Posted on behalf of a concerned contributor.
In Friday's New Zealand Herald we learn that the goverment has built a "crime crystal ball". The insights it yields by analysing data about victims of crime and criminals is being used to guide policy, on the assumption that it will show where money can be invested for the best results.
Unfortunately, we don't know very much about the model, or the IDI (Integrated Data Infrastructure). Lots of questions could be asked:
Who validates the model? Who decides what is and is not causal? Is the data going in reliable? Why are PWC doing this as commercial work rather than public servants? Where is the transparency? Why should we trust the model?
These are live issues. In the US, such models have been shown to encode racism in sentencing decisions. Commercial sensitivity is frequently used as cover to refuse to release official information. Statistics NZ have a commendable approach to security, yet we know that government ministers will override policy intended to protect citizens in order to retaliate to criticism.
The Treasury research mentioned in the article was properly flagged by Treasury as not being reliable enough for policy making because the data used was not consistent across years, yet thanks to Gordon Campbell we also know this hasn't stopped MSD using it to make investment decisions (see "Big Data"). So statistical models are already being used to justify decisions with the appearance of science, even when the people producing them warn they aren't fit for the purpose.
The back story here is a data matching programme of New Zealanders' data which has been going on quietly since Cabinet decided in 2011 that it was a good thing, with very little public debate and no political mandate. Media coverage has focussed on the hoped-for benefits of the "investment" approach, and largely ignored whether those benefits really exist, or how the enabling technology might be abused.
We would probably all learn a lot if people who knew what to ask made requests for official information that would shed light on these issues.