Monday, 9 March 2009

Did post-communist privatisation kill?

An interesting question and one for which the medical journal Lancet argues the answer is "yes". An article, by David Stuckler, Lawrence King and Martin McKee, "Mass Privatisation and the Post-Communist Mortality Crisis: A Cross-National Analysis", Lancet, published online, January 15, 2009, argues that there is a robust correlation between the extent of privatisation and the adult male mortality rate using country-level data for about 24 economies of Eastern Europe and the former Soviet Union.

The paper has been controversial and there are legitimate questions that can be raised about various aspects of the methodology of the article, including the use of country-level data to study death and ownership – phenomena that are inherently microeconomic. Related issues concern possible confounding effects – alternative explanations for the correlation.

But there is a very obvious question to do with causality: How could changing ownership from state to private have raised mortality? The authors of the Lancet article put forward the theory that privatised firms cut employment and then refer to the extensive evidence on the negative impact of unemployment on health to link job loss to mortality. This idea in turn raises the question: Did privatisation systematically lead to substantial job loss? If not, then the causal mechanism of the paper breaks down and the article's results are open to question. Note that the Lancet article provides no evidence on this question.

So did privatisation lead to substantial job loss? In a column at John S. Earle says the answer is a clear "no". The column, "Mass privatisation and mortality: Is job loss the link?", looks at the results of a study forthcoming in the Economic Journal. This study is "Employment and Wage Effects of Privatisation: Evidence from Hungary, Romania, Russia, and Ukraine" by David J. Brown, John S. Earle and Almos Telegdy.

Earle discusses, in his column, differences between the data used in the Lancet paper and the EJ paper and the advantages of the EJ paper's dataset:
Our analysis is not at the country level, as in the Lancet article. The problem with such aggregated data is that a variety of confounding influences may explain the results – just the sort of issues that have heated up the blogosphere, but that may never be resolved simply because they cannot be measured. Instead, our analysis uses data on nearly every manufacturing firm inherited from the socialist period in four major transition economies, Hungary, Romania, Russia, and Ukraine.

The firm is the level at which decisions on employment are made, and with our data we directly observe ownership, employment, and many other variables. Equally important, we observe firms for many years (up to 20 years in these databases), so we can follow the path of employment and other variables for long periods both before and after privatisation takes place. We also observe firms that are never privatised, which together with those that are not yet privatised (but will be) can form a control group in examining the effect of privatisation on employment within a particular industry and year. The ability to compare firms within industries and years – apples with apples, rather than apples with oranges – is another benefit of analysing data at the level of the decision-maker, rather than the aggregate.
The important result in the EJ study that Earle notes his column is
Analysing these data with several statistical methods to control for possible biases due to selection of firms for privatisation, we find no evidence that privatisation systematically lowers firm-level employment.
Figure 1, which comes from the column, shows results from two methods.
One incorporates firm fixed effects to control for selection bias in the level of employment, and the other adds firm-specific trends to control for selection bias in the growth of employment (labelled “without trends” and “with trends” in the figure, respectively). The estimated effects of privatisation to domestic owners are generally positive, and where they are negative the magnitudes are very small and usually statistically indistinguishable from zero. The estimated effects of foreign privatisation are almost always positive, large, and statistically significant, generally implying a 10-30% expansion of employment following the foreign acquisition.

The estimated foreign privatisation effect in Romania is the largest negative value, but it is only -7.1%, and statistically insignificantly different from zero. In the country with the most (in)famous mass privatisation, Russia, the domestic privatisation effects are positive, and, when estimated with trends, the effect is the largest of any of these four countries. Analysis of the long time series in the data shows that the absence of negative employment effects of privatisation is the consequence neither of delayed restructuring several years after privatisation nor of pre-privatisation downsizing, which is negligible in these economies.
Earle goes on to look at the possibility that the reason for the lack of job losses is that privatisation simply doesn't really matter for firm behaviour – new private owners do not restructure and therefore do not lay off workers. He writes
Our [Brown, Earle and Telegdy] research investigates this possibility by decomposing the employment effects of privatisation into two components, which we label “productivity” and “scale” effects. Holding the firm’s scale – its level of production – constant, an increase in productivity tends to lower employment. Holding constant the level of productivity, an increase in scale tends to raise it.
The empirical analysis of these mechanisms finds that privatisation tends to raise both productivity and scale. The results are displayed in Figure 2, which again comes from the column.
This figure shows that
Both effects are much larger in firms privatised to foreign investors, with 10-25% increases in productivity, and 15-40% increases in scale. The dominance of the scale over the productivity effect implies the positive impact of privatisation on employment that we observe.
and that
Privatisation to new domestic owners in Hungary and Romania also yields positive productivity and scale effects, but they are smaller (6-10%) than the corresponding foreign effects, and the productivity effects slightly dominate the scale effects, resulting in the very small negative impacts of privatisation on employment in these cases. The productivity and scale effects of domestic privatisation are also positive but very small in Ukraine, and they nearly exactly cancel, leaving a tiny positive impact on employment. Domestic privatisation in Russia is the outlier, with negative estimated effects on both productivity and scale, but the drop in productivity exceeds the fall in scale, resulting in a positive net employment impact.
The important overall result is that in no cases are there substantial job loses showing up in the data and thus the causal link put forward by the authors of the Lancet article is not supported by the firm-level data.

Earle ends his colum by saying
Of course, it is possible that some other link, not suggested by the article and unrelated to employment outcomes, could explain the observed privatisation-mortality correlation at the country level. Our analysis suggests that further progress on this question would benefit from analysis of data at the level where the action occurs – individual data in the case of death and firm data in the case of privatisation.

No comments: