Sarah O'Connor Jobs and politics: the perils of data journalism

Here’s a simple question: which of Britain’s parliamentary constituencies have seen the biggest job market recoveries since the coalition government took office in 2010?

The answer, I thought, might well generate a news story in the week of the UK general election. So I downloaded a time-series of the number of Jobseeker’s Allowance benefit claimants in each constituency. (I used JSA claimant data because, when you’re looking at small geographical areas, they’re far more accurate than survey-based measures of employment and unemployment.)

Then I ran into a dilemma. I thought I’d blog about it, because it’s a reminder that data journalism requires you to make judgements all the time – and those judgements can dramatically change the nature of the story you end up telling. At the FT we try to give readers information and insight without bias (particularly during hyper-partisan election campaigns) so this is the sort of thing we think hard about.

The dilemma was this: what’s the best way to measure the fall in unemployment between 2010 and 2015? First I ranked the constituencies by the percentage change in the number of JSA claimants. The top of the table looked like this:

 

It looks like a cracking story. Of the 30 constituencies that have seen the most dramatic labour market improvement over the past five years of Tory-led government, 28 are Conservative. But there’s a problem. You would expect to see bigger percentage changes when the numbers you start from are low. And Conservative constituencies like these had relatively low unemployment to start off with.

Then I ranked the constituencies by the percentage point change in their claimant count rate (the proportion of the population on JSA). This time, the top of the table looked like this:

That looks like a cracking story too. But it has the opposite problem: you would expect to see bigger percentage point falls when the percentages you start from are high. And Labour constituencies like these had relatively high unemployment to start off with.

Neither of these methods is wrong. Both are perfectly reasonable, in fact. Yet the stories they tell are polar opposite. It didn’t seem right to pick one or the other. Instead I zeroed in on one place where something dramatic seemed to be happening (Liverpool) and went there with a notebook to find out more.

For me, this exercise was a useful reminder that data journalism is no more inherently objective than any other sort of journalism – and that you should beware people who tell you otherwise (or fail to tell you about the judgements that underpin their analyses.)

My colleague John Burn-Murdoch has kindly put the constituency data into a nifty sortable table, so you can play around with it yourself and look up your own constituency. Click on the grey headers to sort by the different variables.

Download the full dataset.

Note: I’ve adjusted the data to include jobseekers who’ve been transferred from JSA onto Universal Credit, a new benefit that is rolling out in some areas.