Surveys generally include other questions related to trust. For instance, the WVS asks the “fair question”: “Do you think most people would try to take advantage of you if they got the chance, or would they try to be fair?” The GSS includes the trust question and the fair question, and adds the “help question”: “Would you say that most of the time people try to be helpful, or that they are mostly just looking out for themselves?” These different questions are sometimes used to build indexes intended to provide alternative measures of trust or get an average indicator of moral values or civic capital (Tabellini, 2010; Guiso et al., 2011).
Although most of the surveys directly ask questions about generalized trust based on evaluations, there have also been attempts to measure trust with questions on expectations about what would happen in a given concrete situation. One of the most well-known examples is the “lost wallet” question used in the Gallup World Poll: “If you lost a wallet or a purse that contained items of great value to you, and it was found by a stranger, do you think it would be returned with its contents, or not?” However, this question is limited to a small number of surveys so far and the hypothetical nature of the question prevents it from being a real behavioral measure.
In the case of institutional trust, questions are traditionally formulated through a common heading: “Do you have confidence in your …?” (The space is filled in with an entity taken from a list of institutions, such as “government” or “congress.”) Possible answers are generally “yes/no/don’t know,” or a response on a scale from 0 to 10. The surveys generally ask questions about different, mainly public, institutions (e.g., the parliament, the courts, the government, the armed forces). Some questions also refer to those who are in charge of implementing the policies (e.g., civil servants, police officers, MPs). As noted by Delhey, Newton, and Welzel (2011), institutional trust can vary depending on the institution, and so it is recommended that questions are asked for each specific institution, rather than attempting to measure institutional trust as a single construct, or combine several questions into a trust index. Nevertheless, the OECD Guidelines on Measuring Trust, using a principal component analysis of different types of trust questions in the World Values Survey, find three main factors for institutional trust: trust in nongovernmental institutions (major companies, banks, universities, environmental organizations, women’s organizations), trust in political institutions (government of the day, political parties, parliament, civil service), and trust in law and order institutions (armed forces, police, courts). While this analysis confirms the salience of distinct sub-dimensions of institutional trust, it also highlights that many of the finer distinctions often made between different categories of trust are not very informative empirically, and that a relatively narrow range of measures that covers these broad types of institutions will cover the most important aspects (OECD, 2017). Of course, depending on the needs of especially policy-makers, asking institution-specific questions can still be worthwhile.
Behavioral and Experimental Measures of Trust
Survey data supply subjective information—how people judge and feel—which requires caution in use and interpretation. Issues include how individuals interpret the question they are asked, and whether there are systematic differences between groups in that interpretation that might be misinterpreted as differences in the underlying level of trust. For example, individuals who respond that you “need to be very careful” to the trust question could be motivated by a strong aversion to risk (Fehr, 2009; Bohnet and Zeckhauser, 2004; Guiso, Sapienza, and Zingales, 2011). Surveys are generally unable to assess and disentangle the variety of social preferences that can be involved in inter-personal trust such as altruism, reciprocity, or social desirability and reputation, as discussed above.
For this reason, a revolution in experimental economics has led to the development of laboratory experiments designed to elicit a large variety of social behaviors, through protocols such as the “trust game” (described below) or the “public goods game.” These carefully calibrated experiments, which measure the behavior and choices of people, with monetary incentives at stake, not only help disentangle different types of trust but also provide benchmarks against which survey questions can be compared, to determine whether survey questions are measuring actual behavior.
These games focus on a definition of trust that can be directly measured with experimental games, as shown by Fehr (2009). The trust that is measured in these experiments is best thought of as a behavior following Coleman’s concept, according to whom “an individual trusts if he or she voluntarily places resources at the disposal of another party without any legal commitment from the latter, but with the expectation that the act of trust will pay off” (Coleman, 1990). This conception has two elements: a behavioral one, and an expectation that the act of trusting will be of benefit for the person granting it.
In general, these experiments use variants of the “investment game,” also known as the “trust game,” of Berg, Dickhaut, and McCabe (1995). In laboratory experiments, this game is played as follows. In stage 1, the subjects in rooms A and B are each given 10 dollars as a show-up fee. While subjects in room B pocket their show-up fee, subjects in room A must decide how much of their 10 dollars to send to an anonymous counterpart in room B. The amount sent, denoted by M, is tripled, resulting in a total return of 3M. In stage 2, a counterpart in room B is given the tripled money and must decide how much to return to the subject in room A. “Trust in others” is measured by the amount sent initially by the sender. Trustworthiness is measured by the amount sent back by the player in room B.
This framework can be adapted and supplemented with complementary experiments to measure trust as distinct from other attitudes, such as risk aversion, altruism, and reciprocal behaviors, and to distinguish between trusting behavior as a deep-seated preference, and trusting behavior as a function of one’s beliefs about the trustworthiness of others (which can be quickly revised). For example, a positive correlation between the amount sent and the amount returned may reflect a preference for reciprocity. Disentangling altruism from reciprocity may be done by complementing the trust game with the dictator game, where one player has to decide what portion of a sum to share with another player who cannot react to the transfer and has no initial endowment of their own. Experiments along this line by Cox (2004) have demonstrated that reciprocity exists and that the trust motive exists separately from altruism. Other experiments have used measures of risk aversion alongside trust games.
Other studies have used neurobiological methods to measure the role of trust in comparison with preferences with greater precision. Oxytocin, a hormone released especially during breast-feeding and giving birth, is associated with sentiments of affinity and socialization. It is known for deactivating the transmission of feelings of anxiety related to the belief of being betrayed. Kosfeld et al. (2005) evaluated the effect of oxytocin on the pro-social behavior of individuals participating in trust games. The authors proposed additional experiments to distinguish pro-social preferences from risk-taking behavior and from the level of optimism of the participants. The participants in this study were randomly allocated into two groups. The first group inhaled oxytocin through a spray; the second inhaled a placebo and served as the control group. Results from this experiment show that individuals who received oxytocin displayed more trust, and that they continued to behave trustingly in the exchange with others even if the latter did not show any reciprocity. By contrast, other attitudes, such as prudence and risk-aversion, or even beliefs such as optimism in the actions of others, were not affected. Based on this evidence, Kosfeld et al. concluded that the trust game measures veritable preferences for cooperation, rather than risk aversion or anticipations of the others’ actions (see Fehr, 2009, for a survey of experimental measures of trust).
Validity of Trust Measures
The OECD’s Guidelines on Measuring Trust (OECD, 2017) distinguish several criteria for assessing the statistical qua
lity of trust measures, in particular:
• Reliability: the degree to which the measures of trust produce consistent information over time and across different vehicles.
• Face validity: the degree to which a measure is intuitively plausible, measured by the nonresponse rate to the question.
• Convergent validity: whether a measure of trust correlates well with objective measures (e.g., whether self-reported trust correlates with behavioral trust).
• Construct validity: whether a measure behaves as common sense and theory dictate.
General Issues with Existing Survey-Based Measures
There are many practical issues with existing survey-based data on trust that are at least as important, and possibly more so, than the conceptual issues described above.
• Data generally come from nonofficial surveys with very small sample sizes, typically of around 1,000 per country, and sometimes low response rates. This raises concerns about noise-to-signal ratio and nonresponse bias in the sample. The lack of representative samples also makes it very difficult to get a comprehensive description of the level of trust at the local level and to analyze the economic, social, and policy determinants of trust.
• Data coverage is also relatively poor, particularly over time. As an illustration, one of the most used surveys in the literature, the WVS, provides waves only at irregular intervals (every 5 years on average) and the countries covered vary from wave to wave. Lack of time-series data makes it difficult to look at what drives changes in trust. In particular, these databases cannot be used to analyze how policy reforms affect the evolution of trust in others and in institutions.
• Different surveys are very heterogeneous in the question wordings, limiting comparability across surveys. In the case of institutional trust, for instance, questions sometimes refer to similar concepts but use different descriptions, such as “courts” or the “judicial system,” “politicians” or “the government.” Also, different surveys use different response scales. Some surveys rely primarily on a “yes/no/don’t know” response format (GWP), while others surveys such as the ESS and the WVS use longer numeric scales (0–10 or 1–4). This raises a real issue since researchers are forced to rely on different data sources for different groups of countries (WVS for developing countries, and ESS for European countries in the figures used in this chapter) to cope with the poor geographical and time coverage of each survey.
Validity of Survey-Based Measures of Inter-personal Trust
The OECD Guidelines on Measuring Trust (OECD, 2017) find strong evidence for the validity of measures of inter-personal trust: they are consistent across different data sources and over time, their nonresponse rates are relatively low, and they are highly correlated with a large variety of social and economic outcomes. Their policy relevance has also been supported by academic research, as discussed below.
Studies that have analyzed the relationship between survey answers from the generalized trust question and the amount sent in the trust game, an indication of convergent validity, found mixed results. Some studies found that the trust question predicted some aspect of trust behavior, either trustworthiness or trustfulness, but not always the same one (Glaeser et al., 2000; Fehr et al., 2002; Lazzarini et al., 2005; Ermisch et al., 2009). Other studies found differences in the relationship of the trust question to behavior in different countries (Holm and Danielson, 2005). However, differences between the conclusions of the studies may also be due to differences in experimental design, as the designs of the games are not identical between the different experiments. The recently launched TrustLab project, jointly launched by the OECD and France’s Sciences-Po international research university, is the first international database on people’s behaviors, social norms, and preferences, and collects information on trust via survey and experimental measures in a comparable way and on nationally representative samples in different OECD countries (see “TrustLab” sidebar). TrustLab analysis from the first set of countries has found that survey and experimental measures of trust are positively correlated, and that the survey measure of trust, when controlling for other factors, captures altruism and expected trustworthiness of others (Murtin et al., 2018). This confirms the argument that expected trustworthiness of others, rather than one’s own trustworthiness, matters the most for evaluating trust in other people (Fehr, 2009). Overall, the OECD Guidelines therefore conclude that survey questions on interpersonal trust provide valid and reliable information, and there is a strong case for including them in official statistical vehicles.
Validity of Survey-Based Measures of Institutional Trust
The validity of survey-based measures of institutional trust is more mixed, but still positive (OECD, 2017). Although their potential policy relevance is clear and they perform well in terms of construct validity, some interpretation and statistical issues still remain open. Nevertheless, although the evidence base on their validity is not as strong as for measures of inter-personal trust, the OECD recommendation is that these measures should also be collected by official statistics.
Differences in interpretation by respondents can be particularly important with respect to questions on institutional trust. People may interpret “How much do you trust the government?” in several different ways. Do they think the government is competent to deliver services? Do they think the government is honest? Do they think that the government will enact good policies? The answer to these questions may be very different depending on which interpretation the respondent uses. Questions on trust in institutions do not necessarily measure something structural about how well institutions work, since people might answer these questions by thinking about the government in power at the time or about the deep-seated traits of a political system.
In addition, statistics on institutional trust must be used with caution. Should these statistics be used as a measure of people’s perception of institutional trustworthiness, or as a measure of the actual objective level of trustworthiness or transparency? That is, should differences in measures of institutional trust across countries be taken as an indicator, for example, of different levels of corruption? It is difficult to distinguish between beliefs and perceptions on the one hand, and objective measures on the other, especially when individuals are asked about the extent of transparency or corruption of various institutions (Charron, 2016).
So far there has been little evidence on the convergent validity of institutional trust measures. While for inter-personal trust there is a consensus on using generalized trust as the preferred measure, institutional trust covers several dimensions, all of which are of interest (police, banks, and so on). Furthermore, with respect to experimental analyses (against which survey measures might be compared, as in the case of inter-personal trust), the standard trust game does not have an experimental counterpart for the analysis of trust in institutions. However, there are studies that rely on Implicit Association Tests, a method from experimental psychology, to validate institutional trust questions, with promising results (Intawan and Nicholson, 2017). The TrustLab project also includes an Implicit Association Test as experimental measure of institutional trust, and, encouragingly, finds that, controlling for a range of individual characteristics, experimental trust in government is significantly and positively related with survey data on trust in government and trust in the judicial system (Murtin et al., 2018).
Going forward, more consistent and harmonized data will increase the evidence base available and allow researchers to better understand and improve these measures.
TRUSTLAB: MEASURING TRUST AND SOCIAL NORMS THROUGH EXPERIMENTAL TECHNIQUES
TrustLab is an experimental platform developed by Sciences-Po and the OECD to:
• Produce new measures of trust and social norms using a range of techniques.
• Compare trust and social norms across countries and groups of people.
• Understand the drivers of trust at the individual level.
The platform combines experimental an
d nonexperimental techniques. As such, it overcomes some limitations of the experimental approaches used so far, in particular their very small sample sizes, the use of samples that are not nationally representative, and the fact that experimental findings are not linked to comparable survey data.
TrustLab relies on an integrated online platform developed by MediaLab Sciences-Po. In every participating country, a representative national sample of 1,000 people answers a number of traditional survey questions and participates in experimental games providing both behavioral and self-reported information. Games are played with real resources at stake (mean value around 15 euros). Table 10.1 presents the different survey modules and instruments that are used in TrustLab.
Table 10.1. Modules Featuring in TrustLab
StatLink 2 http://dx.doi.org/10.1787/888933840076.
Data from TrustLab provide a rich description of different social norms (trust in others, trustworthiness, altruism, cooperation, reciprocity) and of trust in different institutions (government, parliament, police, judicial system, media, and banks), in addition to some of the potential determinants of trust.
Following a pilot phase in 2016, TrustLab has now been implemented in France, Korea, Slovenia, the United States, Germany, Italy, and the United Kingdom, with a range of academic and governmental partners joining the effort.
Validity of Experimental Measures of Trust
For Good Measure Page 47