Why we need ‘weirdos’ like Dominic Cummings Do we really want policy decisions left in the hands of the innumerate?

Snow’s best novels, in my opinion, are those that deal with the human-shaped fall-out from atomic research, like The New Men. Perhaps the mathematical rather than the physical sciences are more relevant to the 21st Century — I would say that, wouldn’t I — but we are beginning to perceive the human-shaped fall-out from a lack of ability in quantitative reasoning among that caste who set the rules for How Things Are. I described my own clash with an algorithm (over whether or not I should be prescribed a statin) some time ago at UnHerd. Do you qualify for a mortgage, will you get parole, is that a tumour in your spleen – increasingly guidance on these questions will come from algorithms embedded into software.

Are you happy with that? (I think you should be.) But are you happy that the laws relating to that type of activity are being devised and voted on by people who couldn’t tell you the meaning of “prediction model” or “classification algorithm” any more than their predecessors could have answered Snow’s questions about mass and acceleration?

And if you accept my premise — that data science could power the next British industrial revolution — do you want the policies about (for example) which universities should be expanded, and where they should focus their research – do you want that left in the hands of the innumerate?

I’m not exaggerating. Remember the 2012 survey of members of Parliament that showed how few of them could reason statistically? Remember it, and shudder. Nearly a hundred MPs were asked “If you spin a fair coin twice, what’s the probability of two heads?”. Of Conservative MPs, 47% gave the wrong answer. Among Labour, that figure rose to 77%. (The answer, of course, is 25%, since the coin, on its second toss, doesn’t care on which side it landed the first time.)

So when Google’s chief economist said in 2009: “I keep saying the sexy job in the next ten years will be statisticians,” I didn’t join in the laughter; I was cautiously optimistic. In 2016, I spent a sabbatical year in the civil service, writing speeches for a cabinet minister, as part of his extended private office. It gave me first-hand experience of our political class, the elected and the ‘Rolls Royce’ civil service, which I cautiously (I am not as brave as Cummings) described in an article in 2017. It’s worth quoting a bit of it, because it remains relevant to what Cummings is trying to achieve:

A  few years ago, the then prime minister, David Cameron, wanted to improve facility with English among Britain’s immigrant communities. He was chastised by his Left-wing opponents for focusing efforts on women with (mainly) Bangladeshi and Pakistani backgrounds.

In my teabreak one day, I looked up the official statistics which broke down English ability by ethnicity and geography-of-origin, and calculated the odds of lacking English in each of those categories, and compared these between women and men.

It did indeed appear that women with Indian, Bangladeshi and Pakistani backgrounds were between four and five times as likely to lack any English ability, as compared to their menfolk. But I listened in vain to hear mention of such statistics in any political or journalistic commentary.

Government statisticians would have calculated any of those odds ratios for ministers, had the ministers asked for them. But my unofficial calculations took a few minutes prodding with an excel spreadsheet […]. The point is that such a basic facility with numbers should be taken for granted amongst the people who rule us.

Elsewhere in his blogs, Cummings has set out a vision — with references to real-world examples — whereby models to predict policy outcomes are embedded into ministerial ways of working. Thus “Implementing this policy will improve stakeholder outcomes with regard to fairness” — a pretty standard form of civil service submission to a minister — would be replaced by a live screen showing that “Implementing this policy is predicted with 95% probability to make X citizens better off by an average of £Y, plus or minus Z%.” The minister could adjust the policy and in real-time see the impact. Proper cost-benefit analysis could underpin major decisions.

I trained as a statistician and run teams of data scientists, so I have skin in this game. “We need more universities!” shout ministers. “Why? To do what?” aren’t questions they seem disposed to answer. But anyone trying to recruit data scientists into British industry could tell them what we need more of. Even though the universities have expanded like balloons at a children’s party, there aren’t enough young British people with the qualifications we need to find new biological targets to cure diseases, develop new medical tech to diagnose cancer quicker, continue to dominate the financial markets — you name it; China will be opening a university department devoted to it.

There are brilliant places, such as Bristol, whose Centre for Computationally-intensive Statistics and Data Science is investing in the methods the age requires, producing exactly the sort of doctorally-trained data scientists that British companies need. Three cheers to them — they’re not alone (Newcastle comes to mind too, as well as Health Data Research UK). That is the sort of work that government should be curating, expanding, investing in.

Left to its devices, will the civil service comply? Remember that question, when you read a snotty piece about Cummings in the Guardian. More power to his elbow: a data-centric civil service could enable better governance and underpin the industrial strategy the whole country needs. Data science really is as important as Shakespeare, and I’d rather a civil servant was master of the first, than well-read in the second (I’d rather she was both, of course). To build One Nation, dismantle that Establishment indifference to one of our Two Cultures.

---

Graeme Archer is a statistician and writer.

graemearcher