More predictive perils

One of the problems of predictions and probabilities is that the former tend to be wrong (particularly when made by the famous) and the latter are only probable, not certain and highly susceptible to random events.

The blog was reminded of both realities by some new research on predictions by government analysts and recent Victorian political events which could shift the probabilities of a change of government in the State. The reality about predictive failures were exposed by the 2005 release of a 20 year study by Philip Tetlock (see blogs passim) which showed that the performance of experts making predictions was not much better than chance and that the more famous a pundit the more likely they were to be off the money.

Now The Economist  (July 19) has reported some new research by two Canadians from Defence Research and Development Canada published in the Proceedings of the National Academy of Science. It can be found at The authors, David Mandel and Alan Barnes,  looked at 1,500 intelligence assessments produced by a nameless Canadian agency. You can probably guess which area it worked in. They found that the experts were right about three quarters of the time and the explanation for their success seems to because they were cautious about the predictions. Now we know the CIA knew about everything that ever happened from its inception onwards although they never quite got around to making accurate predictions on the basis of this rich intelligence. Just before the Soviet Union collapse they were, famously, making predictions about the strong growth of the Russian economy. And, of course, they notoriously produced ‘intelligence’ to order to justify the Iraq debacle. The Canadian intelligence agencies might well be less hopeless than the CIA but one suspects all intelligence agencies do their utmost to cover up their failures and promote their successes – if only by talking about the great successes they have had and which everybody would appreciate if they were just allowed to tell people about them.

Most importantly though the Mandel-Barnes findings also confirm one important Tetlock finding – that the best at prediction are those who are most cautious in their approach. As The Economist said: “Unlike pundits, who can pontificate from then safety of their armchairs, analysts know that their advice is likely to have consequences in the real world” although one might well respond to that with, what real world for intelligence analysts already?

But, back to the blog’s estimates, based on Bayesian probabilities derived from Nate Silver about the Victorian election. The probabilities may have been significantly influenced by the recent controversy about ALP officials and advisers listening to a journalist’s tape of recorded interviews, including one with former Liberal Premier, Ted Baillieu, which later got emailed around to Liberal Party members by some unnamed Liberal Party people who were factional opponents of the former Premier.

The blog won’t rehearse all the details but a quick look at The Age online will provide that. Basically the ALP, including advisers to the Opposition Leader, committed the unpardonable sin of doing something both dumb and wrong. One or other is bad  – combined they are awful. The blog remembers, when working with the ALP in Opposition in the 1970s, falling into a similar trap when we someone gave us a letter to a Cabinet Minister outlining discussions about a casino. It was a time in the midst of land deals revelations and we were devoting much of our time to creating an impression of crisis in the Government. So we released this new bombshell publicly. Only problem was that it became obvious that this was not a copy but rather an original which was obviously stolen – possibly from a room Opposition staff regularly used as a short cut. Needless to say the backlash, although not as bad as that in the current case, was significant – partly helped by internal ALP opponents tut tutting about it all to sympathetic Gallery reporters.

At the time we made a mistake because we got obsessed with the politics and causing damage to the other side and didn’t spend enough time thinking about all the issues. Probably the current situation has been made worse by the changes in politics since then. These changes have accentuated the trends towards privileging tactical advantage over issues and are seen in the obsessive tactical and policy-free discussions which drive much political strategy development; the concern with short-term tactical advantage; and the general malaise and amorality among many in the political classes from all political parties.

There are two possible outcomes of this latest scandal. First, it might entrench a character issue which could have a major influence in the election. Second, it could just blow over and be largely ignored by the public who will see it as what they expect from a squabble between two groups they hold in contempt – politicians and the media. Following Tetlock, Mandel and Barnes the blog cautiously believes the outcome will probably be midway between the two with a bias towards the second possibility. On balance the voters’ feelings about Tony Abbott and the Federal budget still seem to be a bigger issue, with more legs, than ‘tapegate’. But most importantly, according to an experienced pollster the blog respects the issue may have few legs simply because State politics is nowadays a lower order issue which has less cut through than other things.