TL;DR

Organisations of all types are increasingly making decisions based on data and its analysis, but the rigour involved in this hasn’t yet entered our broader social discourse. I’m hopeful that we all start getting better access to data, and better understanding of the analysis and modelling process so that decisions can be made for the right reasons.

All models are wrong, some are merely useful — Simon Wardley channelling George Box

Background

I spend my days encouraging people to make better decisions based on scientific method and data — collect, analyse, model, hypothesize, experiment — rinse and repeat[1]. My work is just a minuscule part of the overall trend towards running companies on data rather than opinion, and the march towards machine learning[2] and artificial intelligence it brings with it. This makes me very critical of data when it’s put in front of me, and how it gets analysed. I’m going to use a news article I read this morning as an example of bad practice in order to illustrate how things can (and probably will) change for the better.

The News

I’m going to pick apart a no byline piece from the BBC ‘Four-year MOT exemption for new cars proposed’. It’s full of facts and figures, but also has all the hallmarks of a rushed together content farm piece as described in ‘the rest is advertising’.

The proposal

The UK Ministry of Transport (MOT) is proposing that new cars be allowed to go an extra year (4 instead of 3) before their first MOT test. This almost certainly is a decision that’s been made in light of the data. The crucial question here, and one that’s not answered by the article is ‘how many cars fail their MOT test when first presented at 3 years old?’. The MOT people surely know the answer to that question, and that answer no doubt informs the statement that “new vehicles are much safer than they were 50 years ago”.

The irrelevant opinion

The article goes on to present data from an Automobile Association (AA) member poll. Apparently 44% were in favour of the change to 4 years, with 26% against.

It’s pretty clear that those AA members weren’t presented with the data that the MOT has, otherwise I’d expect a vary different outcome.

A question asked with facts presented:

The Ministry of Transport has found that 99.9% of cars presented for their MOT test at 3 years old pass the test, and they’re proposing that new cars now start taking the test after 4 years — does that sound reasonable to you?

Gets a very different answer than:

The Ministry of Transport says that new cars are safer than they were in the past. Do you think the MOT should start at 4 years instead of 3 years like it is now?

My bottom line here is: who gives a rats ass what a bunch of ill informed drivers think — where are the facts driving this decision?

This is not (entirely) the writer’s fault

For sure the writer could have gone back to the Ministry and asked for the fail rate data for cars at 3 years old (and 4 years old etc.), and I’m sure a better article would have resulted. But that’s too much to ask in a world of churning out content and reacting to the next press release or politician’s tweet.

If the Ministry was doing a good job of communicating its proposal perhaps it could have also explained its reasoning, and spoon fed the data with the press release.

What’s this got to do with politics?

Everything is politics — Thomas Mann

With Brexit and Trump’s election 2016 brought a moral panic around ‘fake news’ and the whole concept that one person’s opinion can be more valuable that another person’s fact.

Anti-intellectualism has been a constant thread winding its way through our political and cultural life, nurtured by the false notion that democracy means that ‘my ignorance is just as good as your knowledge.’ — Isaac Asimov

Facts come from data, but it’s easy for the causal link between collected data and presented ‘fact’ to become stretched, especially when statistical methods are being used (which is pretty much all ‘data science’). It’s this bending of fact, particularly in social science such as economics that opened the door to statements like this:

Britain has had enough of experts — Michael Gove

It’s interesting to note that the Brexit Leave campaign made extensive use of data science, along with other modern strategic tools like OODA as described by Dominic Cummings in his ‘how the Brexit referendum was won’. It also seems that we’re dealing with the deliberate introduction of noise into Western political discourse per ‘Putin’s Real Long Game’ and ‘Playtime is Over’.

There is a more hopeful angle though. Peter Leyden argues for a positive refrain in his ‘Why Trump’s Inauguration is Not the Beginning of an Era? — ?but the End’, noting that California might (once again) be ahead of the pack in moving on from celebrity politicians to a more data driven and scientific approach.

From Global Politics to Office Politics

The section above touched on major political events, but it’s worth looking more closely at what happens with data based decision making within organisations. Leaning on my own experience it seems to eliminate lots of office politics.

Don’t bring an opinion to a data fight — Kent Beck

Decisions have traditionally been made based on the Highest Paid Person’s Opinion (HiPPO), and perhaps the heart of office politics has been saying and doing what’s thought to keep the HiPPOs happy. As Andrew McAfee observed in ‘The Diminishment of Don Draper’ the HiPPO is being displaced by data and analytics. This can be very empowering to front line people, and in turn displaces traditional political structures. I think this is for the good, as it seems to make workplaces more pleasant and predictable (rather than confrontational and capricious).

Conclusion

In a world where it seems harder than ever to distinguish fact from fiction it’s on all of us to bring our data and clearly explain our analysis, because that provides facts with provenence, facts that can be understood, facts that can be trusted, facts that can triumph over opinion; and there’s nothing more political than that.

I look forward to better data based journalism in our broader social and political discourse, but I also look forward to what data and data science does to the workplace, because I think less political workplaces are nicer workplaces.

Updated 23 Jan 2017 — I meant to add a link to the London School of Economics series The Politics of Data

This post by Chris Swan was originally made to Medium

Notes

[1] For some insight into the work I’ve been contributing to take a look at my GOTO:London 2016 presentation.
[2] One of the ways I like to think of recent advances in machine learning is that computers are finally doing what we might reasonably expect of them — which mainly boils down to not asking a human a question that the machine can reasonably answer for itself.