(FORTUNE Magazine) – Liberal ideology still sells in some policy areas, but not many citizens wish to buy it when the subject is crime. The Duke plainly lost votes by defending the Massachusetts furlough system, which had produced the Willie Horton story. (Willie was the convicted rapist and murderer who got released and went on to rape again.) What the people want is a hard, hard line on sentencing and parole. Every annual survey by the National Opinion Research Center shows 80% or so judging the courts to be ''not harsh enough'' with crime. The latest disaster for liberal crime policy is in Maryland, where the famous Patuxent Institution is in big trouble over two recent cases. The more egregious concerned a convicted rapist named James M. Stavarakas, whose parole was terminated after he failed a drug test but who then got out anyway on a work-release deal -- and allegedly committed another rape. Patuxent is an experimental institution that accepts only prisoners considered amenable to psychiatric therapy, so its failures come across as stark. The Horton and Patuxent stories remind you of a maddening verity about U.S. prison policies: Prison authorities generally do a rotten job of predicting the post-release behavior of paroled and furloughed convicts. Justice Department data covering 22 states suggest that among young adults who get + paroled, around 70% are rearrested within six years of release. Like Willie Horton and James Stavarakas, every one of those young adults was certified safe before leaving the walls. Curiously lacking in parole and furlough decisions are any quantitative statements about the probability of ''recidivism'' (a.k.a. backsliding) in different cases. Given the torrent of research that has gone into studying recidivism, the omission is remarkable. Criminologists have developed, for example, a Legal Dangerousness Scale (LDS) and other measures with pretty good predictive power. The LDS is a weighted report on the individual's criminal record, past imprisonment, and proclivity for violence, with an adjustment reflecting the severity of the most recent crime. In effect, the LDS can be used as a probability statement -- an expression of the odds that the guy going free will again prove a menace if released. The case for thinking about the guy probabilistically, and getting those odds into the public record, has been argued persuasively by Robert A. Gordon, professor of sociology at Johns Hopkins (and member of an advisory board serving Patuxent). Gordon's argument is clearly not popular among judges and parole board members, and it is easy enough to see why. Just about every candidate for sentencing or parole represents some kind of threat to the public. The recidivist odds may be less than 10%, but they are most unlikely to be zero. The judges naturally do not wish to acknowledge that they are releasing characters who present a gamble. They fear the wrath of the public if they admitted the extent to which they are rolling dice on releases. So the system dishonestly goes on pretending that every individual up for consideration can be labeled ''safe'' or ''dangerous.'' Which results in a ludicrous number of ''safe'' calls. And guarantees that the surveys will go right on showing a demand for tougher courts.