University of Zurich

Department of Economics

Working Paper Series

ISSN 1664-7041 (print)
ISSN 1664-705X (online)

Working Paper No. 352

What Do Cross-Country Surveys Tell Us About

Social Capital?

David Tannenbaum (r) Alain Cohn (r)

Christian Lukas Zünd (r) Michel André Maréchal

Revised version, February 2022

Electronic copy available at: https://ssrn.com/abstract=3639328

 What Do Cross-Country Surveys Tell Us About Social
Capital?
David Tannenbaum ⃝
r Alain Cohn
Christian Lukas Zünd ⃝
r Michel André Maréchal*

February 2022

Forthcoming at Review of Economics and Statistics

Abstract

We assess the predictive power of survey measures of social capital with a new behavioral
data set that examines whether citizens report a lost wallet to its owner. Using data from more
than 17,000 “lost” wallets across 40 countries, we find that survey measures of social capital —
especially questions concerning generalized trust or generalized morality — are strongly and
significantly correlated with country-level differences in wallet reporting rates. A second finding
is that lost wallet reporting rates predict unique variation in the outputs of social capital, such as
economic development and government effectiveness, not captured by existing measures.
Keywords: Social capital, Trust, Honesty, Field Experiment, Surveys

* Authorship order for this paper was randomly determined (denoted by the ⃝ r symbol). Tannenbaum: Department
of Management, University of Utah (email: [email protected]); Cohn: School of Information, University
of Michigan (email: [email protected]); Zünd: Department of Economics, University of Zurich (email: chris-
[email protected]); Maréchal: Department of Economics, University of Zurich (email: [email protected]).
We gratefully acknowledge financial support by the Gottlieb Duttweiler Institute.

Electronic copy available at: https://ssrn.com/abstract=3639328

 Social capital is considered a fundamental factor underlying persistent differences in economic
development (Arrow 1972; Mokyr 2009), and has been linked to country-to-country differences
in growth, innovation, crime, governance, and institutional performance (Algan and Cahuc 2010;
Djankov et al. 2003; Fountain 1998; Knack and Keefer 1997; Lederman, Loayza, and Menendez
2002; Guiso, Sapienza, and Zingales 2004; La Porta et al. 1997; Putnam, Leonardi, and Nanetti
1993; Tabellini 2008). While the concept of social capital has a number of distinct interpretations,1
in this paper we focus on the aspects of social capital thought to be most clearly tied to economic
development and productivity — namely, the shared values, beliefs, and expectations that limit
opportunistic behavior and facilitate cooperation among strangers. This includes the extent that
individuals believe others to be honest and trustworthy, and are willing to cooperate and reciprocate
in kind (Fukuyama 1995; Guiso, Sapienza, and Zingales 2011).
One of the more difficult issues in studying social capital is how to properly measure the
construct. Standard practice across the social sciences has been to use data from large-scale surveys
administered across countries,2 such as the World Values Survey (WVS). While cross-country
surveys have a number of attractive features, this approach also comes with several limitations. For
example, cross-country comparisons of survey data are potentially biased due to cultural differences
in how questions are interpreted (e.g., Laajaj et al. 2019), how participants make use of response
scales (Johnson et al. 2005), and the degree that responses are influenced by social desirability
concerns (Bernardi 2006). That such response biases may vary across countries presents a challenge
for drawing clean comparisons through the use of survey data.3 Layered on top of these concerns is
whether survey responses translate to concrete, meaningful behaviors (Bertrand and Mullainathan
2001; Krosnick 1999). For instance, survey data that measure perceptions of corruption have
been found to correspond poorly with observable levels of corruption, both when looking at
1. Social capital has also been conceptualized as the advantages and opportunities afforded to individuals through
their social networks or membership in certain communities (Bourdieu 1986; Putnam, Leonardi, and Nanetti 1993), as
any form of shared knowledge, norms, and rules among group members (Ostrom 2000), and as social organization that
can not be achieved through individual means (Coleman 1990). While some of these formulations incorporate notions
of reciprocity and trust, they also tend to emphasize the number and strength of social ties to a community. Others
have argued that the concept of social capital should be abandoned, as the construct often includes components that are
conceptually distinct and often empirically uncorrelated with one another (see Bjørnskov and Sønderskov 2013, for a
critical review).
2. In addition to surveys, past research has explored experimental measures of trust in economic games (Glaeser
et al. 2000; Karlan 2005), but experimental laboratory measures are not widely available at the scale needed for
cross-country comparisons (see Fehr et al. 2003 and Bellemare and Kröger 2007 for representative experiments in
Germany and the Netherlands). To what extent behavior in the lab generalizes to the field is still an open question (Levitt
and List 2007). Besides experiments and surveys, researchers have also used behavioral proxies of social capital such
as voter turnout and blood donations per capita (Guiso, Sapienza, and Zingales 2004; Nannicini et al. 2013). However,
these behaviors are subject to different rules and regulations across countries, and for this reason have primarily been
used to examine variation in social capital within rather than across countries.
3. Many survey measures are also conducted using face-to-face interviews, which likely exacerbate the varia-
tion in response bias across countries (e.g., for the WVS, see https://www.worldvaluessurvey.org /AJDocumenta-
tion.jsp?CndWAVE=7).

Electronic copy available at: https://ssrn.com/abstract=3639328

the correspondence within communities (Olken 2009) and across countries (Razafindrakoto and
Roubaud 2010). These limitations raise the possibility that survey measures may have little
correspondence with more concrete measures of social capital.4
In this paper we provide a portable, cross-country behavioral benchmark for studying social
capital. We recently conducted a large international field experiment in which we turned in lost
wallets with varying amounts of money at public and private institutions and measured whether
recipients attempted to return the wallets (Cohn et al. 2019). In total, we deposited 17,303 wallets
in 355 cities spanning 40 countries. Returning a lost wallet implies elements of honesty towards
strangers as well as general prosocial preferences such as altruism, and as such can be seen as a
representative instance of social capital (Gintis 2016).5,6 Much like in large-scale surveys our data
is designed to be portable across countries, but unlike surveys our data examines real behavior
in a naturalistic setting (i.e., recipients are presumably unaware their behavior is observed by
experimenters). In section 3 of the online appendix, we report a number of robustness checks that
support the validity of our lost wallet paradigm, including that wallet reporting rates are unlikely to
be confounded by risk of detection, legal regulations, beliefs about finder’s fees, or cross-country
differences in email usage. Furthermore, Figure 1 illustrates that wallet reporting rates in our
experiment are negatively correlated with objective behavioral proxies of dishonest behavior (such
as cheating on taxes and corruption of public officials) that are available at the regional level for the
US and Italy (see section 4 of the online appendix for details). Together, the results suggest that our
measure captures important variation in social capital across countries.
We compare the performance of a number of widely-used survey measures of social capital
4. More generally, our study adds to the literature on the correspondence between survey responses and behavior
in natural settings. Evidence points to a mixture of both accurate and biased correspondence between the two. For
instance, when survey responses are compared to administrative records, respondents often exaggerate engaging in
a number of socially-desirable behaviors, such having voted in mayoral elections, donating to charity, or reporting
their college grade point average (Kreuter, Presser, and Tourangeau 2008; Parry and Crossley 1950). At the same time,
research has also found that survey measures can often correspond surprisingly well with behavior in natural settings
(Dohmen et al. 2011; Hainmueller, Hangartner, and Yamamoto 2015; Epper et al. 2020; Maréchal et al. 2022; Sunde
et al., in press).
5. In reviewing the literature on different measures of social capital, Paldam (2000) states “the famous wallet-test is
an attempt to measure trust in a more general way: Here N wallets are ‘forgotten’ in public places and the test is how
many that are handed back” (p. 644). Survey data also suggests that returning a lost wallet is viewed as an act of social
capital. For example, when surveying nationally representative samples in the United Kingdom, Poland, and the United
States, we found that the large majority of respondents (89%) viewed keeping a lost wallet as “somewhat inappropriate”
or “very inappropriate” (Cohn et al. 2019).
6. Knack and Keefer (1997) compared trust scores to data from a lost-wallet study conducted by Reader’s Digest.
However, the sample size of the Digest study was considerably smaller than our current data set, at only 400 wallets
total. Furthermore, this data is compromised by potential confounds which our study took explicit steps to remove.
As an example, Reader’s Digest dropped wallets in public spaces, which allows for selection effects (i.e., individuals
who select into the study by deciding to pick up a lost wallet may be different from those who do not). By contrast, in
our study we returned lost wallets to employees at the front desk of different institutions, thereby providing greater
experimental control over who participated in the experiment. For these reasons, our data likely represent a substantial
improvement in comprehensiveness and fidelity compared to previous lost wallet studies.

Electronic copy available at: https://ssrn.com/abstract=3639328

 Figure 1: Wallet Reporting Rates and Dishonesty

USA Italy
100 100

80 80

Wallet Reporting Rate (%) 60 60

40 40

20 20

r = -.370 r = -.686
N = 24 N = 16
0 0
-2 -1 0 1 2 -2 -1 0 1 2
Index of Dishonest Behavior Index of Dishonest Behavior

Notes: The y-axis represents the percentage of recipients in each city reporting a lost wallet in the US (left panel) and Italy (right panel). The x-axis
represents city-level dishonesty index scores, with higher numbers reflecting greater dishonesty. For the US, our dishonesty index was constructed
by extracting the first principal component from (1) the share of self-employed individuals in a city who reports an income in 2009 within US
$500 of the first Earned Income Tax Credit (EITC) kink, as a percentage of individuals with non-zero self-employment income, as a measure of
cheating on taxes (Chetty, Friedman, and Saez 2013), and (2) the number of federal court convictions for corrupt practices between 1976 and 2002
per 10,000 public officials in the state that the city belongs to (Glaeser and Saks 2006). For Italy, our dishonesty index was constructed by extracting
the first principal component from (1) municipality-level rates of compliance or payment of a television licensing fee (Buonanno et al. 2019), (2) the
difference between the cumulative amounts of public money allocated to capital expenditures and existing amounts of physical infrastructure (Golden
and Picci 2006), and (3) historical data on prosecutors’ requests to proceed with a criminal investigation against a member of Parliament (Nannicini
et al. 2013). Lines represent the best fit to the data based on OLS estimation.

to our behavioral benchmark, and document two main findings. The first is that only a subset of
survey measures meaningfully correlate with wallet reporting rates. The fact that only some survey
measures are predictive of social capital (e.g., generalized trust from the World Value Survey)
while other seemingly-similar measures are not (e.g., generalized trust from the Global Preferences
Survey) highlights the importance of benchmarking survey responses against behavioral outcomes.
The second finding is that wallet reporting rates in our data also strongly predict economic outputs
associated with social capital. In fact, wallet reporting rates outperform virtually every survey
measure in predicting country-level differences in gross domestic product, total factor productivity,
and indicators of government effectiveness. In what follows, we explain in greater detail our lost
wallet data and the set of survey measures benchmarked against wallet reporting rates.

A Global Field Experiment

Behavioral data comes from a field experiment we recently conducted consisting of 17,303 lost
wallets in 355 cities across 40 countries (Cohn et al. 2019). Figure 2 provides an overview of the
countries covered in our data set. We typically turned in 400 wallets in the six to eight largest
cities of each country, at one of five institutional settings: (i) banks; (ii) theaters, museums, or other

Electronic copy available at: https://ssrn.com/abstract=3639328

 Figure 2: Coverage of Lost Wallet Data (Blue Countries)

cultural establishments; (iii) post offices; (iv) hotels; and (v) public offices, such as police stations,
courts of law, or town halls. We focused on these institutions because they serve as essential pillars
of civic life, and typically have a public reception area that allowed us to perform the drop-offs.
Experimenters in our study handed over a “lost” wallet (that they ostensibly found outside on the
street) to a front-desk worker and asked them to take custody of the wallet. The experimenter would
then promptly leave without requesting written proof of the transaction. By telling recipients that the
wallet was found outside the building, we prevented possible concerns that the owner might come
back and look for the wallet (or at least provided recipients with plausible deniability of having
received the wallet). After performing the wallet drop-off and exiting the building, experimenters
would immediately record several recipient characteristics and situational factors. In particular,
experimenters made note of the recipient’s gender and approximate age (i.e., whether the person was
40 years or older), and also noted whether other coworkers or bystanders were present during the
transaction (which we use as a proxy for how busy or observable7 the recipient was). Experimenters
also noted whether a computer was visible at the recipient’s desk, which we use as a proxy for lower
effort costs of contacting the owner.
For our wallets we used transparent business card cases, allowing the recipient to inspect its
7. In the online appendix we report additional analyses which suggest that concerns of being observed do not
influence country differences in wallet reporting rates.

Electronic copy available at: https://ssrn.com/abstract=3639328

content without having to open the wallet. Each wallet contained three identical business cards
providing the owner’s contact information, and we typically created three fictitious male owners
for each country using common local names. The wallets also contained a grocery list, a small
dimple key, and (depending on experimental treatment) some money. To signal that the owner
of the wallet was likely a resident, the business cards and shopping list were always provided
in the country’s local language and any money inside the wallet was always in local currencies.
Business cards in each wallet were associated with a unique email address, allowing us to identify
individual wallets that were reported. Our dependent measure was whether a wallet was reported to
its owner by email within 100 days. For greater detail on the study procedure we direct the reader
to the Supplemental Materials of Cohn et al. (2019). Data for the study is publicly available at
https://doi.org/10.7910/DVN/YKBODN.
In the experiment we randomly varied the amount of money in the wallet, with wallets containing
either no money or the equivalent of US $13.45 (adjusted for purchasing power parity). We also
ran additional treatment arms in the United States, United Kingdom, and Poland, including a high
stakes version in which wallets contained the equivalent of US $94.15 and another version that
excluded the dimple key from the wallet. Country-level reporting rates were strongly correlated
across experimental conditions — for example, the rank-order correlation between the $13.45
condition and the condition without money was 0.939 — so we combine data across conditions
for the present analysis. The results we report below are virtually unchanged when we restrict our
analysis only to wallets containing no money, or only to wallets containing money.

Survey Measures of Social Capital

Table 1 provides an overview of the survey measures we benchmark against wallet reporting rates.
We selected these measures based on their prominence and use in the past literature, their degree of
fit with our operationalization of social capital (i.e., values and beliefs that encourage cooperation
among strangers and limit free-riding), and availability of data for public use and suitability for
cross-country comparisons.
Generalized Trust. Our first two survey items attempt to measure general tendencies to trust
others. Such measures logically relate to both social capital and to our measure of wallet reporting
rates, as in equilibrium people should only be willing to trust if others are honest. The most popular
measure of this sort (and perhaps the most widely used measure in the social capital literature) is
the “generalized trust” measure used in both the World Values Survey (WVS) and European Values
Study (EVS), which asks respondents whether most people can be trusted or not. Both the WVS
and EVS are large-scale, representative surveys that ask a standardized set of questions in repeated

Electronic copy available at: https://ssrn.com/abstract=3639328

cross-sections.8 One exception in our data set is Kenya, where we use generalized trust responses
from the Afrobarometer (the question and response format is identical to that in the WVS/EVS).
Our second measure of generalized trust (“GPS trust”) comes from the Global Preferences Survey,
which was collected within the framework of the 2012 Gallup World Poll (Falk et al. 2016; Falk
et al. 2018). The GPS was designed to measure key economic preferences (such as risk, time, and
social preferences) with nationally representative samples. The GPS trust measure asks respondents
whether they assume others have “only the best intentions”, and is a strong predictor of trusting
behavior in incentivized trust games. In total, 39 of the 40 countries in our lost wallet data overlap
with survey data from the WVS/EVS, and 36 of the 40 countries overlap with the GPS.
Generalized Morality. Our next two measures focus on norms and beliefs of appropriate
conduct and behavior towards others beyond one’s immediate family, kinship, or social group.
This generalized, or non-parochial, sense of morality has been thought to be closely linked to
social capital because it fosters large-scale cooperation (Tabellini 2008, 2010). Our first measure
(“generalized morality”) comes from the WVS/EVS and is based on prior work by Tabellini (2008).
For this measure respondents indicate whether “tolerance and respect for other people” is one of
the top five qualities children are encouraged to learn at home (from a list of ten possible qualities
such as obedience, hard work, or feelings of responsibility).9 Our second measure (“Universal
Moral Values”) is based on work by Enke (2019) and captures the strength of “universal” as
opposed to “particularist” moral values taken from the Moral Foundations Questionnaire (Graham
et al. 2011). This survey measure is based on a theoretical framework from moral psychology
which distinguishes between impersonal (universal) values such as fairness and individual rights —
which facilitate cooperation among strangers — and in-group-specific (particularist) values such as
in-group loyalty and respect for established hierarchies. The MFQ data comes from a sample of
self-selected respondents who chose to complete the MFQ at www.yourmorals.org between 2008
and 2018. In terms of coverage, data from 38 countries in our lost wallet data overlap with the
generalized morality measure from the WVS/EVS, and data from 35 countries overlap with the
universal moral values measure.
8. The WVS consists of seven waves collected from 1981 to 2020. The EVS consists of five waves collected from
1981 to 2017. For all measures from the WVS/EVS, we combine responses across survey waves as country-level
responses since our set of items have a high degree of temporal stability. In section 7 of the online appendix, we report
results when only using responses from the survey wave closest in time to when our lost wallet data was collected for
each country. Using this alternative data set we find similar (and if anything, slightly stronger) results to those reported
here.
9. Tabellini’s index of generalized morality varies across papers. In Tabellini (2008) the index is a composite of the
generalized trust and respect items from the WVS/EVSWorld Values Survey, and in Tabellini (2010) the index also
includes a measure of obedience towards parents and locus of control. We decided to use the single respect item as
our measure of generalized morality because this item (a) appears to most closely resemble the construct of interest
(i.e., has the highest face validity), and (b) allows for cleaner comparisons of generalized trust in terms of predictive
performance. We find similar results to those reported above when using other indices used by Tabellini.

Electronic copy available at: https://ssrn.com/abstract=3639328

 Norms of Civic Cooperation. Our next survey measure involves norms of civic cooperation,
which measure the degree of disapproval for actions that confer a private benefit while imposing
a social cost on others. Such measures directly relate to our conceptualization of social capital as
beliefs and values that limit free-riding. Following Guiso, Sapienza, and Zingales (2011), we extract
the first principal component from three WVS/EVS items which ask opinions about claiming public
benefits one is not entitled to, free riding on public goods, and accepting bribes. For this measure,
data from 37 countries overlap with our lost wallet data.
Prosocial Preferences. The next two survey measures come from the Global Preference Survey
and measure two distinct forms of prosocial preferences: positive reciprocity and unconditional
altruism. Positive reciprocity attempts to measure the propensity to act in a positively reciprocal
way towards others, which is thought to be a key ingredient in facilitating cooperation and limiting
free-riding (i.e., conditional cooperation; Fischbacher and Gächter 2010; Rustagi, Engel, and
Kosfeld 2010). The GPS measure of positive reciprocity is based on two questions. One item asks
respondents to assess how willing they are to return a favor to someone else. The second item
involves a hypothetical scenario in which respondents imagine they are lost in an unfamiliar area
and, after asking a stranger for directions, receive an offer by the stranger to take them to their
destination. Respondents are then asked which of six presents they would give to the stranger as
a “thank you,” with the presents varying between 5 and 30 euros. In the GPS the two items are
aggregated and given roughly equal weights (0.485 and 0.515, respectively).
The GPS also provides a measure of unconditional altruism, a form of prosocial preferences
that has been empirically linked to honest behavior in general (Cappelen, Sørensen, and Tungodden
2013; Kerschbamer, Neururer, and Gruber 2019; Maggian and Villeval 2016) and wallet reporting
rates in particular (Cohn et al. 2019). The first item asks respondents how willing they would be
to give to good causes without the expectation of anything in return. The second item involves a
hypothetical scenario where the respondent receives an unexpected windfall of 1,000 euros, and
asks them how much of the windfall they would donate towards a good cause. In the GPS the items
receive weights of 0.365 and 0.635, respectively, before being combined. For both measures, data
from 36 countries overlap with our lost wallet data.
Returning a Lost Item. Lastly, we include a recent measure (“Return Lost Item”) from the
2019 Lloyd’s Register Foundation World Risk Poll conducted by Gallup. The poll consists of
interviews with nationally representative samples of over 150,000 respondents from 142 countries
and territories. We use a single question that asks respondents how likely a stranger would be to
return a small lost item if found, from “not at all likely” to “somewhat likely” to “very likely.” Since
the measure bears a close resemblance to our behavioral measure of social capital — returning a
lost wallet — we include this item as a way to evaluate the strength of other survey measures (which
attempt to measure more general values or beliefs) in predicting wallet reporting rates.

Electronic copy available at: https://ssrn.com/abstract=3639328

Results

Survey Measures and Wallet Reporting Rates

We report coefficient estimates and p-values from ordinary least squares (OLS) regressions where
we regress wallet reporting behavior on country-level variables of social capital. Observations are
coded as 100 when a wallet was reported and 0 otherwise, and all survey measures are standardized
at the country-level to have a mean of zero and standard deviation of one. With this coding scheme,
regression coefficients can be interpreted as the percentage point difference in reporting rates
associated with a one standard deviation change in the explanatory variable. For all models we also
include fixed effects for treatment condition, institutional setting, and all recipient and situational
characteristics recorded during the wallet drop-off. We adjust standard errors for clustering at the
country-level, and also adjust p-values to control for the false discovery rate10 (i.e., the proportion of
significant results expected to be false due to multiple hypothesis testing; Benjamini and Hochberg
1995; Benjamini and Yekutieli 2001). In the online appendix, we also show that our results are
robust when excluding our set of covariates from the analyses, or when using probit rather than
OLS models.
Figure 3 graphically displays the country-level correlations between survey measures of social
capital and wallet reporting rates. First, and perhaps unsurprisingly, our strongest predictor of
reporting a lost wallet is the “return lost item” (panel H) that serves as our benchmark correlation
due to its close resemblance to our outcome measure. Based on our OLS specification, a one
standard deviation increase in “return lost item” scores is associated with a 10.65 percentage point
increase in wallet reporting rates (p < 0.001; see Table A4 in the online appendix for the full
regression results). To give a sense of the magnitude for this relationship, the coefficient represents
a 0.54 standard deviation in the total variation found in wallet reporting rates across countries.
Of the remaining seven survey measures of social capital, the two strongest predictors are
the generalized trust11 and generalized morality measures from the WVS/EVS (panels A and
C, respectively). Based on our OLS models in Table A4, a one standard deviation increase in
generalized trust is associated with an 9.5 point increase in wallet reporting rates (p = 0.003), and
10. We use Benjamini-Hochberg adjusted p-values to control the false discovery rate to be less than or equal to 0.05,
which are calculated as follows: Let m be the total number of tests conducted and i be the ranking of p-values in the set,
from smallest to largest. The adjusted p-value for a test is either the observed p-value multiplied by m/i or the adjusted
p-value associated with any lower-ranked raw p-value, whichever is smaller.
11. We were also able to examine the role of measurement error for our estimates of generalized trust from the
WVS/EVS, as we had near-identical measures of both generalized trust and wallet reporting rates (as suggested by
Gillen, Snowberg, and Yariv 2019). Using an “obviously related” instrumental variables (ORIV) regression for the
subset of countries in which we can apply the correction (n = 30), we find that the country-level correlation between
generalized trust and wallet reporting rates increases from 0.58 (without correction) to 0.69 (with correction). Full
details on our ORIV specification are provided in the online appendix.

Electronic copy available at: https://ssrn.com/abstract=3639328

 Figure 3: Wallet Reporting Rates and Measures of Social Capital

A B C D
100 r = 0.604 100 r = 0.024 100 r = 0.612 100 r = 0.461
Wallet Reporting Rate (%)

Wallet Reporting Rate (%)

Wallet Reporting Rate (%)

Wallet Reporting Rate (%)

N = 39 N = 36 N = 38 N = 35
80 80 80 80

60 60 60 60

40 40 40 40

20 20 20 20

0 0 0 0
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
Generalized Trust GPS Trust Generalized Morality Universal Moral Values

E F G H
100 r = 0.392 100 r = 0.050 100 r = -.214 100 r = 0.645
Wallet Reporting Rate (%)

Wallet Reporting Rate (%)

Wallet Reporting Rate (%)

Wallet Reporting Rate (%)

N = 37 N = 36 N = 36 N = 39
80 80 80 80

60 60 60 60

40 40 40 40

20 20 20 20

0 0 0 0
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
Civic Cooperation Norms Positive Reciprocity (GPS) Altruism (GPS) Return Lost Item

Notes: Scatterplots display the country-level relationship between wallet reporting rates and (A) generalized trust from
the World Values Survey (WVS) and European Values Study (EVS), (B) generalized trust from the Global Preferences
Survey (Falk et al. 2018), (C) generalized morality (“respect and tolerance for others”) from the WVS/EVS, (D)
universal moral value scores from the Moral Foundations Questionnaire (Enke 2019; Graham et al. 2011), (E) an index
of norms of civic cooperation from the WVS/EVS (Guiso, Sapienza, and Zingales 2011), (F & G) positive reciprocity
and altruism scores from the Global Preferences Survey, and (H) expectations about having a lost item returned from the
World Risk Poll. For each graph the y-axis represents wallet reporting rates in a given country (from 0-100%) and the
x-axis represents the explanatory variable (standardized at the country-level to have a mean of 0 and standard deviation
of 1). Lines represent the best fit to the data based on OLS estimation. The upper-left corner of each panel reports the
country-level correlation between the outcome and predictor variable, as well as the number of countries in the analysis.

a one standard deviation increase in generalized morality is associated with a 10.2 point increase
(p < 0.001). Remarkably, the predictive power of these two measures is nearly identical to that of
our “return lost item” measure. After generalized morality, we observe positive but relatively weaker
correlations for universal moral values (from the Moral Foundations Questionnaire) and norms
of civic cooperation (from the WVS/EVS). Based on our OLS models, a one standard deviation
increase in universal moral values and norms of civic cooperation are associated with, respectively, a
6.8 and 6.4 percentage point increase in wallet reporting rates (both are significant at the 5% level).
Lastly, all three measures from the GPS — trust, positive reciprocity, and altruism — fare
relatively poorly in predicting wallet reporting rates. Coefficients for the three items are small and

10

Electronic copy available at: https://ssrn.com/abstract=3639328

not statistically distinguishable from 0 (p-values range from 0.193 to 0.949). The null result for
generalized trust from the GPS may seem puzzling in light of the relatively strong performance
of the generalized trust measure from the WVS/EVS. However, we note that the trust question
from the GPS was selected based on its performance to predict (first-mover) trusting behavior in an
incentivized trust game, and past research has found trust game behavior to be weakly correlated
with generalized trust from the WVS/EVS (Glaeser et al. 2000; Karlan 2005; Lazzarini et al. 2005).
In light of these findings, it is less surprising that trust from the WVS is predictive of wallet reporting
rates while trust from the GPS is not.
We pause to note that respondents in our lost wallet data were not representatively sampled
from each country’s population. Respondents were front-desk employees at public and private
institutions, and we performed wallet drop-offs in larger cities (with populations of 100k or greater)
within each country. By contrast, our survey measures come from data sets with nationally
representative samples (with the exception of the Moral Foundations Questionnaire). In section 8
of the online appendix we report results from a robustness test in which we compare coefficients
for our WVS/EVS items with a restricted sample from the WVS/EVS that more closely resembles
our front-desk workers based on demographic proxies available in the survey data. In particular,
we restrict our sample to employed respondents who lived in a city with a population of 100k or
greater. We find qualitatively similar results when using this restricted sample: generalized trust
and generalized morality remain our two strongest survey measures of wallet reporting rates, while
norms of civic cooperation is a relatively weak predictor. Thus, differences in sample characteristics
do not appear to meaningfully bias our results.

Wallet reporting rates as a predictor of economic and institutional perfor-

mance
Of ultimate interest to economists is the ability of social capital measures to explain variation in
economic development. As such, we compared the lost wallet data to our set of survey measures
in predicting economic and institutional performance. Unlike our previous analyses we now treat
wallet reporting rates as a right hand side variable and, to facilitate comparison, we aggregate all
measures at the country-level and standardize these measures (including wallet reporting rates) to
have a mean of zero and standard deviation of one. Our aim here is not to establish causality, but
rather to examine the extent that our behavioral measure contains new information in explaining
variation in economic development.
Table 2 illustrates the additional predictive value of wallet reporting rates, over and above
each survey measure of social capital, in explaining country-level heterogeneity in economic and
institutional performance. Column 1 reports the bivariate relationship between each measure and

11

Electronic copy available at: https://ssrn.com/abstract=3639328

GDP per capita,12 and column 2 reports the multivariate model where wallet reporting rates are
included alongside each survey measure. The same exercise is repeated for total factor productivity
in columns 3 and 4 of the table.
Two patterns clearly emerge from the analysis. First, wallet reporting rates explain substantial
variation in economic productivity above existing survey measures. For instance, when predicting
GDP per capita, adding wallet reporting rates alongside generalized trust increases the R-squared by
nearly 50% (from 0.43 in the bivariate model to 0.63 in the multivariate model). Second, for the
multivariate models, wallet reporting rates outperform (in terms of coefficient size) nearly every
survey measure of social capital, and often substantially so. Wallet reporting rates also outperform
the “return lost item” from the World Risk Poll.
The last four columns of Table 2 show that wallet reporting rates also explain unique variation in
institutional performance across countries. Columns 5–8 of the table report the same analysis used
in the previous paragraph, but with World Bank government effectiveness ratings and a behavioral
measure of institutional efficiency (i.e., the proportion of incorrectly addressed international mail
that is returned; Chong et al. 2014) as outcome variables.13 Similar to our economic productivity
measures, we find that wallet reporting rates explain substantial variation in institutional performance
not captured by existing survey measures of social capital.
To quantify the relative contribution of wallet reporting rates in explaining economic outputs, we
performed a series of pairwise dominance analyses (Azen and Budescu 2003; Budescu 1993). This
procedure decomposes the total R-squared from a multivariate model into the relative contribution
provided by each variable in the model, with contribution weights standardized to sum to one. We
conducted a dominance analysis for all 32 comparisons provided in Table 2 (eight multivariate
models for each of our four outcome variables). Results are reported in Table S4 in the online
appendix. We find that wallet reporting rates contribute the majority of variance explained in 29 of
the 32 models. In 23 of the comparisons, wallet reporting rates outperform its survey counterpart
by more than a factor of two. Thus, a country’s propensity to report a lost wallet appears to
contain considerable new information above existing survey measures in explaining cross-country
differences in economic performance.
12. We conducted our lost wallet experiment from 2013 to 2016, so we use GDP and total factor productivity indices
from 2017. Data are from Penn World Tables
13. The World Bank has five other indices of governance quality: (1) voice and accountability, (2) political stability
and absence of violence, (3) regulatory quality, (4) rule of law, and (5) control of corruption. Compared to government
effectiveness ratings, we find even more pronounced effects of wallet reporting rates in predicting these other indices.

12

Electronic copy available at: https://ssrn.com/abstract=3639328

Conclusion
In this paper we use lost wallet reporting data from Cohn et al. (2019) to assess the predictive power
of survey measures of social capital commonly used in economics, and establish two stylized facts.
First, some survey measures of social capital, such as generalized trust and generalized morality,
are strongly correlated with country differences in the tendency to report a lost wallet. This finding
suggests that skepticism over the use of survey data to measure social capital may be unwarranted
(see also Bjørnskov 2021). It is also reassuring that the most widely used survey measure of
social capital — the generalized trust measure from the WVS/EVS — is highly predictive of
wallet reporting rates. Other measures, such as prosocial preference measures from the recently
developed Global Preferences Survey, performed relatively poorly in predicting wallet reporting
rates. Additional empirical examination may be needed to establish whether such measures can
serve as useful proxies of a country’s social capital.
A second finding is that lost wallet reporting rates explain additional variation in economic and
institutional performance across countries, suggesting that this measure contains unique information
not captured by existing survey measures of social capital. When feasible, researchers may wish to
use lost wallet reporting rates from our data, alone or in combination with existing survey measures,
when examining the economic outputs of social capital.

13

Electronic copy available at: https://ssrn.com/abstract=3639328

References
Algan, Yann, and Pierre Cahuc. 2010. “Inherited trust and growth.” American Economic Review
100 (5): 2060–92.

Arrow, Kenneth J. 1972. “Gifts and exchanges.” Philosophy and Public Affairs 1 (4): 343–362.

Azen, R, and DV Budescu. 2003. “The dominance analysis approach for comparing predictors in
multiple regression.” Psychological Methods 8 (2): 129–148.

Bellemare, Charles, and Sabine Kröger. 2007. “On representative social capital.” European Eco-
nomic Review 51 (1): 183–202.

Benjamini, Yoav, and Yosef Hochberg. 1995. “Controlling the false discovery rate: A practical and
powerful approach to multiple testing.” Journal of the Royal Statistical Society. Series B 57 (1):
289–300.

Benjamini, Yoav, and Daniel Yekutieli. 2001. “The control of the false discovery rate in multiple
testing under dependency.” Annals of Statistics, 1165–1188.

Bernardi, Richard A. 2006. “Associations between Hofstede’s cultural constructs and social desir-
ability response bias.” Journal of Business Ethics 65 (1): 43–53.

Bertrand, Marianne, and Sendhil Mullainathan. 2001. “Do people mean what they say? Implications
for subjective survey data.” American Economic Review 91 (2): 67–72.

Bjørnskov, Christian. 2021. “Civic honesty and cultures of trust.” Journal of Behavioral and
Experimental Economics 92:101693.

Bjørnskov, Christian, and Kim Mannemar Sønderskov. 2013. “Is social capital a good concept?”
Social Indicators Research 114 (3): 1225–1242.

Bourdieu, Pierre. 1986. “The forms of capital.” In Handbook of Theory and Research for the
Sociology of Education, edited by J.G. Richardson. New York: Greenwood Press.

Budescu, David. 1993. “Dominance analysis: A new approach to the problem of relative importance
of predictors in multiple regression.” Psychological Bulletin 114 (3): 542–551.

Buonanno, Paolo, Matteo Cervellati, Sara Lazzaroni, and Giovanni Prarolo. 2019. “Political his-
tory, fiscal compliance and cooperation: Medieval social contracts and their legacy.” CEPR
Discussion Paper No. DP14214.

Cappelen, Alexander W, Erik Ø Sørensen, and Bertil Tungodden. 2013. “When do we lie?” Journal
of Economic Behavior & Organization 93:258–265.

14

Electronic copy available at: https://ssrn.com/abstract=3639328

Chetty, Raj, John N Friedman, and Emmanuel Saez. 2013. “Using differences in knowledge across
neighborhoods to uncover the impacts of the EITC on earnings.” American Economic Review
103 (7): 2683–2721.

Chong, Alberto, Rafael La Porta, Florencio Lopez-de-Silanes, and Andrei Shleifer. 2014. “Letter
grading government efficiency.” Journal of the European Economic Association 12 (2): 277–
298.

Cohn, Alain, Michel André Maréchal, David Tannenbaum, and Christian Lukas Zünd. 2019. “Civic
honesty around the globe.” Science 365 (6448): 70–73.

Coleman, James S. 1990. Foundations of Social Theory. Cambridge, MA: Harvard University Press.

Djankov, Simeon, Edward Glaeser, Rafael La Porta, Florencio Lopez-de-Silanes, and Andrei
Shleifer. 2003. “The new comparative economics.” Journal of Comparative Economics 31 (4):
595–619.

Dohmen, Thomas, Armin Falk, David Huffman, Uwe Sunde, Jürgen Schupp, and Gert G Wagner.
2011. “Individual risk attitudes: Measurement, determinants, and behavioral consequences.”
Journal of the European Economic Association 9 (3): 522–550.

Enke, Benjamin. 2019. “Kinship, cooperation, and the evolution of moral systems.” The Quarterly
Journal of Economics 134 (2): 953–1019.

Epper, Thomas, Ernst Fehr, Helga Fehr-Duda, Claus Thustrup Kreiner, David Dreyer Lassen, Søren
Leth-Petersen, and Gregers Nytoft Rasmussen. 2020. “Time discounting and wealth inequality.”
American Economic Review 110 (4): 1177–1205.

Falk, Armin, Anke Becker, Thomas Dohmen, Benjamin Enke, David Huffman, and Uwe Sunde.
2018. “Global evidence on economic preferences.” The Quarterly Journal of Economics 133
(4): 1645–1692.

Falk, Armin, Anke Becker, Thomas J Dohmen, David Huffman, and Uwe Sunde. 2016. “The pref-
erence survey module: A validated instrument for measuring risk, time, and social preferences.”
Netspar Discussion Paper No. 01/2016-003.

Fehr, Ernst, Urs Fischbacher, Bernhard Von Rosenbladt, Jürgen Schupp, and Gert G Wagner. 2003.
“A nation-wide laboratory: Examining trust and trustworthiness by integrating behavioral
experiments into representative survey.” CESifo Working Paper Series.

Fischbacher, Urs, and Simon Gächter. 2010. “Social preferences, beliefs, and the dynamics of free
riding in public goods experiments.” American Economic Review 100 (1): 541–56.

15

Electronic copy available at: https://ssrn.com/abstract=3639328

Fountain, Jane E. 1998. “Social capital: Its relationship to innovation in science and technology.”
Science and Public Policy 25 (2): 103–115.

Fukuyama, Francis. 1995. Trust: The Social Virtues and the Creation of Prosperity. New York: Free
Press.

Gillen, Ben, Erik Snowberg, and Leeat Yariv. 2019. “Experimenting with measurement error:
Techniques with applications to the caltech cohort study.” Journal of Political Economy 127
(4): 1826–1863.

Gintis, Herbert. 2016. Individuality and Entanglement: The Moral and Material Bases of Social
Life. Princeton, New Jersey: Princeton University Press.

Glaeser, Edward L, David I Laibson, Jose A Scheinkman, and Christine L Soutter. 2000. “Measuring
trust.” Quarterly Journal of Economics 115 (3): 811–846.

Glaeser, Edward L, and Raven E Saks. 2006. “Corruption in America.” Journal of Public Economics
6 (90): 1053–1072.

Golden, Miriam, and Lucio Picci. 2006. “Corruption and the management of public works in Italy.”
In International Handbook on the Economics of Corruption, 457.

Graham, Jesse, Brian Nosek, Jonathan Haidt, Ravi Iyer, Spassena Koleva, and Peter Ditto. 2011.
“Mapping the moral domain.” Journal of Personality and Social Psychology 101 (2): 366–385.

Guiso, Luigi, Paola Sapienza, and Luigi Zingales. 2004. “The role of social capital in financial
development.” American Economic Review 94 (3): 526–556.

. 2011. “Civic capital as the missing link.” In Handbook of Social Economics, edited by
J. Benhabib, A. Bisin, and M. O. Jackson, 1:417–480. Amsterdam: North-Holland.

Hainmueller, Jens, Dominik Hangartner, and Teppei Yamamoto. 2015. “Validating vignette and
conjoint survey experiments against real-world behavior.” Proceedings of the National Academy
of Sciences 112 (8): 2395–2400.

Johnson, Timothy, Patrick Kulesa, Young Ik Cho, and Sharon Shavitt. 2005. “The relation between
culture and response styles: Evidence from 19 countries.” Journal of Cross-Cultural Psychology
36 (2): 264–277.

Karlan, Dean S. 2005. “Using experimental economics to measure social capital and predict financial
decisions.” American Economic Review 95 (5): 1688–1699.

Kerschbamer, Rudolf, Daniel Neururer, and Alexander Gruber. 2019. “Do altruists lie less?” Journal
of Economic Behavior and Organization 157:560–579.

16

Electronic copy available at: https://ssrn.com/abstract=3639328

Knack, Stephen, and Philip Keefer. 1997. “Does social capital have an economic payoff? A cross-
country investigation.” Quarterly Journal of Economics 112 (4): 1251–1288.

Kreuter, Frauke, Stanley Presser, and Roger Tourangeau. 2008. “Social desirability bias in CATI,
IVR, and web surveys: The effects of mode and question sensitivity.” Public Opinion Quarterly
72 (5): 847–865.

Krosnick, Jon A. 1999. “Survey research.” Annual Review of Psychology 50 (1): 537–567.

La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer, and Robert W Vishny. 1997. “Trust
in large organizations.” American Economic Review 87 (2): 333.

Laajaj, Rachid, Karen Macours, Daniel Alejandro Pinzon Hernandez, Omar Arias, Samuel D
Gosling, Jeff Potter, Marta Rubio-Codina, and Renos Vakis. 2019. “Challenges to capture the
big five personality traits in non-WEIRD populations.” Science Advances 5 (7): eaaw5226.

Lazzarini, Sergio G, Regina Madalozzo, Rinaldo Artes, Jose de Oliveira Siqueira, et al. 2005.
“Measuring trust: An experiment in Brazil.” Brazilian Journal of Applied Economics 9 (2):
153–169.

Lederman, Daniel, Norman Loayza, and Ana Maria Menendez. 2002. “Violent crime: Does social
capital matter?” Economic Development and Cultural Change 50 (3): 509–539.

Levitt, Steven D, and John A List. 2007. “What do laboratory experiments measuring social
preferences reveal about the real world?” Journal of Economic Perspectives 21 (2): 153.

Maggian, Valeria, and Marie Claire Villeval. 2016. “Social preferences and lying aversion in
children.” Experimental Economics 19 (3): 663–685.

Maréchal, Michel André, Alain Cohn, Jeffrey Yusof, and Raymond Fisman. 2022. “Whose pref-
erences matter for redistribution? Cross-country evidence.” Working paper, University of
Zurich.

Mokyr, Joel. 2009. The enlightened economy: An economic history of Britain, 1700-1850. 64–119.
New Haven, Connecticut: Yale University Press.

Nannicini, Tommaso, Andrea Stella, Guido Tabellini, and Ugo Troiano. 2013. “Social capital and
political accountability.” American Economic Journal: Economic Policy 5 (2): 222–50.

Olken, Benjamin A. 2009. “Corruption perceptions vs. corruption reality.” Journal of Public
Economics 93 (7-8): 950–964.

Ostrom, Elinor. 2000. ““Social capital: a fad or a fundamental concept?”” Social capital: A multi-
faceted perspective 172 (173): 195–198.

17

Electronic copy available at: https://ssrn.com/abstract=3639328

Paldam, Martin. 2000. “Social capital: one or many? Definition and measurement.” Journal of
Economic Surveys 14 (5): 629–653.

Parry, Hugh J, and Helen M Crossley. 1950. “Validity of responses to survey questions.” Public
Opinion Quarterly 14 (1): 61–80.

Putnam, Robert D, Robert Leonardi, and Raffaella Y Nanetti. 1993. Making Democracy Work:
Civic Traditions in Modern Italy. Princeton, New Jersey: Princeton University Press.

Razafindrakoto, Mireille, and François Roubaud. 2010. “Are international databases on corruption
reliable? A comparison of expert opinion surveys and household surveys in sub-Saharan Africa.”
World Development 38 (8): 1057–1069.

Rustagi, Devesh, Stefanie Engel, and Michael Kosfeld. 2010. “Conditional cooperation and costly
monitoring explain success in forest commons management.” Science 330 (6006): 961–965.

Sunde, Uwe, Thomas Dohmen, Benjamin Enke, Armin Falk, David Huffman, and Gerrit Meyerheim.
in press. “Patience and comparative development.” Review of Economic Studies.

Tabellini, Guido. 2008. “Institutions and culture.” Journal of the European Economic Association 6
(2-3): 255–294.

. 2010. “Culture and institutions: economic development in the regions of Europe.” Journal
of the European Economic Association 8 (4): 677–716.

18

Electronic copy available at: https://ssrn.com/abstract=3639328

 Table 1: Overview of Survey Measures

Measure Source No. of Description:

Overlapping
Countries
Generalized WVS, EVS, & 39 Average response by country to the question “Generally speaking,
Trust Afrobarometer would you say that most people can be trusted or that you need to
be very careful in dealing with people?” (0 = need to be very
careful, 1 = most people can be trusted).

Trust (GPS) Global 36 Average response by country to the question “I assume that
Preference people have only the best intentions.” Responses can range from 0
Survey (does not describe me at all) to 10 (describes me perfectly).

Generalized WVS, EVS 38 Fraction of respondents by country who select “tolerance and
Morality respect for other people” as one of their answers to the question
“Here is a list of qualities that children can be encouraged to learn
at home. Which, if any, do you consider to be especially
important?” Respondents could select up to five qualities from a
list of ten.

Universal yourmorals.org 35 Average difference by country between “universal” and

Moral Values “communal” moral values from the Moral Foundations
Questionnaire (Graham et al. 2011).

Civic WVS, EVS 37 First principal component (extracted at the country-level) from
Cooperation how justifiable respondents thought it was to (1) claim
government benefits to which one is not entitled, (2) avoid paying
a fare on public transit, and (3) accepting a bribe in the course of
one’s duties. Responses could range from 0 (always justifiable) to
10 (never justifiable).

Positive Global 36 Average response by country to (1) a self-assessment of

Reciprocity Preference willingness to return a favor to someone else, and (2) a
(GPS) Survey hypothetical choice scenario involving a gift exchange in return
for help from another individual. Items were individually z-scored
and then given weights of .485 and .515, respectively.

Altruism Global 36 Average response by country to (1) a self-assessment of

(GPS) Preference willingness to give to good causes without expecting anything in
Survey return, and (2) a hypothetical donation decision. Items were
individually z-scored and then given weights of .365 and .636,
respectively.

19

Electronic copy available at: https://ssrn.com/abstract=3639328

 Table 2: Predictive Value of Wallet Reporting Rates
Log GDP Log Productivity Government Letter Grade
per capita (TFP) Effectiveness Efficiency
(1) (2) (3) (4) (5) (6) (7) (8)
Generalized 0.464∗∗∗ 0.220∗∗∗ 0.131∗∗ 0.025 0.590∗∗∗ 0.381∗∗∗ 0.077∗∗ −0.013
trust (0.093) (0.067) (0.049) (0.053) (0.077) (0.084) (0.041) (0.040)
Wallets 0.404∗∗∗ 0.182∗∗∗ 0.347∗∗∗ 0.148∗∗
(0.084) (0.057) (0.082) (0.050)
N 39 39 38 38 39 39 39 39
R2 0.428 0.634 0.180 0.407 0.574 0.700 0.078 0.263

Trust (GPS) 0.319∗∗ 0.308∗∗ 0.061 0.050 0.333∗∗ 0.321∗∗∗ −0.016 −0.019
(0.140) (0.099) (0.072) (0.050) (0.113) (0.065) (0.050) (0.039)
Wallets 0.496∗∗∗ 0.197∗∗∗ 0.523∗∗∗ 0.133∗∗
(0.086) (0.048) (0.064) (0.043)
N 36 36 34 34 36 36 36 36
R2 0.196 0.616 0.035 0.383 0.197 0.628 0.003 0.213

Generalized 0.384∗∗∗ 0.147∗∗ 0.194∗∗∗ 0.144∗∗∗ 0.590∗∗∗ 0.397∗∗∗ 0.080∗ −0.012

morality (0.055) (0.067) (0.029) (0.031) (0.065) (0.078) (0.036) (0.047)
Wallets 0.393∗∗∗ 0.083∗ 0.320∗∗∗ 0.153∗∗
(0.099) (0.040) (0.087) (0.056)
N 38 38 37 37 38 38 38 38
R2 0.366 0.598 0.473 0.522 0.589 0.694 0.083 0.268

Universal moral 0.263∗∗ 0.032 0.179∗∗∗ 0.105 0.165 −0.116 0.118∗∗ 0.069
values (0.099) (0.078) (0.044) (0.059) (0.116) (0.076) (0.041) (0.041)
Wallets 0.522∗∗∗ 0.167∗∗∗ 0.635∗∗∗ 0.111∗∗∗
(0.095) (0.049) (0.086) (0.034)
N 35 35 35 35 35 35 35 35
R2 0.153 0.593 0.335 0.545 0.046 0.540 0.219 0.360

Civic 0.193∗ 0.004 0.116∗∗∗ 0.056 0.301∗∗ 0.093 0.092∗ 0.042

cooperation (0.083) (0.083) (0.037) (0.054) (0.113) (0.101) (0.043) (0.055)
Wallets 0.484∗∗∗ 0.154∗∗ 0.530∗∗∗ 0.129∗∗
(0.080) (0.051) (0.074) (0.052)
N 37 37 36 36 37 37 37 37
R2 0.090 0.567 0.166 0.396 0.152 0.548 0.109 0.287

Positive 0.118 0.095 −0.020 −0.027 0.096 0.072 0.009 0.003

reciprocity (GPS) (0.121) (0.092) (0.053) (0.045) (0.100) (0.064) (0.040) (0.044)
Wallets 0.499∗∗∗ 0.201∗∗∗ 0.527∗∗∗ 0.133∗∗
(0.102) (0.047) (0.084) (0.044)
N 36 36 34 34 36 36 36 36
R2 0.027 0.450 0.005 0.367 0.017 0.454 0.001 0.209

Altruism (GPS) 0.051 0.160 −0.009 0.026 0.040 0.155 −0.033 −0.006
(0.116) (0.964) (0.052) (0.050) (0.119) (0.085) (0.037) (0.037)
Wallets 0.541∗∗∗ 0.205∗∗∗ 0.566∗∗∗ 0.132∗∗
(0.108) (0.049) (0.075) (0.043)
N 36 36 34 34 36 36 36 36
R2 0.005 0.480 0.001 0.366 0.003 0.485 0.015 0.209

Return lost 0.446∗∗∗ 0.211 0.182∗∗∗ 0.070 0.542∗∗∗ 0.315∗∗ 0.091∗∗ 0.001
item (0.077) (0.143) (0.023) (0.046) (0.080) (0.106) (0.036) (0.062)
Wallets 0.366∗∗ 0.157∗∗ 0.355∗∗∗ 0.142∗∗
(0.161) (0.063) (0.106) (0.062)
N 39 39 37 37 39 39 39 39
R2 0.369 0.513 0.351 0.466 0.477 0.594 0.110 0.263
Note: OLS estimates with robust standard errors in parentheses. Outcome variables are log GDP per capita, log total factor productivity (relative to
the United States), government effectiveness ratings from the World Bank, and the proportion of incorrectly addressed international mail from a
country that is returned to sender (Chong et al. 2014). All explanatory variables are aggregated at the country-level and standardized to have a mean
of zero and standard deviation of one. Significance levels after correcting for the false discovery rate (Benjamini and Hochberg 1995; Benjamini and
Yekutieli 2001): ∗ p ≤ 0.10, ∗∗ p ≤ 0.05, ∗∗∗ p ≤ 0.01.

20

Electronic copy available at: https://ssrn.com/abstract=3639328