Growth regressions and data revisions in Penn World Tables

Paul Atherton, Simon Appleton, Michael Bleaney

Research output: Journal PublicationArticlepeer-review

3 Citations (Scopus)


Purpose: Penn World Tables (PWT) data on output measured at international prices are the data most frequently used in cross-country growth regressions. These data are subject to revision, and the amendments can be substantial for a minority of countries, although negligible for most. The purpose of this paper is to investigate the effect of data revisions on research results using the data. Design/methodology/approach: Using Hanushek and Kimko's analysis of the relationship between growth and schooling quality and Sala-i-Martin's tests of model selection, the authors investigate how much the results of cross-country growth regressions vary if the most recent vintage (6.2) of PWT data is used, rather than the previous vintage (6.1). Findings: The variation is substantial enough to imply significant differences in research results using different vintages of the PWT data. Practical implications: The results reinforce the case for examining the sensitivity of growth regressions to outliers, which may be subject to subsequent data revision that might substantially affect the conclusions. Originality/value: Previous research has identified significant revisions between successive vintages of PWT growth data, but has implied that this is not likely to affect the results of cross-country growth regressions based on long-run averages rather than on annual data. The findings suggest that this is not necessarily the case.

Original languageEnglish
Pages (from-to)301-312
Number of pages12
JournalJournal of Economic Studies
Issue number3
Publication statusPublished - Aug 2011
Externally publishedYes


  • Data revisions
  • Economic growth
  • Education
  • Growth
  • Model selection

ASJC Scopus subject areas

  • Economics, Econometrics and Finance (all)


Dive into the research topics of 'Growth regressions and data revisions in Penn World Tables'. Together they form a unique fingerprint.

Cite this