is a normal part of most adults’ lives: in a 2011 government survey, two-thirds of men and 54% of women claimed (admitted?) to have had a drink in the previous week. For anyone other than a very occasional drinker, I think a month would be plenty. Admittedly, alcohol is less normal that it has been – in 1998, a similar survey put the drink-in-previous-week figures at 75% of men and 59% of women; in another recent survey, only 39% of under-16s reported ever having drunk alcohol, the lowest figure this survey has ever recorded. But it’s normal for all that; after all, giving up alcohol for a month wouldn’t be something you do for charity if it were something you’d do anyway.

So how about going cold turkey for four months?

People who repeatedly commit alcohol-related crime will be forced to wear ankle tags that monitor if they are still drinking, under a year-long pilot scheme. The “sobriety tags”, to be worn around the clock, will enforce abstinence by measuring a person’s perspiration every 30 minutes and testing to see if it contains alcohol. If any trace is found, an alert will be sent to the offender’s probation officer and they can then be recalled to court, where they may be resentenced or face sanctions such as a fine.

It is anticipated that up to 150 offenders will be fitted with the tags. They will be banned from drinking alcohol for up to 120 days, and the tag will test them to see if they flout the ban. Offenders will be screened before being tagged, and the scheme will not be used on people who are alcohol-dependent and require specialist support. The scheme, being introduced by the mayor of London, Boris Johnsons, builds on a similar scheme in the US and aims to reduce alcohol-related reoffending, ease pressure on the police and courts, and make streets safer.

A salutary initiative which will help some potentially dangerous problem drinkers clean up their act (the view of Deborah Orr in the Guardian)? Or should we be more sceptical?

2. Your weakness is none of my business

The London pilot was provided for in section 77 of LASPO; on its completion the scheme will either be rolled out nationally or repealed altogether. (Section 77(7) is an interesting bit of drafting; I don’t remember seeing an “if this doesn’t work, let’s just forget about it” clause in an Act of Parliament before.)

There are two ways of reading the pilot: it depends whether you see it as having a good chance of success, or as being highly likely to fail. My initial reaction was the latter. If you put me (or, I venture to suggest, you, dear reader) on a total four-month alcohol ban, the result would be a foregone conclusion; as for inflicting a scheme like this on repeat alcohol offenders, that would just be setting them up to fail sooner rather than later. Repeat offenders are people with problems, if that’s not stating the blindingly obvious – you don’t persist in putting yourself at risk of arrest if you’ve got lots of other, less liberty-threatening options. Repeat alcohol-related offenders, specifically, are more likely than not to be people with a problematic relationship with alcohol – and, since alcohol problems tend to involve compulsive activity and impaired judgment, people with alcohol problems are more likely than not to breach controls imposed on them. (Although if candidates for this scheme do have drink problems, they’ll have to deal with them on their own. A nasty detail here is that the scheme excludes anyone whose problems amount to dependency, hence qualifying them for support as well as for control measures.)

If the scheme works as described – if an alarm goes off every time a tagged offender has a drink, leading automatically to the decision that the offender’s alcohol abstinence order has been breached – the scheme will “reduce alcohol-related reoffending [and] ease pressure on the police and courts” by one means only: by making 150 offenders’ lives a bit less pleasant, subjecting them to stress and anxiety for a while, then recalling them to court and rubber-stamping a prison sentence. (Admittedly, the scheme is not predicated on prison sentences for breach; alternative sentences are available, for example a fine. But the idea that imposing fines on repeat alcohol-related offenders might deter them from reoffending rests on rather unrealistic assumptions about human nature.) Overall, its effect on those 150 people will be to interfere with their lives, imposing technologically-mediated humiliation and harassment, before stigmatising them as failures and disrupting their lives still further. This isn’t rehabilitation, it’s sadism.

If the scheme is intended to work as it is described here, it is being undertaken either naively (in which case it will fail horribly) or cynically. And if it is being undertaken cynically, it represents a peculiarly debased and unlawlike use of the criminal justice system: the imposition of behavioural conditions on an offender, in place of a punitive sentence, in the expectation that the offender will breach them and incur a punitive sentence for the breach. This use of criminal penalties will be familiar to anyone who recalls the history of the ASBO: I remember cases where rowdy youths were barred from particular streets, a graffiti-sprayer from using public transport, a suspected drug-runner from sitting in the front passenger seat of a car – in each case, with the force of a criminal offence. In each case, the offender has (ostensibly) had the option of modifying his behaviour and walking free; in each case, the behaviour prohibited is legal and normal for other citizens; in each case, the behaviour modification requirements are impossible to comply with, or not without intrusions into the offender’s daily life so extensive as to greatly reduce his freedom to lead a normal life. Most importantly, in each case the offender is held responsible for any breach and for the penalty attached to it, thus incurring the stigma of indiscipline and recidivism as well as that of law-breaking. Barring repeat drink offenders from drinking – by law – is only a more direct and straightforward version of the same logic. The inference is that problem drinkers – like anti-social youths – are now seen as a subject population, a minority group needing to be controlled for the sake of the majority, and whose own rights can be disregarded.

3. Computer says No

If the scheme is intended to work as described here… But that may be a big If. The Alcohol abstinence and monitoring requirement was added to LASPO after lobbying from a group of American professionals with experience of a similar scheme in South Dakota; one member is Keith Humphreys, a Stanford professor of mental health policy and former White House drug advisor. In the South Dakota scheme, Humphreys writes,

[alcohol-related repeat offenders were sentenced] to mandatory sobriety coupled with daily testing. Those who tested positive were arrested immediately and given a modest sanction, such as a night in jail. In the 24/7 sobriety model, a slow, inconsistent and capricious system of monitoring offenders was transformed into one that provided swift, certain and fair sanctions for drinking.

The effects were dramatic. A stunning 99.4 per cent of the over five million breathalyser tests administered to offenders have been negative. Despite the punishment for drinking being modest, its certainty and swiftness motivated many previously recalcitrant offenders to change their ways.

Research by the RAND Corporation – a US-based non-profit global policy think tank – found that 24/7 sobriety dropped repeat drink driving arrests by 12 per cent. The same study also yielded a pleasant surprise: domestic violence arrests dropped by 9 per cent, despite not being a focus of the programme. … This week, under the leadership of Mayor Johnsons and his team, a pilot of the programme will be launched in South London. Leaping the pond will come with some challenges, particularly around delivering sanctions swiftly within the constraints of British law, but local tailoring of innovations is always an essential part of making them spread.

It works, in other words. The figures are in: not only do offenders comply with the stay-straight provisions more than 99% of the time, the effects can be seen in the crime figures.

What should we make of this? I’ve got three comments, of increasing generality; let’s say, one technical, one analytical and one philosophical. The technical comment is this: I can’t believe it. I’ve looked at the South Dakota 24/7 Sobriety Program Evaluation Report (PDF here; more information and links here); it tells me that, when 4,009 drink driving offenders were ordered to report to a judge twice a day to be breathalysed, for periods averaging three months, 98.1% of those tests were negative and only 0.3% positive (1.3% of tests did not take place because the participant was excused; 0.3% of tests were unauthorised no-shows). Pace Humphreys, this isn’t “a stunning 99.4 per cent of … over five million breathalyser tests”, for two reasons. Firstly, the 4,009 offenders in the sample took 817,926 tests (an average of 204 apiece); I’m not sure where the ‘five million’ figure comes from. Secondly, the figure of 99.4 per cent includes the 1.3 per cent of tests ‘excused’ and excludes the 0.3 per cent of no-shows. The real positive rate is the total of positive tests as a proportion of the tests taken, or 98.1/(98.1+0.3), which works out as a (stunning) 99.7% pass rate. In the course of their participation in the programme, 2,659 of the 4,009 participants – slightly less than two-thirds – had no negative tests at all. (These are convicted drink drivers, remember, being breath-tested twice a day for a period of months.) Of the remaining 1,350, between 584 and 802 (between 14% and 20%) had precisely one negative test; the number who had four or more negative tests is somewhere between 131 (one in 31) and 267 (one in 15). (The published figures lump together unauthorised missed tests and negative tests as ‘Failures’, making it difficult to derive the exact number of participants who had a certain number of negative tests.)

These figures seem to represent a level of behaviour modification which I find, literally, incredible. The programme’s advocates argue that the novelty of this outcome is related to the novelty of its key innovation, the use of those “swift, certain and fair sanctions” – if people know that one drink will infallibly earn them a night in the cells, they’re strongly motivated to stay clean. Maybe so (I’ll say a bit more about this approach further down). Even so, purely based on my knowledge of human nature, I find it very hard to imagine any combination of rewards and incentives having a 99.7% success rate in modifying entrenched behaviour, in any population (imprisoned populations included). Or perhaps I should say, based on my knowledge of British (European?) human nature. I have just as hard a time imagining the participants in such a scheme meekly turning up to be breathalysed 98.4% of the time – again, whatever the incentives and whoever (and wherever) the participants were – but that certainly appears to have happened in South Dakota. It’s a cultural difference, perhaps; when it comes to criminal justice agencies, the British (and British law-breakers in particular) just don’t respect their authority – or not the way that South Dakotans do. (“I thought they were big on liberty over there?” – my wife.)

With regard to the London pilot the point about showing up is moot – the testing is to be carried out by an ankle bracelet, replacing the stigma and inconvenience of twice-daily reporting with the stigma and humiliation of wearing a shackle that spies on you. But the point about the near-as-dammit 100% negative test rate, and the seemingly total behavioural transformation brought about by the programme, is absolutely key – and I think there’s a genuine case for saying that, like a Communist bloc election result, these figures are just too good. A clue of sorts is provided by the RAND research referred to by Humphreys – presented, in the form of a paper by Beau Kilmer, to the ISSDP conference mentioned earlier. In the published version of the paper, Kilmer and his colleagues (Nancy Nicosia, Paul Heaton and Greg Midgette) write:

More than 17,000 individuals participated in the 24/7 Sobriety Project between 2005 and 2010 and their tests indicated that there were approximately 2.25 million days without a detected alcohol violation. This does not mean that there was absolutely no drinking on those days. Rather, it provides support for a reduction in the incidence of heavy drinking among a population with a history of problem drinking.

Emphasis added. Perhaps the South Dakota breathalysers weren’t all that sensitive; perhaps the scheme operated, in practice, like a benign version of the Philadelphia Lie Detector, scaring offenders into cleaning up their act without, for the most part, actually detecting whether they had done or not. The tiny minority of negative tests might be accounted for by the tiny number of occasions when participants turned up with a really substantial amount of alcohol in their system, e.g. still drunk from the night before. The use of breathalysers to carry out the twice-daily test raises the question of what level of blood alcohol concentration constituted a positive result; if the standard DUI level of 0.08% was being used, this would be entirely compatible with participants maintaining a relatively normal level of consumption of alcohol – which in turn would make that 99.7% pass rate considerably less surprising.

This kind of approach is not planned for the London trial, however. The Alcohol Abstinence Monitoring Requirement Toolkit produced by the Mayor’s Office (downloadable here) notes:

[LASPO] allows the court to specify that the offender cannot drink more than a specified amount of alcohol (expressed as the proportion of alcohol in any one or more of the offender’s breath, blood, urine, sweat or by some other means); thus allowing for the possibility of minimal drinking rather than abstinence. For the purpose of the pilot, this provision will not be used and complete abstinence will be enforced.

Sure enough, the statement to be signed by participants in the London trial states baldly “I must not drink any alcohol until my requirement ends.” This, however, raises the question of evidence. The most widely-cited figures for the South Dakota scheme appear to be based on a sample of 4,009 participants who submitted to a twice-daily breath test, which (according to Kilmer et al) did not return a positive result for low levels of alcohol. Generalising these results from South Dakota to South London – a vastly different setting both culturally and legally – is problematic enough. Using a ‘low consumption’ success story to justify a ‘total abstinence’ trial – in a different country, with a different legal system – is problematic in a more fundamental sense; it’s not comparing like with like.

4. He’s got ’em on the list

Still, the effectiveness of the South Dakota pilot is proven; there’s that 12% drop in repeat DUI arrests that we heard about, and the 9% drop in domestic violence arrests. (According to Kilmer et al, after being implemented the programme was rapidly extended to cover arrests for offences other than DUI – specifically including domestic violence – so it’s not quite correct for Humphreys to describe the fall in domestic violence arrests as a ‘pleasant surprise’, a side-benefit of a programme focusing on drink driving.) You can’t argue with the numbers.

Well, maybe not. But on reading the Kilmer et al paper it turns out that the ‘12% drop’ isn’t anything like as simple as a comparison between year 1 and year 2 in area A, or for that matter between area A and area B in year 2. Although the initial five-county pilot of the scheme might have lent itself to the second approach, analysis was made more complicated by the rapid and uncontrolled adoption of the scheme in other counties. The writers’ approach was to make a virtue of necessity, taking arrest data from the 66 counties of South Dakota – all of which had adopted the scheme by the end of 2010 – and calculating the overall extent to which the arrest rate was affected by the introduction of the scheme. The scheme was defined as having been introduced at the point where the numbers in the scheme in a given month first equalled 25% of the county’s average DUI arrests. The much quoted 12% drop is actually an incident rate ratio of 0.883 (p < 0.05), derived from a Poisson regression of five parameters. The outcome (arrest rate) for a given county and time period is assumed to be given by the sum of α, the effects of the scheme itself; β, other known factors associated with the county; γ, “unobservable characteristics of each county that are fixed over time”; δ, “fixed effects for each month in the sample”; and ε, an undefined variable associated with both county and time period (and not referred to in the text).

More data is better than less; in principle a 0.883 regression coefficient is actually more reliable, and hence more informative, than an eyeball comparison of two figures showing a 12% drop. (If you find this counter-intuitive, join the club.) The authors’ approach is also an ingenious way to circumvent the muddying of the waters brought about by the wide adoption of the scheme. But concerns remain, particularly given the relative weakness of the results (the coefficients relating to repeat DUI and domestic violence arrests are the only programme-related results to rise to the level of p < 0.05). What difference might alternative estimates for ‘county’ and ‘month’ effects (β and δ) have made? How did the researchers arrive at values for the unobservable γ or the undefined ε – and what difference might they, in turn, have made? There are also some signs of data-mining and cherry-picking, both in the data presented and in its interpretation. The authors report “no statistically or substantively significant effect of [the programme] on first-time DUI arrests” and “suggestive evidence that it may have reduced reported traffic crashes involving men aged 18 to 40 years”. The coefficient for crashes involving 18- to 40-year-old males is 0.956 (95% CI 0.909-1.006); the figure for first-time DUI is 1.062 (0.955-1.181). One may be positive or null but is probably weakly negative; the other may be negative or null but is probably weakly positive; and neither of the two is statistically significant. In other words, the evidence for an effect on first-time DUI is no weaker than the crash-related evidence, but the effect in this case is positive – which is to say, on an uncharitable reading, this is “suggestive evidence” that the programme may have increased first-time DUI arrests. The single strongest result reported has nothing to do with the programme: according to the data, the Sturgis Rally (an annual motorcycle rally held in and around Sturgis SD) has an effect on repeat DUI arrests of 1.477 (1.330-1.641) and on crashes of 1.642 (1.293-2.086), in both cases with p < 0.001.

But let’s assume for the sake of argument that the figures are good. Let’s take it as given that Kilmer et al have shown that the effect of introducing twice-daily-breathalyser-monitored sobriety in county X in month m will be that county X‘s repeat DUI arrest figures, averaged over months m+1…m+n, are 12% lower than they were when averaged over months m-1…mn. Now what? Or rather, now why? What’s the mechanism? I ask this not because failure to say how the effect occurred would call the effect into question – it wouldn’t – but because Kilmer et al do seem to have a specific mechanism in mind. Consider the threshold used to define when the programme had been implemented – when the numbers in the scheme in a given month equalled 25% of the county’s average monthly DUI arrests – and the alternative threshold suggested later in the paper, five scheme participants for every 10,000 population. South Dakota has a population of 825,000, of whom 17,000 were on the programme between 2005 and 2010. This is over 2% of the entire population, and a considerably higher proportion of the population typically involved in DUI offences; Kilmer et al note that, in some counties, more than 10% of men aged between 18 and 40 participated in the programme at some point. Lastly, consider the expansion of the programme following the passage in 2007 of South Dakota House Bill 1072:

The unanimous passage of House Bill 1072 dramatically expanded the 24/7 program. The bill went into effect July 1, 2007, and provided funds to counties that wanted to adopt the program. The new law allowed judges to order anyone they believed had an alcohol problem, pre- or post-conviction, to participate in the program. The law also changed rules for those who lost their license for a repeat DUI offense. It had previously been possible for some of these individuals to receive a permit to drive only to and from work, but these permits were now conditional on 24/7 participation.

So participation could be ordered by a judge without the individual involved being convicted of anything. Moreover, continuing participation could be made a condition for holding a (restricted) driving licence, making participation in the programme more or less a life sentence.

On first reading of the Kilmer et al paper I wondered if a measure for long-term rehabilitation and reform was being assumed to be successful on the basis of a confusion between outcome effects and programme effects. In other words, I wondered if the reduction in repeat DUI arrests might not be a sign of problem drinkers taking the (enforced) opportunity to turn their life around – as in Orr‘s optimistic account – but merely of drinking being temporarily suppressed by the imposition of a stressful and demeaning twice-daily reporting requirement. On re-reading it, I’m not convinced that the authors are even thinking in terms of outcome effects. Got an alcohol problem? Right – you’re on the programme; no drinking until I say so. You’ll get off the programme eventually; if you start causing trouble then, guess what – you’re back on the programme. No alcohol problem now!

5. As sloshed as Schlegel

In technical terms, I think the individual test data can’t possibly show what it appears to show, and hence that it doesn’t give reliable guidance for the London pilot. Analytically, I think the programme as described by Kilmer et al is one of mass behaviour modification rather than criminal justice – the idea seems to be, not to use the programme to help former offenders reform, but to put as many people as necessary through the programme for as long as necessary to bring crime down. This leads into my philosophical concerns about the programme – concerns which remain even if we assume the validity of the test data and the crime drop figures.

Humphreys describes the scheme as delivering “swift, certain and fair sanctions for drinking”; Kilmer and colleagues describe the South Dakota programme as combining “frequent monitoring with swift, certain and modest sanctions for violations”. Swift and certain: you test negative, you’re off to the cells, no ifs, no buts, no phone call, no duty solicitor. What’s the problem? Maybe there isn’t a problem at all – maybe the problem’s with our way of thinking, or the presuppositions of our system?

The key principle in disposals like this is certainty: offenders need to know that a breach will result in instant and decisive penalty. Our criminal justice system resists such practice. The pilot’s weakness is that it could take weeks for action to be brought against offenders who drink while tagged. If an offender breaches the sobriety order, they will first have to be returned to court where further sanctions can be imposed. Imprisonment is only likely to result from persistent non-compliance. This isn’t swift and sure justice, but neither is it the Mayor’s fault – it’s a fundamental weakness in our system of probation which needs fixing.

Opinion polling for the GLA in 2011 found that sobriety orders were popular with the public (well over two thirds supported the idea). Doubtless some will disagree. I recall the horror of a prominent politician, partial to a shandy, on being told that an electronic device might be used to monitor someone’s alcohol consumption. And I wonder how long it will be before the civil libertarians complain that sobriety tags are an egregious breach of human rights…

Thus Nick Herbert MP (Con). (It’s interesting, in passing, to see the reference to ensuring an “instant and decisive penalty”, emphasis added; Humphreys and Kilmer et al both stressed that the near-automatic sanctions for non-compliance were modest.) I think this line of argument needs to be resisted. It’s fundamental, not to “our criminal justice system” but to the rule of law, that the law governs us as free individuals. We obey the law because the law is worth obeying, not because we have personally been commanded to do certain things – still less because we have been threatened with adverse consequences if we do not. To break the law is also a free choice, and one which may be justifiable. To be put on trial is thus to enter a field of judgment, to have our actions ‘tried’ (tested) against the standard of the law; the possibility always exists – however remote it may be in the day-to-day operation of the courts – that it will be the law that blinks. However clear and unambiguous the law may appear to be, the outcome of a trial is never certain; a judge’s guidance may be overruled by a jury verdict; the law itself may be amended by an appeal court judgment.

To say that we are governed by law, then, is to say two things. The first is that the law is a system of rules with certain characteristics – universality, comprehensibility, followability – which make it possible for each one of us to arrange our lives so that we obey it: the law respects our freedom. The second is that those who break the law are not only brought to account but allowed to put their case, protected by certain fundamental safeguards – the rules of procedural justice – which exist to ensure that nobody is unjustly criminalised. The law respects us as free and rational citizens, when we break it as well as when we obey it.

The ‘sobriety tag’ scheme is unlawlike twice over. On one hand, the idea of “fair sanctions for drinking” tends to suggest that drinking should be sanctioned; I wouldn’t have thought this was a route the USA wanted to go down, again. At best it suggests that drinking should be banned for certain, arbitrarily chosen people – as unlawlike a proposal as you could wish for, effectively substituting the rule of individual law-enforcers for the rule of law. On the other, contra Herbert, closing the gap between infraction and sanction is not “swift and sure justice”, or not for any meaning of the word ‘justice’ associated with the law. Any kind of automaticity in declaring somebody a law-breaker – including but not limited to the technological automation delivered by alcohol shackles – subjects the law-breaker to the law, without defence or excuse, in a way that is corrosive to respect for the law (which we are presumably asking those law-breakers to carry on obeying in other areas of their life).

Perhaps the most telling part of Herbert’s comment is his sneering reference to a critical politician being “partial to a shandy”. Herbert himself may be a teetotaller, but he must be aware that the great majority of his fellow-citizens are, in fact, “partial to a shandy” – and as such might have well-grounded concerns in response to the development of technology that enforces total abstention. The unstated minor premise, underlying that two-thirds vote as well as Herbert’s complacency, is that alcohol shackles will never be a threat to us. Herbert’s confident that these measures will only be applied to them, that unruly minority whose behaviour needs controlling – and that we can trust law-enforcers to identify the people to control and modify their behaviour in the right way. (Deborah Orr’s message is much the same, dressed in more liberal language.)

But this isn’t the rule of law. It’s using the law, but using it as a means of controlling a deviant minority (problem drinkers). It’s using the law – in Hegel’s image – like a man raising a stick to a dog.

Feuerbach bases his theory of punishment on threat and thinks that if anyone commits a crime despite the threat, punishment must follow because the criminal was aware of it beforehand. But what about the justification of the threat? A threat presupposes that a man is not free, and its aim is to coerce him by the idea of an evil. But right and justice must have their seat in freedom and the will, not in the lack of freedom on which a threat turns. To base a justification of punishment on threat is to liken it to the act of a man who lifts his stick to a dog. It is to treat a man like a dog instead of with the freedom and respect due to him as a man. But a threat, which after all may rouse a man to demonstrate his freedom in spite of it, discards justice altogether.
Hegel‘s Philosophy of Right, thesis 99.