DPF has a post up citing Not PC fisking The Spirit Level contention that in wealthy stable democratic countries there’s a relationship between income inequality and life-expectancy. Creswell and DPF’s argument is that the famous chart that illustrates this is flawed:
Because if you add Hong Kong, South Korea and the Czech Republic the regression line changes:
Which is true – but ignores that fact that the point of The Spirit Level was to compare policy outcomes between stable democratic countries: South Korea was a military dictatorship until the late 1980s, the Czech States were part of the Soviet Empire, Hong Kong is not a democratic country, it’s a city-state run as a Special Administrative Region of China.
DPF’s post accuses the Spirit Level authors of cherry-picking states: but looking at the countries they’ve thrown into the second chart you have to ask yourself why they’ve added the Czech Republic but not Poland. Aren’t they pretty similar? Well it turns out Poland has high income inequality (8.8) and a low life expectancy (75 years). Why add South Korea and not Thailand? Income inequality in Thailand is 68 years, income inequality is 12.6. So while the first graph picks countries in order to test a hypothesis the countries added to the second appear to have been very carefully cherry-picked purely to move the regression line.
I don’t care. I’d rather a shorter life in freedom in Somalia than a longer life in captivity in a socialist hell-hole like Finland you commie!
Sincerely yours
J Galt
Unemployed
Comment by Guy Smiley — August 18, 2010 @ 6:55 am
How far are we going back with these third variable explanations? Spain and Portugal were fascist dictatorships until the 70s. Greece and Ireland were very poor until the 80s. Israel is unique for many reasons. I’m not sure how much we should blame events 20+ years ago for a country’s failure to benefit from inequality. Hong Kong might not be a full nation state but it is undeniably a society – The Spirit Level is subtitled ‘why more equal societies do better’. The UN and OECD treat it as a nation state for the purposes of data collection and it self-evidently has a distinct character from China.
Thailand and Poland aren’t in my analysis because they are much poorer than Portugal (which is the poorest nation in The Spirit Level). The rest of them are there because they are either roughly as wealthy as Portugal or wealthier. There are various sources for finding per capita GDP and countries move up and down the rankings quite a bit but these two are as good as any: http://siteresources.worldbank.org/ICPINT/Resources/Atlas_2005.pdf and http://hdr.undp.org/en/media/HDR06-complete.pdf. The second of these is where Wilkinson and Pickett got their inequality figures from. On page 335, you’ll see that Poland’s inequality is 5.6, not 8.8, ie. fairly average (W & P use the 20/20 figure not 10/10). You’ll also see that it has a much lower GDP than any of the countries above. Thailand is far too poor for absolute poverty not to influence the findings, which is the purpose of only looking at rich countries.
If you think more countries should be in the analysis, have a look at Peter Saunder’s work (Beware False Prophets). He uses a broader range of nations and comes to much the same conclusion as me. Alternatively, you might think there are already too many countries. In that case remove Portugal, Greece, Israel and New Zealand (note that all the poorest countries are also ‘less equal’). Again, the correlations will break down. Really, the only way to replicate The Spirit Level’s results is to use the same dates, sources, measures and arbitrary cut off that the authors do.
Comment by Christopher Snowdon — August 18, 2010 @ 7:07 am
But Danyl … that second graph better maps onto my ideological preconceptions about the world. Doesn’t that make it truthier than the first one?
Comment by Andrew Geddis — August 18, 2010 @ 7:12 am
Just as truthy as the first one Lady Penelope.
Comment by dribble — August 18, 2010 @ 7:27 am
“Thailand and Poland aren’t in my analysis because they are much poorer than Portugal”
That seems arbitrary (because it is). Where’s Mexico, Argentina? You included the Czech Republic, but not Slovakia. Slovenia, but not Croatia or Serbia. Looks like cherry-picking (because it is). Well done.
‘Spain and Portugal were fascist dictatorships until the 70s. Greece and Ireland were very poor until the 80s.’ Thanks. Who cares?
Comment by Guy Smiley — August 18, 2010 @ 7:38 am
Dum de doo, the spirit level is cherry picking too.
Comment by dribble — August 18, 2010 @ 7:49 am
and… a simple linear regression is probably the greatest dumbification of a complex issue you could perform.
i think they call it statistics by excel.
Comment by che tibby — August 18, 2010 @ 7:58 am
For life expectancy, I’d exclude the Czech Republic, Slovenia, Singapore, Hong Kong, Spain, Portugal and Greece, because life expectancy is affected by things that happened many decades ago, when none of these countries were first-world democracies.
However, the Spirit Level does this analysis with the same group of countries for lots of measures other than life expectancy (and finds the same pattern for most of these measures), and the list of countries it uses is probably appropriate for most of them.
Comment by kahikatea — August 18, 2010 @ 9:28 am
Really, the only way to replicate The Spirit Level’s results is to use the same dates, sources, measures and arbitrary cut off that the authors do.
My response to this – and your broader accusations – is that if you excluded Singapore from the original Spirit Level graph then they’d have found a much stronger trend line. And if they were cherry picking data points then that’s what they’d have done, with some sophistry thrown in explaining that it wasn’t a real democracy. Instead it’s in there because it fits their criteria, which are very transparent and reasonable. I still don’t quite understand what your criteria are.
Comment by danylmc — August 18, 2010 @ 9:35 am
“I still don’t quite understand what your criteria are.”
SELECT * FROM societies WHERE inequality=high AND wealth=high ?
L
Comment by Lew — August 18, 2010 @ 9:51 am
I think that the most sensible comment about this whole debate that I have seen in the last couple of days is Che’s at 7:58am. Even if regression was a sensible way of looking at this issue, my guess is that the goodness of fit statistic in both graphs above must be fairly low.
Comment by MikeG — August 18, 2010 @ 10:00 am
@ danylmc
Removing Singapore wouldn’t make the trend-line “much stronger”. It would only move the r-squared from 0.20 to 0.22 and it would make it more obvious to the casual reader that the line is going down thanks to Japan and Portugal.
My criteria is the same as W & P’s – study the rich, market economies. They just miss some. I used Portugal as my benchmark only because W & P believe it fits the bill. I personally think that Portugal’s relative poverty explains why it sometimes performs less well. But if Portugal fits the bill, so do several other countries.
Any definition of ‘rich market economies’ is bound to be somewhat arbitrary, but W & P’s definition is excessively arbitrary (ie. top 50 as measured back in 2002). It’s odd to include Portugal and Greece but not the Czech Republic and South Korea, let alone Hong Kong. Either they should all be in, or none of them should be in. Either way, many of the graphs fall apart. I’ve already posted two sources for all countries by GDP. If you think I should have included some others I’d be interested to hear which ones, why, and what difference you think they would make.
BTW, if the correlations were robust in the first place, quibbling about a few countries here or there wouldn’t make any difference. It’s only because they’re so fragile that it matters. Also, they DO exclude Singapore from several other graphs where it would have made a big difference.
Comment by Christopher Snowdon — August 18, 2010 @ 10:00 am
Regression (including simple trend lines) is highly sensitive to outliers. You have to be pretty careful how you interpret regression results, although the plots are so pretty and the statistics so impressive that caution tends to get thrown to the wind. Inferring causation is also very tricky. The better test would be longitudinal analysis, not cross-sectional.
Not that I’ve read the book … although at least that means my opinion is unbiased
Comment by Vibenna — August 18, 2010 @ 10:02 am
[...] }); }I agree with Dim Post that the choice of countries to add to the choice of countries made in the Spirit Level is a bit [...]
Pingback by TVHE » Spirit level: A more fundamental concern — August 18, 2010 @ 10:02 am
Whatever the graphs……..the Dominant economic philosophy now – Neo-Liberalism/Neo-Con’ has a social and financial control agenda and it is designed to drive the Rich v Poor gap . And it does. Look all around for the results. The sales tax – GST/VAT drives the gap even harder…despite what Nats are saying – vs Income tax cuts. Neo-lib policy tricks the middle classes and then bleeds them into poverty . For social control, destroying social structure and a satisfied middle class is essential now.
Comment by bobberesford.com — August 18, 2010 @ 11:32 am
Look, let’s just increase the sample to any country with a GDP of above $14000 USD per year. Then we’ll have a clear indication.
Comment by georgedarroch — August 18, 2010 @ 11:46 am
Not sure why $14,000 should be the magic number but, as it happens, Peter Saunders did pretty much what you asked in his report Beware False Prophets: <a href="http://www.policyexchange.org.uk/assets/Beware_False_Prophets_Jul_10.pdf"http://www.policyexchange.org.uk/assets/Beware_False_Prophets_Jul_10.pdf
Comment by Christopher Snowdon — August 18, 2010 @ 11:50 am
I like this graph, from the kiwiblog thread: http://i1010.photobucket.com/albums/af230/RRM22/SpiritLevel01_thumb2.jpg
The truth revealed!
Comment by Repton — August 18, 2010 @ 12:03 pm
I like this graph, from the kiwiblog thread: http://i1010.photobucket.com/albums/af230/RRM22/SpiritLevel01_thumb2.jpg
The truth revealed!
But you still have to remove Portugal to make that work.
Cherry picking once again!
Or you could claim Portugal is a Nessie dropping
Comment by eszett — August 18, 2010 @ 1:14 pm
I picked $14,000 as approximately half NZ’s per-capita income. Move down towards $5-10,000 and you’re too far from NZ to be able to make meaningful comparisons. Any figure will do, if you can show why it’s reasonable.
Comment by georgedarroch — August 18, 2010 @ 1:18 pm
Vibenna is quite correct that regression is highly sensitive to outliers. In fact, my biggest criticism of the Spirit Level, is not that it uses regression lines (which are pretty much meaningless in this case) but that they tackle the issue of explaning outliers in only the most superficial of ways. It is obvious that the differences between, say, Japan and Portugal are massively larger than a simple measure of inequality.
Comparing statistics between nations is an extremely dubious pastime at the best of times, simply because nations often measure their stats in different ways. Trying to find simple correlations is an exercise in futility or ideology (depending on your selection bias).
Even the much more comparable data between US states ignores the vast complexities of geography, politics and history that underlie statistics such as infant mortality and longevity.
Comment by MacDoctor — August 18, 2010 @ 1:23 pm
Pish posh. Spirit Level good, Not PC/Kiwiblog bad.
Comment by expatexpat — August 18, 2010 @ 1:28 pm
if the graph was drawn plotted from 0 – not 76 – it might look a little like noise.
Comment by NeilM — August 18, 2010 @ 1:51 pm
You can prove anything with facts.
Comment by Roger Parkinson — August 18, 2010 @ 5:32 pm
I agree with MacDoctor that deriving conclusions from this data is pointless, especially given the frequent outliers. However, the thing that got me interested was the very different mortality rates between the Scandanavian countries. Any thoughts on this?
Comment by Owen — August 18, 2010 @ 5:42 pm
Danish bacon.
Comment by expatexpat — August 18, 2010 @ 6:16 pm
a straight line is the shortest distance between the facts and my beliefs.
not say that there isn’t merit in not having huge income disparities but this graph is not a great argument for that. It raises more quesions than it answers.
Comment by NeilM — August 18, 2010 @ 7:18 pm
In lieu of reading the book, I looked at the Equality Trust website, on which the criticism that correlation is not causation is answered with “but there are lots of correlations!”, and on which the P-values for all variables magically scrape in under the 0.05 threshold, and on which they handwave away race even after noting that differences between African-American and Hispanic health in the US can’t be explained by income.
Is there *any* rigour in their work? Even their paper in the BMJ is just a bunch of correlations. All the hard work — building plausible complex models, carefully considering and eliminating alternative explanations — is at best left to the citations. As some kind of leftist, I’d hate for people to think that this shoddy work is our best argument against inequality.
Comment by bradluen — August 18, 2010 @ 8:56 pm
Sigh!
Lies, damn lies, and statistics.
Again.
Comment by peterlepaysan — August 18, 2010 @ 9:35 pm
Re: 25
As a resident of Scandinavia, my (unscientific) observation would be that the Danes definitely drink more, smoke more and exercise less than their Northern neighbours. All things that until recently they were actually proud of, and in fact mocked the Swedes for living long, but boring lives.
Comment by EuroKiwi — August 18, 2010 @ 10:01 pm
Danes definitely drink more, smoke more and exercise less than their Northern neighbours
…and proximity to Germany can’t help.
Comment by Phil — August 19, 2010 @ 10:58 am
Trying to be as objective as possible, I would have thought both Singapore and Hong Kong are inappropriate to include as they are essentially cities. In both cases the society spills across the border. And correct me if I am wrong, but I would imagine a lot of the poorer parts of the society do not live in the cities themselves.
Comment by Matthew — August 20, 2010 @ 6:10 pm