The Internet is expected to be an important source of economic growth in the 21^st century. The Congressional Budget Office¹ predicts the U.S. economy will grow at an annual rate of 2.1 percent over the coming decade—an increase of 0.9 percent over U.S. growth for the period 1974 to 1995. Varian et al.² estimate that the Internet will account for 48 percent of this increase in growth. In a similar vein, Litan and Rivlin³ discuss research estimating Internet-driven productivity gains in U.S. manufacturing of 0.2 and 0.4 percent per year. Since the Internet dramatically reduces the cost of transmitting information, the costs associated with the distribution of goods and services between businesses, between businesses and consumers, and between businesses and their employees are reduced as well, accounting for these expected gains in productivity.

Whether predictions regarding the contribution of the Internet to economic growth come to pass depends upon whether people and firms choose to adopt the Internet and how fully they embrace the idea of conducting business over it. The degree to which people and firms adopt Web-based activities will depend on how willing they are to accept the greater anonymity and associated possibilities for opportunism inherent in Web-based transactions. This willingness may, in turn, depend on how much people trust each other. If trust does influence Internet adoption, it will have an indirect impact on economic growth rates among nations through its influence on the adoption of this growth-enhancing technology.

In addition to the possibility of an indirect impact of trust on growth, there is evidence that trust directly impacts economic growth and growth rate differences across countries. Prior to the late 1990s, economic growth rates were explained almost exclusively in terms of labor and capital endowments and differences in how these endowments are augmented by capacities for technological change. Differences in the prosperity of nations or regions relative to others are, in some cases, difficult to explain in terms of differences in these standard economic variables. During the 1990s, spurred largely by observations and arguments put forth by social theorists like Fukuyama⁴ and Putnam et al.,⁵ economists investigated the possibility that differences in economic growth might stem directly from differences in the extent to which members of different cultures were willing to trust each other. The arguments in favor of this possibility are straightforward. Almost all transactions involve some opportunities for misrepresentation, non-compliance, or outright fraud. Detailed contracts, extensive monitoring of performance, and litigation are means of discouraging such behaviors, but they are all costly to implement. Almost all transactions involve some opportunities for misrepresentation, non-compliance, or outright fraud. Empirical evidence suggests that mutual trust is an efficient substitute for these enforcement mechanisms. For example, Dyer and Chu⁶ examined differences in procurement costs in 453 supplier-automaker relationships in the U.S., Japan, and South Korea. Procurement costs incurred in situations where the suppliers trusted automakers the least were five times higher than those in which the suppliers trusted automakers the most, while the costs associated with negotiating contracts and post-contractual disputes were double.

Trust appears to have significant returns at the macroeconomic level as well. Knack and Keefer,⁷ for example, found that a very simple measure of how trusting inhabitants of different countries are was a significant explanatory variable in regressions of average annual growth rates in per capita income from 1980 to 1992. Moreover, the impact was very large—a 10 percent increase in the measure of trust translates into an increase of 0.1 percent in economic growth—a sizable increment, given world average growth rates of 1 to 3 percent in the latter half of the 20^th century.

The fact that trust directly impacts economic growth through reductions in transaction costs, coupled with the possibility that it may impact growth indirectly to the extent that it impacts Internet adoption rates, raises a troubling possibility: namely, that low-trust countries, the majority of which tend to be of low and middle income, will take a double hit in terms of economic growth in the coming years. They may be penalized for low trust by incurring higher transaction costs and by lower adoption rates of growth-enhancing technology. Knack and Keefer's⁷ findings suggest that the first effect, higher transaction costs, will surely come to pass. Whether the second, lower Internet adoption rates, does as well depends upon whether trust does, in fact, encourage Internet adoption. Our objective in this paper is to determine whether this proposition is true. To presage our findings, it is. This result would seem to suggest that efforts to increase trust in low and moderate trust countries are in order. Unfortunately, we show that the returns for any such policy will be greater for high-trust rather than for low-trust countries, so that differences in trust among countries, will promote an increasing digital divide between them. To the extent that contributions the Internet makes to economic growth accrue disproportionately to high trust countries, this digital divide will translate into a developmental divide.

Data

The specifics of our analyses of the impact of trust on Internet adoption are dictated by the availability of trust measures for different countries. In their examination of whether trust directly influences economic growth rates, Knack and Keefer⁷ used responses to a question involving trust posed to thousands of respondents from 29 countries with market economies in the 1981 and 1990–1991 World Values Survey (WVS).⁸ The question was, “Generally speaking, would you say that most people can be trusted, or that you can't be too careful in dealing with people?” Knack and Keefer took the percentage of respondents from each country who answered that people could be trusted as a measure of how “trusting” that country's populace was.⁹ Then they conducted regression analyses examining the impact of this measure of trust on average annual growth in per capita income for 1980 to 1992. They found that trust contributes significantly to economic growth, particularly in poorer countries without developed legal enforcement systems.¹⁰

The growth rates in Knack and Keefer⁷ were averages over the period 1980–1992. To minimize endogeneity problems, specifically, the possibility that economic growth rates have an impact on levels of trust, they computed trust values based on 1980 WVS responses where possible and 1990 responses otherwise. Knack and Zak¹¹ provide trust measures derived from responses to the 1995 WVS for 17 of the 29 countries used in Knack and Keefer⁷ and 1990 values for 11 of the others. (No recent trust measure is available for Nigeria, the 29^th country in the Knack and Keefer study.) Given that the Internet was not commercialized until 1995, endogeneity is not an issue in our analyses, so we use the most recent 1995 data where possible and 1990 values otherwise. None of the results reported in the ensuing sections are particularly sensitive to whether we employ a combination of values, or exclusively 1990 values. Values for this trust variable for each country in Knack and Keefer's original study (excluding Nigeria), as well as values for all other independent and dependent variables considered in our analyses are shown in Table 1.

For each of the 28 countries for which we had a trust measure, we tried to collect two measures of Internet penetration. The Organisation for Economic Cooperation and Development (OECD) provides data on the percentage of households with Internet access in 1999 and/or 2000 for 17 countries. To maximize available degrees of freedom, we combined this data, taking the average for countries with 1999 and 2000 data and the single year data for the remaining countries, to create data on the percent of households with Internet access for 1999–2000, denoted “IP1.”¹² OECD also provided data on the number of Internet subscribers per 100 inhabitants in 2000 for 22 of these countries (denoted “IP2”).¹³

The literature on the determinants of technology adoption suggests a number of economic, demographic, and infrastructural factors that might influence Internet adoption. A number of economic, demographic, and infrastructural factors might influence Internet adpotion. Economic theory suggests that the quantity of a product that is demanded depends on its own price, the buyers' income, and the price of substitutable and complementary goods. For our measure of income, we computed the average per capita national income for our sample of countries by averaging data provided by the World Bank for the period 1995–1999.¹⁴ This variable is denoted “Income.” Our measure of Internet access price, denoted “Int. Price,” is the average price of 20 hours of Internet access for 1995–2000 in dollars adjusted for purchasing power parity, as computed by OECD.¹⁵

In addition to variables suggested by economic theory, there are a host of demographic characteristics that have been found to influence the adoption of new technologies. Young people, those with more education, and those who are more cosmopolitan are all more disposed to new technologies. To examine the role of age, we collected data on the percentage of the population 60 and older, as reported by the United Nations.¹⁶ We denote this variable “Age.” The impact of education on adoption is captured by the variable “Education,” which reports the average number of years of schooling among the population 25 and older, and is taken from Barro and Lee.¹⁷ As a measure of cosmopolitanism, we average data from the World Bank on the urban population as a percent of the total population for years 1995 through 1999. This variable is denoted “Urban.”

In addition to explanatory variables generally found to influence the adoption of new technologies, there are others associated with the specific characteristics of the Internet. To use the Internet, one must have a personal computer or other device and a means of connecting to the Web—a phone line or an alternative. As such, PC usage/availability and the level of infrastructure development as measured by main phone lines are other reasonable candidates for explaining Internet penetration. Our measure of PC penetration was derived from the estimated number of self-contained computers designed to be used by a single individual per 1000 inhabitants, obtained from the World Bank World Development indicators for the years 1995 through 1999.¹⁸ Data on each country was averaged over the five-year period to construct the average PCs per 1000 inhabitants, denoted “PC.” To gauge the ability of people in different countries to connect to the Internet, we collected data on the average number of telephone mainlines per 1000 population for the period 1995–1999 reported by the World Bank¹⁹ for each of our sample countries. This variable is denoted “Lines.”

Analysis

As a first attempt at testing the proposition that trust is an important factor in Internet adoption, we consider the simple linear regressions and scatter plots of IP1 and IP2 with respect to trust as shown in Figures 1 and 2. In the case of IP1, the correlation with trust explains 64 percent of the total variation in Internet adoption.

Figure 1    Figure 2

For Internet subscribers per 100 inhabitants (IP2), shown in Figure 2, the data point for South Korea is not plotted, as it would constitute an extreme outlier and would not be a fair comparison with other countries. The reason for this is that South Korea has the largest proportion of Internet subscribers in the sample (23/100) but a trust value slightly below the mean (30 versus 36). South Korea's front-runner position in terms of Internet subscribers has been attributed to the coincidence of a number of factors,^20–22 most notably overcapacity in fiber-optic cable and a government policy promoting competition among Internet access providers. Fiber-optic overcapacity has been absorbed through provision of broadband Internet providing connection speeds roughly 20 times those achieved through traditional phone lines. Moreover, given the competition among providers and the peculiarities in the way charges for traditional phone usage are calculated, this broadband access is provided at low prices, roughly comparable to service over phone lines. When South Korea is dropped from the IP2 series, the fit of the regression shown in Figure 2 is comparable to that obtained using IP1.

These simple univariate linear regression results support the contention that trust is an important determinant of Internet adoption although, as noted earlier, there are a host of economic, demographic and infrastructural variables that might explain adoption as well. To flesh out what the determinants of Internet adoption are and rule out the possibility that the observed contribution of trust to adoption of this technology is spurious, we conducted multivariate regressions on IP1 and IP2. Because our dependent measures are proportions, we subjected both to the inverse-logit transformation F^-1(y) = ln(y/1 – y). Here F is the cumulative distribution function for the logistic distribution and F^-1 is its inverse. The transformed dependent variables are regressed against the relevant independent variables by using ordinary least squares regression.^23,24

In light of the relatively small number of countries for which we have complete data, compared to the large number of potential explanatory variables, two sets of regression results are reported for each dependent measure. In the first set, all relevant regressors are run against the corresponding dependent variables, and the results are examined to see whether trust is a significant factor when all other potentially relevant variables are controlled. In the second set of regressions, a stepwise procedure is employed to examine whether our trust variable explains Internet adoption across countries in equations containing only statistically significant explanatory variables.²⁵

Regressions of IP1 and IP2 are of the following general form:

F^-1(IP1(2)) = ₀ + ₁Trust + ₂Income + ₃Int.Price + ₄Age + ₅Education + ₆Urban + ₇Lines + ₈PC

where ₀ is the intercept term and the remaining _i's are the values of the partial derivatives of the dependent variable resulting from unit changes in the independent variables, all else being equal. For regressions of IP1 (shown in the first two columns in Table 2), Lines, PC, and Trust enter at better than the .05 significance level in the “all regressor” estimation. In the stepwise regression, Lines and PC enter significantly at the .05 level, while Trust and Income (with an unanticipated negative sign) enter at p = .055 and p = .071, respectively. The proportion of the total variation in IP1 accounted for in these estimations is quite high, with adjusted R² (coefficient of determination) for both equal to .87.

The “all regressor” estimations for IP1 and IP2 both exhibit high multicollinearity. This is not surprising, given the small number of observations compared to the number of independent variables and the relatively high correlation between many of the independent variables. Multicollinearity is not a problem in the stepwise regressions reported. Residuals in all of the regressions reported tend to be randomly dispersed.

Regression results for IP2 (shown in the center two columns of Table 2) reflect some similarities to those obtained for IP1 but also important differences. Regarding the similarities, Lines and Income are selected as significant explanatory variables in both “all regressor” and stepwise regressions. PC is not, however, significant in explaining IP2 nor does Trust enter as significant in either of the IP2 equations. Instead, Education enters significantly in the stepwise regressions of IP2. The adjusted R²s for these equations, .70 and .69, are high, although lower than those for IP1.

Many of the discrepancies between results obtained for IP1 and IP2 are due to the presence of South Korea in the IP2 series. Withholding South Korea from the estimation of IP2 produces several consequences as shown in the right-hand columns in Table 2. First, the fit of the equations to the data improves substantially—making them comparable to those obtained using IP1. Second, the importance of Average PCs per 1000 (p = .137 versus p = .823) increases, although this variable is still shy of significance. Third, education becomes an insignificant explanatory variable in the stepwise as well as “all regressors” estimation. Finally, trust becomes a statistically significant explanatory variable in both regressions.

In summary, regression results obtained for the average percentage of households with Internet access suggest that Internet adoption depends not only upon technological preconditions—PCs and phone lines—but also on trust. If we are willing to exclude South Korea as an anomaly from observations of Internet subscribers, the results obtained using IP2 corroborate the importance of trust and phone lines.

Our findings regarding the importance of needed infrastructure are consistent with results reported in Hargittai²⁶ and Robison et al.²⁷ in which the number of main phone lines per 1000 inhabitants was found to be an important explanatory variable in regressions of Internet hosts per 1000 inhabitants across nations.²⁸ Diez-Picazo²⁹ reports regression results from an analysis of pooled cross-sectional and time series data on hosts per 1000 inhabitants, in which the number of personal computers per capita in the previous year enters significantly. Internet adoption depends not only upon technological preconditions— PCs and phone lines— but also on trust. Finally, there is some evidence consistent with the importance of trust. In their analysis of hosts, Robison et al.²⁷ found that the level of “political openness,” (an index measuring how democratic different countries are in terms of elective government and constitutional constraints on political power), positively influences Internet penetration. It seems reasonable to expect that people in societies characterized by “fair” institutions will be more willing to trust than people living in societies in which the government is less accountable. Knack and Keefer⁷ report regression results to this effect.

To the extent that Internet usage promotes economic growth, our findings would seem to suggest that policy makers, particularly those in low-trust countries, should consider formulating programs to increase trust. Whether this is advisable depends first on the extent to which the crude measure of trust we use really reflects differences in how much people trust in different cultures. If it does, the next question concerns what to do—what programs can a government implement to encourage trust? Finally, there is the question of impact—assuming trust-enhancing policies exist, what kind of return can a society expect to receive by investing in them? We address each of these questions in turn.

Measuring trust. Trust is clearly a difficult variable to measure, and it is natural to ask whether responses to the simple survey question contained in the WVS provide a good measure. An obvious issue here concerns what people have in mind when they respond to the WVS survey question. The hope is that the responses reflect a general willingness to put oneself at risk or a general expectation regarding others, and not a willingness to trust some specific group or to trust others in a specific circumstance. To the extent that the WVS question is silent regarding groups or circumstances, the latter seems less likely. Moreover, Knack and Keefer⁷ note that the correlation between the WVS question concerning trust in one's family members and the general trust question is low. They also discuss evidence from a Readers Digest study reported in The Economist (June 22, 1996) in which wallets containing 50 dollars and the owner's address were “lost” in 14 European and 12 U.S. cities. The percentage of wallets returned by country correlates highly (.67) with the WVS-based trust measure.

The Knack and Keefer trust measure also tends to agree with results from experiments comparing how trusting people from different countries are when playing simple trust games. In these games, one player (the sender) is given some amount of money, for example, ten dollars, and may send any portion of it to a second player (the receiver). Any amount sent is increased by a known multiple (e.g., doubled) before it is given to the receiver. The receiver then decides how much, if any, to send back to the sender. The amount sent by the sender is a measure of trust, while the amount sent back is a measure of trustworthiness. Willinger et al.³⁰ find Germans more trusting than the French in these games, while Buchan et al.³¹ find that mainland Chinese participants (with a value of 56 for the Knack and Kiefer trust measure) are more trusting than U.S. participants who are, in turn, more trusting than Japanese and Korean players. All of these orderings, except for the ranking of the U.S. above Japan, are consistent with the ordering reflected in the Knack and Keefer measure.

Recent studies that compare subjects' survey responses with their behavior in trust games have produced conflicting results. Glaeser et al.³² examined the extent to which Harvard undergraduates' responses to the WVS trust question predicted the amount they sent to a counterpart in a trust game. They found that responses to the trust question didn't predict the amounts sent (i.e., how trusting players are) but did predict amounts sent back when respondents were in the position of the receiver (i.e., how trustworthy they are). Fehr et al.³³ conducted a similar study in the context of a representative survey of German households. They report the opposite results—that responses to the WVS question are a significant predictor of trusting but not trustworthy behavior in the trust game.

Building trust—an open question. To the extent that trust impacts economic growth directly by reducing transaction costs and indirectly by encouraging Internet adoption, policies aimed at increasing trust would seem in order. What these policies are and, indeed, whether they exist, depends on the factors that lead people to trust others. It may be that people responding to the question of whether others can be trusted answer affirmatively because they live in societies where formal mechanisms (e.g., property rights and legal statutes) and/or informal conventions (e.g., widely shared norms regarding the sanctioning of unfair or unethical behavior) assure that in potentially contentious situations it is, in fact, best for the parties involved to behave cooperatively.^34,35 Such environmental factors are subject to influence through policies. In this vein, Zak and Knack³⁶ examine prospects for increasing trust (and thus growth) through measures designed to build civic culture, enhance contract performance, increase freedom of association, reduce income inequality, and raise educational levels.

An alternative, and not mutually exclusive, reason people in some nations may be more trusting than others is because they are simply psychologically or culturally predisposed to expect others to behave benevolently.^37,38 It is not obvious what sorts of policies might be pursued to implement changes in such cultural propensities. The fact that WVS responses regarding trust are highly correlated over time (e.g., from 1980 to 1990 and 1995) may suggest that these cultural propensities are quite stable and not amenable to either unintended or intended manipulation.^39,40

Yet a third, and again not mutually exclusive, interpretation of responses to the WVS question about trust is that it indicates not only an attitude regarding willingness to trust people but also a willingness (or unwillingness) to trust technologies. The product adoption literature⁴¹ classifies groups of individuals according to their propensity to adopt new products. “Innovators” are characterized as venturesome and as risk-taking, whereas those people in the “late majority” and “laggard” groups are described as skeptical and suspicious. To the extent that these attitudes toward risk and propensities to suspicion apply generally, as assumed, for example, in the standard models of decision-making under uncertainty, propensity to trust and willingness to adopt new technologies will be positively correlated.^42,43 Once again it is not obvious how general propensities of this type would be altered through governmental policies.

The comparative static analysis of trust and Internet adoption. Determining what policies to pursue to promote trust within a country is a difficult proposition. In contrast, ascertaining the impact of such policies and how the impact varies across countries is fairly straightforward. To demonstrate this, imagine that all countries invest an equal amount of funds in policies to promote greater trust and receive the same proportionate increment to their trust score as a consequence. To calculate the impact of these proportionate changes in trust on Internet adoption rates, we use the models obtained from the stepwise regression exercises for IP1 and IP2 (excluding South Korea). For each dependent variable y_i, let y*_i = F(x_i) be our predicted value. In this case, the proportional impact on y resulting from a percentage change in trust (i.e., the elasticity of y with respect to trust) is:

Notice that under the logistic distribution, the estimated trust elasticity for any country is simply the estimated coefficient for trust multiplied by that country's level of trust. The estimated elasticities of Internet penetration with respect to trust for all countries except South Korea are shown in Table 3, where countries are sorted from low to high in terms of their trust levels, with the mean responses shown at the bottom. This sorting of the scores highlights the basic implication of this comparative static exercise regarding how increases in trust translate into increases in adoption: High-trust countries will benefit proportionately much more from their investments in trust than will low-trust countries.

To see how these results translate in terms of growth rates in Internet adoption, suppose each country adopts a policy that improves its trust scores by 5 percent per year.⁴⁴ For a country with the average number of Internet subscribers (IP2), this policy produces the series of growth rates depicted by the center line in Figure 3. As depicted, the growth rate in Internet subscribers increases from approximately 4 percent to 6¼ percent. This translates into an increase from a current subscription level of 13 percent to a subscription level of 21 percent by 2010. In Norway, the most trusting country in the sample, trust reaches 100 percent by the year 2010 with an associated Internet subscription level increasing from 16 percent to 35 percent. In contrast, for the lowest-trust country, Brazil, this policy only stimulates the growth rate from 0.35 percent to 0.5 percent over the 10-year period. The impact of this 10-year policy of 5 percent annual growth in trust is to increase Internet subscription from 1.6 percent to 1.67 percent!

Figure 3

Whether it makes sense for countries to promote Internet adoption through policies to enhance trust or through investment in main phone lines depends upon how the costs of the different policies compare with their relative benefits. Our analyses enable us to characterize the benefits side of this equation. To demonstrate, note that the impact of a unit change in the level of trust on our dependent measures is given by:

Similarly, the impact of a unit change in the number of main phone lines is given by _LinesF(x_i). These expressions indicate a property of the logistic model; namely, that countries with larger predicted levels of Internet adoption reap larger absolute benefits from unit changes in any independent variable. The ratio of the benefits accruing from a unit change in trust versus a unit change in main lines is simply the ratio their corresponding regression coefficients, _Trust/_Lines.⁴⁵ As such, to justify investments in trust so as to increase Internet subscribers (our IP2 measure) by 1 unit (1 percent), the cost of doing so must be less than 77 percent (i.e., _Trust/_Lines equals 0.023/0.030) of the cost of increasing Lines by 10 units. Similar computations can be made for our other dependent measures with respect to their relevant policy variables.

Conclusions

Trust has been found to have a direct influence on economic growth across countries through its impact on the cost of transactions. In this paper, we hypothesized that trust may also have an indirect impact on economic growth across nations with the Internet impacting growth rates and trust impacting adoption of the Internet. Our results suggest that trust does, in fact, influence Internet adoption. Since low-trust countries tend to be low- or middle-income countries, this will result in a digital divide between these countries and higher-trust, higher-income ones. To the extent that the level of Internet adoption influences economic growth, this digital divide will translate into a developmental divide.

How large this divide will be is, at present, unknowable. It seems safe to assume that any growth dividend accruing from the Internet increases at least linearly as Internet adoption rises. If network effects are relevant, then the relationship between Internet penetration and a growth dividend will be stronger for greater levels of adoption. While policies designed to encourage trust among low-trust nations would seem to be a means of mitigating these digital and developmental divides, the implications of our comparative static analyses are not encouraging: High-trust countries benefit more from such policies. Of course, it is possible that there are policies that might effectively and significantly increase trust at low cost. Further research to understand the implications of trust measures will be needed to determine what such policies might entail.

Cited references and notes