THE DYNAMICS OF TRUST: ADJUSTMENT IN INDIVIDUAL TRUST LEVELS TO CHANGES IN SOCIAL ENVIRONMENT Daniela Scidá Master Thesis CEMFI No. 1101 March 2011 CEMFI Casado del Alisal 5; 28014 Madrid Tel. (34) 914 290 551. Fax (34) 914 291 056 Internet: www.cemfi.es This paper is a revised version of the Master's Thesis presented in partial fulfillment of the 200x-200x Master in Economics and Finance at the Centro de Estudios Monetarios y Financieros (CEMFI). I am very grateful to David Dorn for his great supervision. I appreciate very useful comments from all the professors both during presentations and private discussions. I specially thank Manuel Arellano for his advice. I thank my classmates for all the support during these two years at CEMFI, and specially Ignacio H. Lopez, L. Alonso Villacorta, Florina R. Silaghi, Ruxandra Ciupagea, Maria Ines Berniell, Rebeca Anguren Martin, Inmaculada González Pérez, Alba Diz Tomé, Catalina Campillo, and Andrea F. Popescu. I also thank my family and friends in Argentina for their great support along these two years. Errors and omissions are exclusively my own responsibility.
Master Thesis CEMFI No. 1101 March 2011 THE DYNAMICS OF TRUST: ADJUSTMENT IN INDIVIDUAL TRUST LEVELS TOCHANGES IN SOCIAL ENVIRONMENT Abstract There has been a large interest in economics to study trust. By trust people normally refer to the typical survey question in the European Social Survey or World Value Survey Generally speaking, would you say that most people can be trusted, or that you can't be too careful in dealing with people?. Previous literature has found that trust is important both at an aggregate and at an individual level. However, this literature usually ignores the dynamics of trust and treats it as a condition with which people are born with. This study analyzes how trust adjusts when a person's social environment changes, by looking at immigrants and children. The two main findings are that immigrants tend to adjust their trust to that of the host country over time, and children tend to adjust from their parents' trust level towards that of the population as they get older. These results provide first evidence on the dynamics involved in individuals' trust. Daniela Scidá Brown University daniela.scida@hotmail.com
1 Introduction There has been a large interest in economics to study trust. By trust people usually refer to a typical survey question available in the World Value Survey, European Social Survey, or General Social Survey: Generally speaking, would you say that most people can be trusted, or that you can't be too careful in dealing with people?. In addition, people also have in mind the following question: Do people trust others in contracts even if they are not enforceable?. Previous literature has found that trust is important both at an aggregate and at an individual level. On the one hand, trust is found to be important for country's economic outcomes like economic growth and development. For instance, in an inuential paper Knack and Keefer (1996)[10] found that a country's level of trust is indeed correlated with its rate of growth, and this correlation persists even after controlling for the quality of law enforcement (Zak and Knack, 2001[16]). Also, trust is found to be important for individual economic outcomes such as income. For instance, Guiso et al. (2009)[2] found that there is a humpshaped relationship between trust and income. This means that neither low trust nor high trust are desirable in terms of economic outcomes. People who trust too little will give up trade and prot opportunities too often, and those who trust too much will be cheated too often. As a result, the economic outcomes of both types of people will be harmed. Since trust is so important for individuals, it is natural to wonder what determines their trust. However, previous literature has ignored the dynamics of trust, and treated it as a condition with which people are born. In contrast, the contribution of the present analysis is to consider that trust is at least partly formed from the interaction with other people. Indeed, we can think that the mechanism through which people adjust their trust is by interacting with others. Therefore, this study analyzes how individuals' trust is generated and depends on people around them. More precisely, the analysis is structured into two approaches: the Immigrants Approach and the Children Approach. In the rst approach, I study how individuals' trust depends on home country trust and host country's trust levels using a sample of immigrants 1. That is, using data on trust from the European Social Survey (ESS)[3] and World Value Survey 1 Along this study to refer to the country in which the immigrant was born I will use either the term home country or country of origin. To refer to the country to which the person has immigrated, I will use either the term host country, or current country, or new country. 1
(WVS)[15] I selected a sample of immigrants in countries of the ESS. I analyzed whether these immigrants adjusted their trust level towards that of their new country departing from the mean trust level in their country of origin, and taking into account how many years the individual has been living in the host country. The main result in this approach provides evidence that immigrants, on average, tend to adjust completely to the mean level of trust of the host country after living there for more than 20 years. In the second approach, I study how individuals' trust depends on parents trust and country's trust levels. Thus, I selected a sample of native children 2 from the ESS, and analyzed whether they tend to adjust their trust level away from that of their parents and towards that of the population as they get older. The main result from this approach is that over time children adjust completely their trust level to the mean level of trust of the society in which they live and interact. The rest of this study is organized as follows. Section 2 introduces some basic ideas on the dynamics of trust. Then, Section 3 describes the data used in both approaches. Section 4 presents a detailed explanation on the Immigrants Approach and Children Approach, and the results. Finally, Section 5 concludes. 2 The Dynamics of Trust There is a large literature on Cultural Economics. This literature seeks to analyze people's behaviour, by exploiting characteristics that a person inherits from her culture of origin. For instance, Fernandez & Fogli (2009)[5] study the work and fertility outcomes of second-generation American women. They argue that these women, born and raised in the United States, face the same markets and institutions but they potentially dier in their cultural heritage as reected in their parents' country of origin. Tabellini (2008)[14] studies the role of culture as a channel of historical inuence within countries. In an intent to go beyond the general claim that culture matters, or that informal institutions are important, he estimates the eect of specic cultural traits, traditionally regarded as favorable to economic growth and to the eective functioning of democratic institutions. 2 By children we should understand for the rest of this study, the sons and/or daughters of a person of any age, and not just kids. In particular, I will select them to be between 16 and 40 years old. The reason will become clear in Section 4. 2
In line with these ideas, it has been found that trust is important both at an aggregate and at an individual level. On the one hand, trust is found to be important for country's economic outcomes like economic growth and development. For instance, in an inuential paper Knack and Keefer (1996)[10] found that a country's level of trust is indeed correlated with its rate of growth, and this correlation persists even after controlling for the quality of law enforcement (Zak and Knack, 2001[16]). Also, trust is found to be important for individual economic outcomes such as income. For instance, Guiso et al. (2009)[2] found that there is a hump-shaped relationship between trust and income. This means that neither low trust nor high trust are desirable in terms of economic outcomes. People who trust too little will give up trade and prot opportunities too often, and those who trust too much will be cheated too often. As a result, the economic outcomes of both types of people will be harmed. However, this previous literature has ignored the dynamics of trust, and treated it as a condition with which people are born. In contrast, the contribution of the present analysis is to consider that trust is at least partly formed from the interaction with other people. Guiso et al. (2008)[7] build an OLG model based on the cultural transmission of beliefs. They argue that economics models are generally silent on how people acquire priors, and posit that intergenerational cultural transmission plays a big role in this formation. In their model, children absorb the prior from their parents and after experiencing the real world, they transmit it (updated) to their children. Thus, it is natural to think that the mechanism through which people adjust their trust is by interacting with others. An individual who interacts with another person will be aected by this person's behavior. That is, she will inuence the individual's trust by either honoring or violating the implicit trust promise. We could picture two general scenarios. If a person meets a lot of trustworthy people, then, most likely, she will adjust up her own trust because there exists a reward in trusting others 3. By contrast, if a person meets a lot of untrustworthy people, then she will become less trusting. The main hypothesis here is that a person's trust should adjust to the trustworthiness of people around her. We could think of people's trustworthiness as an objective probability of compliance with contracts, while trust can be thought as the individual assessment of that probability. In other words, trustworthiness is a virtue or characteristic of a person, while trust is the individual perception of oth- 3 For instance, that you can sign contracts in a world where contracts are not necessary enforceable. 3
ers trustworthiness. An issue is that we do not measure people's trustworthiness in the data, but rather people's self reported level of trust 4. However, with a lot of interaction between people in the population, we could expect that population's trust adjusts eventually to trustworthiness of people. That is, we could think that population's trust is a function of population's trustworthiness. At the same time, an individual's trust can be thought as a function of population's trustworthiness. As a result, an individual's trust would be a function of population's trust 5. The way to contrast this trust formation hypothesis is to test whether people put in a new environment adjust in trust to it. For this purpose I will conduct two exercises based on the following questions: Do immigrants adjust their trust level, departing from home country's trust, to that of the host country? Do children adjust their own trust level, departing from that taught by their parents, towards that of the population? The ideal test would consist of changing the set of people with whom a person interacts, and study how her level of trust behaves in response to this dierent set of partners. However, we cannot say exactly who interacts with whom, though we can do some experiments. An immigrant used to interact with people from her home country, and now she does it with those of her new country. A child use to interact more with her parents, and as she gets older she does it with the whole population. Hence, a main hypothesis can be derived: Hypothesis I: If there is a change in the people with whom you interact, your level of trust will adjust from your old set of partners' trust level towards that of the new set of partners. In addition, a second hypothesis can be derived regarding dierences in speed of adjustment that depend on the sign of the gap between the average trust level in the old environment and the average trust level in the new one: Hypothesis II: If you are a low trust person, you will interact less with others. Thus, it will take more time for you to adjust. In contrast, a high trust person will interact very frequently, and therefore gure out very quickly others trustworthiness. 4 This corresponds to the answer to a survey question on trust available in the European Social Survey or World Value Survey such as Generally speaking, would you say that most people can be trusted, or that you can't be too careful in dealing with people? in a scale from 0 to 10. 5 See Sapienza et al. (2008)[12] for a detailed discussion on trust and trustworthiness. Also see Tabellini (2005)[13] & Tabellini (2008)[14]. 4
This second hypothesis may raise some concerns like whether a person that changed from a high trust environment to a low trust one could adjust faster even though lower trust people in the new environment may refuse to interact with her. We could think of several features of interaction between people that could explain why this is not likely the case. If we think of incomplete contracts or not enforceable contracts, it may be natural to think that trust is needed from both parties, but it doesn't have to. There may be dierent parties that initiate the interaction, and thus dierent types of interactions. Namely, interactions in which both need to trust, in which only the rst party (i.e. higher trust person) needs to trust, or in which only the second party (i.e. lower trust person) needs to trust. The possible outcomes for all of these cases are that either the other party accepts the interaction and complies, or accepts and cheats, or does not accept at all. For instance, let's imagine you are an student renting a room in an apartment in a lower trust country where you are moving in. It could be that, since you are renting from other students, then you don't have to pay money in advance but promise you will rent the room, and the other party that she will book it for you. In this case, trust is needed from both sides. In contrast, it could also be possible that trust is needed just from one of them. For instance, she could ask you to send a month rent in advance to book the room, with just the promise that she will keep it for you. In this later case, even though the second party is from a lower trust country she doesn't have to rely on the other person's trust. Hence, she will be willing to interact with the other person even though she is a low trust person. As a result, even if people from the new environment have lower trust levels they will still get involved in interactions with the outsider, though less likely in cases where their own trust is also needed. 3 The Data 3.1 The Sources In order to study the dynamics in the formation of trust, this study relied on two sources of data: the European Social Survey (ESS) and the World Value Survey (WVS). The ESS and WVS data sets provide for each country information on individuals social values, cultural norms and behavior patterns. 5
The ESS is a biennial cross-sectional survey covering over 30 nations, most of them European. The survey has been conducted four times, the rst round was elded in 2002-2003, and the fourth in 2008-2009. The number of countries varies by round: 22 in the rst, 26 in the second, 25 in the third, and 31 in the fourth. For each country the ESS provides information on a representative sample of about 2,000 individuals. 6 Overall, the total number of individuals surveyed varies from 40,000 to 49,000 depending on the round. The WVS is also a cross sectional survey, that nowadays covers over 60 countries. It was rst administered in 1981, 10 years latter the second wave took place, and further waves followed the second wave at intervals of approximately 5 years. 7 The number of countries varies depending on the wave: 22 in the rst, 42 in the second, 54 in the third, 62 in the fourth, and 57 in the last one. In each country the survey was administered to about 1,000 to 3,500 individuals, with an average in the fth wave of about 1,450 interviews per country and a worldwide total of about 83,000 interviews. 3.2 The Empirical Strategy 3.2.1 The Immigrants Approach For the Immigrants exercise, this study used the ESS as main source of information for host countries, and the WVS as source of information only for countries of origin. The ESS is used to build the immigrants' sample, since it allows to identify people that were not born in the country (i.e. immigrants). In addition, for these people, it has information on their country of origin and asks the question of trust with intensity in answers in all its rounds. However, the ESS has a key limitation. Since these immigrants have country of origin from all over the world, the ESS does not provide information on the average level of trust in all countries of origin, only for some European countries. Therefore the need of an additional data set. I thus used the WVS as an auxiliary data set to compute the average level of trust in the countries of origin of the immigrants' sample. The reason for using the WVS just as an auxiliary data set is that it does not allow to distinguish between immigrants and natives, nor it has information on the country of origin of the 6 The size of the sample varies depending on the country's population. 7 The ve waves correspond to the periods 1981-1984, 1989-1993, 1994-1999, 1999-2004, and 2005-2008 6
person interviewed. Hence, I selected from the ESS those people that report not to have been born in the country where the survey was administered (i.e. immigrants) 8, and that immigrated at age 16 or more 9. For these people we have, at an individual level, the answer to the trust question of the ESS which is explained in detail in Subsection 3.3. We also have information on their host country, and their country of origin. Thus, using data from the ESS I computed the average level of trust in each of the host countries. Then, for those countries of origin for which we have data on the WVS, I computed the average level of trust in each of them. There are some European countries of origin that do not have data from the WVS but they do on the ESS. 10 Therefore, in order to increase the number of countries of origin for which we have average level of trust information, I calculated the average level of trust in these European countries from the ESS. Finally, I attached to each individual in the immigrants' sample the corresponding average level of trust of her host country, and her origin country. 11 For the purpose of increasing the size of the immigrants' sample I pooled the last three rounds of the ESS: round 2 (2004/2005), round 3 (2006/2007), and round 4 (2008/2009). This data is only cross-sectional, thus, we cannot follow individual people over time. We can only observe people at dierent ages. The results will be aected if there are strong cohort eects, that is, if overall country trust changes overtime. In the ESS we can check if this is or not the case. I tested the equality in mean trust by country between dierent ESS rounds. The results 8 The ESS asks the person if s/he was born in the country or not, and also asks in which country the person was born (i.e. country of origin). 9 In all the immigrants' specications, for the sake of the analysis and comparison between people living in the current country dierent number of years, I dropped people that immigrated at an age below 16 years old. Thus, for instance, we know a person that immigrated 10 years ago would have at least 26 years old. Also, this ensure that the person has been suciently exposed to the culture of the country of origin. 10 The European countries not available in the WVS are: Austria, Belgium, Czech Republic, Denmark, Estonia, Greece, Hungary, Ireland, Iceland, Luxembourg, Portugal, and Slovakia. In addition there is Israel which is also present in the ESS and not in the WVS. 11 The main reason for using the ESS for host countries instead of the WVS, relies on the fact that there are many host countries that are not present on the WVS. If we use the WVS we would either have many missing values, or the need to also use the ESS to complete the data. Therefore, the criteria adopted was to use the data set that have most of the countries present, if not all of them. As a result we have that host country data on trust comes from the ESS while origin country data on trust comes mainly from the WVS. A possible consequence could be dierences in mean level of trust between host and origin, that do not correspond to actual dierences. However, using the countries available in both data sets, I conducted some tests to check whether there are signicant dierences in the mean and variance of the variable used in each data set, and the tests indicated that overall there are no evidence of such dierences. 7
are shown in Table 1. TABLE 1. Tests of equality of mean by country across rounds Country Tests of equality of mean (P-value) R2 = R3 R2 = R4 R3 = R4 AT Austria 0.6497 --- --- BE Belgium 0.8186 0.3237 0.2917 BG Bulgaria --- --- 0.9653 CH Switzerland 0.2672 0.8471 0.4529 CY Cyprus --- --- 0.1906 CZ Czech Republic --- --- --- DE Germany 0.6028 0.9103 0.0936 DK Denmark 0.9204 0.5124 0.2369 EE Estonia 0.1821 0.0302 0.2219 ES Spain 0.0091 0.9356 0.0954 FI Finland 0.4301 0.1613 0.9846 FR France 0.7623 0.8949 0.6345 GB United Kingdom 0.5937 0.5960 0.8470 GR Greece --- --- --- HU Hungary 0.3743 0.5244 0.6555 IE Ireland 0.3415 --- --- IL Israel --- --- --- IS Iceland --- --- --- LU Luxembourg --- --- --- NL Netherlands 0.2625 0.6113 0.5681 NO Norway 0.2757 0.8364 0.1491 PL Poland 0.8067 0.6138 0.8177 PT Portugal 0.4324 0.0542 0.0021 RU Russian Federation --- --- 0.0299 SE Sweden 0.3287 0.8822 0.5971 SI Slovenia 0.3003 0.1979 0.4734 SK Slovakia 0.9480 0.0140 0.0978 TR Turkey --- --- --- UA Ukraine 0.2977 --- --- All 0.5722 0.2950 0.3051 As we can see, overall we cannot reject the null hypothesis that the mean level of trust for each country is the same across rounds 12. Moreover, when we pool all the countries together, the null hypothesis cannot be rejected either. That is, this result provides some evidence that there are no strong cohort eects across rounds. Therefore, it is more plausible to compare people with dierent ages thinking only of time adjustments. With regard to the WVS data, I used the last wave (2005/2008). The reason is that wave 5 of the WVS is the rst wave in which one of the questions of 12 I also tested the equality of standard deviation by country between dierent ESS rounds. Overall the null hypothesis was not rejected. 8
interest regarding trust values has answers with intensity instead of just being a yes or no question. This is very important in order to distinguish intensity in trust across countries, and thus to analyze adjustments paths that vary depending on the country of origin and the host country. Following this strategy, the raw immigrants' sample consisted of 11,364 individuals. After selecting people that immigrated at least at age 16, the sample got reduced to 8,322 observations. However, since we do not have data on trust levels for all countries of origin, we are left with 5,786 observations in the nal sample. This data corresponds to 29 host countries and 64 countries of origin from all over the world 13. 3.2.2 The Children Approach For the children's exercise, this study only used ESS data since mean level of trust in country of origin is not used in this approach. Thus, we are no longer interested in mixing the two data sets. Following the Immigrants Approach strategy, I pooled round 2, round 3 and round 4 of the ESS, and selected people age 16 to 75. More specically, I split the sample in two and selected a children's sample, which corresponds to natives 14 between 16 and 40 years old, and a parents' sample, which corresponds to people between 41 and 75 years old 15. With this strategy, we are left with a nal sample of 120,015 observations, and 29 ESS countries 16. Given that there are some missing values for the trust question and 13 The host countries are: Austria, Belgium, Bulgaria, Switzerland, Cyprus, Czech Republic, Germany, Denmark, Estonia, Spain, Finland, France, United Kingdom, Greece, Hungary, Ireland, Israel, Iceland, Luxembourg, Netherlands, Norway, Poland, Portugal, Russian Federation, Sweden, Slovenia, Slovakia, Turkey, and Ukraine. The countries of origin are: Burkina Faso, Egypt, Ethiopia, Ghana, Morocco, Mali, Rwanda, South, Africa, Zambia, China, Cyprus, Georgia, Indonesia, Israel, India, Iran, Islamic, Republic of Jordan, Japan, Korea, Republic of Malaysia, Thailand, Turkey, Vietnam, Australia, Bulgaria, Czech Republic, Hungary, Moldova, Republic Of Poland, Romania, Russian Federation, Slovakia, Ukraine, Argentina, Brazil, Chile, Colombia, Peru, Trinidad and Tobago, Uruguay, Canada, Mexico, United States, Denmark, Estonia, Finland, United Kingdom, Ireland, Iceland, Norway, Sweden, Spain, Greece, Italy, Portugal, Serbia, Slovenia, Austria, Belgium, Switzerland, Germany, France, Luxembourg, and Netherlands. 14 For the children's sample I only used natives to avoid immigration eects in the adjustment of trust analysis. 15 The maximum age for parents was chosen such that the oldest person in the children's sample can have parents 35 years older than them. The need of splitting the sample in two will become clear in Section 4. 16 These countries are the same ESS countries as in the previous approach. 9
region in which the person lives 17, the nal samples used in the regressions had 41,338 observations in the children's case, and 69,421 in the parents' case. 3.3 The Trust Variables The ESS asks in particular the following two trust questions: (Q1) Generally speaking, would you say that most people can be trusted, or that you can't be too careful in dealing with people? in an scale from 0 to 10, where 0 is you can't be too careful, and 10 is most people can be trusted. (Q2) Do you think that most people would try to take advantage of you if they got the chance, or would they try to be fair? in an scale from 0 to 10, where 0 is most people would try to take advantage of me and 10 is most people would try to be fair. Both questions have answers with intensity between 0 and 10, which is important for analyzing adjustment paths. However, in the WVS only the second question (Q2 ) is asked with intensity in the answers. Though Q1 is usually used in the literature, from Table 2 we can see that Q2 is very similar. Column (1) and (3) show the mean level of trust for each question in each of the ESS countries. From these columns we can see that the means in both questions are quite close. The table also displays the correlation coecient between the two questions, by country, using observations at individual level. These correlation coecients range from 0.42 to 0.66, while the overall correlation coecient is 0.58 when we pool all countries together. The right part of the table shows the mean trust by country, for those countries of the ESS also available in the WVS, and a comparison between Q2 in the ESS and in the WVS. As we can see, overall, the dierence between the two questions is small (i.e. -0.04 in average). Finally, the bottom panel of Table 2 displays correlation coecients using observations at country level. As expected, these coecients in all the cases are quite high (i.e. close to 0.9). 17 Standard errors in these regressions were clustered at a regional level, thus we lose those observations with missing values on the region. 10
TABLE 2. Comparison of mean trust levels by country in ESS and WVS trust questions Country ESS Q1 ESS Q2 Corr. Coeff. (*) WVS Q2 Diff. (1) Mean (2) Sd. (3) Mean (4) Sd. (1) & (3) (5) Mean (6) Sd. (3) - (5) AT Austria 5.14 2.38 5.83 2.31 0.51 --- --- --- BE Belgium 4.97 2.25 5.76 2.08 0.48 --- --- --- BG Bulgaria 3.41 2.68 4.38 2.60 0.59 4.61 2.19-0.23 CH Switzerland 5.70 2.16 6.44 1.98 0.45 7.00 1.99-0.56 CY Cyprus 4.35 2.63 4.85 2.10 0.65 4.53 2.46 0.32 CZ Czech Republic 4.28 2.40 5.23 2.27 0.58 --- --- DE Germany 4.84 2.33 5.82 2.13 0.46 5.55 2.22 0.27 DK Denmark 6.90 2.09 7.30 1.88 0.55 --- --- --- EE Estonia 5.31 2.20 5.61 2.28 0.52 --- --- --- ES Spain 4.98 2.07 5.32 2.01 0.51 5.17 1.97 0.15 FI Finland 6.51 1.87 6.83 1.79 0.54 6.63 2.00 0.20 FR France 4.48 2.23 5.76 2.15 0.42 6.12 2.11-0.36 GB United Kingdom 5.29 2.20 5.69 2.08 0.50 6.01 2.23-0.32 GR Greece 3.82 2.43 3.68 2.21 0.66 --- --- --- HUHungary 4.18 2.40 4.61 2.46 0.54 --- --- --- IE Ireland 5.63 2.43 6.01 2.34 0.47 --- --- --- IL Israel 5.29 2.42 5.49 2.36 0.53 --- --- --- IS Iceland 6.37 2.29 6.90 1.82 0.45 --- --- --- LU Luxembourg 5.02 2.46 5.64 2.50 0.50 --- --- --- NL Netherlands 5.85 2.07 6.30 1.81 0.55 6.47 1.70-0.16 NO Norway 6.69 1.87 6.95 1.82 0.54 7.28 1.88-0.33 PL Poland 3.93 2.39 4.74 2.41 0.43 4.66 2.49 0.08 PT Portugal 3.87 2.30 4.86 2.18 0.53 --- --- --- RU Russian Federatio 3.95 2.71 4.90 2.67 0.47 5.40 2.78-0.50 SE Sweden 6.22 2.11 6.61 1.96 0.54 7.17 1.88-0.57 SI Slovenia 4.17 2.54 4.87 2.46 0.53 5.53 2.45-0.66 SK Slovakia 4.16 2.41 4.63 2.35 0.61 --- --- --- TR Turkey 2.95 2.73 3.65 2.90 0.47 4.75 2.59-1.10 UA Ukraine 4.29 2.69 4.71 2.48 0.59 5.35 2.37-0.64 ALL countries 4.98 2.48 5.59 2.35 0.58 WVS countries 5.04 2.48 5.69 2.31 0.57 5.73 2.40-0.04 Correlation coefficients using country level observations Corr. Coeff. Q1 ESS - Q2 ESS Corr. Coeff. Q1 ESS - Q2 ESS Corr. Coeff. Q2 ESS - Q2 WVS Corr. Coeff. Q1 ESS - Q2 WVS 0.96 (for all countries in ESS) 0.97 (for countries both in ESS and WVS) 0.91 (for countries both in ESS and WVS) 0.88 (for countries both in ESS and WVS) (*) Correlation coefficients in this column are calculated using observations at an individual level Hence, we can conclude from this table that there is some evidence on the similarity between Q1 and Q2. Consequently, to avoid mixing the questions in a same exercise, this study used Q2 for the Immigrants Approach, since it is the only question available in both data sets with intensity in the answers, and Q1 for the Children Approach 18. 18 I also tried using Q2 for the Children Approach, and the results were similar. 11
3.4 Descriptive Statistics In order to have a broader idea of how the main variables in the nal sample look like, Table 3 displays some descriptive statistics. Panel A shows descriptive statistics for those variables that are present in both approaches, while Panel B and C show them for variables specic to Immigrants Approach and Children Approach respectively. In particular, for the Immigrants Approach, we can observe that an average individual has a level of trust of 5.65, with an standard deviation of 2.35. In addition, from Panel B, we can notice that the average trust among countries of origin is a bit smaller than that of host countries. This suggests that an average individual tends to migrate from a low trust country to a high trust country. Also, the ESS provides information on the number of years that an immigrant has been living in the country 19. As we can see, 26% of the people has been living in the country 0 to 10 years, 17% of the people have been living in the country 11 to 20 years 20, and 57% of the people have been living in the country for more than 20 years. Finally, we can group the countries of origin by world region to have a broader idea of where our immigrants come from. This is shown in the lower part of Panel B. Most of them come from European countries (i.e. 80%), in particular half of these European immigrants come from East Europe. Then, 7% come from Asia, and another 7% from Africa, while only 4% and 2% come from Latin America and North America respectively. With respect to Australia-Oceania, it represents just a 0.04% of the immigrants' sample. Now, with regard to the Children Approach, it is worth noting that an average children has a level of trust of 5.04, with an standard deviation of 2.37. In particular, as we can observe from Panel C, this is similar to the estimated average trust for parents (i.e. 5.05), and to the average trust among all countries (i.e. 4.94). This is consistent with the fact that the percentage of children that have parents with higher trust than the country is a bit higher than the percentage 19 The ESS asks the question how long ago did you rst come to live in [country]?, and the possible answers are: within last year, 1-5 years ago, 6-10 years ago, 11-20 years ago, more than 20 years ago. To have more balanced categories and ease the analysis later on, the three rst categories were pooled together under the category name 0 to 10 years ago. 20 It is worth to notice that having only 17% of the people living in the country between 11 and 20 years could potentially be a problem for our analysis since this percentage is quite small with respect to the other two categories. Ideally, to avoid these issues, we would need the exact number of years an immigrant has been living in the country. Unfortunately, this information is not available. Thus, one should be careful when interpreting the results for this middle group. 12
of children with relatively lower trust parents (i.e. 57% vs. 43%). TABLE 3. Descriptive statistics for Immigrants Approach and "Children" Approach All rounds Immigrants "Children" Variable Obs Mean Std. Dev. Min Max Obs Mean Std. Dev. Min Max A. General variables Individual Trust 5,786 5.65 2.35 0 10 41,338 5.04 2.37 0 10 Age 5,742 53.86 15.17 17 99 41,338 28.07 7.39 16 40 Male 5,778 0.45 0.50 0 1 41,314 0.48 0.50 0 1 Primary education 5,658 0.15 0.36 0 1 40,370 0.08 0.27 0 1 Secondary education 5,658 0.43 0.50 0 1 40,370 0.60 0.49 0 1 More than secondary educ. 5,658 0.42 0.49 0 1 40,370 0.32 0.47 0 1 Father's primary education 5,035 0.40 0.49 0 1 41,338 0.22 0.42 0 1 Father's secondary education 5,035 0.37 0.48 0 1 41,338 0.52 0.50 0 1 Father's more than secondary educ. 5,035 0.22 0.42 0 1 41,338 0.25 0.43 0 1 Mother's primary education 5,279 0.47 0.50 0 1 41,338 0.24 0.43 0 1 Mother's secondary education 5,279 0.39 0.49 0 1 41,338 0.53 0.50 0 1 Mother's more than secondary educ. 5,279 0.15 0.35 0 1 41,338 0.23 0.42 0 1 B. Immigrants variables Mean trust in origin country 5,786 5.41 0.74 3.63 7.95 Mean trust in current country 5,786 5.79 0.68 3.63 7.38 Lived in country 0 to 10 years 5,695 0.26 0.44 0 1 Lived in country 11 to 20 years 5,695 0.17 0.38 0 1 Lived in country + 20 years 5,695 0.57 0.49 0 1 Asia 5,786 0.064 0.24 0 1 Australia-Oceania 5,786 0.004 0.06 0 1 East Europe 5,786 0.411 0.49 0 1 West Europe 5,786 0.176 0.38 0 1 North Europe 5,786 0.107 0.31 0 1 South Europe 5,786 0.109 0.31 0 1 Africa 5,786 0.069 0.25 0 1 Latin America 5,786 0.038 0.19 0 1 North America 5,786 0.022 0.15 0 1 C. "Children" variables Mean trust in country 41,338 4.94 1.00 2.94 6.94 Estimated trust of parents 41,338 5.05 0.35 3.43 5.65 Parent's sample age 69,421 55.97 9.66 41 75 4 The Results In this section I will discuss in detail the specications that were estimated in each of the two approaches, their interpretation, and the main results. 4.1 The Immigrants Approach The underlying idea in this approach is that an individual's trust in the host country is the result of an adjustment process departing from the trust level in her origin country. That is, the level of trust of an immigrant that arrives to a new country is a function of the home country trust, and the more she stays and 13
interacts in her new country, the more she will adjust her trust towards that of the host country trust. As discussed in Section 3, this exercise is based on a sample of immigrants, dened as people that were not born in their current country and that immigrated at age 16 or more. The baseline specication, which is estimated by OLS clustering errors by individuals with same country of origin that live in the same current country 21, is the following: T rust icot = α + γt rusto io + δ 1 Lived2 ict + δ 2 Lived3 ict +θ 1 (Diff ico Lived1 ict ) + θ 2 (Diff ico Lived2 ict ) (1) +θ 3 (Diff ico Lived3 ict ) + β X it + η t + υ icot, where T rust icot is the level of trust of individual i in host country c with country of origin o interviewed in round t of the ESS, α is a constant term, T rusto io is the average trust in origin country o of individual i 22, Diff ico is the dierence in average trust between host country and origin country of individual i 23, Lived1 ict is a dummy that takes value 1 if the individual has been living in the host country between 0 and 10 years at the time of the interview t, Lived2 ict is a dummy that takes value 1 if the individual has been living in the host country between 11 and 20 years at the time of the interview t, Lived3 ict is a dummy that takes value 1 if the individual has been living in the host country more than 20 years at the time of the interview t. The interaction terms between Diff ico and Lived1 ict, Lived2 ict or Lived3 ict take into account that the adjustment towards the host country's level of trust may vary depending on how many years the individual has been living there. X it is a vector that contains a quartic on individual's age and individual's demographic controls that can aect trust 24, and η t controls for ESS round xed 21 The idea behind the choice of clusters in the errors is that we could think, for instance, that an Argentinean that goes to live to Italy is not the same as an Argentinean that goes to live to Spain in terms of unobservables, but two Argentinean living in Spain are similar. 22 I also tried using the mode instead of mean trust, and the results were similar. 23 Mean trust in the host country was calculated using only natives, to avoid spurious correlation and to have a cleaner measure of trust in the host country. 24 Demographic controls includes: a dummy for male gender, two dummies for father's secondary and more than secondary education (the omitted category is primary education), two dummies for mother's secondary and more than secondary education (the omitted category is primary education), and two dummies for secondary and more than secondary education of the 14
eect. According to our hypothesis the coecient of T rusto io (i.e. γ) should be positive and high, since this coecient reects how much of an individual's trust was inuenced by the country of origin at the moment the individual arrived to the new country. That is, for instance, a person that comes from a high trust country is expected to exhibit high trust when she has just immigrated to the host country. Moreover, we could expect this initial trust level to be quite close to that of the home country in which the individual has previously lived and interacted. In addition, our hypothesis suggests that as a person interacts with other people in the host country, she will eventually learn about others' trustworthiness, and adjust over time her own trust to that typical of the new country. Hence, the longer a person lives in a country the more she should adjust to the new country's trust level. The total adjustment experienced by a person that has been living in the country 0 to 10 years, 11 to 20 years, and more than 20 years is given by the coecients on the interaction terms between Diff ico 25 and Lived1 ict, Lived2 ict or Lived3 ict (i.e. θ 1, θ 2, and θ 3 respectively). Hypothesis I then implies that θ 1, θ 2, and θ 3 should be positive, and in particular θ 1 < θ 2 < θ 3. In other words, for instance, a person that has been living in a host country for more than 20 years is expected to have adjusted more to the trust level of this host country than a person who has been living there for 10 years or less (i.e. θ 1 < θ 3 ). We could also expect that a person that has been living in the current country for more than 20 years would have adjusted almost completely in terms of trust (i.e. θ 3 = 1). The results corresponding to this rst exercise are shown in Table 4. From column (1) to (4) the dierent variables of model (1) are added step by step. Hence, column (1) to (3) correspond to simplied versions, while column (4) corresponds to the full specication as stated in equation (1). Thus, in order to interpret the results I will focus on column (4). individual (the omitted category is primary education). 25 The Diff ico variable could be thought as the distance in trust units that a person has to travel in order to reach the trust level of her host country. The direction can be up or down depending on whether Diff ico is positive or negative. For instance, a positive Diff ico means that the current country trust level is higher than that of the home country. 15
TABLE 4. Immigrants Approach: using difference between current and origin country mean trust levels Dependent variable: Individual trust in current country VARIABLES (1) (2) (3) (4) (5) Mean trust in origin country (TrustO) 0.9649*** 0.9784*** 0.9830*** 0.9856*** 1.0000 (0.0557) (0.0529) (0.0520) (0.0585) (0.0000) Diff. in trust: current country - origin country (Diff) 0.7666*** (0.0531) Lived in current country 11 to 20 years (11 to 20 years) -0.0444-0.1166-0.0393-0.0380 (0.1097) (0.1222) (0.1303) (0.1307) Lived in current country 20+ years (20+ years) -0.1003-0.1854-0.0600-0.0599 (0.0880) (0.1291) (0.1502) (0.1501) Diff * 0 to 10 years 0.7112*** 0.7285*** 0.7173*** 0.7249*** (0.0709) (0.0719) (0.0750) (0.0689) Diff * 11 to 20 years 0.5506*** 0.5611*** 0.6330*** 0.6392*** (0.0830) (0.0831) (0.0853) (0.0807) Diff * 20+ years 0.9129*** 0.9274*** 0.9469*** 0.9549*** (0.0616) (0.0600) (0.0619) (0.0535) Male -0.0967-0.1060-0.1057 (0.0632) (0.0707) (0.0707) Secondary education father -0.0385-0.0401 (0.0905) (0.0900) More than secondary education father -0.0337-0.0363 (0.1098) (0.1090) Secondary education mother 0.2436*** 0.2416*** (0.0897) (0.0898) More than secondary education mother 0.1551 0.1541 (0.1289) (0.1288) Secondary education 0.0161 0.0154 (0.0965) (0.0963) More than secondary education 0.3935*** 0.3924*** (0.1162) (0.1157) ESS round FE Yes Yes Yes Yes Yes Quartic on age No No Yes Yes Yes Tests of coefficients (p-value reported) TrustO = 1 0.528 0.683 0.744 0.806 --- Diff * 0 to 10 years = Diff * 11 to 20 years 0.099* 0.087* 0.415 0.408 Diff * 0 to 10 years = Diff * 20+ years 0.016** 0.017** 0.009*** 0.009*** Diff * 20+ years = 1 0.158 0.227 0.392 0.400 Observations 5,786 5,695 5,647 4,694 4,694 R-squared 0.061 0.065 0.067 0.087 0.098 All models include a constant. Model (5) is estimated restricting to 1 the coefficient of mean trust in origin country of model (4) (test reported in table). The omitted category in number of years lived in current country is 0 to 10 years. The omitted category in all the education groups of variables is primary education. Robust standard errors in parentheses are clustered by individuals with same country of origin living in same current country. *** p<0.01, ** p<0.05, * p<0.1 First of all, as expected, we can observe that the coecient of T rusto io is positive and not statistically dierent from one, according to the test reported in the bottom part of the table. In addition, the coecients of the interaction terms are all positive and signicant at 1% level. In particular, as expected, a person that has been living in the country for 10 years or less has not fully adjusted to the level of trust of the new country (i.e. the coecient is around 0.7 and statistically 16
dierent from 1). Though the interpretation for people that have been living in the country between 11 to 20 years is not quite clear, we cannot reject the null hypothesis that a person that has been living in the country between 0 and 10 years has experienced a similar adjustment than one that has been living in the country for 11 to 20 years 26. In contrast, the coecient on the third interacted term is greater than the other two and not statistically dierent from one. This suggests that after living in the country for more than 20 years a person has fully adjusted her trust towards that of her current country. With respect to the coecients of the control variables, it is worth mentioning that more educated people or people with more educated mothers tend to report higher levels of trust. Finally, column (5) shows the same specication as column (4) but restricting the coecient of T rusto io to one, as the tests suggested. Given that after constraining this coecient the results are similar, this modied specication will be used in the remaining of this section. Summing up, Table 4 suggests that an immigrant tends to adjust her level of trust over time departing from that of her country of origin towards that of her new country. 4.1.1 From Low to High & High to Low It is reasonable to think that the adjustment path may not be the same for a person that comes from a low trust country and goes to a high trust country, than for a person that comes from a high trust country and goes to a low trust country. In particular, hypothesis II suggests that a person that comes from a low trust country will interact less with other people, and thus need more time to adjust. In contrast, a person that comes from a high trust country, will interact very frequently, and therefore gure out more quickly others' trustworthiness. Based on these ideas, I also estimated by OLS an extended version of the baseline model (1) in which I allowed for the Diff ico coecient to dier depending on whether Diff ico is positive (movements from lower trust to higher trust countries) 26 This can be due to the fact that, as mentioned in the Data Section, the percentage of people living in the country between 11 to 20 years is quite small with respect to the other categories. 17
or negative (movements from higher trust to lower trust countries): T rust icot = α + γt rusto io + δ 1 Lived2 ict + δ 2 Lived3 ict +θ 1 (Diff LH ico Lived1 ict ) + θ 2 (Diff LH ico Lived2 ict ) +θ 3 (Diff LH ico Lived3 ict ) + θ 4 (Diff HL ico Lived1 ict ) (2) +θ 5 (Diff HL ico Lived2 ict ) + θ 6 (Diff HL ico Lived3 ict ) +β X it + η t + υ icot, where Diff LH ico (Diff HL ico ) is the dierence in average trust level between current and origin country of an individual i that moved from a lower (higher) trust country relative to the host country 27. The interpretation is similar to that of model (1) with the exception that now we allow for dierent speeds of adjustments depending on the direction of the movement: from low to high or high to low. More precisely, we should still expect the coecient of T rusto io to be positive and close to 1, and the coecient on the interacted terms to be positive. Now, with regard to the new dynamics, our hypothesis suggests that the speed of adjustment should be higher for a person that comes from a higher trust country than for a person than comes from a lower trust country. For instance, a person that has been living in a host country for more than 20 years and comes from a higher trust country is expected to adjust more quickly than one that has been living in the country for a similar number of years but comes from a lower trust country (i.e. θ 3 < θ 6 ). The results are shown in Table 5. As we can see, in earlier periods (i.e. for people that has been living in the country 0 to 10 years, or 11 to 20 years) it is not clear that there exists a dierent adjustment speed for people that come from lower vs. higher trust countries. In contrast, the coecient of Diff HL ico for people that have been living in the country for 20+ years is higher (i.e. 1.13) than the one of Diff LH ico (i.e. 0.87). 27 The rest of the variables are dened as in the baseline specication. 18
TABLE 5. Immigrants Approach: using difference between current and origin country mean trust levels, split up into movements from low trust to high trust countries (LH) and from high trust to low trust Dependent variable: Individual trust in current country VARIABLES (1) (2) (3) (4) Mean trust in origin country (TrustO) 1.0000 1.0000 1.0000 1.0000 (0.0000) (0.0000) (0.0000) (0.0000) (+) diff. in trust: current country - origin country (Diff low to high) 0.7467*** 0.8101*** (0.0577) (0.0608) (-) diff. in trust: current country - origin country (Diff high to low) 0.8568*** 0.8020*** (0.1026) (0.1018) Lived in current country 11 to 20 years (11 to 20 years) -0.0328-0.0565 (0.1667) (0.1802) Lived in current country 20+ years (20+ years) 0.0029 0.0463 (0.1367) (0.1945) Diff low to high * 0 to 10 years 0.7251*** 0.7635*** (0.0950) (0.1030) Diff low to high * 11 to 20 years 0.5510*** 0.6942*** (0.1219) (0.1270) Diff low to high * 20+ years 0.8059*** 0.8690*** (0.0792) (0.0859) Diff high to low * 0 to 10 years 0.7194*** 0.6554*** (0.1450) (0.1439) Diff high to low * 11 to 20 years 0.5762*** 0.5508*** (0.1742) (0.1630) Diff high to low * 20+ years 1.1624*** 1.1301*** (0.1612) (0.1718) ESS round FE Yes Yes Yes Yes Quartic on age No Yes No Yes Demographic controls No Yes No Yes Tests of coefficients (p-value reported) Diff low to high * 0 to 10 years = 1 0.004*** 0.022** Diff low to high * 0 to 10 years = Diff low to high * 11 to 20 years 0.268 0.664 Diff low to high * 11 to 20 years = Diff low to high * 20+ years 0.065* 0.241 Diff low to high * 20+ years = 1 0.015** 0.128 Diff high to low * 0 to 10 years = 1 0.053* 0.017** Diff high to low * 0 to 10 years = Diff high to low * 11 to 20 years 0.512 0.621 Diff high to low * 11 to 20 years = Diff high to low * 20+ years 0.017** 0.022** Diff high to low * 20+ years = 1 0.314 0.449 Diff low to high * 0 to 10 years - Diff high to low * 0 to 10 years = 0 0.977 0.595 Diff low to high * 10 to 20 years - Diff high to low * 10 to 20 years = 0 0.920 0.553 Diff low to high * 20+ years - Diff high to low * 20+ years = 0 0.101 0.261 Observations 5,786 4,758 5,695 4,694 R-squared 0.084 0.094 0.088 0.099 All models include a constant. All models are estimated restricting to 1 the coefficient of mean trust in origin country. A person is considered to go from a low (high) trust to a high (low) trust country if the difference "mean level of trust in current country minus mean level of trust in origin country" is positive (zero or negative). The omitted category in number of years lived in current country is 0 to 10 years. Demographic controls includes: a dummy for male gender, two dummies for father's secondary and more than secondary education (the omitted category is primary education), two dummies for mother's secondary and more than secondary education (the omitted category is primary education), and two dummies for secondary and more than secondary education of the individual (the omitted category is primary education). Robust standard errors in parentheses are clustered by individuals with same country of origin living in same current country. *** p<0.01, ** p<0.05, * p<0.1 This suggests that after living for a long time period in the new country, people that came from higher trust countries have adjusted completely to the new 19