Advertisement
Research Article

International Monetary Fund Programs and Tuberculosis Outcomes in Post-Communist Countries

  • David Stuckler mail,

    To whom correspondence should be addressed. E-mail: ds450@cam. ac.uk; after October 2008, david.stuckler@aya.yale.edu

    Affiliation: Faculty of Social and Political Sciences, University of Cambridge and King's College, Cambridge, United Kingdom

    X
  • Lawrence P King,

    Affiliation: Faculty of Social and Political Sciences, University of Cambridge and Emmanuel College, Cambridge, United Kingdom

    X
  • Sanjay Basu

    Affiliation: Department of Epidemiology and Public Health, Yale University, New Haven, Connecticut, United States of America

    X
  • Published: July 22, 2008
  • DOI: 10.1371/journal.pmed.0050143

Reader Comments (2)

Post a new comment on this article

A reanalysis of the effects of IMF programs on tuberculosis mortality

Posted by plosmedicine on 31 Mar 2009 at 00:34 GMT

Author: Ban Cheah
Position: Analyst/Programmer
Institution: Westat
E-mail: cheah234@yahoo.com
Additional Authors: None
Submitted Date: February 06, 2009
Published Date: February 9, 2009
This comment was originally posted as a “Reader Response” on the publication date indicated above. All Reader Responses are now available as comments.

The panel data estimates of the authors of the study "International Monetary Fund Programs and Tuberculosis Outcomes in Post-Communist Countries" do not include a correction for possible serial correlation in the errors. Once this serial correlation is taken into account it is likely that the estimate of the effect of IMF programs in the panel data regressions would no longer be statistically significant.

I have tried to replicate the study using two different corrections to the standard errors: 1) Arellano-Bond estimates 2) multilevel models, and have found that the coefficient on IMF Programs is no longer statistically significant.

Details of my replication study can be found here:
http://sites.google.com/s...

Competing interests declared: I do not have any competing interests to declare.

RE: A reanalysis of the effects of IMF programs on tuberculosis mortality

plosmedicine replied to plosmedicine on 03 Apr 2009 at 13:45 GMT

Author of comment: David Stuckler
Institution: Oxford University, Christ Church; Research Fellow, Department of Sociology
Email: david.stuckler@chch.ox.ac.uk

This comment was emailed to the PLoS Medicine staff on 23 March 2009.

We are pleased that Dr. Cheah has replicated some of our findings using different data and modelling approaches, but unfortunately he makes a few mistakes in his analysis. First, he misreads our paper and suggests we did not address the issue of serial correlation, but actually we clustered our standard errors as described in the paper, which is the standard approach for dealing with serial correlation (“xtreg ,cluster(countryid)”, see tables 1-4, also Wooldridge 2001). However, we further applied a series of alternative approaches as robustness checks, such as including a lagged dependent variable or correcting for the serial correlation explicitly (i.e., estimating rho and using GLS approaches), as we discussed in the paper. Second, as is well-known, the issue of serial correlation only affects the standard errors, not the size of the estimated coefficients in the standard model (Wooldridge 2002). Hence the fact that his adjustments for serial correlation led to significant attenuation in the coefficients indicates an error in his analysis. In particular, one of the statistical estimators he uses, xtabond, has a severe finite sample bias and poor precision (Arellano and Bond 1991), which could significantly dilute effect sizes and artificially inflate standard errors. It is noteworthy that none of the coefficients is significant in his model, including the established associations between GDP and tuberculosis incidence in this region (WHO 2008). Others have noted that some autocorrelation corrections can have “dire consequences for the reliability of inference based on such models” (McGuirk and Spanos 2002) and offer “a simple message to autocorrelation correctors: Don’t” (Mizon 1995). In sum, we applied the best available modelling approaches to these issues, and our findings are quite robust to the possibility of serial correlation.

References

1. Wooldridge, J. 2005. Introductory Econometrics: A Modern Approach. 3rd edition. Southwestern College Publishers: U.S.

2. Wooldridge, J. 2002. Econometric Analysis of Cross-Sectional and Panel Data. MIT Press.

3. Arellano, M and Bond, SR. 1991. Some tests of specification for panel data:
Monte Carlo evidence and an application to employment equations, Review of Economic Studies v58: 277-297.

4. Mizon, GE. 1995. A simple message for autocorrelation correctors: Don’t. Journal of Econometrics v69: 267-88.

5. McGuirk and Spanos 2002. The linear regression model with autocorrelated errors: just say no to error autocorrelation. Paper presented at the American Agricultural Economics Association.

6. World Health Organization (WHO). 2008. Global tuberculosis control— surveillance, planning, financing. Available: http://www.who.int/tb/pub.... Accessed 31 March 2008.

No competing interests declared.

RE: A reanalysis of the effects of IMF programs on tuberculosis mortality

plosmedicine replied to plosmedicine on 03 Apr 2009 at 13:48 GMT

Author of comment: David Stuckler
Institution: Oxford University, Christ Church; Research Fellow, Department of Sociology
Email: david.stuckler@chch.ox.ac.uk

This comment was emailed to the PLoS Medicine staff on 23 March 2009.

We are pleased that Dr. Cheah has replicated some of our findings using different data and modelling approaches, but unfortunately he makes a few mistakes in his analysis. First, he misreads our paper and suggests we did not address the issue of serial correlation, but actually we clustered our standard errors as described in the paper, which is the standard approach for dealing with serial correlation (“xtreg ,cluster(countryid)”, see tables 1-4, also Wooldridge 2001). However, we further applied a series of alternative approaches as robustness checks, such as including a lagged dependent variable or correcting for the serial correlation explicitly (i.e., estimating rho and using GLS approaches), as we discussed in the paper. Second, as is well-known, the issue of serial correlation only affects the standard errors, not the size of the estimated coefficients in the standard model (Wooldridge 2002). Hence the fact that his adjustments for serial correlation led to significant attenuation in the coefficients indicates an error in his analysis. In particular, one of the statistical estimators he uses, xtabond, has a severe finite sample bias and poor precision (Arellano and Bond 1991), which could significantly dilute effect sizes and artificially inflate standard errors. It is noteworthy that none of the coefficients is significant in his model, including the established associations between GDP and tuberculosis incidence in this region (WHO 2008). Others have noted that some autocorrelation corrections can have “dire consequences for the reliability of inference based on such models” (McGuirk and Spanos 2002) and offer “a simple message to autocorrelation correctors: Don’t” (Mizon 1995). In sum, we applied the best available modelling approaches to these issues, and our findings are quite robust to the possibility of serial correlation.

References

1. Wooldridge, J. 2005. Introductory Econometrics: A Modern Approach. 3rd edition. Southwestern College Publishers: U.S.

2. Wooldridge, J. 2002. Econometric Analysis of Cross-Sectional and Panel Data. MIT Press.

3. Arellano, M and Bond, SR. 1991. Some tests of specification for panel data:
Monte Carlo evidence and an application to employment equations, Review of Economic Studies v58: 277-297.

4. Mizon, GE. 1995. A simple message for autocorrelation correctors: Don’t. Journal of Econometrics v69: 267-88.

5. McGuirk and Spanos 2002. The linear regression model with autocorrelated errors: just say no to error autocorrelation. Paper presented at the American Agricultural Economics Association.

6. World Health Organization (WHO). 2008. Global tuberculosis control— surveillance, planning, financing. Available: http://www.who.int/tb/pub.... Accessed 31 March 2008.

No competing interests declared.