I've held forth a few times to keep jurist-listers apprised of the
existence and progress of our sentencing support project.  As you may
recall (and can certainly recall or catch up on from the web link at the
end of this message), I've been promoting the use of technology to give
us better information at sentencing, probation violation, and release
hearings as to which dispositions seem to correlate (or not) with
reduced criminal behavior by which offenders.  Direct work on this
project is now ongoing for over four years, although its roots in my
life probably date from 1968 or so when I was working on a law review
article on capital punishment and had the Calif. Dept. of Corrections
run some statistics on paroled felons.

Our idea is to gather operational data from all criminal justice -
related sources into a "data warehouse," and to provide query tools to
users.  The tools allow us (judge,  attorney or probation or corrections
or pretrial release officer) to select the offender whose case is before
us and the charge we're dealing with.  Using user-modifiable defaults,
the software gathers data on offenders who are like the one before us in
criminal history, age, and other attributes, and data on sanctions
(including jail, treatment, and anything else we've regularly used) and
criminal events after those sanctions were imposed.  The software then
uses the data so selected to display a bar chart showing the success
(again, measured by a user-modifiable measure of "success" which is tied
to some flavor of recidivism) of all dispositions used commonly enough
on this cohort of offenders to generate useful data.

We have a team refining the software which consists of a technical lead
person recently hired by Multnomah County (Portland, OR), a deputy DA, a
public defender, the project agency head, and myself.  We've produced a
demonstration to take around the state on a CD-ROM, which I've shown to
judges, corrections types, and politicians.

The big news is that for the first time anywhere this tool has actually
been run against real data in a courtroom.  Yes, limitations abound --
the "program" data available so far limits the tool to a slice of
misdemeanor probations, and to few enough of them to produce sufficient
numbers only for the most common of offenders and crimes.  And we have a
long way to go to bring this tool to the point where it will give us
good data on most cases, and the ability to produce truly useful
information on predicting what works best on whom.  But this stage -
actually running the program against real data - marks the beginning of
a new approach to criminal justice just as surely as the flight at Kitty
Hawk began a new mode of transportation.  The Wright Brothers' success
at Kitty Hawk was announced in a telegram like this:

  SUCCESS FOUR FLIGHTS  THURSDAY MORNING ALL AGAINST TWENTY-ONE MILE
WIND STARTED FROM LEVEL WITH ENGINE POWER ALONE  AVERAGE SPEED THROUGH
AIR THIRTY-ONE MILES LONGEST 57 SECONDS  INFORM PRESS HOME  CHRISTMAS

    On my first courtroom use of sentencing support software, I entered
the case number of an African American man who was coming to be
sentenced for Theft II.  I selected Theft II from the charges suggested
by the software based on its access to the data showing the counts in
the complaint.  Since we have very limited program data yet available to
this system, and relatively few African Americans in our jurisdiction
(over represented, of course, in the criminal justice system, but still
few), and fewer still with no criminal history at 36 years of age, I had
to broaden the "offender profile" to include "white" as well as "African
American" males between 18-40 with or without criminal history to get
enough data for the program to give me a bar chart.  A few weeks back, a
programmer took out the custody sanction display capability, but
restoring that is a top priority because we certainly need to compare
jail and programs (and jail with and jail without programs, and vice
versa). What I got was bars representing the correlations between
"success" and three commonly used components of sentences - each of two
theft "talk" programs and alternative community service (ACS).  What
appeared from the bars was that for the fairly broad profile in
question, when sentenced for property crimes, one theft talk program's
referees were more likely to "succeed" than another, and that one's
referees were more likely to "succeed" than those referred to ACS.

When I changed the outcome measure from "any conviction for a property
crime within three years of sentencing" to "any conviction for a violent
or property crime within three years of sentencing" and clicked the
"recalculate" button, I waited a few seconds, and got new bars --
showing a substantial reversal in greater success as between the theft
talk programs and a small drop in success for ACS.

We are just scratching the surface here, and it's taken far longer than
I had hoped.  I expect with this level of crude offender profile, all the
results probably "mean" is that one theft talk program is getting harder
cases than the other (consistent with my anecdotal information about how
those who decide direct their charges to one or the other).  I don't
have an explanation for the shift in outcomes measured by violent
crime.  But I'm confident that when we soon (I hope and expect) add
enough data to do fairly careful profiles of offenders to compare -- so
that we compare people of like criminal history and age, for example --
we will get information truly useful in doing a far better job of
sentencing than we now do.

More importantly, the tool should focus the issue of "what works"
sufficiently to give it a place at the table it has generally been
denied -- at least in the courtroom.

If you're interested, I'm keeping updates on the web page I maintain
about this project; I should have the newest screen shots on line within
a week.

Michael Marcus
Multnomah County (Portland) OR USA
smartsentencing.com