Abstract
In their October 2011 ASR article, “How General Is Trust in ‘Most People’? Solving the Radius of Trust Problem,” Delhey, Newton, and Welzel pioneer the measurement of trust radius, defined as the width of the circle of people among whom a certain trust level exists. One key finding of their study is that trust radius correlates so strongly with out-group trust level that the two are essentially identical constructs; this clashes with standard theory’s finding that trust level and trust radius are distinct trust constructs. This result proves to be erroneous, however, due to a mistaken label. I seek to correct this labeling error and find that trust radius is, in fact, a separate trust construct. Trust radius differs from established trust level constructs such as level of trust in most people and level of out-group trust, not only theoretically but empirically as well.
Trust is widely recognized as a key factor in the functioning of societies (Coleman 1990; Cook 2001; Hardin 2002; Sztompka 1999). Theory identifies trust radius and trust level as two distinct but equally quintessential components of trust (e.g., Fukuyama 1995). Trust level concerns the intensity with which people trust each other, this can be high or low. Individuals’ trust level matters because it determines their willingness to cooperate with others. Trust radius, on the other hand, concerns the width of the circle of people among whom a certain trust level exists, this can be broad or narrow. Trust radius matters because it determines with whom individuals are willing to cooperate, thus affecting the reach of productive social exchange within society (Fukuyama 2001).
Empirical studies of trust—its determinants, correlates, and consequences—focus on trust level, for which data and measures are readily available, and pay only scant attention to trust radius. The classic trust item—asking respondents whether “most people can be trusted” or “you need to be very careful in dealing with people”—concerns the level, high or low, of respondents’ trust in “most people.” Other items ask respondents how much they trust specific groups, notably the out-group (e.g., level of trust in other nationalities). However, even out-group trust level needs to be distinguished conceptually from trust radius. Out-group trust concerns the intensity of trust in individuals who fall outside the “inner-circle,” whereas trust radius concerns the scope of trust, that is, how wide a person’s circle of trusted individuals is. Hence, even when out-group trust is high, people can still reserve higher levels of trust for a relatively narrow group of individuals.
Delhey, Newton, and Welzel (2011) (henceforth DNW) took an enormous step forward with the development of a method for quantifying trust radius. However, one chief finding of DNW (pp. 798–800) is that trust radius correlates so strongly with out-group trust level (r = .94) that the two are essentially identical constructs and not distinct. When re-analyzing the same data, I found this to be incorrect, however, and the actual correlation proves to be much lower (r = .381). DNW have been led in the wrong direction because they mislabeled an earlier version of their out-group trust measure as the trust radius measure, logically resulting in a high correlation between the two (Christian Welzel, e-mail, October 9, 2012). The same error is repeated in Table 2 of DNW (p. 799), which considers associations between the incorrect trust radius measure and various civicness indicators known to correlate with trust level (e.g., gender equality and democracy). That is, Table 2 does not consider associations between indicators of civicness and trust radius but between indicators of civicness and out-group trust.
Here I address DNW’s labeling error, focusing on the key conceptual issue of trust level and trust radius being distinct trust constructs (or not). I relate measured trust radius to three established trust level measures (level of trust in most people, level of out-group trust, and—not yet mentioned—level of in-group trust) and find that trust radius is clearly distinct from trust level, not only theoretically (cf. Fukuyama 1995, 2001) but also empirically. This re-analysis of DNW’s data was catalyzed by some inconsistencies in their country data and sample statistics. Hence, to prevent future researchers from being led astray, the online supplement (http://asr.sagepub.com/supplemental) details these further issues in DNW’s groundbreaking study and presents corrected country data and sample statistics.
Quantifying Trust Radius
The “trust in most people” survey item has become the preferred measure in empirical trust research. DNW (p. 787), as well as many others before them (e.g., Hardin 2002), note that this classic item leaves the term “most people” unspecified. DNW’s approach to quantifying trust radius is theoretically grounded in the idea that interpretation of the term “most people” varies across countries and has different degrees of in-group versus out-group connotation. When thinking about “most people,” do respondents mostly consider individuals from their in-group or do they also consider individuals outside their in-group? DNW make the intuitively plausible assumption that the degree of in-group/out-group connotation of the trust in most people item relates to trust radius in such a way that one can use it to get to a measure of trust radius.
Practically, DNW employ two item batteries included in the fifth wave of the World Values Survey (World Values Survey Association [WVS] 2009): one that measures level of in-group trust, and one that measures level of out-group trust, which they combine with the trust in most people item. In-group trust is measured by averaging scores on three items that ask respondents how much they trust (1) their neighborhood, (2) people they know personally, and (3) their family. Similarly, out-group trust is measured by averaging scores on three items asking respondents how much they trust (1) people they meet for the first time, (2) people from another religion, and (3) people from another nationality.
The first step in DNW’s measurement method is to assess the degree of in-group/out-group connotation of the trust in most people item. DNW do so by empirically modeling trust in most people as a function of individuals’ level of in-group trust and level of out-group trust. Models are estimated by country, rendering a coefficient for in-group trust and for out-group trust for each country in the sample.
In the second step, DNW subtract the coefficient for in-group trust from the coefficient for out-group trust to obtain the measure of a country’s trust radius (rescaling the resulting number to fit a 0–1 scale). Subtracting the two coefficients captures the intuition that, if out-group trust is the stronger predictor of trust in most people, the radius of trust is wide, meaning that the term “most people” does not merely include the in-group but also the out-group to a large extent; moreover, if in-group trust is the stronger predictor of trust in most people, the radius of trust is narrow and the term “most people” includes the out-group only to a limited degree.
The empirical analysis employs the same procedure and data as DNW. Data as well as additional information on the WVS and the included questionnaire items are available from the WVS’s website (http://www.worldvaluessurvey.org).
Trust Radius Vis-À-Vis Trust Level
A key question DNW address is “whether one of the components of the radius measure—out-group trust—could be used alone as a surrogate for the radius” (p. 798). They report a strong correlation between out-group trust level and trust radius, “an important and genuinely new finding” that would lead to the conclusion “that the level of out-group trust is a valid proxy for the trust radius” (p. 800). Considering that the actual correlation (r = .381) is much lower than DNW reported (r = .94), I re-assess the relations between trust radius and out-group trust, adding two other trust level measures (level of trust in most people and level of in-group trust). As they are all concerned with the same level component of trust (see above), I expect strong intercorrelations between the three trust level measures. Trust radius should correlate comparatively weakly with the trust level measures, although most strongly with out-group trust level.
Results confirm my expectations (see Table 1). While trust level measures are strongly interrelated, trust radius, as measured using DNW’s method, is highly distinct from these established trust level measures, just as standard trust theory would have it. Hence, contrary to DNW’s reported results, my results show that trust radius and trust level are indeed separate trust constructs.
Correlations between Measured Trust Radius and Measures of Trust Level
Note: Standard deviations are in parentheses. Correlations are robust correlations (bootstrapped with 10,000 repetitions). Following Cohen’s (1988) classification of effect sizes, weak correlations (|r| > .1), moderate correlations (|r| > .3), and strong correlations (|r| > .5) are in italics,
p < .05; **p < .01 (because I have clear hypotheses concerning the sign of the correlations, I use one-tailed significance tests).
Conclusions
The importance of trust—as a factor fostering beneficial societal outcomes and as a subject in the social sciences—is difficult to overestimate. DNW fill an essential gap in the trust literature, extending its usual focus to consider trust radius as the second quintessential feature of trust, after its level. This comment corrects the key error in DNW: that level of out-group trust is highly correlated with and, therefore, a valid proxy for trust radius. My analysis and empirical evidence show that measured trust radius is, in fact, highly distinct from trust level measures, just as standard trust theory would have it. DNW’s measurement method itself may also be criticized, but such a critique is beyond the scope of the present comment. I look forward, however, to future research exploring alternative approaches, for instance, using other measures of in-group trust and out-group trust.
Footnotes
Acknowledgements
The author thanks Christian Welzel and Jan Delhey for their helpful discussion. Thoughtful comments by Robbert Maseland and Kees van Veen are gratefully acknowledged.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
