Fraser Nelson's misleading chart scales
Blog reader Matt Raven has drawn my attention to one of the graphs in this blog post by Fraser Nelson on The Spectator's Coffee House blog on Friday.
Now, before I start I want to make it clear that I am a big fan of Fraser Nelson. He is an excellent columnist and is often able to cut through the crap and get to the real issues especially when it comes to figures and statistics. He has been forensic in his dissection of government figures and long may he continue to do this sort of important work.
Unfortunately his high standards seem to have eluded him a little in one of the graphs in the post referred to. I have reproduced the relevant section here:
Look at the number of jobs in the private sector and split it down by immigration status:So foreign-born workers account for all of the net job creation since 1997.
Now I am not arguing with the figures or his conclusion that foreign-born workers account for all of the net job creation since 1997. I am sure Fraser and the ONS have been rigorous about these and I will take it as read that this is correct. What I am arguing with (and what Matt pointed out to me) is the presentation of these figures. Fraser has the two lines superimposed on top of each other which invites direct comparison but he uses different starting points for the scales on each one.
The scale for the UK born workers runs from 17.8 million to 19.4 million on the left hand side and the scale for the foreign born workers runs from 1.4 to 3 million on the right hand side. I can see what he is trying to do (contrast the relative changes) but I think that this choice of scales would give someone who only glanced at this chart and didn't look closely enough to notice the disparity in scales the impression that the foreign-born figures have actually surpassed the UK-born figures as the UK-born figures have dropped off vertiginously at the end of last year.
This seems potentially very misleading to me and unneccesary. Fraser has a good enough argument without resorting to visual distortions like this.
I have tried to come up with a fairer way to represent this data. Of course I do not have the source data but I printed out the chart and by pinpointing the plot points on it I have been able to reverse engineer an approximation of this data. I have then produced my own versions of it.
Firstly, in order to demonstrate that I am using approximately the same data as Fraser, here is the data presented in a similar way to how he did it:
Secondly I adjusted the scales so that they are both the same running from 0 million to 20 million. This is now what the updated chart looks like.
This gives quite a different impression. The relative sizes of the different types of labour force are much clearer. This is fairer but of course you cannot see the changes in the data so easily this way so if I had been doing Fraser's post I would have also included two more charts with his original scales but with each data set on separate charts which would have made it clearer that they were different scales. They would then clearly be "explosions" of sections of the final chart above. This could even be represented by circling the data sets and pointing an arrow from them to the two sub-charts.
Putting them on the same one was destined to cause confusion in my view. His point is strong enough without this sort of potentially misleading representation.
Posts
9 comments:
Mark, thanks for your kind words. I used that scale as my point was not about the absolute level of immigrants in the private sector workforce (which has doubled from 7% to 14%). My point was about the change - the new jobs.
My fear is that the UK labour market has a dysfunctionality wired into it whereby when the economy expands it sucks in immigrants as opposed to tackling our own worklessness. This, I believe, has been the story since 1997. The scales I used do make that point quite strongly.
But I agree with you that charts and statistics can be tweaked and manipulated to amplify certain points - just as words can be chosen to make sentences give a certain meaning. I don't mean to present my work as Just Down From Mount Sinai truth - it's just another way of looking at the data. I think the blogosophere is uniquely well placed to bring such scrutiny to bear, and I'm delighted that we can have debates like this. They'd never have happened in the MSM.
Hi Fraser. Thanks for stopping by!
I understand that data can be presented in all sorts of ways but I just felt that in this case there is a real danger that someone casually dipping in and out of your post (I don't know if it was also published in the magazine version of TS) could get quite a misleading impression of what the chart was saying.
Anyway, I agree with you that we need to do more to tackle unemployment in this country. It has been a long term failure of this government (and others before it) and appears to be systemic.
Oh, and as I said please keep up the good work!
No matter how the figures are presented they seem to back up the figures on immigration uncovered by Frank Field and Nick Saomes I think it was.
Since 1991 they discovered that 2.4 million immigrants had been allowed to settle in the UK. One of the most interesting parts of these figures is that of that amount 2.14 million came to England. This means that Scotland Wales and Northern Ireland between them only took 260,000 in 18 years or so. Looked at another way it equates to the other home nations taking in 277 people per week ever week for 18 years and England taking 2286 people per week.
The disturbing part about this is public spending under the Barnett Formula is Scotland is approximately £1600 per annum per head higher in Scotland than it is in England whilst England is taking in 10 times more immigrants. Looked at in this way one could begin to make a case for gerrymandering by a government run by a Scottish Cabal. You can begin to see some of the reasons why the white working class of England, newly rediscovered by the political classes, are prepared to vote for the BNP.
Where “someone who only glanced at this chart and didn't look closely enough to notice the disparity in scales” was misled, the blame lies with that someone. Surely there are limits to the aid and warnings that need be given to the careless, indolent and/or thick before allowing them access to graphs? Maybe technology can come to our aid such that only those who have passed an exam in graph reading would be given a password to enable graphs in internet articles to be displayed. This is a non-story and I will not disguise my disappointment at seeing it on your hitherto pleasing blog.
Sorry to hear you feel that way CRM. I enjoy your comments and contributions. I would not have posted this if I did not think there was a serious point to make.
I tend to be a bit pernickerty about the display of data in charts. I still remember being taught when I was doing stats in school about how putting the scale from something other than 0 can be very misleading and you need to tread carefully.
However much you might hope that people study things like charts long enough to fully understand what they are saying, people are often busy and there are 5 different charts in Fraser's piece, all requiring some time taken to digest them. People could be forgiven for not fully interpreting the final chart given the two different scales charted against each other and being left with a wrong impression of the relative sizes of the different groups mentioned.
The fact that Fraser chose to present his data in this way gives his opponents (of which I hope I have made clear I am not one) an opportunity to accuse him of attempting to distort the data.
When I did my work on the safe seats/ expenses scandal correlation a couple of months ago, I could have changed the bar graph so that the scale ran from just below where the lowest bar started. that would have given the visual impression of the disparity I discovered being even more stark. I resisted the temptation because it would have left me open to the charge of trying to distort the data to exaggerate my case. I had enough on my plate defending the methodology and source data I had used without having to defend the way I had presented it so I just went for a straight vanilla bar-chart that ran from 0 in the y-axis.
I hope you understand where I am coming from.
Yes, I do understand. I just think the limits of aid are appropriately drawn earlier. On the political capital point, anyone venturing to criticise would have to explain FN’s graph, thereby reinforcing his point whilst showing (in a self-defeating move) that with modest attention from the reader, his graph is fair enough.
Separately, thank you for your kind words: they are appreciated. :-)
This is the typical presentation of statistics by someone who wishes to present a particular viewpoint; I learnt long ago to look at graphs and other diagrammatic representations for false zeros, etc.
Another favourite is to say that some event "doubles the risk" of an occurrence. True, maybe. But if the original risk was one in a million, maybe not of great concern, but is the original risk was one in ten, doubling the risk could be of considerable concern.
Motto: Don't believe any statistics unless they are presented by a statistician who doesn't have an axe to grind!
This is the typical presentation of statistics by someone who wishes to present a particular viewpoint; I learnt long ago to look at graphs and other diagrammatic representations for false zeros, etc.
Another favourite is to say that some event "doubles the risk" of an occurrence. True, maybe. But if the original risk was one in a million, maybe not of great concern, but is the original risk was one in ten, doubling the risk could be of considerable concern.
Motto: Don't believe any statistics unless they are presented by a statistician who doesn't have an axe to grind!
I probably agree with the use of chopped scales on this one - as Fraser has done - rather than rebasing both to zero, since that draws attention to the comparative changes, which was the aim.
I'd hope that someone taking the time to read a piece with 5 graphs in it would not read the piece and skim one of the graphs.
My improved version, if I did one, would make label the graph "change in employment" rather than "numbers employed" and start both scales at zero - i.e., measure the delta.
Post a Comment