em posted an update 2 weeks, 4 days ago
Just reading this https://www.theguardian.com/money/2018/apr/02/shocking-disparity-in-pension-income-revealed-by-latest-hmrc-data ,
That means the City of London tops the UK table for pension income, yet just eight miles away lies the area with the UK’s second-lowest pension income. In the London borough of Barking and Dagenham, the mean average pension income that older people had to manage on was £12,800.
For a kick off I cannot see why anyone would be shocked that a pensioner in the Square Mile has a much larger pension than someone in Barking or Stoke, like DOH.
However what does shock me is that a rag like the Guardian which claims to do top quality journalism refers to mean average, when I would think a mode average would be much more revealing.
I have no evidence but I would suspect that ex ford workers and the odd millionaire are raising that mean, way above the mode.
I’d imagine the pension difference correlates well with the income difference when they were working and how much they invested in it. It’s a non article by the guardian; it is a pre local election bit of anti rich elite vote chasing on behalf of momentum.
Mode is so unfashionable, though…
I wonder what the correlation is between Guardian reading and pension.
I’m going to guess that those on a low pension don’t read the Guardian…
“I have no evidence but I would suspect that ex ford workers and the odd millionaire are raising that mean, way above the mode.”
The ex ford workers are in the area with a low mean. Given the size of the Ford plant to the town, I would have thought the median mode and mean for Dagenham wouldn’t be much different as so many of the pensioners would have worked there.
The City of London has expensive housing and Dagenham is a cheaper place to live. None of this data is particularly surprising. If you want the raw data I expect it’s on the HMRC website and you can work out the median and mode if you want to.
Whether the mean, median, or the mode, is best suited to interpret the data really depends on the distribution of the data. So I am not sure how you can make such a statement without having any idea about the distribution.
My understanding of income distribution in the UK tells me they tend to approach a log-normal distribution with a low sigma, albeit with a thick upper tail. Whether you use the mean, median, or mode, it shouldn’t make too much difference, but the median or mode should be marginally better in this case. I’m just assuming that pensions follow a similar distribution as income which could be completely wrong.
However you have to consider that in an article written for the general public, a journalist may decide to use the mean even if it’s not scientifically the very best measure, simply because most of the audience can make sense of the mean, but may not make much sense of the median or the mode.
Sometimes it’s better to give your audience sub-optimal metrics they can understand well, than giving them optimal metric they are likely to misunderstand and misinterpret. As long as they tell the same story. The important thing is to always highlight the caveats of your metrics, which they have not done.
I this particular case the story is really about group differences, if you’ve got a really fat tail in the distribution then the mean median and mode could all be completely meaningless in interpreting group differences !
A better way to present this data would have been to use a simple plot.