Greetings, Folks

In this morning's Independent: "Computer program to help set
jail sentences".
-----------------------
"Britain is ... to use a computer to help decide sentences of criminals
... A computer program called Oasys wil give danger scores to more than
200,000 people a year, indicating how much risk they pose to the public
and how likely they are to reoffend.

"The scores will be based on factors that include an individual's family
background, address, education and literacy, peer group, past offending,
"attitude" and responses to questions by probation officers.

"The results will help to determine their sentences and how they are
dealt with. The system will also help the Home Office to establish a
national offender database."
-----------------------

Well, I can follow the last bit. All's fair in love and war, and when
you want a national database.

For the rest, some questions occur to me (and, I'm sure, to you too),
and I wonder if anyone reading this actually knows the answers to any
of them.

1a. What data will form the basis of the information on which the
    output of "Oasys" will be based?

1b. What statistical advice was input into 1a?

2a. What computational procedures (no doubt involving statistical
    decision methodology) and criteria are used by "Oasys"?

2b. What statistical advice was input into 2a?

3a. Is there an estimated error rate for "decisions", or scores,
    enerated by this system?

3b. What statistical advice was input into 3a?

4.  Is this the thin end of a wedge introducing new principles into
    the administration of justice?
[As far as evidence of _guilt_ is concerned, with one or two clear
exceptions such as Riotous Assembly, the behaviour of other people
is not a priori relevant evidence in law of your own guilt, unless
a direct association between your own behaviour and that of specific
individuals can be established as a matter of fact. Of course guilt
is not an issue here: the system is intended for use on the already
convicted. But hitherto sentencing policy has operated on much the
same lines: your own history and propensities, and possibly those of
specific people known to be your associates, may be taken into account;
but "being black and living in Brixton" has not of itself been a formally
admissible factor (whatever the private motives of judges may be).
In brief, "I am not as other men" has been a legitimate stance,
unless explicitly undermined in the particular case. Nor am I talking
about use of "class information" to focus _suspicion_ in detection and
investigation of crime: that is fair enough, provided it is subsequently
backed up by direct and specific evidence.]

5a. In the event that there are no satisfactory answers to 1a, 1b,
    2a, 2b, 3a and 3b, might we be looking at a system heavily
    based on subjective prior judgement?

5b. Is there any element of professional statistical opinion involved
    in 5a?

Well, one could go on. Please send any comments privately and I will
try to summarise (judiciously ...) to the list.

Specific information from people who know what's really going on will
be especially welcome.

Best wishes to all,
Ted.

--------------------------------------------------------------------
E-Mail: (Ted Harding) <[log in to unmask]>
Fax-to-email: +44 (0)870 284 7749
Date: 18-Aug-00                                       Time: 11:04:12
------------------------------ XFMail ------------------------------


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%