The most important comment on the Giles v Piketty debate yet

Posted on

My friend, and occasional co-author, Howard Reed has I think, published the most important commentary yet on the Chris Giles v Thomas Piketty spat on which I have previously commented. It has just been published in the Guardian.

Howard has analysed all the original data sources used for the Uk by Piketty and Giles and has reworked them, using his own encyclopaedic knowledge of this subject. His conclusion is:

To summarise, Chris Giles's investigation of Piketty's data has uncovered some errors and inconsistencies which Piketty will hopefully address in future work. This shows the importance of quality assurance and third party checking of all results from statistical analysis — particularly when they involve spreadsheets, where it is very easy to make errors.

However, Giles then goes on to make a very serious error of his own in handling the UK data: he treats changes in the way wealth inequality is measured over the decades as if they were real changes in the underlying distribution of wealth. This error leads him to the misleading conclusion that wealth inequality fell in the UK between 1980 and 2010, whereas in fact it has increased (although not by quite as much as Piketty's published results would suggest).

Or, as I'd put it: Chris Giles forgot to correct for the fact that the way in which data was collected in the UK since the 1970s changed several times and made no adjustment for this. As Howard says in the piece:

Chris Giles's initial inquiry into Piketty's results was caused by the discrepancy between Piketty's estimate of UK top 10% wealth share in 2010 (71%) and the ONS estimate from the Wealth and Assets Survey (44%).

However, the vast majority of this discrepancy (23 percentage points out of 27) is, as I have shown in this post, due to data discontinuities.

As a result Giles claimed there was no evidence the rich are getting richer in the UK. As Howard puts it:

To believe that the Giles series represents an accurate picture of the evolution of wealth inequality in the UK over the last 50 years, one would have to believe that the wealth share of the top 10% really did fall by 12 percentage points during the 1970s, and by another 11 percentage points between 2005 and 2006.

Does anyone really believe this? Of course not. Instead, changes to the way wealth inequality is measured have caused our estimate of wealth inequality to jump downwards at various specific points in time — 1974, 1976 and 2006 in Graph 2. And it is these measurement changes which are driving Giles's results.

Or as I put it last week: there are data errors, interpretations and prejudice. I think Howard has settled this debate and I am glad to have fallen on the right side.