British Journal of Political Science

http://journals.cambridge.org/JPS

Additional services for British Journal of Political Science:

Email alerts: Click here

Subscriptions: Click here
Commercial reprints: Click here
Terms of use : Click here

Empirical Linkages Between Democracy and Economic Growth

John F. Helliwell

British Journal of Political Science / Volume 24 / Issue 02 / April 1994, pp 225 - 248
DOI: 10.1017/S0007123400009790, Published online: 03 March 2009

Link to this article: http://journals.cambridge.org/abstract_S0007123400009790

How to cite this article:

John F. Helliwell (1994). Empirical Linkages Between Democracy and Economic Growth. British Journal of Political
Science, 24, pp 225-248 doi:10.1017/S0007123400009790

Request Permissions : Click here

Downloaded from http://journals.cambridge.org/JPS, IP address: 130.216.129.208 on 23 Apr 2015

BJ.Pol.S. 24, 225-248 Copyright © 1994 Cambridge University Press
Printed in Great Britain

Empirical Linkages Between Democracy and

Economic Growth
JOHNF.HELLIWELL*

Using cross-sectional and pooled data for up to 125 countries over the period from 1960 to 1985,
this article evaluates the two-way linkages between democracy and economic growth. The
effects of income on democracy are found to be robust and positive. The effects of several
measures of democracy and personal freedoms on growth are assessed in a comparative growth
framework in which growth of GDP per adult depends negatively on initial income levels, as
implied by the convergence hypothesis, and positively on rates of investment in physical and
human capital. Adjusting for the simultaneous determination of income and democracy makes
the estimated partial effect of democracy on subsequent economic growth negative but in-
significant. This nonsignificant negative effect is in any case counterbalanced by the positive
indirect effect that democracy exerts on growth via education and investment. The general result
of the growth analysis is that it is still not possible to identify any systematic net effects of
democracy on subsequent economic growth.

There is a longstanding presumption that democracy takes root and survives

where levels of economic development and education are high.1 There is more
recent literature dealing with the question of whether democracy is a luxury
that must be paid for in terms of subsequently slower increases in living
standards.2 Where linkages between democracy and economic growth have
been established, they more often than not support the notion that democracy
does pose a cost in terms of subsequent reductions in growth rates. As a third
wave of democracy has spread in the 1980s, particularly in eastern Europe, a
systematic assessment of available global evidence seems well worth while. Is
it better that economic development and reform take precedence over the
*
* Department of Economics, University of British Columbia, Harvard University, and National
Bureau of Economic Research. I am grateful for inspiration, advice and helpful comments from
Alberto Alesina, Andre" Blais, Michael Bruno, Dick Cooper, Marc Gaudry, Sam Huntington, Andy
Moravcsik, Bob Putnam, David Sanders, Jim Stock, Hugh Ward and anonymous referees. I am also
grateful for the research support of the Social Sciences and Humanities Research Council of Canada,
and of the Harvard Center for International Affairs.
1
The earlier evidence is surveyed in Seymour Martin Lipset, 'Some Social Requisites of Democ-
racy: Economic Development and Political Legitimacy', American Political Science Review, 53
(1959), 69-105.
2
Many empirical studies of the effects of democracy on economic growth are reviewed in L.
Sirowy and A. Inkeles, 'The Effects of Democracy on Economic Growth and Inequality: A Review',
Studies in Comparative International Development, 25 (1990), 126-57. The extent to which the
results depend on country-specific factors is emphasized in S. Haggard, Pathways from the Peri-
phery: Tlte Politics of Growth in the Newly Industrializing Countries (Ithaca, NY: Cornell
University Press, 1990), chap. 10.
226 HELLIWELL

spread of democracy (as in China) or that democracy should precede economic

reforms (as in most of Central and Eastern Europe), or are the two
types of change independent? There will always be limitations on the
applicability of previous experience to current problems, but the issues
raised are of sufficient importance, and the range of previous experience
broad enough, to encourage an attempt to review the current state of the
global evidence.
This article takes advantage of the increasing availability of comparable data
for economic growth and income levels in most of the world's national econ-
omies, and combines it with some of the available data categorizing political
rights and civil liberties in an equally large number of economies. In order to
examine the possible effect of the level of economic development on political
democracy, a sample of 125 countries is used. This sample represents the
largest number of countries for which it is possible to obtain comparable
measures of per capita real incomes3 and regular assessments of the extent of
political and civil rights. These initial results are then confirmed using other
measures of democracy and more complex estimation techniques for a smaller
sample of countries. In order to assess the effects of democracy on subsequent
economic growth, the sample is reduced to ninety-eight countries and a model
of comparative growth is estimated for the period 1960 to 1985. The sample is
reduced for other tests which make use of earlier estimates of the relative status
of political democracy among about ninety countries in 1960 and 1965. The
fact that the results using different sample sizes are very similar suggests that
the use of smaller numbers of countries, where necessary, does not pose major
estimation problems.
The empirical analysis is divided into two sections. To study the effects of
income levels on the character of the political system, annual indices of politi-
cal and civil liberties for the years 1976-85 are regressed on logarithmic levels
of real GDP per capita. Supplementary tests for the importance of schooling
levels are also carried out, since education is frequently considered to play an
important role in enabling the choice of a democratic form of government.
These results are shown in Tables 1 and 2, and are described in Section I. The
main analytical tool used to assess the effects of democracy on economic
growth is an empirical framework which examines comparative growth per-
formance over the 1960-85 period in a way that allows simultaneously for (a)
convergence in the rates of growth of per capita GDP, (b) possible returns to
scale, and (c) international differences in investment rates in human and physi-
cal capital. By adding measures of democracy and political freedoms, it is then
possible to assess the extent to which the political system has any systematic
3
The income measure used is real gross domestic product (GDP) per capita, converted at
purchasing power parity exchange rates, using data compiled by national statistical agencies, with
the collaboration of the United Nations and the OECD. The features of the Mark V release of the
data used in Tables 1 and 2 are described in R. Summers and A. Heston, "The Penn World Table
(Mark 5): An Expanded Set of International Comparisons, 1950 to 1988', Quarterly Journal of
Economics, 106 (1991), 327-68.
 Linkages Between Democracy and Economic Growth 227

influence on current and subsequent growth performance. This is undertaken in

Section II, and in Tables 3 and 4.

I. ARE THERE ECONOMIC PREREQUISITES TO DEMOCRACY?

Introducing his seminal cross-national study, Seymour Lipset observed 'per-
haps the most widespread generalization linking political systems to other
aspects of society has been that democracy is related to the state of economic
development.'4 For a sample of forty-eight countries, Lipset examined the
cross-sectional correlations between regime type and mid-century measures of
economic development. He attempted to avoid the complications of mixing
political cultures by dividing his sample into two main groups: twenty-eight
European and English-speaking countries and twenty Latin American coun-
tries. Within the first group, he found that the average per capita income was
more than twice as high in the thirteen stable democracies than in the fifteen
unstable democracies and dictatorships. Among the Latin American countries,
all had average incomes less than any of the countries in the first sample, but
the seven democracies or unstable dictatorships had average incomes about 40
per cent higher than the thirteen stable dictatorships (with substantial income
overlap in these two categories). Qualitatively similar results were obtained
using measures of industrialization, education and urbanization as alternative
measures of economic development. There were two key problems with Lip-
set's analysis, however. First, because his measures of economic development
derived from the period after that used to classify political regimes, Lipset's
correlations were unable to establish whether it was economic development
which led to democracy or vice versa. Secondly, the Second World War, while
no doubt to some extent a consequence of the lack of democracy in many of the
European countries, also led to destruction of their economic capacities, hence
providing a correlation between low postwar incomes and low prewar levels of
democracy that could not be used directly to support the conclusion that low
levels of income lead to low levels of democracy.
Another thirty years of history, better measures of comparative real incomes,
more regular and systematic measures of democracy, and a much larger sample
of countries available for analysis suggest that it is now appropriate to revisit
these issues. The analysis conducted here is based on measures of economic
development and democracy covering 125 countries for each year from 1976
through 1985 - a total of 1,250 observations. Economic development is meas-
ured by average per capita real income. The measure of democracy is obtained
by transforming measures of political rights and civil liberties published annu-
ally by Gastil.5 The index takes the value of zero for a country with no political

4
Lipset, 'Some Social Requisites of Democracy', p. 75.
3
The data are described in R. D. Gastil, 'The Comparative Survey of Freedom: Experiences and
Suggestions', Studies in Comparative International Development, 25 (1990), 25-50. Gastil's separ-
ate indices for political rights and civil liberties are each on a scale from 1 to 7, with 1 representing
228 HELLIWELL

and civil rights, and the value 1.0 for a country with full measures of both types
of rights. The two components of the democracy index, political rights and civil
liberties, are highly correlated with one another (r = 0.93), and give indis-
tinguishable results if used separately in the regressions reported below, so at
this stage it will be sufficient to use the combined index.
Table 1 shows the results of several regressions explaining variations in
political freedoms, among countries and over time, by real per capita GDP
converted at purchasing power parity exchange rates. Initially, separate regres-
sions were run for each year, but since the coefficients were insignificantly
different from year to year, it was possible to increase the efficiency of the
estimates by stacking the observations to form a single sample of 1,250 obser-
vations. Equation 1 shows the simplest form of the regression, in which about
42 per cent of the variance among countries in the freedom index is explained
by variations in per capita incomes. The coefficient of 0.2 on the logarithm of
per capita GDP suggests that a 10 per cent increase in per capita income raises
the predicted value of the democracy index by 2 points on a 100-point scale.
The notion that different cultures may give rise to sharply different degrees of
democracy, even at equivalent levels of income, is tested in Equation 2, using
a series of 'regional' dummy variables. The Equation 2 results show that, even
after adjusting for the effects of different levels of per capita income, the degree
of democracy is sharply higher in OECD countries, sharply lower in six oil-
dependent countries of the Middle East,6 slightly lower in Africa and slightly
higher in Latin America, with the base of comparison being the remaining
countries, mainly in Eastern Europe and Asia. These geographic/cultural dif-
ferences raise to 63 per cent the fraction of variation explained by the equation.
The estimated effect of income drops from 0.20 to 0.12 when the regional
variables are added, reflecting the strong correlation between regions and aver-
age incomes, with the OECD countries in particular being both richer and more
democratic. However, since the income effect remains very significant, the
equation still shows that the strong correlation between democracy and income
is not simply due to the fact that the OECD countries are richer and more
democratic than most other countries.
Tests were also conducted to see if the effect of income on democracy varied
by region or by level of income in some way more complex than that captured
by Equations 1 and 2. These tests involved testing for slope coefficients that
differed by income or region, and testing the log-linear model against quadratic
and cubic functions of income. Only in the case of the OECD region did the

(F'note continued)
the highest levels of rights, and 7 the lowest. Summing the two indices, as Gastil does in his more
recent work, gives a measure that takes the value 2 for the most democratic and 14 for the least
democratic systems. This is linearly transformed to make an index for the probability of freedom
(PFR), ranging from 0 for no freedoms to 1.0 for fully democratic systems. If FR is the 2 to 14 index,
then PFR = (14 - FR)/12.
6
These are Bahrain, the United Arab Emirates, Iran, Iraq, Kuwait and Saudi Arabia.
TABLE 1 Effects of Income and Education on Democracy, 1976-85
Equations

0) (2) (3) (4) (5) (6)

No. of observations 1,250 1,250 980 980 10 X 125 10X98
Estimation OLS OLS OLS OLS Iterative Iterative
method stacked stacked stacked stacked Zellner Zellner
Constant - 1.132 - 0.508 - 1.11 -0.344 - 0.323 - 0.808
(20.46) (6.46) (13.46) (4.05) (2.30) (5.30)
Coefficients Co
inGDP 0.205 0.122 0.194 0.097 0.098 0.145
(28.08) (12.47) (14.87) (7.93) (5.55) (6.29)
OECD 0.334 0.314 0.366
(16.22) (13.07) (6.94)
MEOIL - 0.370 - 0.356
(14.86) (4.47)
Africa - 0.100 -0.114 - 0.134
(4.91) (5.39) (3.11)
Latin America 0.079 0.055 0.123 a
(3.46) (2.54) (2.53)
Secondary school 1.857 0.884 3.08
5.
a
(4.46) (2.52) (3.82) 81
R2 0.419 0.634 0.579 0.660 0.57-0.68 0.52-0.61
D.W. 0.21 0.29 0.26 0.29 1.66-2.18 1.6-2.0
S.E.E. 0.256 0.203 0.221 0.196 0.19-0.21 0.21-0.23
Notes: Absolute values of / statistics arc in parentheses. For Equations 1 to 4, these are estimated using H. White's heteroscedasticity-consistent estimator. The six Middle 3
East oil exporters are not in the 98-country sample, and hence MEOIL docs not appear in Equations 3 , 4 and 6. The ranges for statistics below Equations 5 and 6 reflect
variations among the ten cross-sectional equations. Equations 5 and 6 arc estimated using Zellner's SUR method, and all estimation uses K. White's SHAZAM programme.
Sec H. White, 'A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity', Econometrica. 48 (1980), 149-70; A. Zellner, 'An
Efficient Method for Estimating Seemingly Unrelated Regressions and Tests for Aggregation Bias', Econometrica, 30 (1960), 54-78; and K. J. White. SHAZAM User's
Reference Manual 7.0 (Toronto and London: McGraw Hill, 1993). to
230 HELLIWELL

data suggest that a separate slope coefficient should be included in the model
rather than the separate intercept shown in Equation 2 - but the difference was
not significant.7 The tests for non-linearity did not reveal anything preferable
to the log-linear form, thus casting some doubt on the notion of there being a
threshold level of income above which there is a sharp increase in the probabil-
ity of a country being democratic.8
Equations 3 and 4 show the effect of adding the secondary school enrolment
rate, measured as a fraction of the adult population, to Equations 1 and 2
(because of missing data the number of cases is reduced to 98). Dahl and others
have hypothesized that the apparent link between economic development and
democracy may be partly explained by the fact that literacy, education and
communication all increase the effective demand for democracy.9 The
significant additional effect of schooling in Equations 3 and 4 shows that
education and economic development are distinct enough that their separate
contributions to the attainment of democracy can be estimated. In Equation 3,
the coefficient on the education variable suggests that an increase of 1 per cent
in the percentage of the working age population that is in secondary school
raises the democracy index by 1.85 points (if the index is measured on a scale
of 100). The effect drops to 0.88 when the regional effects are allowed for in
Equation 4, reflecting the large differences among the regions in their rates of
schooling, with the less democratic continents also having lower average
schooling.
Although the stacking of the 1,250 observations in principle provides a much
larger sample, and hence more powerful estimates of the effects, the year-
to-year changes in the democracy index and in relative incomes are small
compared to the differences among countries. Thus there is a strong year-
to-year correlation of the error terms for each country, which shows up as a

7
The R2 rises from 0.634 to 0.635 if the OECD has a separate slope rather than intercept in
Equation 2. The coefficient on lnGDP falls from 0.122 to 0.120, with an additional income effect
of 0.037 for the OECD countries.
8
If the level and the square of per capita GDPs arc used as explanatory variables, both are strongly
significant, in a 125-country cross-sectional equation for the 1985 Gastil index, with the coefficients
being positive for the level and negative for the quadratic term. However, if an artificial encompass-
ing model is set up, containing these two variables and the logarithm of real per capita GDP (the
variable that is used in the equations reported in Table 1), statistical tests show that there is no
significant difference in explanatory power between the quadratic and the logarithmic models, with
the data preferring the logarithmic model to the quadratic model. The P-value of the Wald chi-
square test for excluding the two variables of the quadratic model is 0.25, while it is 0.21 for
excluding the logarithmic variable, with the tests in both cases being of restricted versions of the
encompassing equation. If the equation is augmented by adding the separate intercept for the Middle
East oil producers, the preference for the logarithmic over the quadratic form is much stronger, with
a P-value for the Wald test of excluding the quadratic variables being 0.43, compared to 0.03 for
excluding the logarithmic variable. Since the peak of the quadratic is very near the top of the range
of GDP per capita, at the levels of Sweden and Australia, the exponential and quadratic forms give
very similar predicted values for most countries.
9
R. A. Dahl, Polyarchy: Participation and Opposition (New Haven, Conn.: Yale University
Press, 1971), pp. 74-5.
 Linkages Between Democracy and Economic Growth 231

very low Durbin-Watson statistic, if the sample is stacked with the ten
observations for each country grouped together, as is the case for Equations 1
to 4. The result of this is that the sample is not really as big as it appears to be,
and the significance of the coefficients is overstated. This can be rectified by
re-estimating the equations with the sample of 1,250 split into ten equations of
125 observations each, with coefficients constrained to be the same in each
equation, and estimated by an interactive procedure that takes into account the
loss of information implied by the errors being very similar in each of the ten
annual equations. Equations 2 and 3, the equations for the large sample with
regional effects and for the smaller sample with GDP and schooling, are esti-
mated by this method and shown as Equations 5 and 6. The /-statistics are
substantially reduced. The GDP and regional effects remain highly significant
in Equation 5, as do the GDP and schooling effects in Equation 6. Equations
5 and 6 show that the regional factors together contribute more than schooling,
but that schooling goes some distance in explaining the variations in democ-
racy not captured by differences in GDP per capita.10 In addition, as will be
shown below, education plays an important role in explaining long-term GDP
growth, and hence the levels of GDP per capita, thus providing a second
channel whereby education affects democracy.
The analysis so far has made use of Gastil's index of political freedom, since
it is available on a consistent basis for many years and many countries. Bollen
has prepared alternative measures for a smaller sample of countries, and has
surveyed some of the issues involved in developing quantitative measures of
democracy.11 Bollen's index, which he has published for 1960 and 1965, is an
equally weighted sum of six component indexes, three relating to popular
sovereignty (fairness of elections, election of chief executive and election of
the legislature) and three to political freedoms (press freedom, freedom of

10
This conclusion needs to be treated with some caution, as inclusion of the OECD variable in
Equation 6 removes the significance of the education variable, evidence of the fairly high correlation
between the two variables. Thus to some extent education and the complex of factors that define
members of the OECD are competing for explanatory power, with the OECD variable adding to the
equation by slightly more than the schooling variable.
11
In surveying the issues, Bollen concludes that it is important not to confound political liberties
and political rights with political stability. He argues that the former two comprise an appropriate
measure of political democracy, while the latter is not. The Gastil measures accord with Bollen's
preferences, by focusing on political rights and freedoms rather than political stability, and in
providing measures whose changes might themselves provide an index of stability. Bollen's views
and Gastil's measures both seem to embody key features of Dahl's Polyarchy dimensions of open
competition and widespread participation, with individuals protected in their rights to express their
political opinions, and free to form parties and to compete in binding elections by unintimidated
voters. Both of Gastil's component indices seem relevant, since civil rights and political freedoms
are in many respects mutually supportive. There is no evidence that either component has more
influence than the other, as Gastil's two component indices give indistinguishable results when used
separately. See K. A. Bollen, 'Issues in the Comparative Measurement of Political Democracy',
American Sociological Review, 45 (1980), 370-90; and K. A. Bollen, 'Political Democracy:
Conceptual and Measurement Traps', Studies in Comparative International Development, 25
(1990), 7-24.
232 HELLIWELL

group opposition, and lack of government sanctions against political oppo-

sition). Since the Gastil index of political freedoms has been linearly converted
into a scale with zero representing lack of political freedoms and 1.0 full
freedoms, it should be directly comparable with the Bollen measures.12
Comparing the Bollen index for 1960 with the Gastil indices for 1976 and
1985, several differences are readily apparent. The Bollen index is unimodal,
with nineteen countries at 0.95 or above, two-thirds of the countries above
0.50, only six countries below 0.25, and none below 0.10. The Gastil index is
bimodal, with modes at both ends: sixteen countries are rated at 1.0, and
nineteen countries are below 0.10. As shown in the data appendix, the mean of
the Bollen index is 0.68, compared to 0.46 and 0.52 for the Gastil 1976 and
1985 measures. The simple correlation between the 1960 Bollen index and the
1976 Gastil index is 0.59, and the Spearman rank correlation is 0.62. The
Bollen index also appears to be more volatile over time than the Gastil index,
as the correlation between the 1960 and 1965 values, for countries that appear
in both, is lower than between any pair of the Gastil indices. Without overlap-
ping observations, it is not possible to analyse the differences further, and even
general conclusions are hard to reach, since the 1960s may well have been a
more volatile period than that from the mid-1970s to the late 1980s.13 In any
event, it is clear that the Bollen and Gastil indices are quite independent
attempts to measure fairly similar concepts of political democracy, so that any
attempts to use the Bollen data to confirm the results from the Gastil data are
likely to be useful.14
Equations 1 and 2 of Table 2 show the results of using the ninety observa-
tions for the 1960 Bollen index as the dependent variable for the re-estimation
of Equations 1 and 6 of Table 1. The coefficient of income, in Equation 1, is

12
The biggest factor limiting the comparison is that the Gastil indices do not start before the
mid-1970s, with the result that there is no overlap in the time periods covered by the Gastil and
Bollen indices. In addition, the Bollen index for 1965 is only available for ninety of the ninety-eight
countries for which full data are available from 1960 through 1985, and the 1960 index has several
fewer observations. To provide as full as possible a sample of the state of democracy for the
beginning of the growth period, 1965 values were used to fill out the 1960 sample to ninety
countries.
13
Huntington reports that the number of democratic states fell from thirty-six in 1962 to thirty
in 1973 and has sincerisenagain, tofifty-eightin 1990. Measured as percentages of the total number
of states, Huntington calculates that democracies fell from 32.4 percent in 1962 to 24.6 percent in
1973 and rose to 45 per cent in 1990. See S. Huntington, The Third Wave: Democratization in the
Late Twentieth Century (Norman: University of Oklahoma Press, 1991).
14
There are also statistical grounds for being glad to have the unimodal Bollen index available
to check the results that use the bi-modal Gastil index as a dependent variable. The Gastil index, like
many of the dichotomous measures of democracy, is not normally distributed about its mean. In
addition, both indices are bound in the range between zero and 1.0. To avoid the risks of biased
standard errors that might arise from the implied non-normality, all OLS equations estimated with
either of the democracy indices as the dependent variable makes use of H. White's procedure for
estimating standard errors consistently in the presence of heteroskedastic residuals. See H. White,
*A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedas-
ticity', Econometrica, 48 (1980), 149-70.
 Linkages Between Democracy and Economic Growth 233

slightly lower than using the Gastil index, and the equation explains 30 per cent
of the cross-sectional variance of the Bollen index. This is strongly supportive
of the results based on the Gastil index. Adding the schooling variable for 1960
(Equation 2) lowers the income coefficient and gives a strong coefficient
(r = 4.2) on the schooling variable. These results are also close to those of
Equation 6 in Table 1, which implies that the Gastil and Bollen results are
consistent with one another, bearing in mind the different scaling of the two
education variables.15
The results so far seem to indicate a fairly strong influence from GDP per
capita to the level of political rights and freedoms. But the data may be reveal-
ing correlation rather than causation. Are there ways of checking against the
possibility that something else might be determining democracy, and then
democracy is facilitating the attainment of high levels of income? Some direct
tests of the influence of democracy on economic growth are reported in the next
section. If we find there are no positive effects of democracy on economic
growth, then the effects we find here of income on democracy are not likely to
be the result of reverse causation. But it is possible to do some preliminary tests
to guard against those possibilities, using the 98-country sample for which
determinants of the rate of economic growth are available. One simple check
is to estimate the 1985 equation splitting the 1985 income variable into two
parts, the 1960 level of real GDP per adult and the change between 1960
and 1985, and then to re-estimate the equation using instrumental variables
for the determinants (other than the level of democracy) of post-1960
growth. If the positive effect we have found flowing from income to democracy
is being inflated because something else is determining democracy, and
democracy is then helping economic growth, then the application
of estimation techniques designed to remove simultaneous equation bias
in the coefficient on post-1960 growth will lead to a reduction in the estimated
effect of post-1960 income growth on democracy. If, on the other hand,
there is a negative feedback from democracy to growth, then re-estimation
should be expected to lead to an increase in the estimated effect of post-
1960 income growth on the level of democracy. Table 2 shows the results
of these tests.
To provide a starting point, Equation 3 shows the results of regressing the
1985 Gastil index on 1985 real GDP per adult for the 98-country sample.
Equation 4 shows the result of splitting the income variable into 1960 income
and subsequent growth. The coefficient on 1960 income is almost twice that on
the variable for post-1960 growth, and the tests below the equation reveal the

15
Although the estimated coefficient for schooling is much smaller for the Bollen equation, it is
necessary to bear mind that the 1960 schooling variable is measured as a fraction of population of
school age, while the education variable from the growth equation is measured as a fraction of the
adult population. When this scale adjustment is made, the coefficients are insignificantly different
from one another. If regional variables are added to Equations 1 and 2 of Table 2, they are not
significant.
 to
u>

TABLE 2 Effects of Income and Education on Democracy for I960 and 1985
x
Equations r
(1) (3) (4) (5) (6)
IWEl

No. of observations 90 90 98 98 98 98
Estimation method OLS OLS OLS OLS IV IV
Dependent variable Bollen 1960 Bollen 1960 Gastil 1985 Gastil 1985 Gastil 1985 Gastil 1985
Constant - 0.524 0.144 - 1.545 - 1.751 - 1.750 - 1.70
(3.29) (0.62) (12.31) (13.78) (9.95) (8.59)
Coefficients
InGDPa60 0.158 0.055 0.289 0.271
(7.97) (1.69) (17.04) (10.02)
Sch60 as prop of age group 0.492
(4.18)
lnGDPa85-lnGDPa60 0.148 0.345
(2.84) (3.80)
InGDPa85 0.256 0.282
(16.16) (12.98)
/-value of coefficient differences 2.38 0.72
Probability of equality (Wald) 0.017 0.412
R2 0.298 0.359 0.620 0.638 0.613 0.576
S.E.E. 0.220 0.210 0.216 0.211 0.218 0.228
Notes: Absolute values of/ statistics, in parentheses, arc for Equations 1 to 4 adjusted for hctcrosccdasticity. Durbin-Watson statistics are not reported as they have no use in cross-sectional
regressions unless the observations have been organized in groups to be tested for homogeneity. Equations 5 and 6 are estimated by instrumental variables, using 1960 values for GDP, investment,
schooling, and sample average scale and population growth as instruments. The income variable used is the logarithm of real GDP per adult. The probability tests for Equations 4 and 6 are
the probabilities of the coefficients on 1960 income and post-1960 growth being equal, using Wald's chi-square statistic.
 Linkages Between Democracy and Economic Growth 235

difference to be statistically significant, with the probability of the result being

due to chance estimated at less than 2 per cent. If this difference between the
coefficients is due to there being a negative feedback from democracy to
subsequent growth, then re-estimation by appropriate simultaneous equation
methods should move the two income coefficients closer together. Equation 6
shows that this indeed happens. When instrumental variable estimation is used
to re-estimate Equation 4, the two coefficients switch relative sizes, and the
difference between them becomes insignificant.16 We would also expect, if
there is a negative feedback from democracy to subsequent growth, that the
re-estimation of Equation 3 by instrumental variables would raise the estimated
effect of income on democracy. This turns out to be the case, as shown by
Equation 5, although the increase is not large enough to be statistically
significant.
The above tests suggest that the results reported in Tables 1 and 2, which
show a positive effect flowing from income to democracy are not due to a
positive effect that flows from democracy to growth. Since re-estimation to
avoid simultaneous equation bias raises rather than lowers the estimated effect
of income on democracy, we can be sure that the positive effect of income on
democracy is not due to positive feedbacks from democracy to economic
growth. On the contrary, since the attempts to purge the bias have led to an
increase in the estimated positive effect of income on democracy, the results
suggest that the feedback from democracy to growth may be negative, with
democracy possibly having a negative partial effect on subsequent economic
growth. The next section assesses this possibility more directly.

II. DEMOCRACY AND ECONOMIC GROWTH

Previous studies on the effects of democracy on economic growth have been

surveyed by Sirowy and Inkeles.17 They distinguish three perspectives on the
topic: a conflict perspective, in which economic growth is seen to require an
authoritarian regime to implement the kinds of policies needed to facilitate
rapid growth; a compatibility perspective\ which argues that democracies are as
capable as authoritarian regimes of combining redistribution and growth in
such a way as to broaden markets and achieve economic expansion; and a
sceptical perspective, which doubts any systematic linkage between democ-
racy and growth. Of the thirteen studies they survey, three find an unqualified

16
Instrumental variables regression is a single-equation procedure for removing simultaneous
equations bias by employing instruments for each of theright-hand-sideendogenous variables in the
equation. Eligible instruments should be free of any correlation with the error terms in the equation
being estimated, yet be closely correlated with the variable for which they are acting as an instru-
ment. The instruments used in this study mainly comprised 1960 observations of a closely related
variable. Thus Bollen 1960 was used as an instrument for Gastil 1976, 1960 schooling rates were
used as an instrument for average 1960-85 schooling rates, and 1960 investment rates were used as
instruments for average 1960-85 investment.
17
See Sirowy and Inkeles, "The Effects of Democracy on Economic Growth and Inequality".
236 HELLIWELL

negative effect of democracy on growth, four find a negative effect in some

circumstances and regime types, and six find no relationship.18 Sirowy and
Inkeles attribute the discrepancies among the results to differences in time
period, country coverage, and uneven matching of political and economic
measures. They are especially critical of the lack of a clearly specified growth
model in which the effects of democracy can be assessed and the general failure
to account for other key factors, many of which might be presumed to be
correlated with democracy, that influence economic growth. In view of the
mixed nature of the evidence, and the availability of longer samples of compar-
able growth experience, it seems appropriate to try to make a systematic
attempt to test the relationship anew.
The empirical analysis of the effects of democracy on economic growth
which is developed here starts with an extended form of the Solow growth
model.19 This has been augmented by Mankiw, Romer and Weil to include
human capital accumulation, with real output determined as a Cobb - Douglas
function of physical capital, human capital and efficiency units of labour:20
no = m'wytMouo)1 -*-p co
where y is the level of real output, H is the stock of human capital, L is the stock
of labour (growing at rate n), K is the stock of physical capital (depreciating at
rate <5) and A is the level of technology, growing at the constant rate g. The
coefficients imply constant returns to all factors taken together, and hence
diminishing returns to any combination of physical and human capital. If Sk is
the fraction of output invested in physical capital, and st, is the fraction invested
in human capital, then in the steady state the log of output per capita is:
\n[Y(t)/L(t)] = + gt - ((a + p)/(\ - a - /?))ln(/i + g + 5)
1IL4(0)
+ (a/(l - a - P))\n(sk) + (/?/(l - a - P))\n(sh) (2)
This framework is extended to include the possibility of what Mankiw et al.
call 'conditional convergence', whereby if each country starts at some level of
output that differs from its steady state value, there will be convergence to-

18
Three studies outside the range of their survey report some evidence of positive effects of
democracy on growth; Pourgerami, Kormendi and Meguire, and Grier andTuIlock. However, since
their measures of democracy were taken late in the growth period under survey, these studies are
open to the risk of reverse causation. Attempts to make appropriate adjustments will be reported
later in this section and in Table 3. See A. Pourgerami, 'The Political Economy of Development:
A Cross-national Causality Test of the Development-democracy-growth Hypothesis', Public
Choice, 58 (1988), 123-41; R. C. Kormendi and P. G. Meguire, 'Macroeconomic Determinants of
Growth', Journal ofMonetary Economics, 16(1985), 141-63; and K.B. Grier and G.Tullock, 'An
Empirical Analysis of Cross-National Economic Growth, 1951-1980', Journal of Monetary Eco-
nomics, 24 (1989), 259-76.
19
See R. M. Solow, 'A Contribution to the Theory of Economic Growth', Quarterly Journal of
Economics, 70 (1956), 65-94, and Technical Change and the Aggregate Production Function',
Review of Economics and Statistics, 39 (1957), 312-20.
20
See G. Mankiw, D . R o m e r and D. Weil, *A Contribution to the Empirics of Economic Growth*.
Quarterly Journal of Economics, 107 (1992), 407-37.
 Linkages Between Democracy and Economic Growth 237

wards the steady state growth path for that country. This need not imply that
all countries have the same equilibrium level of income per capita (the level of
A can clearly be different across countries, based on variations in natural
resources, institutions, and other factors unrelated to the stocks of human and
physical capital) or even the same growth rate, since the equilibrium growth
rate for each country will depend on its population growth and investment in
human and physical capital.21 The Solow model augmented for human capital
accumulation predicts that the rate of convergence of each country towards its
steady state growth path will be at the proportional rate A, where

X = (/i + s + <5)(1 - a - P). (3)

The log difference between current income per effective worker and that in any
given earlier period 0 is thus given by

ln(y(0) - lnO'(O)) = (1 - <?-*)(a/(l - a - P))\n(sk)

+ (1 -e" ; j )(/J/(l - a - p))\n(sh)
- 0 ~ <?"*)((« + / W -OL-P))
^ (4)

Applied by Mankiw et al. to a cross-sectional sample of the growth experience

of ninety-eight countries from 1960 to 1985, this equation seemed to fit the
experience of the developing as well as the industrial countries. There was
evidence of conditional convergence towards the steady state growth rate
for the whole sample of countries, as well as for the more restricted sample
of industrial countries.22 Their results also showed that allowing for the

21
The technology index A is none the less assumed to have the same exogenous growth rate in
each country. Alternative convergence models assume that there is also convergence in the rates of
growth of the efficiency indices, thus giving international transfers of knowledge a key role to play
in the convergence process. Tests of equal versus converging growth rates for the efficiency indices
strongly favour the latter, as reported by J. F. Helliwell and A. Chung, 'Macroeconomic Conver-
gence: International Transmission of Growth and Technical Progress', in P. Hooper and J. D.
Richardson, eds. International Economic Transactions: Issues in Measurement and Empirical
Research (Chicago: University of Chicago Press, 1991), pp. 388-436.
22
Convergence of growth rates among the current industrial countries has also been shown by
Abramovitz, Maddison, Dowrick and Nguyen, and Baumol, among others. The Baumol results were
queried by De Long because of the possibility that the tests were likely to be biased by focusing only
on the countries that ended up rich. This difficulty is largely avoided by the use of nearly complete
samples of countries in subsequent studies, including Mankiw et al. and in this article. See M.
Abramovitz, "The Catch-up Factor in Postwar Economic Growth', Economic Inquiry, 28 (1990),
1-30; A. Maddison, Phases of Capitalist Development (Oxford: Oxford University Press, 1982); S.
Dowrick and D.-T. Nguyen, 'OECD Comparative Economic Growth 1950-85', American Econ-
omic Review, 79 (1989), 1010-30; W. Baumol, 'Productivity Growth, Convergence and Welfare:
What the Long-Run Data Show', American Economic Review, 76 (1986), 1072-85; and J. B. De
Long, 'Productivity Growth, Convergence and Welfare: Comment', American Economic Review,
78(1988), 1138-54.
238 HELLIWELL

TABLE 3 Effects of Democracy on Economic Growth,

1960-85
Equations

(1) (2) (3) (4)

No. of observations 90 90 90 90
Estimation method IV IV IV(a) IV(b)
Constant 2.631 2.641 1.784 1.889
(2.06) (3.75) (3.09) (1.53)
Coefficients
Scale 0.057 0.055 0.055 0.071
(2.06) (1.99) (1.95) (2.03)
lnGDPa60 - 0.407 -0.391 - 0.423 -0.312
(5.39) (5.24) (5.21) (2.12)
Invest - (/i + g + <5) 0.238 0.284 0.215 0.358
(1.58) (1.74) (1.34) (1.31)
School -
(n + g + <5) 0.420 0.419 0.414 0.454
(4.52) (4.57) (4.44) (4.12)
Bollen 1960 - 0.200
(1.11)
Gastil 1976 0.109 - 0.624
(0.58) (0.70)
0.445 0.458 0.439 0.336
0.329 0.301 0.331 0.360
Notes: Absolute values of /-statistics are in parentheses. The dependent variable in all equations is the growth in
real GDP per adult from I960 to 1985. i.e. lnGDPa85 - lnGDPa60. Instrumental variables used in all equations
include primary and secondary schooling and investment rates in 1960, 1960 GDPa, and scale, all in log form,
and n + g + S. Equations 2 and 4 also use Bollen 1960 as an instrument, while Equation 3 uses Gastil 1976, and
hence deliberately leaves in the possibility of simultaneous equations bias in the estimation of the effect of
democracy on growth.

accumulation of human capital lowered the estimated coefficient on physical

capital to a level that was consistent with capital's share in output, and hence
with the Cobb-Douglas assumption of constant returns to scale. Mankiw et al.
interpreted their results as a vindication of the augmented Solow model, and an
implicit rejection of the increasing number of models built on the assumption
that knowledge spillovers create the likelihood of increasing returns to scale at
the national level.23

23
Romer and Lucas, among others, have presented models assuming increasing returns at the
national level. Alternative endogenous growth models by Grossman and Helpman assume econom-
ies of scale and knowledge spillovers mainly at the industry level, which need not imply returns to
scale at the national level. See P. M. Romer, 'Are Non-Convexities Important for Understanding
Growth?*, American Economic Review, 80 (1990), 97-103; R. E. Lucas, 'Why Doesn't Capital
Flow from Rich to Poor Countries?', American Economic Review, 80 (1990), 92-6; and G. M.
Grossman and E. Helpman, Innovation and Growth in the Global Economy (Cambridge, Mass.:
MIT Press, 1991).
 Linkages Between Democracy and Economic Growth 239

Equation 1 of Table 3 shows the results of estimating Equation (4) using

the sample of ninety countries for which the growth data and the Bollen
index are both available, and the same 1960 to 1985 growth period analysed
by Mankiw et a/.24 To test for the possible existence of economies of
scale at the national level, a measure of each economy's sample-
average GDP is added to the estimation equation. Because results reported
later in this article confirm that both schooling and investment are influenced
by both income and democracy, all of the estimated growth equations
in Table 3 are estimated by instrumental variables, using schooling and
investment rates at the beginning of the 1960-85 growth period as instru-
ments for the average income data used in the growth equations. The
results show a strong conditional convergence effect,25 the expected
effects of the investment rates in physical and human capital and from
population growth, 26 and also evidence of slight economies of scale.27
The model therefore seems to provide an appropriate test-bed for investigating
the effects of democracy on growth. The main problem is that the Gastil data
are only available from the mid-1970s, while the growth period that is being
studied starts in 1960. As it has already been established that income levels
exert a positive influence on democracy, the use of a measure of democracy
from the middle or end of the sample period runs the risk that a possible
negative effect of democracy on growth would be masked by the reverse effect
of income level on democracy. This difficulty can be dealt with in two ways.
First, it is possible to use the Bollen index of political democracy for 1960,

24
The dependent variable is the logarithm of real G D P per adult in 1985 minus the logarithm of
real income per adult in 1960, following Mankiw et al. The independent variables are the same a s
in Equation 4 , using the logarithm of the gross investment rate to measure the fraction of output
invested in physical capital and the logarithm of the percentage of the adult population in secondary
school to proxy for the share of output devoted to investment in human capital. T h e results reported
in Table 3 impose the coefficient restrictions implied by Equation 4 in the text, that the coefficient
on the population growth term should be equal to the negative of the sum of the coefficients on the
investment and education variables. Tests show that the restriction is easily accepted, leads t o a
slightly higher explanatory power (after taking account of the degree of freedom saved) and does
not change any of the results materially, as reported by Mankiw et a/."The income data are from the
Mark IV version of the Summers and Heston data set. See Mankiw et a/., ' A Contribution to the
Empirics of Economic Growth', and R. Summers and A . Heston, ' A N e w Set of International
Comparisons of Real Product and Prices: Estimates for 130 Countries, 1950 to 1985', Review of
Income and Wealth, 34 (1988), 1-25.
23
This is shown b y the significant negative coefficients o n the variable measuring initial real
income per adult, which imply that countries with lower average incomes at the beginning of the
growth period had faster growth rates in the subsequent twenty-five years, once account is taken of
differences in the rates of investment in physical and human capital.
26
T h e investment and education variables both have the expected positive sign, although the
investment rate is not significant when instrumental variables estimation methods are used.
27
This is shown by the positive coefficient on the variable measuring the average scale of each
economy during the sample period. Subsidiary tests show that the result for economies of scale is
based entirely on the experience of the OECD countries, and depends on the use of sample-average
scale rather than initial scale in the equation. No evidence of scale economies appears when the
experience of the developing countries is separately assessed.
240 HELLIWELL

which is clearly free of the risk of positive feedback from post-1960 economic
growth. Secondly, we can make use of the Gastil measures and attempt to allow
for the possible effects of simultaneous equations bias by appropriate esti-
mation techniques.
Equation 2 of Table 3 shows the effect of adding the Bollen index to the
cross-sectional growth equation for ninety countries. The coefficient suggests
a negative effect of democracy on growth, although the effect is not statistically
significant at the usual levels. If the coefficient were to be taken at face value
it would imply that a 20-point increase in the Bollen index, for example a shift
from the democracy level of Libya (31) to that of Senegal (49), as they were
assessed by Bollen in 1960, would have reduced 1985 GDP per adult by 4 per
cent. The 95 per cent confidence bands for the estimate of the change in 1985
GDP range from about + 4 per cent to — 12 per cent, so not too much should
be made of the specific estimate. It should be noted, however, that this esti-
mated negative effect is in addition to the slowing effect that richness perse has
on subsequent economic growth in the convergence model being estimated
here.
Turning to the Gastil index, the situation becomes even more cloudy, as
there are no measures as far back as 1960, and the later measures are likely to
be contaminated by reverse causation. Equation 3 shows the effects of adding
the Gastil index for 1976, about halfway through the growth period. In this
equation the Gastil 1976 index is treated as though it were truly exogenous, and
it is included in the list of instrumental variables. The coefficient is positive,
small and weakly determined: the 95 per cent bounds range from — 0.20 to
•+• 0.34. When the equation is re-estimated using the Bollen 1960 index as an
instrument for Gastil 1976, as shown in Equation 4, the effect turns fairly large
and negative, but is still not significantly different from zero. The fact that the
estimated effect turns negative using simultaneous equations estimation is,
however, what would be expected if there were a positive effect of income on
democracy and a negative reverse effect from democracy to subsequent econ-
omic growth.
A recent study by Pourgerami develops a measure of democracy based on
Amnesty International reports of human rights violations, and finds that more
democracy (i.e. few infringements of civil liberties) is good for growth.28 Since
Pourgerami's democracy measure relates to the end of his 1965-84 estimation
period, the study is open to Sirowy and Inkeles* criticism of mismatched
timing, and is susceptible to reverse causation.29 To check for this possibility,
the Pourgerami democracy index was used in the Table 3 growth equation for
the seventy-six countries for which both sets of data are available. The results
show a positive but insignificant effect which approaches zero when the Bollen

28
See Pourgerami, 'The Political Economy of Development*.
29
See Sirowy and Inkeles, 'The Effects of Democracy o n Economic Growth and Inequality',
p. 137.
 Linkages Between Democracy and Economic Growth 241

1960 index is used as an instrumental variable for Pourgerami's democracy

measure.
Studies by Kormendi and Meguire and Grier and Tullock both use dichoto-
mous transformations of Gastil's 1978 index of civil liberties in equations for
GDP growth.30 Kormendi and Meguire find a weak positive effect, on 1950-77
GDP growth, of a dichotomous variable equal to 1.0 for the nineteen countries
(almost all in the OECD) with the highest levels of civil liberties. Grier and
Tullock find a stronger negative effect from a dichotomous variable with the
value 1.0 for twenty-seven countries, twenty-one of which are in Africa, with
the lowest levels of civil liberties in a pooled time-series cross-section equation
covering per capita growth for 113 countries from 1950 to 1981. Since the
Gastil index of civil liberties is highly correlated with the matching index of
political rights (r = 0.91 for the 1976 measures), and hence with the combined
index of democracy, these results may seem to run counter to most of the other
evidence showing, if anything, a weak negative effect of democracy on sub-
sequent growth.31 However, once again the apparent contrast seems to be due
primarily to reverse causation, as both studies use an index of civil rights at the
end of the growth period under review which is likely to have been influenced
by the already-established positive linkage from income levels to political and
civil rights.
To assess the extent to which these previous findings might have been
contaminated by reverse causation, I constructed four dummy variables which
took the value 1.0 for the countries which scored lowest (a) on the Gastil civil
rights index for 1978, (b) on the Gastil combined index for 1978, and (c) on the
two corresponding versions of the Bollen index for I960. 32 These variables
were then added to Equation 1 of Table 3 - the basic estimation of the growth
model by instrumental variables (results not reported here). None of the dum-
mies produced the sort of significant positive effect on growth hypothesized by
Grier and Tullock or by Kormendi and Meguire. Indeed, when either of the
Bollen dichotomous variables was used, either representing itself or used as an

30
See Kormendi and Meguire, 'Macroeconomic Determinants of Growth', and Grier and Tul-
lock, 'An Empirical Analysis of Cross-National Economic Growth, 1951-1980'.
31
There is a possibility that civil liberties and political rights have different effects on economic
growth, with the former encouraging the movements of people and ideas likely to foster growth and
the latter posing greater risks of short-term policy choices leading to instability of the type emphas-
ized in R. Dombusch and S. Edwards, eds. The Macroeconomics of Populism in Latin America
(Chicago: University of Chicago Press, 1991). However, when the difference between the 1976 civil
liberties and political rights indices was added to the ninety-country growth equation of Table 3 , it
took an insignificant negative coefficient, casting doubt o n the idea that civil rights are more
growth-inducing than political rights.
32
Following Grier and Tullock, I used all countries with values equal to 6.0 or 7.0 for the Gastil
index of civil rights in 1978. T h e analogous cutoff for the freedom index was 12.0 or more for the
sum of the civil rights and political freedoms measures. In both cases, I constructed a Bollen
dichotomous index covering the same number of countries. This involved a Bollen index of 0.49 or
below for the twenty-two countries with the lowest values of civil rights, and 0.48 or below for the
twenty-four countries with the lowest values for the combined Gastil index.
242 HELLIWELL

instrument for simultaneous equations estimation of the impact of one of the

Gastil measures, the effect was negative - opposite to that found by Grier and
Tullock and consistent with the other results reported in Table 3. The essential
reason for Grier and Tullock's result, therefore, seems to be reverse causation.
Democracy was very much in flux between 1960 and 1976, with almost half of
the countries in the bottom civil liberties group in 1960 being out of that group
by 1976, being replaced by other countries that had ranked relatively high in
1960. Among those which slid back after 1960 were two countries (Chile and
Ghana) in Huntington's group of second-wave democracies that were subject
to reversals in the fifteen years after I960, 33 and others that had not been
classified as democracies in 1960, but which became even less democratic after
1960. The simple correlation between the Bollen 1960 and the Gastil 1978
dichotomous indexes is only about 0.15, showing partly the differences there
can be between different assessments, but revealing even more the great
changes in political and civil liberties that occurred in many countries
in the 1960s and 1970s. Thus the dichotomous indexes for civil liberties
and political rights, when purged of the effects of post-1960 changes,
seem to support the results shown in Table 3: higher initial levels of the
democracy measures seem to have, if anything, a weak negative effect
on subsequent growth. It might be more appropriate to say that the aggregate
evidence does not support any significant linkage between the level of
democracy and subsequent economic growth.34 To go further would probably
require making distinctions among various types and features of democratic
and undemocratic regimes.
It has been argued by Olson that mature democracies may be likely to suffer
a slowdown in growth because of a slow buildup in the powers of special
interest groups whose successful claims for special treatment reduce the
growth of the economy as a whole.35 If this is the case, we might expect to find
slower growth in the older democracies, after adjusting for initial income levels
and the other factors determining economic growth, including the current level
of democracy. A rough test of this hypothesis was undertaken by constructing
a qualitative variable which took the value of 1.0 for each of Huntington's ten
countries that have been continuously democratic since early in the twentieth
century.36 When added to Equation 2 of Table 3, the variable yielded a negative
coefficient, as the maturation hypothesis would suggest, but the coefficient was

33
See Huntington, Vie Viird Wave, p. 14.
34
A similar conclusion is reported by Cooper, based on cases studies of eighteen major develop-
ing countries. See R. N. Cooper, Economic Stabilization in Developing Countries (New Haven,
Conn.: Yale University Press, 1991), pp. 74-5.
33
See Mancur Olson jr. Vie Rise and Decline of Nations: Economic Growth, Stagflation and
Rigidities (New Haven, Conn.: Yale University Press, 1982).
36
The countries are Australia, Canada, Finland, Iceland, Ireland, New Zealand, Sweden,
Switzerland, United Kingdom and the United States (see Huntington, Vie Viird Wave,
p. 14).
 Linkages Between Democracy and Economic Growth 243

so insignificant that there was a drop in the overall explanatory power of the
equation.37
It is probably reasonable to regard these estimates of the effects of
democracy on economic growth as being consistent with the broad pattern
of earlier results surveyed by Sirowy and Inkeles. In essence, the effects
identified are generally negative - though they are neither uniform
nor very strong. However, the results reported thus far all assess the
partial effect of democracy on subsequent economic growth, taking the
rates of schooling and investment as given. What of the possibility that
democracy, once acquired, helps to establish conditions that encourage higher
levels of schooling and investment, and thereby increases economic growth
via these indirect channels? The first step in assessing this possibility
is to estimate the effects of democracy on subsequent rates of schooling
and investment. The appropriate tests are performed in Table 4, using the
Bollen 1960 index to reduce the dangers of reverse causation. The equations
are estimated both with and without the inclusion of the 1960 levels of per
capita GDP, which are likely to have a strong positive effect on the schooling
rate, and possibly also on the investment rate. The democracy index is seen to
have a positive effect on subsequent schooling and investment rates, with the
estimated effect becoming smaller and less significant when account is taken
of the effects of initial income levels. Indeed, as Equation 4 shows, when prior
GDP levels are taken into account, the level of democracy appears to exert no
significant influence on education levels.
What are the net effects of democracy on growth when the direct negative
effects are combined with the positive indirect effects? Looking first at the
direct effects of democracy on subsequent growth, the coefficient on the Bollen
index in Table 3 suggests that if a completely undemocratic country in 1960
had instead been fully democratic the logarithm of 1985 GDP per adult would
have been lower by 0.200, roughly equal to a 0.8 per cent reduction in the
average annual growth rate from 1960 to 1985. However, the estimated indirect
positive effects are slightly larger, being 0.149 for investment and 0.136 for
schooling.38 Combining the direct and indirect effects.would suggest a weak
positive effect of 0.085. All three channels are estimated with great impreci-
sion, so that the only appropriate conclusion is that no significant net effect can
be shown on the basis of these results.

37
The coefficient on the dummy variable covering the ten rich democracies was - 0.065, with
a /-value of - 0.4, and the adjusted R2 fell from 0.458 to 0.444.
38
The investment effect is the product of the coefficient on investment in Equation 2 of Table 3
and the coefficient on the Bollen index in Equation 2 of Table 4. The schooling effect uses the
schooling coefficient from Equation 2 of Table 3 and the Bollen coefficient from Equation 4 of
Table 4.
244 HELLIWELL

TABLE 4 Effects of Democracy on Investment and Education,

1960-85
Equations

(1) (2) (3) (4)

Dependent Investment Investment Schooling Schooling

variable rate rate rate rate
No. of
observations 90 90 90 90
Estimation method OLS OLS OLS OLS
Constant - 2.545 -4.180 - 4.338 - 8.730
(19.44) (11.37) (18.24) (15.62)
Coefficients
lnGDPa60 0.258 0.695
(4.69) (8.30)
Bollen 1960 1.023 0.524 1.672 0.326
(5.72) (2.71) (5.14) (1.11)
R2 0.263 0.405 0.222 0.561
S.E.E. 0.443 0.398 0.805 0.605
Notes: The dependent variables are the logarithms of the investment and schooling rates described in the data
appendix. They are averages over the 1960-85 period used for the growth equations. The independent variables are
the I960 Bollen democracy index and the logarithm of 1960 real GDP per adult.

III. C O N C L U S I O N S A N D I S S U E S FOR F U R T H E R RESEARCH

The data surveyed here support strongly the notion that countries at higher
income levels are more likely to have democratic forms of government. It has
been shown that this positive effect does not appear to be the result of reverse
causation: estimates of the reverse effect of democracy on subsequent growth
indicate that this feedback is more likely to be negative than positive. This
conclusion is reinforced by the fact that models which allow for the
simultaneous determination of democracy and growth show positive effects of
income on democracy that are larger than those estimated by more simple
methods. These results tie in with the evidence that countries starting with
lower levels of per capita income have higher initial growth rates which
subsequently tend to slow down as income levels converge with those of richer
countries. One possible component of this slowdown is that countries adopt
democratic forms of government during the development process. It is still
unclear whether the adoption of a democratic government contributes to
growth rate convergence by reducing the subsequent growth of the democratiz-
ing countries.
It is relatively uncontentious to suggest that increasing levels of education
and income are likely to increase citizen demands for many things, including
the range of political and civil freedoms that characterize democratic systems.
 Linkages Between Democracy and Economic Growth 245

It is less clear, however, how or why certain features of democratic government

might help or affect subsequent growth. It is almost certainly the case that some
aspects of democratic systems are more helpful to subsequent growth than
others.39 Alesina and Rodrik, for example, argue that democracies with initially
unequal distributions of income will have lower growth than democracies
with more even distributions:40 the large group of the enfranchised poor
in the first case will vote for high taxes on capital, which will then lead to
lower investment and hence lower growth rates of GDP.41 It is also clear,
from the evidence of several waves and reverse waves of democratization,
analysed most recently by Huntington, that there are many reasons why
countries have adopted democratic systems, and then in many cases lost
them and sometimes tried again.42 Analysis of the dynamics of these waves
and reversals might well help to illuminate the subsequent effects of democracy
as well the strong link from income to democracy. The numbers of countries
involved, and the range of experiences they illustrate, offer at least some
hope for enriching understanding of the links between political systems and
economic performance.
Although the analysis in this article has treated all countries equally,
differing only in their values for the variables under investigation, there may be
many country-specific or culture-specific factors that influence the linkages
between democracy and economic growth, and which may be obscured in a
study based on large samples of countries. To provide just one example, recent
research on the Asian economies, for which the convergence model of growth

39
Studies have found a negative linkage between economic growth and the instability of govern-
ment (Alesina etal.) and the frequency of assassinations and coups (Barro), although Londregan and
Poole have found that the significance of the negative effect of coups on subsequent growth becomes
slight when the two-way linkages between coups and economic growth are jointly estimated. T h e
results of Alesina etal suggest that the two-way linkages between political instability and economic
growth are unaffected by the level of democracy as measured by a three-valued index of democracy.
Using the 1960 values for their index in the Table 3 equation gives similar results to those using the
1960 Bollen index, although the negative effect is less significant using the Alesina etal. index. See
A . Alesina, S. Ozler, N . Roubini and P. Swagel, 'Political Instability and Economic G r o w t h ' , NBER
Working Paper, No. 4 1 7 3 (Cambridge: National Bureau of Economic Research, 1992); R. J. Barro,
'Economic Growth in a Cross-Section of Countries', Quarterly Journal of Economics, 106 (1991),
4 0 7 - 4 4 ; J. Londregan and K. Poole, 'Poverty, the Coup Trap, and the Seizure of Executive Power',
World Politics, 42 (1990), 151-83.
40
See A. Alesina and D. Rodrik, 'Distributive Politics and Economic Growth', CEPR Discussion
Paper, No. 565 (London: Centre for Economic Policy Research, 1991).
41
Presumably it is important to make the distinction between pre-tax and post-tax distributions
of income. One of the influential strands of thinking arguing that democracy will be bad for growth
adopts the position that democratic governments will be more likely to undertake redistributive
policies that lead to higher tax rates and otherwise discourage savings, labour supply and capital
accumulation. For examples, see P. T. Bauer, Equality, the Tliird World, and Economic Delusion
(London: Weidenfield & Nicholson, 1981), S. Huntington and J. Nelson, No Easy Choice: Political
Participation in Developing Countries (Cambridge, Mass.: Harvard University Press, 1976) and E.
Wecde, 'The Impact of Democracy on Economic Growth: Some Evidence from Cross-National
Analysis', Kyklos, 3 6 (1983), 2 1 - 3 9 .
42
See Huntington, Tlxe Tliird Wave.
246 HELLIWELL

fits poorly, reveals a significant negative partial effect of democracy on sub-

sequent growth.43 In general, there is a considerable amount of unexplained
international variation in both economic growth and political institutions and
practices to allow for the analysis of more specific factors.
What does the evidence presented in this article offer by way of answer to
the questions posed at the outset: are there economic prerequisites to democ-
racy? Does democracy help or hinder subsequent economic growth? Is the
Chinese sequencing better than the Russian one?
The results in the first section of this article confirm a robust positive
relation between the level of per capita income and the adoption of democracy.
There appear to be no clearly defined thresholds or prerequisites - just
a strong tendency for democracy to become the chosen and maintained form
of government as countries get richer and as education levels increase. These
results suggest that democracy has an intrinsic value that is increasingly
sought after as populations become better off and better educated. With
respect to the impact of democracy on subsequent economic growth,
the evidence in this article pours cold water on the notion that introducing
democracy is likely to accelerate subsequent growth. The statistical fine-
tuning undertaken to account for reverse causation increased the estimated
likelihood of a negative direct effect of democracy on subsequent economic
growth. This said, this negative effect was statistically non-significant
and appeared to be offset by indirect positive influences flowing through
investment and education. An optimistic interpretation of the overall
results would thus be that democracy, which is generally considered to
offer considerable non-economic benefits, is available at little cost in terms
of subsequent lower growth. Even more optimistic is the further result
that the economic cost of democracy drops, and may even be eliminated,
when account is taken of the positive effects that democracy appears to have
on subsequent education and investment, both of which tend to increase
economic growth.
On the sequencing question, the evidence does not indicate any magical or
'most favourable' time for democratization, beyond the basic result that the
probability of a country being democratic rises steadily with its level of in-
come. Thus democracy delayed is not likely to be democracy lost, as long as
the alternative interim system of government is able to deliver continuing
increases in living standards.

41
This partial negative effect only appears if allowance is made for the positive effects of
investment and openness, both of which may in turn be positively influenced by the level of
democracy. Differences in education levels do not appear to help explain variations in growth rates
among the Asian economies. See J. F. Helliwell, 'International Growth Linkages: Evidence From
Asia and the OECD", in T. Ito and A. Kreuger, eds, Macroeconomic Linkage: Saving, Exchange
Rates and Capital Flows (Chicago: University of Chicago Press, 1993).
 Linkages Between Democracy and Economic Growth 247

DATA APPENDIX
This appendix shows the means, standard deviations and some correlations for three
measures of democracy and the variables used in the growth regressions, for the
ninety-country sample defined by the availability of the Bollen data and employed for
most of the equations presented in Tables 2 to 4. The three democracy indices are the
Bollen index (from Bollen 1980) for 1960 and the Gastil indices for 1976 and 1985. The
Gastil indices combine data for political rights and civil liberties to form an index
bounded by 0 and 1.0, with zero meaning least democratic, just as is the case for the
Bollen index. The income per capita figures are measured in thousands of 1980 inter-
national dollars per adult, with GDPa60 being the antilog of the variable used in the
equation estimates reported in Tables 2 to 4. Similarly, the other variables are the
antilogs of the variables used in the estimation, with country scale converted to billion
dollars at 1980 rates, average growth reported in the dimensions of annual average
percentage growth in real income per adult, schooling as secondary school enrolment
as a fraction of the adult population, investment as a fraction of total GDP, and

TABLE A Some Statistical Values

Mean Standard deviation

GDPa60 3130.0 2941.0

GDPa85 5466.0 5362.0
Bollen60 0.683 0.262
Gastil76 0.460 0.346
Growth 1.73% 1.77
Scale 80.9 245.0
Investment 0.178 0.081
School 0.055 0.035
71 + g + S 0.072 0.0087

Correlations

GDPa60 GDPa85 Bollen60 Gastil76

GDPa85 0.889
Bollen60 0.572 0.579
Gastil76 0.783 0.804 0.595
Gastil85 0.739 0.754 0.569 0.877

Growth Scale GDPa60 Investment School

Scale 0.151
GDPa60 0.124 0.449
Investment 0.577 0.215 0.529
School 0.480 0.323 0.673 0.653
n + g +S - 0.222 - 0.233 ~ 0.479 - 0.366 - 0.340
248 HELLIWELL

n + g + 5 (the unlogged form of n + g + <5, average annual population growth, as a

proportion, plus a constant to represent depreciation and average technical progress).
The growth equation data are the same as those used and published by Mankiw et ai,
being drawn from the Summers and Heston 1988 publication, with the addition of
World Bank data for schooling and demography.44 The 1960 instruments for schooling
are based on World Bank data, while those for 1960 investment are from the Summers
and Heston Mark 5 data set used for the 125-country sample analysed in Table I.45 The
right-skewness apparent from the means and standard deviations of the raw data disap-
pears with the logarithmic transformation used prior to estimation.

44
See Mankiw, Romer and Weil, *A Contribution to the Empirics of Economic Growth'; Sum-
mers and Heston, *A New Set of International Comparisons of Real Product and Prices'.
45
See R. Summers and A. Heston, 'The Penn World Table (Mark 5): An Expanded Set of
International Comparisons, 1950-1988', Quarterly Journal of Economics, 106 (1991), 327-68.