RoC*Rol Formula and Briefings

Dear Department of Corrections,

My understanding is that the Corrections department uses a program
called 'RoC*Rol' in predicting the chance that an offender will
reoffend. I would like to request under the Official Information
Act the following information about this system:

1) the source code of the RoC*Rol program

2) if it isn't included in the answer to request (1); the formula
that the RoC*Rol program uses to generate a score

3) if it isn't included in the answer to request (1) or (2), the
different variables that RoC*Rol uses (the article 'Big Data is
Watching You' on Stuff mentions 14 variables, but the 'What is
RoC*RoI' section in the CPS operations manual only mentions 8)

4) any reports, memos, studies, briefings or other documents that
analyse the effectiveness of RoC*Rol

5) any reports, memos, studies, briefings or other documents that
analyse the relationship between RoC*Rol scores and the chances of
the Parole Board granting parole (I'm guessing that the Parole
Board may have some of this information, so I am sending an OIA
request for this to them also)

If any documents are withheld from my request I would like to
please know the titles and dates of those documents.

Although not required under the OIA, I understand it can sometimes
be helpful to explain the reason for a request: I have two primary
reasons.

Firstly, I am law student at Canterbury University with an interest
in Criminal Law, and I would find it interesting to see what
factors are assessed in determining whether an offender will
reoffend, and the weights given to those factors. For example, it
is often said that alcohol and other drugs have a strong
relationship with crime: I'd be interested to know if attending a
treatment program effects the RoC*Rol score at all. This is the
reason for requests 1, 2, and 3.

Secondly, I find it a bit scary that a computer program is making
this kind of assessment as opposed to a trained human. This is the
reason for requests 4 and 5.

Feel free to contact me if you wish me to clarify my request.

Yours faithfully,

Joshua Grainger

Dear STURMFELS, Emma (WELLHO),

I would much prefer to receive a scanned electronic response unless
postal delivery is absolutely necessary. If the files are too large
to send via email then there is an upload facility on the FYI
website for submitting a response (I believe you can access it it
at http://fyi.org.nz/upload/request/rocrol_...)

FYI is a public website, which means that others get to scrutinize
and look over documents requested, increasing accessibility and
advancing open government. This is along with the costs saved by
not having to pay for postage and printing.

Yours sincerely,

Joshua Grainger

Dear STURMFELS, Emma (WELLHO),

My attention has been drawn to the Ombudsmen guidelines on this
matter, which state:

"Where the information requested is contained in a document,
section 16(2) requires that information shall be made available in
the way preferred by the requester unless to do so would:
 Impair efficient administration;
 Be contrary to any legal duty in respect of the document; or
 Prejudice the interests protected by sections 6, 7 or 9 of the
Act and in the case of section 9 there is no countervailing public
interest."

May I ask which of these criteria apply to the decision here?

It seems to me that if material is being photocopied or printed,
then if this would be no harder than scanning the material: indeed
often the same machine is used for both scanning and photocopying
in many offices.

If this is a problem of technical difficulties due to the size of
the file, I'd note that something other OIA officers have done when
responding to FYI requests is to upload the file to a web service
and then send the link, such as using the service
https://www.yousendit.com/

I apologise for the difficulties: I am not trying to be difficult,
it is just that if am trying to make these documents accessible to
others who are interested.

Yours sincerely,

Joshua Grainger

Dear Info@Corrections,

Thank you very much for your reply. I do appreciate it, in
particular the information that static variables as are used, and
the rather large sample size.

However, your reply seems to only address questions 4 and 5 of my
request (reports about RoC*Rol), rather than 1-3 (the source-code,
algorithm, and variables of RoC*Rol). Although your answer was
useful, I am still hoping that I will receive a reply to questions
1 through 3 of my request.

Will more information be forthcoming in the future, or was this
response anticipated to be an answer to the whole of my OIA
request?

Yours sincerely,

Joshua Grainger