How impressions of public employees’ warmth and competence influence trust in government

Abstract

A central argument in the performance-trust literature is that the performance of public service delivery shapes citizen trust in government. However, studies have failed to address to what extent single interactions with public employees affect citizen trust in government. Building on insights from social psychology, this article argues that citizens spontaneously make impressions of the warmth (e.g., friendliness) and competence (e.g., effectiveness) of the public employee with whom they interact. Utilizing two large-scale randomized survey experiments, conducted in Denmark and the United States, this study demonstrates that impressions of warmth and competence in a concrete interaction with a public employee have important effects on general trust in civil servants. The findings further provide evidence that a single case experience can have implications for trust in broader administrative and political institutions of government. Thus, this article shows that even a single bureaucratic encounter can potentially have wide-ranging implications.

Acknowledgments

I would like to thank Martin Baekgaard, Lene Aarøe, Lasse Laustsen, and the anonymous reviewers for their insightful feedback on previous drafts of this article. I am also thankful for the comments and suggestions provided by participants at APSA 2019 and EGPA 2019.

Disclosure statement and data availability

The author has no conflicts of interest to declare. Study 2 was pre-registered at OSF: https://osf.io/mbz65. Data and code supporting the results presented in the paper can be found in Harvard Dataverse: https://doi.org/10.7910/DVN/8HB1OZ

Notes

1 This approach to trust in government sometimes also includes incumbent office holders. However, this article focuses only on governmental institutions as such and not specific government actors.

2 The warmth dimension is sometimes also labeled with other names as, for instance, “morality” (Wojciszke 2005) or “integrity” (McCurley and Mondak 1995) in the literature. However, the central point is that in all studies the dimension refers to intentions.

3 The hypotheses in the pre-registration mentioned trust in governmental institutions in general and did not include the sub dimensions of administrative and political institutions. To ensure clarity in the analyses, it is specified in the article’s hypotheses that these sub dimensions of governmental institutions are studied. Importantly, I follow the general guidelines for open science, as the proposed difference between political and administrative institutions is exploratory since it was not directly stated in the pre-registration.

4 Selection bias might also be at play in observational designs if characteristics (e.g., general optimism or ideological orientation) explain how citizens evaluate both government performance and trust in government.

5 Specifically, Turkprime’s platform was used to collect responses and only allowed workers with 100+ HITs approved and an approval rate > 95.

6 Importantly, regression analyses (reported in Table B2 and B3 in the Online Appendix) suggest that the experimental groups are balanced on background characteristics (on the 5% level).

7 In the Danish version of the survey, respondents were asked about the national government in Copenhagen.

8 For instance, ESS does not ask about trust in the public administration, but this item is included in Eurobarometer surveys. Additionally, the use of standard survey items made it possible to use the translated versions of these standard items in the Danish version of the survey.

9 Trust in civil servants was (as the only item) asked using different wording. Trust in civil servants is generally not measured in the literature by using a scale but rather by asking about agreement to a statement. Specifically, the question from the International Social Survey Programme (ISSP) asking to what extent respondents agree with the following statement (on a five-point scale) is employed in the article: “Most civil servants can be trusted to do what is best for the country.”

10 The principal component analysis for the Danish respondents extracts two factors with Eigenvalue > .88, and a scree plot also supports two factors (not shown). Furthermore, the rotated factor loadings (oblique oblimin) are consistent with the expected pattern of administrative and political trust. However, the second factor for the US respondents has Eigenvalue = .56.

11 An instructional attention check (Oppenheimer, Meyvis, and Davidenko 2009; Ejelöv and Luke 2020) was also included in the survey to measure more general attention. The substantial results from the analyses are not altered by excluding respondents that did not pass this general attention check (results not shown).

12 Supplementary analyses show that treatment is correlated with specific attention and that there are significant differences (on the 5% level) in gender and left-right self-placement between attentive and non-attentive subjects (results not shown). While this indicates that attentive respondents are not a random subset of the full sample and, thus, that some selection process is at play, it is not evidence of a problem.

13 Given the strong correlation between outcome measures (see Online Appendix B), I also tested whether results are substantially different if they are analyzed used MANOVA instead. The overall results are not altered (see Online Appendix C).

14 As in study 1, Turkprime’s platform was used to collect responses and only workers with 100+ HITs approved and an approval rate > 95 were allowed. Furthermore, respondents from study 1 were excluded to take part in the survey.

15 Technically, the protocol from Burleigh et al. (2018) was used to screen out non-US workers in Qualtrics. To further ensure data quality in the sense of only studying attentive subjects, an instructional attention check was used (Oppenheimer, Meyvis, and Davidenko 2009; Ejelöv and Luke 2020). The check is different from the one used in study 1 but modeled on the check used in Hauser and Schwarz (2016). (See Online Appendix for the full wording.)

16 A randomization check revealed that the experimental groups are generally balanced on covariates. Only in a few cases, there are significant differences at the 5% level, which is not surprising given the large number of tests (see Table G2 in the Online Appendix).

17 As in study 1, the manipulations of warmth and competence are correlated, but, importantly, to a much lesser extent in study 2. Difference in perceived competence (between warmth conditions) = .05, p < 0.05; difference in perceived warmth (between competence conditions) = .08, p < 0.001. These differences are substantially small and practically impossible to avoid.

18 To ensure comparability across studies, the statement “Most civil servants can be trusted” is used in the main analyses. In addition, this item is used in both studies as it is the validated standard measure used in the ISSP. Using an alternative question asking “How much do you personally trust civil servants,” which is in line with the other trust items, generally does not alter the results. Only the effect of competence impressions becomes marginally insignificant (p = .11), while the effect of warmth stays the same both in terms of substantial and statistical significance (p < .05).

19 Several additional exploratory analyses are performed (cf. the pre-registration) (see Section I in the online appendix for results).

Additional information

Notes on contributors

Frederik Godt Hansen

Frederik Godt Hansen ([email protected]) is a PhD student at the Department of Political Science, Aarhus University. His current research examines the effects of citizen–state interactions on citizen trust in government. His research interests also include experimental methods, causal inference, and behavioral public administration.