Promise and Pitfalls in the Use of "Secondary" Data-Sets: Income Inequality in OECD Countries As a Case Study

TLDR: This paper examined the role of secondary data sets in empirical economic research, taking the field of income distribution as a case study, and illustrated problems faced by users of "secondary" statistics, showing how both cross-country comparisons and time-series analysis can depend sensitively on the choice of data.

Abstract: This paper examines the role of secondary data-sets in empirical economic research, taking the field of income distribution as a case study. We illustrate problems faced by users of "secondary" statistics, showing how both cross-country comparisons and time-series analysis can depend sensitively on the choice of data. After describing the genealogy of secondary data-sets on income inequality, we consider the main methodological issues and discuss their implications for comparisons of income inequality across OECD countries and over time. The lessons to be drawn for the construction and use of secondary data-sets are summarized at the end of the paper.

Figures

Table 8), for 1929, 1935/36 and 1941, Fitzwilliams (1964, p. 5, Table 4) for 1944-1961, and Radner and Hinrichs (1974, p. 21, Table 3) for 1964, 1970, 1971; data for 1962 and 1963 ignored because based on extrapolations. – CPS families: Bureau of the Census (1999, Table F-4): total gross money incomes, for families; some minor breaks in underlying data are ignored, but a major break in 1993 is shown. – CPS families and unrelated individuals: Bureau of the Census (1998, p. B6, Table B3): total gross money incomes, for families and unrelated individuals; some minor breaks in the underlying data are ignored, but a major break in 1993 is shown.

Figure 4

Figure 6

Table 5): elaboration on data from EVS for household equivalent disposable incomes, weighted by persons (excluding foreign family heads); OECD equivalence scale. – GSOEP Panel: Hauser (1996, Table 1): elaboration on data from GSOEP for household equivalent disposable incomes, weighted by persons (excluding foreign family heads); OECD equivalence scale.

Figure 2

Figure 3

Full text

Temididiscussione
del Servizio Studi
Promise and Pitfalls in the Use of “Secondary” Data-Sets:
Income Inequality in OECD Countries
by A. B. Atkinson and A. Brandolini
Number 379 - October 2000

Thepurposeofthe“Temididiscussione”seriesistopromote the circulationofworking
papers prepared within the Bank of Italy or presented in Bank seminars by outside
economists with the aim of stimulating comments and suggestions.
The views expressed in the articles are those of the authors and do not involve the
responsibility of the Bank.
Editorial Board:
A
NDREA BRANDOLINI,FABRIZIO BALASSONE, MATTEO BUGAMELLI, FABIO BUSETTI, RICCARDO
CRISTADORO, LUCA DEDOLA, PATRIZIO PAGANO, PAOLO ZAFFARONI; RAFFAELA BISCEGLIA
(Editorial Assistant).

PROMISE AND PITFALLS IN THE USE OF “SECONDARY” DATA-SETS:
INCOME INEQUALITY IN OECD COUNTRIES
by Anthony B. Atkinson
*
and Andrea Brandolini
**
Abstract
Secondary data-sets have come to play an increasing role in empirical economic research.
This paper examines the major new secondary data-set assembled by Klaus Deininger and Lyn
Squire (DS) at the World Bank. We concentrate on its coverage of the OECD countries. We have
particularly in mind the user of income inequality statistics who does not wish to go back to the
original data. In order to motivate the analysis, we first present two examples of the problems
which may arise, showing how both cross-country comparisons and time-series analysis may
depend sensitively on the choice of data. Section 3 of the paper sets the DS data-set in the
historical context of earlier exercises in assembling comparative information on income inequality.
In Section 4, we consider the methodological issues which arise in the use of income distribution
data and their relation to the different sources of evidence. In Section 5, we discuss their
implications for the comparison of income inequality across OECD countries, and the use of
dummy variables to allow for definitional and data differences. Section 6 is concerned with
changes in income inequality over time, and the establishment of consistent series for individual
countries. The lessons to be drawn for use of secondary data-sets in the field of income distribution
are summarised at the end of the paper.
JEL classification: D31, C80.
Keywords: personal income distribution, secondary data-sets.
*
Nuffield College, Oxford.
**
Bank of Italy, Research Department.

Contents
1. Introduction.....................................................................................................................9
2. Two case studies............................................................................................................13
2.1 Comparison of income inequality in OECD countries...............................................13
2.2 Trends over time in the Netherlands.........................................................................17
3. The Deininger-Squire data-set in historical perspective ...................................................19
4. A bewildering variety of estimates..................................................................................23
4.1 Definitions...............................................................................................................23
4.2 Sources ...................................................................................................................25
4.3 Processing...............................................................................................................26
4.4 Conclusion ..............................................................................................................28
5. Dealing with data differences..........................................................................................28
6. Changes in income inequality over time ..........................................................................32
6.1 United Kingdom......................................................................................................33
6.2 United States...........................................................................................................35
6.3 Canada....................................................................................................................37
6.4 France.....................................................................................................................39
6.5 West Germany.........................................................................................................40
6.6 Conclusions.............................................................................................................42
7. Overall conclusions ........................................................................................................43
Appendix: Time series for Gini coefficients in selected countries .........................................46
References..........................................................................................................................50

1. Introduction
1
Secondary cross-national data-sets have come to play an increasing role in empirical
economic research. The past decade has seen, for instance, widespread use of the
international national accounts data assembled by Summers and Heston (1991). Here our
focus is on the major new secondary data-set on income inequality assembled by Deininger
and Squire (1996) at the World Bank. Described as “the largest possible”, the Deininger and
Squire data-set draws together more than 2,600 observations on Gini coefficients and, in
many cases, quintile shares from a wide variety of studies covering 135 developed and
developing countries for the years 1947-1994. The statistics were selected by requiring that
they be from national household surveys for expenditure or income, that they be
representative of the national population, and that all sources of income or expenditure be
accounted for, including own-consumption. Further, Deininger and Squire identify in their
data-set a “high quality” subset of nearly 700 observations for 115 countries, not more than
one per country per year, which they label “accept” for the guidance of users.
The construction of this data-set by Deininger and Squire (referred to below as the DS
data-set) is a remarkable achievement, and they deserve full praise for allowing all interested
researchers free and easy access to the data.
2
It has been already used, just to give a few
examples, by Deininger and Squire themselves (1998) to test the hypothesis of Kuznets
(1955) on the relationship between inequality and growth, by Bénabou (1996) to study the
convergence of inequality across countries, by Vanhoudt (1997) to assess the effect on
inequality of aggregate economic variables and labour market policies, by Checchi (1998) to
examine the relationship between inequality in incomes and inequality in educational
1
We are very grateful for their most helpful comments on the first version to Francois Bourguignon,
Sam Bowles, Andrea Cornia, Klaus Deininger, Karen Gardiner, Raffaela Giordano, Giorgio Gobbi, Richard
Hauser, John Hills, Sampsa Kiiski, John Micklewright, Rosa Mulè, Brian Nolan, Thomas Piketty, Tim
Smeeding and Lyn Squire. We thank Roland Bénabou and Christina Romer for supplying copies of the data
they used, and Hans de Kleijn of CBS (Statistics Netherlands) for supplying data for the Netherlands. The
views expressed here are solely those of the authors; in particular, they do not necessarily reflect those of the
Bank of Italy.
2
The data-set is available from World Bank’s web-site, at the address: http://www.worldbank.org/html/
prdmg/grthweb/dddeisqu.htm.

Frequently asked questions

Q1. What are the contributions in this paper?

This paper examines the major new secondary data-set assembled by Klaus Deininger and Lyn Squire ( DS ) at the World Bank. In order to motivate the analysis, the authors first present two examples of the problems which may arise, showing how both cross-country comparisons and time-series analysis may depend sensitively on the choice of data. Section 3 of the paper sets the DS data-set in the historical context of earlier exercises in assembling comparative information on income inequality. In Section 4, the authors consider the methodological issues which arise in the use of income distribution data and their relation to the different sources of evidence. In Section 5, the authors discuss their implications for the comparison of income inequality across OECD countries, and the use of dummy variables to allow for definitional and data differences. The lessons to be drawn for use of secondary data-sets in the field of income distribution are summarised at the end of the paper. 

Q2. What is the effect of removing the dummy variable?

But with the Gottschalk and Smeeding data10 in place of the DS data, the inflation variable ceases to be significantly different from zero, and the R2 drops to 0.11 (column 5).11 (The Gottschalk and Smeeding data relate to disposable income, so that no dummy variable is included for the gross/net distinction.) 

Q3. How did they compute the Gini coefficients?

In practice, they computed the Gini coefficients from parametric Lorenz curves estimated on available grouped data by using POVCAL, a programme designed at the World Bank. 

Q4. What is the role of the secondary data-set?

In their view, all sources are imperfect, and ideally a secondary data-set should includeinformation which can be used to assess the reliability of the observations.