Abstract
A growing body of empirical work suggests that identifying the actors formally tasked with implementing policy can focus attention away from incumbent politicians. We examine the effects on blame attribution and voting intention of (a) the identifiability of a responsible policy worker (administrator), and (b) the evaluability of the policy work or outcome (policy failure), in the context of programs at two federal agencies (loans by the Small Business Administration and inspections by the U.S. Department of Agriculture). Using a set of online survey experiments with 1105 US adults, we find that the evaluability of a (negative) outcome generally reduces voting intention, but that the identifiability of a policy worker (administrator) tends to shift blame away from the incumbent politician and thus to increase voting intention. These experimental findings provide at least partial support for our theoretical expectations.
Introduction
In representative government, the vote is a blunt instrument, but the dominant view in empirical political science is that its potency lies in sanctioning incumbents (e.g., Kam et al., 2020). In political behavior generally, Achen and Bartels (2017: 118) contend that “when they are in pain [voters] are likely to kick the government” unless no “cultural construction” allows them to “take out their frustrations on other scapegoats.” However, a growing body of empirical work suggests a different mechanism when bureaucracy is involved: changing the actors formally tasked with implementing policy can focus attention away from incumbent politicians. Hood (2011: 19) calls this an agency strategy of blame avoidance and it motivates a growing empirical literature. Using a survey experiment in England, for example, James et al. (2016) find that an agency strategy does shift blame away from local politicians for poor street quality. By contrast, Mortensen (2013) shows evidence that the Norwegian program that centralized healthcare was associated with less blame for central government in media accounts. We advance this literature by empirically examining the effect of shared responsibility for policy implementation on political accountability.
Theory
Drawing on the theory of representative government (e.g., Manin, 1997), Bertelli (2016) argues that two critical factors sharpen or weaken voters’ retrospective evaluations of politicians: (a) whether a voter can identify a “policy worker,” i.e., the bureaucrat, contractor, or other actor responsible for implementing policy; and (b) whether a voter can evaluate their “policy work,” i.e., the results or outcomes of the policy. Both identification and evaluation prove difficult to assess in much policy work, rendering accountability an important, but elusive, democratic value. Bertelli (2016) hypothesizes that the first step of the voter’s accountability heuristic relates to the question of identifiability and is similar to the recognition principle in the “fast and frugal” literature from social psychology (Gigerenzer and Goldstein, 1996). If the policy worker is identifiable, a voter can assign accountability to the worker or the worker’s organization. However, if the policy worker is not identifiable, the voter’s locus of blame or credit for policy work defaults to the politician. In the second step, the voter must consider the issue of evaluability, which corresponds to the discrimination rule (Gigerenzer and Goldstein, 1996). That is, given information cues from the media or elsewhere about the policy results or outcome, does the policy work seem acceptable or unacceptable? If the cues cannot discriminate between an acceptable and an unacceptable outcome, responsibility again defaults to the politician.
This theory is consistent with a view of representative government in which votes are cast retrospectively and are influenced by past policy choices and their connections with incumbent politicians. Two things are important to emphasize. First, the default option is to hold the politician accountable, and information cues can shift blame away from politicians and onto policy workers. Second, it treats the perception of accountability (blame) and voting intention (sanction) as separate concepts. This stands in contrast with empirical public administration studies that focus solely on blame. For instance, James et al. (2016: 93) ask whether politicians are “deserving of blame” and Mortensen (2013) codes newspaper articles for their blame (credit) of national or subnational institutions in negative (positive) articles.
Hypotheses
We operationalize this claim through two sets of hypotheses. To begin with, the influences of identifiability (factor 1) and evaluability (factor 2) on perceived accountability, i.e., whether to blame the policy worker or the politician, are as follows:
In addition, with respect to retrospective accountability, i.e., incumbent voting intention, we test the following claims:
These hypotheses were preregistered in the Open Science Framework on July 10, 2019, prior to data collection (DOI DOI:10.17605/OSF.IO/MYRGQ). We should point out, however, that the wording of H4 was mistakenly reversed in the preregistered version of the hypothesis and has been corrected above.
Experimental design and method
The design is a 2 × 2 factorial experiment, with Factor 1 being the identifiability of the policy worker (administrator) and Factor 2 capturing the evaluability of the policy outcome. Thus, there are four conditions:
Low identifiability, Low evaluability
Low identifiability, High evaluability
High identifiability, Low evaluability
High identifiability, High evaluability
Each respondent received two vignettes, with both vignettes constructed using the same 2 × 2 design. One vignette described a Small Business Administration (SBA) program that provides low-interest loans to businesses, homeowners, and renters that can be used to repair or replace real estate, personal property, equipment, and other assets lost or damaged by a natural disaster. The other vignette described a US Department of Agriculture (USDA) program to modernize government inspection and regulation of meat, poultry, and processed egg products prepared for distribution in stores and restaurants. The order of the two vignettes was randomized for each respondent.
In both vignettes, factor 1 (identifiability) is operationalized by naming (versus not naming) the administrator in charge of the program. Factor 2 (evaluability) is operationalized by specifying (versus not specifying) a concrete outcome of the program. The outcome for the SBA program is a backlog of loans (25%) waiting to be processed. The outcome for the USDA program is news reports of an outbreak of E. coli in the supply of ground beef. Thus, both outcomes are negative events or policy failures, on which we focus because both theory and evidence suggest the widespread importance of negativity bias in political behavior (Soroka, 2014) and blame avoidance (Hood, 2011).
We assess two outcomes (dependent) variables: a measure of perceived accountability for the policy event, and the intention to vote (or not) to re-elect the politician who voted for the policy. Both questions asked for responses on a 0–100 scale and were worded as follows:
[Perceived accountability] Where would you place responsibility for the performance of this program? From 0=The representative in Congress who voted for the program, to 100=The administrators who implemented the program.
[Vote intention] How likely would you be to vote for this representative of your district in the next congressional election? From 0=Very unlikely, to 100=Very likely.
These questions were asked in the order above and immediately after presentation of each of the experimentally varied vignettes.
Data were obtained from an online sample of n=1105 US adults, with responses gathered through invitations sent to the Qualtrics research panel during July 11–16, 2019. Representative sampling quotas were established for region, sex, age, and race, based on national estimates from the American Community Survey (see Tables B1–B2 in the supplementary material for the descriptive statistics). Our set of experiments was the first of three experimental modules embedded in the same online questionnaire. Data were analyzed (unweighted) with Stata 16.
Results
Figures 1 and 2 present the results for the SBA vignette experiment graphically, with the corresponding OLS regressions provided in Table 1. The regressions employ −.5, .5 effect coding of the factors in place of traditional dummy coding so that the regression coefficients and related tests capture main effects. In addition to the usual unstandardized regression coefficients, we also present y-standardized coefficients as a measure of effect size.
Blame of policy worker, SBA vignette.
Vote for representative, SBA vignette.
SBA.
With respect to blame attribution, the results suggest that the identifiability of the policy worker, the Director of the Office of Disaster Assistance, had only a slight positive effect that was not significant statistically. Similarly, the evaluability of the policy outcome, the 25% of loan applications waiting to be processed, did not have an effect on blame attribution. But with respect to voting intention, in contrast, both identifiability and especially evaluability influenced responses. Specifically, when the Director of the Office of Disaster Assistance was identified, respondents were more likely (by 3 scale points, or .13 SD) to vote for the re-election of the representative. And when the negative outcome was evaluable (25% of loan applications waiting to be processed), respondents were less likely (by more than 5 scale points, or .22 SD) to vote to re-elect the representative. There was, however, no interaction effect.
Figures 3 and 4 present the results for the USDA vignette experiment, with the corresponding regressions provided in Table 2. With respect to blame attribution, identifiability of the policy worker, the Undersecretary for Food Safety, leads to a slight increase in blame attribution, although the effect is not significant statistically. The evaluability of the policy event, a news report of an outbreak of E. coli in the supply of ground beef, results in a decline (of nearly 4 scale points, or .14 SD) in blame of the administrator, which is significant at the p<.05 level. There is no evidence of an interaction effect. With respect to voting intentions, however, both the identifiability effect, the evaluability effect, and the interaction effect are significant statistically. Identifiability of the policy worker, the Undersecretary for Food Safety, results in a greater intention to vote to re-elect the representative (by over 6 scale points, or .24 SD). The evaluability of the policy event, the news report of an E. coli outbreak, leads respondents to be less likely to vote to re-elect the representative (by nearly 9 scale points, or .33 SD). Interestingly, however, the positive and statistically significant interaction term suggests that the identifiability of the policy worker reduces the negative effect of the policy failure on voting intentions (by 10 points, or .39 SD).
Blame of policy worker, USDA vignette.
Vote for representative, USDA vignette.
USDA.
Although we did not preregister hypotheses related to subgroup differences, we ran some exploratory analyses of experimental effects by political knowledge, ideology, and party identification (see Tables C1–C6 in the supplementary material). The clearest findings from these subgroup analyses related to political knowledge, measured using a series of five factual questions about the US government (e.g., Carpini and Keeter, 1993). Specifically, we found consistently larger experimental effects of identifiability, evaluability, and their interaction for the more politically knowledgeable participants in our study.
Discussion
Our study has some noteworthy limitations. To begin, we used hypothetical vignettes in an online survey that people may respond to differently than they would to real-life politicians and policy failures. Thus, our study suffers from a lack of mundane realism. We also examined our hypotheses in the context of just two federal agencies, SBA and USDA, and two specific policy failures. How voters may respond to other kinds of policy failures, involving other government agencies or even other levels of government, may be different. Finally, although we used sampling quotas for age, gender, race, and region based on American Community Survey statistics, participants in our study still came from a voluntary, online sample of US adults and thus may not be representative of the US adult population. Despite these limitations, our findings do still shed some initial empirical light on the hypotheses in Bertelli (2016) and invite additional investigations of this theory.
Results from both the SBA vignette and USDA vignette do not support H1, which predicted that identifiability (of the policy worker) will be positively associated with assigning responsibility to the policy workers who implemented the program. The coefficients were positive, as expected, but small in magnitude and insignificant statistically. Our results also do not support H2, which predicted that evaluability (of the outcome) is positively associated with assigning responsibility to the policy workers. Indeed, in the USDA vignette, evaluability of a negative outcome (an E. coli outbreak) was negatively associated with blame of the policy worker—and thus positively associated with blame of the politician. And with respect to H3, the interaction hypothesis, we found no evidence of an interaction effect on the assignment of responsibility or blame.
Regarding retrospective accountability, as measured by voting intentions, our results fall more in line with the theoretical expectations. As predicted by H4, the results of both the SBA and USDA vignette experiments show that identifiability (of the policy worker) is positively associated with the likelihood of voting to re-elect the policymaker (the politician) who established the program in law. Consistent with H5, findings from both the SBA and USDA vignette experiments show that evaluability (of a policy failure) is negatively associated with the likelihood of voting to re-elect the politician. With respect to the interaction hypothesis H6, the findings from the SBA vignette were null. However, the USDA vignette results are consistent with the expectation that the interaction of identifiability and evaluability has a positive effect on the likelihood of voting to re-elect the policymaker. We speculate that the stronger effects in the USDA vignette could be attributable to the fact the negative outcome, an E. coli outbreak in the supply of ground beef, was likely more salient to respondents. In any event, this finding suggests that, given an evaluable policy failure, the clear identifiability of a policy worker shifts electoral accountability away from the policymaker, enhancing the likelihood of voting to re-elect the incumbent.
Supplemental Material
Heuristics-Accountability-Supplement-2020-07-06 – Supplemental material for Heuristics and political accountability in complex governance: An experimental test
Supplemental material, Heuristics-Accountability-Supplement-2020-07-06 for Heuristics and political accountability in complex governance: An experimental test by Anthony M. Bertelli and Gregg G. Van Ryzin in Research & Politics
Footnotes
Authors’ note
The study was approved by Rutgers University, Arts and Sciences IRB #2019000960 in May 2019.
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Supplemental materials
Carnegie Corporation of New York Grant
This publication was made possible (in part) by a grant from the Carnegie Corporation of New York. The statements made and views expressed are solely the responsibility of the author.
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.
