Portal de Periódicos da CAPES

Did mass privatisation really increase post-communist mortality? – Authors' reply

2010; Elsevier BV; Volume: 375; Issue: 9712 Linguagem: Inglês

10.1016/s0140-6736(10)60160-2

1474-547X

David Stuckler, Lawrence King, Martin McKee,

Employment and Welfare Studies

We have watched with interest the increasing sophistication of attempts to discredit our paper, many at fora where we were not present, so we are grateful that we can finally respond. These criticisms have included misrepresentation of basic mortality data and a series of letters from leading advocates of privatisation that was, in turn, gratuitously offensive, epidemiologically uninformed, and factually wrong. Unfortunately, these two letters continue on this path, with manipulation of data in ways that can be interpreted as owing more to the pursuit of preconceived beliefs than to a search for scientific truth. As Christopher Gerry and colleagues note in their webappendix, “Our goal here is not to establish per se what does cause mortality. Rather, we are concerned to demonstrate that there is no evidence in the data used by Stuckler et al that mass privatisation resulted in increased mortality.” Given this, we are surprised that neither sees any need to declare a conflict of interest as authors on both letters have received funds from organisations supporting privatisation. In attempting to show that we are wrong, they employ biologically implausible assumptions, commit fundamental methodological errors, and thus generate findings that are inconsistent with published data. These measures meet established criteria for “data torture”,1Mills J Data torturing.N Engl J Med. 1993; 329: 1196-1199Crossref PubMed Scopus (166) Google Scholar as summarised in the panel and in our webappendix, where we address all of their points in detail.PanelDefinitions of “data torture”1Mills J Data torturing.N Engl J Med. 1993; 329: 1196-1199Crossref PubMed Scopus (166) Google Scholar and selected examples of such from letters by Gerry and colleagues and by Earle and GehlbachInterpreting every result as confirming a major hypothesis •Earle and Gehlbach state that, because the 2-year lag of mass privatisation has a negative coefficient, “one could as easily conclude that privatisation lowered as raised mortality in the former Soviet Union.” However, this finding is consistent with our theory, reflecting an artifactual rebound from the labour market shock of privatisation.•Their suggestion that privatisation of already overstaffed and underproductive firms increased employment and decreased productivity is inconsistent with other authoritative sources such as the European Bank for Reconstruction and Development as well as our individual-level analyses (see webappendix).8European Bank for Reconstruction & DevelopmentTransition report: ten years of transition. EBRD, London1999Google ScholarLack of biological plausibility •Neither group of authors offers a plausible alternative explanation. Both transform data without biological justification. Their disregard of immediate effects and invoking of 2-year lags is contrary to existing epidemiological evidence.2Perlman F Bobak M Assessing the contribution of unstable employment to mortality in posttransition Russia: prospective individual-level analyses from the Russian Longitudinal Monitoring Survey.Am J Public Health. 2009; 99: 1818-1825Crossref PubMed Scopus (44) Google Scholar, 9Leon DA Saburova L Tomkins S et al.Hazardous alcohol drinking and premature mortality in Russia: a population based case-control study.Lancet. 2007; 369: 2001-2009Summary Full Text Full Text PDF PubMed Scopus (307) Google Scholar Gerry and colleagues implausibly find that the only determinant of the post-communist mortality crisis is mortality in the preceding year.•Earle and Gehlbach generate biologically implausible short-term associations between income and cancer. Gerry and colleagues generate an implausible lack of a short-term association between war and mortality.Lack of reporting of number of data comparisons made •Earle and Gehlbach fail to present the full extent of comparisons made in trying to test their hypothesis that we are incorrect. In at least one case, they attempt to recode our war variable but fail to confirm their hypothesis (that we are wrong), but do not present it.Dropping subjects without biological justification •Earle and Gehlbach drop significant subjects without appropriate justification (ie, removing each country's mortality experience over the period studied from the analysis of over 80% of the mortality variations we sought to explain), and Gerry and colleagues also remove subjects from the analysis (ie, using the lag of mortality rates). They propose no biological justification for these steps, nor do they justify the initial year chosen for estimating long-term mortality trends.•Both groups effectively drop exposures and outcomes without justification. They fail to adhere to scientific conventions in estimating contemporaneous and lagged effects simultaneously in finite-distributed lag models. Instead, both rely on an ad hoc specification, failing to apply standard joint F tests or a Bayesian Information Criterion or a biological justification to determine the appropriate number of lags.Inappropriate classification of exposure and disease •Earle and Gehlbach use the wrong exposure variable (successful privatisation, all types; misclassifying >50% of the exposures and dropping 36% of the population cases of mass privatisation). Both Earle and Gehlbach and Gerry and colleagues erroneously remove the contemporary effect of privatisation from analysis, changing the study's hypothesis and removing country mortality experiences (>70% of mortality data). Earle and Gehlbach also use an incorrect disease measure in their regional analysis (crude death rates).See webappendix for more details and examples. Interpreting every result as confirming a major hypothesis •Earle and Gehlbach state that, because the 2-year lag of mass privatisation has a negative coefficient, “one could as easily conclude that privatisation lowered as raised mortality in the former Soviet Union.” However, this finding is consistent with our theory, reflecting an artifactual rebound from the labour market shock of privatisation.•Their suggestion that privatisation of already overstaffed and underproductive firms increased employment and decreased productivity is inconsistent with other authoritative sources such as the European Bank for Reconstruction and Development as well as our individual-level analyses (see webappendix).8European Bank for Reconstruction & DevelopmentTransition report: ten years of transition. EBRD, London1999Google Scholar Lack of biological plausibility •Neither group of authors offers a plausible alternative explanation. Both transform data without biological justification. Their disregard of immediate effects and invoking of 2-year lags is contrary to existing epidemiological evidence.2Perlman F Bobak M Assessing the contribution of unstable employment to mortality in posttransition Russia: prospective individual-level analyses from the Russian Longitudinal Monitoring Survey.Am J Public Health. 2009; 99: 1818-1825Crossref PubMed Scopus (44) Google Scholar, 9Leon DA Saburova L Tomkins S et al.Hazardous alcohol drinking and premature mortality in Russia: a population based case-control study.Lancet. 2007; 369: 2001-2009Summary Full Text Full Text PDF PubMed Scopus (307) Google Scholar Gerry and colleagues implausibly find that the only determinant of the post-communist mortality crisis is mortality in the preceding year.•Earle and Gehlbach generate biologically implausible short-term associations between income and cancer. Gerry and colleagues generate an implausible lack of a short-term association between war and mortality. Lack of reporting of number of data comparisons made •Earle and Gehlbach fail to present the full extent of comparisons made in trying to test their hypothesis that we are incorrect. In at least one case, they attempt to recode our war variable but fail to confirm their hypothesis (that we are wrong), but do not present it. Dropping subjects without biological justification •Earle and Gehlbach drop significant subjects without appropriate justification (ie, removing each country's mortality experience over the period studied from the analysis of over 80% of the mortality variations we sought to explain), and Gerry and colleagues also remove subjects from the analysis (ie, using the lag of mortality rates). They propose no biological justification for these steps, nor do they justify the initial year chosen for estimating long-term mortality trends.•Both groups effectively drop exposures and outcomes without justification. They fail to adhere to scientific conventions in estimating contemporaneous and lagged effects simultaneously in finite-distributed lag models. Instead, both rely on an ad hoc specification, failing to apply standard joint F tests or a Bayesian Information Criterion or a biological justification to determine the appropriate number of lags. Inappropriate classification of exposure and disease •Earle and Gehlbach use the wrong exposure variable (successful privatisation, all types; misclassifying >50% of the exposures and dropping 36% of the population cases of mass privatisation). Both Earle and Gehlbach and Gerry and colleagues erroneously remove the contemporary effect of privatisation from analysis, changing the study's hypothesis and removing country mortality experiences (>70% of mortality data). Earle and Gehlbach also use an incorrect disease measure in their regional analysis (crude death rates). See webappendix for more details and examples. John Earle and Scott Gehlbach, after replicating our findings, introduce three implausible and erroneous manipulations to our exposure variable, outcome data, and hypothesised mechanism. First, as a result of misreading all three places in our paper where we describe our methods, they miscalculate our main explanatory variable, creating errors in the timing of exposure and misclassifying 36% of the cases of mass privatisation (overall 56·3% misclassification), thus failing to test our hypothesis. Second, without a biological justification and neglecting evidence from this region that mortality peaks around the time of unemployment,2Perlman F Bobak M Assessing the contribution of unstable employment to mortality in posttransition Russia: prospective individual-level analyses from the Russian Longitudinal Monitoring Survey.Am J Public Health. 2009; 99: 1818-1825Crossref PubMed Scopus (44) Google Scholar they dismiss the possibility of contemporaneous effects of mass privatisation on health. Indeed, given evidence that workers' stress rose in anticipation of privatisation, adverse causal effects could have occurred in the period before privatisation.3Aslund Å How Russia became a market economy. Brookings Institution, Washington, DC1995Google Scholar, 4Lieberman I Kopf DJ Privatization in transition economies: the ongoing story. Emerald Group, Bingley2007Google Scholar, 5Kokh A The selling of the Soviet Empire. SPI Books, New York1998Google Scholar Nonetheless, Earle and Gehlbach use an ad hoc specification of lagged effects, a method which, according to one statistical textbook, “opens the researcher to the charge of data mining”.6Gujarati D Basic econometrics. McGraw-Hill, New York1995Google Scholar Third, in a situation where mortality rates were undergoing fluctuations that were unprecedented in a peacetime era,7Cornia GA Paniccia R The mortality crisis in transitional economies. Oxford University Press, Oxford2000Crossref Google Scholar they remove each country's mortality trend during the 1990s by adding 27 control variables (misreported as removing “long-term trends”), producing attenuation bias, and eliminating more than 80% of variations in mortality we sought to explain (see figure 1, webappendix). This is equivalent to removing patients with adverse cardiac outcomes from a randomised controlled trial of a cardiovascular drug without justification. Gerry and colleagues, in addition to committing mistakes similar to those of Earle and Gehlbach, set up a straw man by falsely attributing to us the view that mass privatisation was the only cause of fluctuating mortality when it clearly was not (as shown in more than 180 papers on post-communist mortality we have collectively coauthored). First, our model explains the three cases they cite (Armenia, Georgia, and Czech Republic) for which there is an inconsistency in the timing of mortality peaks: Armenia was subject to a blockade creating widespread shortages of food and energy during its war with Azerbaijan; Georgia was struggling with a large influx of refugees and economic chaos after the conflict in South Ossetia; and the Czech Republic had very high levels of social capital mitigating rises in mortality from mass privatisation. Second, they incorrectly state that Russia's mass privatisation programme was “announced in December, 1992, and completed in January, 1994.” Those responsible for implementing the programme state, “Vouchers were issued in September 1992” and that the programme officially began in October, 1992, and was completed by July, 1994. Furthermore, many contemporaneous commentators describe the anxieties provoked in the preceding months by preparation for voucher issue.3Aslund Å How Russia became a market economy. Brookings Institution, Washington, DC1995Google Scholar, 4Lieberman I Kopf DJ Privatization in transition economies: the ongoing story. Emerald Group, Bingley2007Google Scholar, 5Kokh A The selling of the Soviet Empire. SPI Books, New York1998Google Scholar Third, in a region where the mortality fluctuations were driven predominantly by alcohol, they adjust for the preceding year's mortality experience as a potential source of confounding, which they implausibly suggest might result from “disease stemming from some past exposure to pollution.” They neglect to point out how this manipulation, in combination with removing our hypothesised mechanism, introduces spurious second-order serial correlation and also makes well established relationships such as war as a cause of mortality disappear. This indicates a serious error in their model. In our webappendix we detail many additional examples where Gerry and colleagues misrepresent our paper, fail to follow statistical reporting and analytical conventions, misreport their own findings, selectively emphasise findings that support their arguments (while ignoring those estimates that corroborate our findings), make factual errors (such as suggesting we did not split our sample into countries that did or did not belong to the Soviet Union when we actually did), and scatter pejorative asides in our direction (accusing us of failing to include pre-1993 data from Slovakia, a country that only came into existence that year). We show our basic findings are robust to all of their criticisms (including controlling for long-term trends and lagged effects). We retain confidence in our original findings that rapid privatisation had a significant role in the short-term rises in mortality among working-age men seen in this region. MM has acted as an adviser to the World Bank, WHO, and European Bank for Reconstruction and Development, advising on health policy in several former communist countries. Download .pdf (.38 MB) Help with pdf files Did mass privatisation really increase post-communist mortality?David Stuckler and colleagues1 have asserted that “mass privatisation programmes were associated with a short-term increase in mortality rates in working-aged men”. We examined their data carefully and explored the assumptions and intuition from which their claim stems. We demonstrate that the findings and methods of Stuckler and colleagues are erroneous. Full-Text PDF Did mass privatisation really increase post-communist mortality?David Stuckler and colleagues1 claim that mass privatisation of enterprises was “a crucial determinant of differences in adult mortality trends in post-communist countries”. We attempted to replicate their results and found that the relationship is not robust. Here we summarise our findings, which are expanded in a webappendix. Because Stuckler and colleagues do not find a positive correlation between privatisation and mortality in central and eastern Europe, but only in the former Soviet Union, we focus on the latter set of countries. Full-Text PDF