Hard on the heels of Monbiot and the Swedish model, it was interesting to find Stumbling and Mumbling commenting on a Spanish/Italian co-production ‘Is the welfare state really harmful for growth?’ (academic paper by Beraldo, Montolio, and Turati).
The authors have found a statistical link, across 19 OECD countries, between higher GDP growth and higher growth of spending on health and education. They reckon their ‘results are consistent with the hypothesis that the contribution of welfare expenditures more than compensates for the distortions caused by the tax system.’
This conclusion would support Labour’s relabeling of welfare spending as ‘investment’, but would run contrary to most mainstream studies that treat such spending largely as consumption.
S&M has a pretty good go at the paper, pointing out that the authors fail to provide any causal mechanism, or to examine the complex lags that you’d expect in the real world. And also the fact that it is health spending- not education- that registers the strongest relationship to growth. Hardly supportive of the old ‘education, education, education’ mantra.
More generally, we shouldn’t be surprised that there’s a statistical association between income growth and welfare spending. It has long been observed that health and education have a high income elasticity, with proportionately richer countries tending to consume proportionately more. The world’s richest economy, the US, spends most on health (14 per cent of GDP).
Just demonstrating this statistical relationship once again proves nothing. The results may be consistent with the authors’ hypothesis, but they may also be consistent with the more usual explanation.
In fairness, deep within the paper, the authors do concede that the direction of causality is ‘inconclusive’ and that they have not addressed the issue. This is obviously a bit of a hole, so to distract attention they immediately treat us to some graphic detail on how they tortured their data attempting to extract a confession of endogeneity. This included ‘…instrumental variable estimation…using GMM…variables transformed in terms of orthogonal deviations, and a full set of time dummies …results based on consistent one-step estimators… dynamic IV estimations…Sargan test…no autocorrelation in the error term…series of differentiated residuals should present a significant first-order correlation…’
Any the wiser? Neither am I, and I once did a whole Masters level course in econometrics at the LSE. These days, if it can’t be explained in English, I generally ignore it.
S&M quite reasonably suggests we should ponder this paper. Having done so, I think we can move on.
I just hope that, through all those dark twisted byways of euro-research funding, we didn’t somehow pay for it.