Zak, P. J., and Knack, S. (2001) Trust and Growth. Economic Journal
Zak, P. J., and Knack, S. (2001) Trust and Growth. Economic Journal
Zak, P. J., and Knack, S. (2001) Trust and Growth. Economic Journal
Published by Blackwell
Publishers, 108 Cowley Road, Oxford OX4 1JF, UK and 350 Main Street, Malden, MA 02148, USA.
Russian Proverb
Adam Smith (1997 [1766]) observed notable differences across nations in the
`probity' and `punctuality' of their populations. For example, the Dutch `are
the most faithful to their word.' John Stuart Mill wrote: `There are countries in
Europe . . . where the most serious impediment to conducting business concerns on a large scale, is the rarity of persons who are supposed t to be trusted
with the receipt and expenditure of large sums of money' (Mill, 1848, p. 132).
Enormous differences across countries in the propensity to trust others survive
today. In Scandinavian cities, bicycles are still commonly left on the street
unlocked and unattended (although anecdotal evidence suggests this practice
has begun to decline). Danish citizens routinely leave small children in
The ideas in this paper came out of discussions by the authors following a conference on the
political economy of growth held at Claremont Graduate University in April, 1997. Thanks go to Tom
Willet and Yi Feng who organised the conference and invited the authors. We have beneted from the
comments of Olugbenga Ajilore, Thomas Borcherding, Joydeep Bhattacharya, Arthur Denzau, Raja
Kali, Rody Manuelli, Hilton Root, Gary Segura, Tridib Sharma, as well as from attendees at the following
conferences and seminars: 2000 Gruter Institute Conference on Evolutionary Biology and Law, 2000
UCLA Conference on International Political Economy, 1999 International Society for New Institutional
Economics, 1999 Latin American Econometric Society, 1998 Public Choice Society, ITAM, UC Santa
Barbara, CSU Northridge, Philadelphia Federal Reserve Bank, The World Bank, The Hoover Institution
at Stanford University, ASU, USC Economics Department, the Center for International Studies at USC,
and Claremont Graduate University's Economics Department, Psychology Department, and Drucker
School of Business. Lastly, we thank two anonymous referees and an editor of this Journal, David de
Meza, for perspicacious suggestions that have improved the exposition. Any errors are the sole
responsibility of the authors. Zak acknowledges the hospitality of the Hoover Institution and the
Economics Department at Arizona State University where part of this work was done.
[ 295 ]
296
[ APRIL
We show that a (Northian) low-trust poverty trap exists for the model in this
paper. If trust is too low in a society, savings will be insufcient to sustain
positive output growth. Such a poverty trap is more likely when institutions
both formal and informal which punish cheaters are weak. Most importantly,
we show that the amount of trust and the existence of a poverty trap depend
critically on the level of social heterogeneity in a society.
The model offers an explanation for the empirical ndings of Easterly
(2000), Easterly and Levine (1997), and Knack and Keefer (1997) that is
grounded in transactions costs rather than on inefcient and unstable government policies. Easterly and Levine (1997) nd that ethnic heterogeneity slows
# Royal Economic Society 2001
2001]
297
1. Theory
Consider an economy with a continuum of innitely-lived consumers. Agents
vary in a potentially large number of ways, including their income, education, ethnicity, religion, etc. We reduce all the ways in which agents can vary
into a single index, with an agent's `type' identied by i distributed over the
positive real line. Though an agent's type may evolve over time, it is xed at
any point in time. Consumers in this economy have standard preferences
and seek to smooth consumption by saving each period, but to access credit
markets they must utilise an investment broker. There is a continuum of
investment brokers who are distinct from consumers but are identied by
the same attributes. Only the broker knows the actual return earned on an
investment, and thus brokers have a moral hazard problem as they have the
opportunity to cheat their clients. Consumers, knowing this, may choose to
# Royal Economic Society 2001
298
[ APRIL
spend time investigating their broker in order to reduce the broker's ability
to cheat.1
Each period, through a random draw, consumers are matched with investment brokers and, at the time the match is made, the broker's type is
unknown, though the distribution of types is known. In the following period,
when the investment is closed out, but prior to its payout the broker's type is
revealed, at which point a decision is made regarding how much time to spend
investigating the broker. Note that, as in the experimental literature, we focus
on the starkest case the degree of trust in a one-shot transaction without
individual-specic reputational effects.2 One of the innovations of the theory is
that there are degrees of trust, rather than having agents choose to trust their
transaction partner completely or not at all as in trust games in the laboratory.
There are two institutional effects, besides an individual's investigation of
the broker's investment, that motivate inherently untrustworthy brokers to
reduce the amount that they cheat. The rst of these is formal institutions.
Formal institutions include investigative agencies, like the Securities and
Exchange Commision, that oversee brokers, as well as the judicial system that
enforces contracts and prosecutes cheaters.3 In the model, punishments take
the form of the partial loss of the broker's fee.
Second, in addition to formal institutions, cheating brokers may face
sanctions due to informal institutions. While Hobbes viewed the government
as the sole source of trust between strangers, J. S. Mill (1848) wrote that `much
of the security of person and property in modern nations is the effect of
manners and opinion' and of `the fear of exposure' rather than `the direct
operation of the law and the courts of justice' (p. 1356). Mill was highly
critical of the English legal system, believing that reputational effects served as
effective substitutes in keeping economic agents honest (p. 444). Informal
sanctions constraining opportunism by agents can include guilt associated with
violating moral norms, `afterlife sanctions' associated with religious dictates,
social sanctions (such as ostracism), and loss of prots through reputational
effects.
All of these informal sanctions depend on, or are facilitated by, social ties.
Moral and religious norms depend on prior socialisation processes; reputa1
The time spent investigating one's broker can be thought of, more generally, as time spent writing
contracts, which can be trivial when one trusts one's broker (when a handshake will sufce), in contrast
to an iron-clad contract specifying all manner of contingencies when one suspects one's broker will
cheat. The model here is related to a large literature on moral hazard and asymmetric information,
especially Tsiddon (1992) who shows that moral hazard can lead to poverty traps (i.e. multiple
equilibria). Also see Hermalin and Katz (1991), Mukherji and Nagarajan (1995), and Phelan (1995).
2
Multiperiod relationships between consumers and brokers will strengthen the results here as
individual (rather than group) reputations will matter. The impact on growth of private information
and reputational effects is examined by Khan and Ravikumar (1998) and Marcet and Marimon (1992).
On long-lasting contractual relationships, see Townsend (1982), Atkeson (1991) and Atkeson and
Lucas (1995).
3
Other formal mechanisms serving to constrain opportunism in nancial markets include regulation (such as nancial disclosure requirements), and private organisations such as credit bureaus, codes
of professional ethics (for example for CPAs), mercantile agencies (such as Dun and Bradstreet), bond
ratings services (such as Moody's), and stock exchange memberships (see Klein, 1997, and Zucker,
1986).
2001]
299
300
[ APRIL
for collective action.7 Informal sanctions are modelled by dening the effective
distance, D(i, j; )UIV d(i, j)=, where > 1. A higher value of indicates that
social institutions reduce the salience of differences across types. When social
restraints are sufciently strong, the effective distance between an investor and
broker of different types is small even if the nominal distance d(i, j) is large.
Now we introduce some notation from which the model will be built. Let c i
be consumption of a type i consumer who earns wage w i , has wealth a i , spends
time h i working in production, and devotes time e i to investigate the return on
his or her investments, with total time normalised to unity. Formal institutions,
denoted p, seek to detect and punish cheating brokers and are funded by a
lump-sum tax, , paid by consumers. Agents have access to an investment
investigation technology, [e i , p, D(i, j; )] :[0, 1] 3 R2 ! [0, 1] which permits
agents to determine the fealty of brokers in reporting investment income, with,
from the discussion above, @=@e i . 0, @=@ p . 0, @=@d(i, j) , 0, and
@=@ . 0. That is, the return from investigation increases with the time a
consumer puts into this activity, as well as with the strength of formal and
informal sanctions, and decreases when the social distance between the
consumer and broker increases. The time spent investigating one's broker, e i ,
will be called diligence. We will further suppose that the increase in one's return
from investigation displays diminishing marginal returns to diligence,
@ 2 =@(e i )2 , 0:
The timing of decisions is as follows. A consumer observes his or her current
wage and expected investment income, with the type of investment advisor last
period being revealed, and chooses an allocation of time between working in
production and investigating last period's broker. Next, the agent works for a
rm and receives the post-investigation return on investment from the previous period's broker. At this point, current period savings is chosen given the
agent's labour income, investment income from the previous period, and the
net-of-cheating expected return on savings from the current period to the next
period. A broker is then randomly assigned to invest the agent's savings. The
type of investment broker to which a consumer is matched is drawn from a
continuous CDF with support on (0, 1) and nite rst and second moments.
A single broker invests all of a consumer's savings.8 The return on an agent's
investment is stochastic because the type of investment advisor (and therefore
the amount of cheating) to which one has been matched is unknown when
savings is chosen. To wit, agent i in this economy maximises lifetime utility by
solving
Max c i ,e i E
7
1
P
t0
t U (c it )
(1)
For example, spatial concentration of ethnic groups may facilitate their organisation.
If one could distribute savings among brokers, then some diversication of risk would be possible.
In addition, markets are incomplete as contingent contracts that would allow consumers to insure
against broker cheating are unavailable. The model is designed in this way so that the effect of trust in
one-to-one transactions is as transparent as possible. Diversication among brokers and partial
insurance would reduce the effect of broker cheating in the model, but the qualitative results of the
model remain unchanged.
8
2001]
301
c it w it h it R t a it ij [e it , pt , Dt (i, j; )] a it1
(2)
1 e it h it
(3)
s.t
Ij
c t w t f1 ij [e it , pt , Dt (i, j; )]gR t a it :
(4)
(5)
ij
@ t
@e it
(6)
w it
R t a it :
(7)
302
[ APRIL
solution to (7) produces the standard inverse demand functions for capital
and labour,
r t F 1 (K t , H t )
(8)
w t F 2 (K t , H t ):
(9)
Equation (9) is total wage expenditures; the wage for a type i worker is
through an allocation relation, w it w t G(i), such that
determined
1 i
0 w t d w t (more will be said about this in the next Section). The gross
yield on savings is R t1 r t1 1 , where r t1 is the interest rate from t to
t 1 and 2 [0, 1] is the depreciation rate on capital.
The state of the economy is a distribution of asset holdings, a.
~ Given such a
distribution, we can dene a competitive equilibrium.
Definition. A competitive equilibrium for the problem dened by (1),
2001]
303
@d(i, j) . 0.
The proposition shows that when wealth is high, one is less likely to count on
formal or informal institutions to reduce cheating. Rather, wealthy agents will
forego income to investigate their brokers and protect their considerable
wealth. These incentives are mitigated when one's wage is high, as time must
be taken off of work to investigate one's broker. Very high wage agents will
simply tolerate cheating as in the predation models of Zak (2000) and Grossman and Kim (1995). Similarly, agents optimally reduce their diligence when
either formal or informal institutions reduce cheating by brokers. Lastly, when
one's broker is revealed to be socially distant, optimal diligence increases since
such a broker has a greater incentive to cheat.
Trust in a society can be dened as the aggregate time that agents do not
spend in verifying others' actions. That is, trust is
1
e it d :
Ht 1
0
In other words, trust is the time agents spend in production rather than
investigating their brokers. This is, of course, not the only denition one can
give for trust, but it is a natural one given the structure of the model.11 Note
that this is an economy-wide measure of trust, not an agent-specic one. This
was done because the analysis that follows characterises the circumstances that
cause societies to have more or less trust, and the empirical tests of the model
follow this tack.
1.3. Income Distribution and Trust
Consider a partition that divides the population into
P N N n. 1n distinct `classes'.
1 2
N
Dene the N -vector E to be fE , E , . . ., E g, with n1 E 1, where n is
the mass of agents of type n. Let us order agent classes by their wages,
11
Trust increases whenever investors' condence that brokers will not cheat them increases, whether
that condence is derived from formal or informal institutions or other sources. Some authors dene
trust more narrowly to exlude any effects of legal mechanisms or formal institutions (e.g., Williamson,
1993; Charny, 1990). Our denition of trust is similar to what Yamagishi and Yamagishi (1994) refer to
as `assurance', an expectation of benign behaviour derived from knowledge of the incentive structure
facing one's trading partner (p. 132).
304
[ APRIL
ij
(11)
ij
(12)
2001]
305
by applying Stein's lemma, where Varfxg is the variance of x.12 Equation (12)
shows that optimal investment depends on both the expectation of brokers
cheating, as well as the variance of cheating by brokers. Since the amount that
brokers cheat is increasing in the distance between the broker and client, the
variance of cheating is increasing in the variance of the distribution of agents.
Proposition 4. Suppose the condition in Proposition 3 holds. Then, an agent's
is decreasing in the variance of the distribution of
optimal investment choice, a it1
agents.
This result is standard in nance when the agent chooses an optimal `portfolio' between risk-free current consumption and risky investment.13 The
import of Propositions 3 and 4 is that heterogeneous societies have lower trust
and therefore lower investment than do homogeneous societies. This occurs
via two mechanisms: in low trust societies, incomes are lower since more time
is spent in diligence (Propositions 1 and 3), and investment is lower since
agents are risk averse (Proposition 4). As long as savings is not too sensitive to
changes in the net interest factor ijR, a decrease in investment lowers capital
formation through the law of motion for the capital stock (10), reducing
output growth in the transitional dynamics.14 Hence, growth is generally lower
in heterogeneous low trust societies.
It is straightforward to show that if a society is sufciently heterogeneous,
investment will be too small to sustain growth (i.e. if ij ! 0 as d(i, j) ! 1).
That is, a low-trust poverty trap exists when social heterogeneity is high. Such a
poverty trap is more likely to exist when formal and/or informal institutions
are weak (a `Northian poverty trap') since both of these reduce investment
returns by raising cheating.
Several additional implications for optimal investment behaviour can be
drawn from condition (11). First, strengthening formal institutions may raise
investment if this policy is not too costly (i.e. if is not too high). In terms of
investment, there is an optimal level of funding for formal institutions if ij is
concave in p. Funding beyond this point will inhibit investment by reducing
the income available to consumers more than it raises the return to investment
by reducing cheating. Second, similar to the argument given above, wage
discrimination reduces investment as aggregate diligence increases (under the
restrictions in Proposition 2) and incomes fall.
Combining the results in the previous subsections, we have shown that
heterogeneous societies, especially those with weak formal and informal
institutions, have lower trust and retarded income growth than less hetero-
12
Stein's lemma states that if random variables x and y are bivariate normally distributed, then
Cov[ g (x), y] E[ g 9(x)] Cov(x, y), providing that the function g (:) is differentiable and some
regularity conditions are met. If ij and c i are not bivariate normally distributed, then (12) approximates (11) by a central limit theorem.
13
See, for example, Huang and Litzenberger (1988, p. 95) for a discussion of this result.
14
Blinder and Deaton (1985) nd robust support showing that savings is insensitive to the interest
rate; see also Deaton (1992, pp. 5975).
306
[ APRIL
2. Empirics
Propositions 1, 2, 3, and 4 produce the ve primary predictions of the model:
(i) Higher trust increases investment and growth;
(ii) Homogeneous societies exhibit higher trust, and thereby investment
and growth;
(iii) Egalitarian distributions of income enhance trust, and thereby raise
investment and growth;
(iv) Discrimination lowers trust, reducing investment and growth;
(v) There is a low-trust poverty trap.
2.1. Measuring Trust
The rst step in testing the model is to identify a reasonable cross-country
measure of trust. We use a measure based on data from the World Values
Surveys (WVS), conducted in several dozen countries in three `waves', 1981,
19901, and 19956 (Inglehart et al., 2000). The measure of trust we use is the
percentage of respondents in each country agreeing that `most people can be
trusted' against the alternative that `you can't be too careful in dealing with
people'. Values range from a low of 5.5% in Peru to a high of 61.2% in Norway.
Surveys typically include between 900 and 2,800 respondents, designed to be a
nationally representative sample. Knack and Keefer (1997) provide empirical
support for the validity of these data, nding, for example, that trust is
strikingly correlated across countries and regions with the number of wallets
that were `lost' and subsequently returned with their contents intact in an
experiment conducted in various European nations and the United States.
Values for trust are also consistent with anecdotal and case study evidence on
trust across countries and regions. For example, values for northern regions of
Italy are higher than for the south, consistent with evidence reported by
Putnam (1993) and others. Values are highest for the Scandinavian countries.
Trust data are available for 41 market economies. Most of these countries
were included in at least two survey waves; we use the earliest observation when
trust is an independent variable, and the latest observation when trust is the
dependent variable. Of these 41 countries, only 29 were included in the
empirical analysis of Knack and Keefer (1997). Values for 9 additional developing countries were obtained from the 19956 survey wave, providing us with a
much more representative sample than the OECD-heavy sample analysed by
15
Other explanations for the relationship between growth and inequality include: redistribution
(Alesina and Rodrik, 1994; Persson and Tabellini, 1994), credit constraints on education (Benabou,
1996), increasing returns and the size of the middle class (Murphy et al., 1989), and the difculty in
choosing a stable set of rules governing property rights (Keefer and Knack, 2001). Perotti (1996) nds
little empirical support for the rst two explanations, the third is only broadly tested, while the fourth
survives a variety of empirical tests.
2001]
307
Knack and Keefer. Two more observations (Greece and Luxembourg) are
taken from Eurobarometer surveys conducted in the 1980s, and another is
from a government-sponsored survey in New Zealand patterned after the
World Values Survey (Gold and Webster, 1990).16
In the model, trust is produced by formal institutions as well as by social
homogeneity and informal sanctions against cheating. Several authors (e.g.
Yamagishi and Yamagishi, 1994) distinguish between two types of trust, one
based on `deterrence' and another based on `benevolence'. The former type
roughly corresponds to trust produced by formal institutions as well as
informal sanctions such as ostracism, while the latter type can be identied
loosely with social homogeneity. The very general phrasing of the WVS survey
question on trust suggests that it is a reasonable proxy for our concept of trust,
which encompasses both deterrence-based and benevolence-based trust. Most
importantly, the question does not contain qualications implying any exclusion of trust derived from the presence of effective legal sanctions. We cannot
rule out the possibility that some respondents may interpret the question to
apply only to interpersonal transactions beyond the reach of the law; to the
extent that this happens, our empirical tests will underestimate the relationship between formal institutions and our denition of trust.
2.2. Trust, Investment and Growth
In Table 1, our trust variable is included in Barro-type cross-country investment
and growth regressions. Dependent variables are investment as a percentage of
GDP, averaged over the period 197092, and average annual growth in per
capita income over the same period, as constructed from Summers and Heston
(1991) data, version 5.6. Other than trust, regressors include 1970 per capita
income, schooling attainment for 1970 (mean years for the population aged
25 and over) from Barro and Lee (1993), and the price of investment goods
for 1970, as a percentage of United States prices (from Summers and Heston,
1991). Table A1 in the Appendix presents means, standard deviations, and
sample sizes for the variables used in the analysis. Correlations among the
independent variables are shown in Table A2 in the Appendix.
Equation 1 shows that investment is higher where incomes are higher, where
investment goods prices are relatively low, and where trust is higher. The
investment/GDP share rises by nearly one percentage point for each sevenpercentage point increase in trust. Equation 2 demonstrates the positive
relationship between trust and growth. In our 41-nation sample, convergence
and (especially) the effects of schooling are weaker than in larger samples.
Higher investment goods prices, relative to US levels, are signicantly and
negatively associated with growth, as expected. Controlling for these inuences, growth rises by nearly 1 percentage point on average for each 1516
Although the trust question asked in these surveys is identical to that in the World Values Surveys,
we cannot rule out the possibility that other differences in survey content and administration may affect
the trust values for these three countries. However, the results reported below are not sensitive to the
inclusion of these three non-WVS observations.
308
Equation
Dependent variable
Method
Constant
GDP per capita
(000s)
Schooling
Attainment
Price of investment goods
Trust
Investment/GDP
Inv/GDP
197092
OLS
21.549
(2.680)
0.817
(0.240)
0.376
(0.486)
0.136
(0.026)
0.133
(0.051)
Growth
197092
OLS
3.998
(0.782)
0.085
(0.099)
0.039
(0.123)
0.043
(0.009)
0.063
(0.020)
Growth
197092
OLS
1.073
(0.719)
0.196
(0.098)
0.091
(0.092)
0.024
(0.008)
0.045
(0.017)
0.136
(0.031)
Trust
Inv/GDP
197092
2SLS
21.503
(3.340)
0.814
(0.361)
0.356
(0.576)
0.136
(0.041)
0.139
(0.093)
Growth
197092
2SLS
4.020
(0.898)
0.084
(0.097)
0.029
(0.155)
0.043
(0.011)
0.060
(0.025)
Inv/GDP
197092
OLS
15.599
(3.649)
1.990
(0.347)
0.544
(0.453)
0.136
(0.024)
0.323
(0.079)
Growth
197092
OLS
2.320
(0.948)
0.245
(0.099)
0.008
(0.110)
0.043
(0.010)
0.117
(0.034)
0.37
1.26
1.88
0.036
(0.011)
0.61
4.44
21.8
0.010
(0.004)
0.50
1.19
1.88
Percent Catholic
Percent Muslim
Percent Eastern orthodox
Trust 3 GDP
R2
SEE
Mean, D.V.
0.55
4.67
21.8
0.43
1.26
1.88
0.57
1.10
1.88
OLS
33.769
(5.940)
1.627
(0.590)
0.486
(1.017)
0.019
(0.055)
0.231
(0.039)
0.226
(0.058)
0.091
(0.085)
0.76
7.95
32.3
0.55
4.67
21.8
[ APRIL
Sample size is 41. White-corrected standard errors are shown in parentheses. Note R2 does not have its usual interpretation in 2SLS. Instruments in 2SLS include
percent Muslim, percent Catholic, and percent Christian Orthodox; p values in test of overidentifying restrictions are 0.19 in equation 5 and 0.34 in equation 6.
Table 1
Trust, Investment and Growth
2001]
309
22
24
230
220
210
0
Trust (residual)
10
20
30
17
In the smaller sample used by Knack and Keefer (1997), trust was no longer signicant in growth
regressions when investment was included as a regressor.
18
Fukuyama's (1995) popular book on trust emphasises constraints on rm scale produced by low
trust; La Porta et al. (1997) provide some empirical support for this view.
19
It is not obvious that reverse causation from growth to trust should lead to an upward bias for the
trust coefcient in OLS. Olson (1963) among others has argued that rapid growth can disrupt
traditional social structures and ties, which can erode trust.
310
[ APRIL
Porta et al. (1997).20 Equation 4 shows results for the rst-stage regression,
which explains 76% of the variation in trust; in the absence of the religion
variables, income, schooling, and investment goods prices explain only 48%.
Two of the three hierarchical religion variables percent Catholic and percent
Muslim are negatively and signicantly associated with trust.
The exogenous component of trust is positively associated with investment
and growth in (Equations 5 and 6, respectively), with coefcients similar in
magnitude to those estimated using OLS in Equations 1 and 2. Trust is
signicant only at the 0.07 level for a one-tailed test in Equation 5, but is highly
signicant in Equation 6. The instruments are valid, as indicated by p values of
0.19 (Equation 5) and 0.34 (Equation 6) from overidentication tests. In
Knack and Keefer (1997), the trust-growth relationship, while generally robust
to specication changes, was somewhat sensitive to the inclusion of two
inuential observations, Korea and Brazil. Neither these nor any other one or
two observations are very inuential in the tests here, which add 12 countries
not in the Knack-Keefer sample. Using the specications of Equations 1 and 2
from Table 1, robust regression techniques which downweight cases with large
residuals produce coefcients for trust that remain signicant at the 0.01 level
in the case of investment, and at the 0.04 level in the case of growth. Deleting
the three non-WVS observations on trust (Luxembourg, Greece, and New
Zealand), the coefcients (and standard errors) for trust are 0.132 (0.061) for
investment, and 0.070 (0.022) for growth.
The trust-growth relationship in Knack and Keefer was also sensitive to the
choice of human capital measures, with the Barro-Lee school attainment
variable used here producing the weakest partial relationship between trust
and growth. Our results are robust to the choice of human capital variables as
well as to the inclusion of policy variables often included in growth regressions,
such as ination, nancial development, and trade intensity.
The negative (but insignicant) coefcient on initial per capita income in
Equation 2 indicates that other things equal, poorer countries grow faster, on
average, than rich. Relative backwardness does not necessarily help every poor
country, however. Investment sufcient for positive growth is facilitated by
trust between economic agents; backwardness then provides a larger advantage
for a high-trust poor nation than for a low-trust poor nation. This implication
of the model predicts a negative coefcient on the interaction term trust 3
GDP per capita in investment and growth regressions. This prediction is borne
out in Equations 7 and 8. For nations with very low trust levels, coefcients on
initial income are large, positive, and statistically signicant indicating that
backwardness yields no growth advantage over rich nations, despite the
presumption of higher returns to capital and the potential for rapid growth.
For trust levels above 25%, the point estimate for the impact of initial income
on growth turns negative, indicating that backwardness is advantageous for
20
La Porta et al (1997) classify these as `hierarchical religions' with inimical effects on interpersonal
trust. Putnam (1993) and Verba et al. (1995) discuss the implications of Protestant-Catholic differences
in hierarchical vs. congregational organisation, lay participation in the clergy, and in congregation size
for the acquisition of civic skills and interpersonal trust.
2001]
311
See www.transparency.de for the data and a description of underlying sources and methodology.
312
[ APRIL
2001]
Table 2.
Formal Institutions, Heterogeneity and Trust
1
Constant
33.520
(8.209)
0.262
(0.798)
1.871
(1.159)
0.465
(0.290)
0.764
(0.158)
17.723
(8.857)
0.258
(0.958)
1.629
(1.095)
0.559
(0.384)
27.287
(12.261)
0.032
(0.887)
2.029
(1.088)
0.608
(0.357)
4.194
(5.858)
0.149
(0.599)
1.592
(0.805)
0.863
(0.271)
9.819
(5.787)
0.194
(0.495)
3.012
(0.845)
7.141
(4.734)
0.820
(0.399)
0.393
(1.026)
3.763
(5.821)
1.478
(0.529)
1.077
(0.897)
53.734
(9.874)
1.030
(0.868
1.485
(1.305)
4.758
(0.906)
5.631
(1.325)
16.477
(5.837)
0.325
(0.076)
homogeneity squared
Economic discrimination
1.067
(0.402)
0.008
(0.003)
Contract enforceability
Corruption index
0.744
(0.232)
0.205
(0.082)
Investor rights
R2
SEE
N
Mean, D.V.
0.69
9.8
36
32.3
0.65
10.2
36
32.4
0.61
10.6
41
32.1
0.72
9.1
38
32.1
0.73
9.3
33
32.2
5.168
(1.005)
3.451
(0.930)
0.74
8.8
37
32.3
4.415
(0.991)
0.342
(0.114)
0.71
9.5
35
32.1
Equation
0.71
9.7
33
32.6
313
314
[ APRIL
2001]
315
Table 3
Trust, Formal Institutions and Growth (197092)
Equation
Constant
1.471
(1.137)
0.297
(0.129)
0.021
(0.112)
0.026
(0.011)
0.112
(0.028)
1.775
(1.075)
0.269
(0.115)
0.090
(0.092)
0.031
(0.011)
0.088
(0.026)
2.474
(1.655)
0.180
(0.156)
0.047
(0.196)
0.047
(0.016)
3.717
(1.565)
0.134
(0.126)
0.117
(0.168)
0.053
(0.015)
3.870
(0.936)
0.133
(0.110)
0.022
(0.154)
0.037
(0.009)
4.054
(0.897)
0.097
(0.090)
0.099
(0.129)
0.042
(0.010)
1.759
(0.719)
0.709
(0.650)
0.36
1.43
33
2.00
0.062
(0.023)
0.50
1.29
33
2.00
0.278
(0.120)
0.065
(0.153)
0.062
(0.025)
0.45
1.28
39
1.91
0.49
1.19
41
1.88
0.043
(0.017)
0.57
1.11
41
1.88
0.31
1.41
39
1.91
The dependent variable is growth. White-corrected standard errors are shown in parentheses.
Table 4
Trust, Heterogeneity and Growth (197092)
Equation
Constant
7.140
(1.960)
0.170
(0.148)
0.103
(0.159)
0.024
(0.013)
0.071
(0.033)
4.756
(1.844)
0.163
(0.137)
0.003
(0.144)
0.030
(0.013)
0.025
(0.030)
6.459
(1.329)
0.031
(0.106)
0.119
(0.139)
0.039
(0.009)
5.030
(1.147)
0.065
(0.097)
0.019
(0.137)
0.043
(0.009)
4.946
(0.936)
0.035
(0.107)
0.124
(0.135)
0.040
(0.009)
4.353
(0.810)
0.088
(0.102)
0.054
(0.126)
0.044
(0.009)
0.031
(0.013)
0.013
(0.012)
0.38
1.32
40
1.86
0.049
(0.020)
0.46
1.25
40
1.86
0.335
(0.132)
0.078
(0.136)
0.060
(0.022)
0.47
1.26
38
1.87
0.32
1.38
32
2.12
0.049
(0.019)
0.40
1.32
32
2.12
0.34
1.39
38
1.87
The dependent variable is growth. White-corrected standard errors are shown in parentheses.
# Royal Economic Society 2001
316
[ APRIL
institutions, social heterogeneity, or discrimination remain signicantly correlated with growth controlling for trust.25
Each proxy for formal institutions, social distance, and discrimination is
signicantly associated with growth in the rst of each pair of regressions.
While the inclusion of trust reduces the coefcient of the property rights index
by about one-fth, that index remains signicantly related to growth (Table 3,
Equations 1 and 2). Unlike our other proxies for formal institutions, the
property rights index is dened explicitly to include government actions
against private agents, namely expropriation of property or repudiation of
contracts by government. It is therefore not surprising that this index remains
signicantly related to growth, controlling for trust between private agents. In
the model, trust is related to the likelihood of being cheated by private agents,
not by government ofcials. Similarly, our survey-based measure (which
inquires about `most people') is likely to be little affected by perceptions of the
trustworthiness of government agents acting in their ofcial capacities.
Results for each of our other proxies for formal institutions, social distance,
and discrimination in Tables 3 and 4 suggest that much of their inuence on
growth occurs through their impact on trust. In the case of contract enforceability (Table 3, Equations 3 and 4), the corruption index (Equations 5 and 6),
income inequality (Table 4, Equations 1 and 2), land inequality (Equations 3
and 4), economic discrimination (Equations 5 and 6), and ethnic homogeneity, the inclusion of trust drastically reduces the coefcient of the relevant
proxy for formal institutions or social distance. Not included in Table 3 or 4 is
the investor rights index. It is not signicant with or without trust in the
regression, but its coefcient is positive and decreases with the inclusion of
trust, consistent with the interpretation that investor rights improve growth by
increasing trust.26 Trust itself is positively and signicantly related to growth in
every case when it is included in growth regressions with a measure of formal
institutions or of social distance. Results in Tables 3 and 4 strongly support the
model's prediction that formal institutions and social homogeneity increase
growth in part by building trust.27
2001]
317
transact for a single period where cheating by the broker is possible. We show
that cheating is more likely (and trust is therefore lower) when the social
distance between agents is larger, formal institutions are weaker, social sanctions against cheating are ineffective, the amount invested is higher, and the
investors' wages are lower. Most importantly, the model shows that the amount
invested decreases as social heterogeneity increases, and when formal and
informal institutions are weaker, adversely impacting income growth. These
implications have strong support in our cross-country empirical work. Trust,
and the social and institutional factors that affect it, signicantly inuence
growth rates. Thus, this research provides a new insight into the way that social
and institutional factors impact economic performance.
The model in this paper generalises to other principal-agent relationships,
for example, creditors and debtors, employers and employees, clients and
consultants, insurers and insured, and retailers and consumers. Further, our
conceptual denition of trust, and our empirical measures, encompass prisoners' dilemmas as well as principal-agent incentive structures.
Several extensions of the model here would be interesting to undertake.
First, the random matching of transacting agents could be relaxed by allowing the probability of a match between two agents to vary inversely with the
social distance between the two, as in Akerlof (1997). In this case segregation
increases, and time devoted to investigating brokers falls. A second extension
along this line is to permit agents to choose whether or not to trade with
each other using a matching technology as in Burdett and Coles (1997).
Again, this would lead to economic segregation. With sufciently extreme
segregation, time spent investigating approaches zero, and trust the
proportion of time spent working approaches one. There are potentially
enormous costs associated with extreme segregation, however, as gains from
specialisation may be severely limited, particularly where there are many
agent groups, or where a scarce resource is concentrated within one agent
group (e.g., Lebanese entrepreneurs in Africa, or Jewish bankers in medieval
Europe).
Taking into account the value of leisure, and of transactions facilitated by
trust that do not enter the national accounts, the model also predicts that trust
should be positively related to subjective measures of well-being across countries or other economic units. J. S. Mill (1848, p. 131) argued that `The
advantage to mankind of being able to trust one another, penetrates into every
crevice and cranny of human life: the economical is perhaps the smallest part
of it, yet even this is incalculable.' We thus would expect that more inclusive
measures of well-being will be associated with trust in the same way that, as we
have shown here, investment and growth improve with trust.
Claremont Graduate University
The World Bank
Date of receipt of the rst submission: November 1998
Date of receipt of the nal typescript: June 2000
# Royal Economic Society 2001
318
[ APRIL
Appendix
Table A1
Descriptive Statistics
Trust (initial value)
Growth, 197092
Investment/GDP, 197092
Per capita income, 1970
Years of education, 1970
Investment goods price, 1970
Property rights index
Corruption index
Investor rights index
Contract enforceability
Income inequality (circa 1985)
Land inequality (circa 1985)
Ethnic homogeneity
Catholic percent
Muslim percent
Christian Orthodox percent
Mean
Std. Dev.
32.2
1.88
21.8
$5,936
5.43
80.7
37.2
6.2
59.7
2.60
37.4
57.9
81.9
46.6
7.2
2.4
15.0
1.57
6.6
$3,610
2.56
19.7
12.3
2.5
12.7
0.65
9.2
15.8
18.2
40.5
21.5
15.3
41
41
41
41
41
41
41
39
37
33
36
36
41
41
41
41
Table A2
Intercorrelations of Table 1 regressors
2
1. Trust
2. Per capita income, 1970
3. Years of education, 1970
4. Investment goods price, 1970
5. Percent Catholic
6. Percent Muslim
7. Percent Eastern Orthodox
0.68
3
0.29
0.81
4
0.60
0.38
0.31
5
0.57
0.05
0.21
0.14
6
0.31
0.41
0.49
0.16
0.41
7
0.19
0.07
0.02
0.04
0.07
0.04
Appendix: Proofs
Proof. [Proposition 1] Implicit differentiation of the optimality condition, produces
@e i =@a i 1 =a i 11 , @e i =@w i 1=(R a i 11 ), @e i =@ p 12 =11 , @e i =@ 2 11 ,
i
and @e =d(i, j) 13 =11 , where m denotes the partial derivative of
ij [e i , p, D(i, j; )] with respect to the m th argument, and mn is the cross-partial of
the m th and n th arguments of ij . Using the restrictions in the proposition proves the
h
results.
2001]
m
m e
n
319
Proof. [Proposition 3] We prove this proposition for a simple mean preserving spread
(MPS) which guarantees that the spread distribution has a higher variance than the
base distribution (Rothschild and Stiglitz, 1971). As in Proposition 2, the third
derivative restriction on ij results in an e i that is decreasing and concave in w i .
Dene 0 as the distribution of consumers before the MPS, and 1 as the distribution
after the MPS. Proposition 1 of Diamond and Stiglitz (1974) characterises the effect of
a MPS on a concave function. Then, under Theorem 1 of Diamond and Stiglitz (1974),
the following holds
1
1
i
e d 1 . e i d 0 :
0
References
Akerlof, G. (1997). `Social distance and social decisions'. Econometrica, vol. 65(5), pp. 100527.
Alesina, A. and Rodrik, D. (1994). `Distributive politics and economic growth'. Quarterly Journal of
Economics, vol. 109, pp. 46590.
Atkeson, A. (1991). `International lending with moral hazard and risk of repudiation'. Econometrica,
vol. 59, pp. 46590.
Atkeson, A. and Lucas, R. E. (1995). `Efciency and equality in a simple model of efcient unemployment insurance'. Journal of Economic Theory, vol. 66, pp. 6488.
Barro, R. and Lee, J. (1993). `International comparisons of educational attainment'. Journal of Monetary
Economics, vol. 32, pp. 36394.
Baumeister, R. F. and Leary, M. R. (1995). `The need to belong: desire for interpersonal attachments as
a fundamental human motivation'. Pyschological Bulletin, vol. 117(3), pp. 497529.
Becker, G. S. (1974). `A theory of social interactions'. Journal of Political Economy, vol. 82(6),
pp. 106393.
Benabou, R. (1996). `Inequality and growth'. in (B. Bernanke and J. Rotemberg, eds.) Macroeconomics
Annual, Cambridge, MA: NBER and MIT Press.
Berg, J., Dickhaut, J. and McCabe, K. (1995). `Trust, reciprocity and social history'. Games and Economic
Behaviour, vol. 10, pp. 12242.
Bergstom, T. C. (1995). `On the evolution of altruistic ethical rules for siblings'. American Economic
Review, vol. 85(1), pp. 5879.
Blinder, A. and Deaton, A. (1985). `The time-series consumption function revisited'. in (W. C. Brainard
and G. L. Perry, eds.) Brookings Papers on Economic Activity 2.
Burdett, K. and Coles, M. G. (1997). `Marriage and class'. Quarterly Journal of Economics, vol. 62(1),
pp. 14168.
Charny, D. (1990). `Nonlegal sanctions in commercial relationships'. Harvard Law Review, vol. 104,
pp. 373467.
Cosmides, L. and Tooby, J. (1992). `Cognitive adaptations for social exchange'. Chapter 3 in ( J. H.
Barkow, L. Cosmides, and J. Tooby, eds.) The Adapted Mind: Evolutionary Psychology and the Generation
of Culture. Oxford: Oxford University Press.
Dawkings, R. (1976). The Selsh Gene. Oxford: Oxford University Press.
Deaton, A. (1992). Understanding Consumption. Oxford: Oxford University Press.
Deininger, K. and Squire, L. (1996). `A new data set: measuring income inequality'. World Bank Economic
Review, vol. 10, pp. 56592.
Diamond, P. A. and Stiglitz, J. E. (1974). `Increases in risk and risk aversion'. Journal of Economic Theory,
vol. 8, pp. 33760.
Easterly, W. (2000). `The middle-class consensus and economic development'. World Bank Policy
Research Working Paper no. 2346.
Easterly, W. and Levine, R. (1997). `Africa's growth tragedy: policies and ethnic divisions'. Quarterly
Journal of Economics, vol. 112(4), pp. 120350.
Frank, R. (1987). `If homo economicus could choose his own utility function, would he want one with a
conscience?' American Economic Review, vol. 77(4), pp. 593604.
# Royal Economic Society 2001
320
[ APRIL
Fukuyama, F. (1995). Trust: Social Virtues and the Creation of Prosperity. The Free Press.
Glaeser, E. L., Laibson, D., Scheinkman, J. and Soutter, C. (2000). `Measuring trust'. Quarterly Journal of
Economics, vol. 115(3), pp. 81146.
Gold, H. and Webster, A. (1990). New Zealand Values Today. Palmerston North, New Zealand: Alpha
Publications.
Granovetter, M. (1985). `Economic action and social structure: the problem of embeddedness'.
American Journal of Sociology, vol. 91(3), pp. 481510.
Greif, A. (1989). `Reputation and coalitions in medieval trade: evidence on the Maghribi traders'.
Journal of Economic History, vol. 49(4), pp. 85782.
Greif, A. (1994). `Cultural beliefs and the organization of society: a historical and theoretical reection
on collectivist and individualist societies'. Journal of Political Economy, vol. 102(5), pp. 91250.
Grossman, H. L. and Kim, M. (1995). `Swords or plowshears? A theory of the security of claims to
property'. Journal of Political Economy, vol. 103, pp. 127588.
Hausman, J. A. (1979). `Individual discount rates and the purchase and utilization of energy-using
durables'. Bell Journal of Economics, vol. 10(1), pp. 3354.
Hermalin, B. E. and Katz, M. L. (1991). `Moral hazard and veriability: the effects of renegotiation in
agency'. Econometrica, vol. 59(6), pp. 173553.
Horowitz, D. L. (1985). Ethnic Groups in Conict. University of California Press.
Huang, C. and Litzenberger, R. H. (1988). Foundations for Financial Economics. Amsterdam: NorthHolland.
Inglehart, R. et al. (2000). `World values surveys and european values surveys, 19811984, 19901993,
and 19951997'. ICPSR Study No. 2790, Inter-University Consortium for Political and Social
Research, Institute for Social Research.
Jazairy, I., Alamgir, M., and Panuccio, T. (1992). State of World Rural Poverty: An Inquiry into its Causes and
Consequences. International Fund for Agricultural Development.
Kandori, M. (1992). `Social norms and community enforcement'. Review of Economic Studies, vol. 59,
pp. 6380.
Kawachi, I., Kennedy, B. P., Lochner, K. and Prothrow-Smith, D. (1997). `Social capital, income
inequality, and mortality'. American Journal of Public Health, vol. 87(9), pp. 14918.
Keefer, P. and Knack, S. (2001). `Polarization, property rights, and the links between inequality and
growth'. Public Choice, forthcoming.
Khan, A. and Ravikumar. B. (1998). `Enduring relationships in an economy with capital'. Working
paper, University of Iowa.
Klein, D. B. (ed.) (1997). Reputation: Studies in the Voluntary Elicitation of Good Conduct. University of
Michigan Press.
Knack, S. and Keefer, P. (1995). `Institutions and economic performance: cross-country tests using
alternative institutional measures'. Economics and Politics, vol. 7, pp. 20727.
Knack, S. and Keefer, P. (1997). `Does social capital have an economic payoff? A cross-country
investigation'. Quarterly Journal of Economics, vol. 112(4), pp. 125288.
La Porta, R., Lopez-de-Silanes, F., Shleifer, A. and Vishny, R. W. (1997). `Trust in large organisations'.
American Economic Review Papers and Proceedings, vol. 87(2), pp. 3338.
La Porta, R., Lopez-de-Silanes, F., Shleifer, A. and Vishny, R. W. (1998). `Law and nance'. Journal of
Political Economy, December, vol. 106(6), pp. 111355
Marcet, A. and Marimon, R. (1992). `Communication, Commitment and Growth'. Journal of Economic
Theory, vol. 58(2), pp. 21949.
McCabe, K. A., Rassenti, S. J. and Smith, V. L. (1997). `Reciprocity, trust and payoff privacy in extensive
form bargaining'. Working Paper, University of Arizona.
Mill, J. S. (1848). Principles of Political Economy. London: John W. Parker.
Mukherji, A. and Nagarajan, N. (1995). `Moral hazard and contractability in investment decisions'.
Journal of Economic Behavior and Oganization, vol. 26, pp. 41330.
Murphy, K., Shleifer, A. and Vishny, R. W. (1989). `Income distribution, market size, and industrialization'. Quarterly Journal of Economics, vol. 104, pp. 53764.
New York Times (1997). `Neglect charge dropped in Danish baby case'. May 22, p. B4.
North, D. C. (1990). Institutions, Institutional Change and Economic Performance. Cambridge: Cambridge
University Press.
Olson, M. (1963). `Rapid growth as a destabilizing force'. Journal of Economic History, vol. 23,
pp. 52952.
Perotti, R. (1996). `Growth, income distribution, and democracy: what the data say'. Journal of Economic
Growth, vol 1(2), pp. 14987.
Persson, T. and Tabellini, G. (1994). `Is inequality harmful for growth? Theory and evidence'. American
Economic Review, vol. 84, pp. 60021.
Phelan, C. (1995). `Repeated moral hazard and one-sided commitment'. Journal of Economic Theory,
vol. 66, pp. 488506.
# Royal Economic Society 2001
2001]
321