2.1 The economy, the bureaucrat, and incentives to do a CBA

The model in this section is an application of the model of delegation developed by Bendor and Meirowitz (Reference Bendor and Meirowitz2004). Consider an economy consisting of $ H $ individuals denoted by superscript $ h=\left\{1,2,\dots, H\right\} $ . Also assume a finite set, $ \left\{1,\dots, J\right\},1\le J\ll H $ of bureaucrats. Besides utility from private goods, individuals derive utility from a non-rival and non-excludable public policy outcome, $ g\in G $ . Utility derived from the public policy outcome is given by a function $ {q}^h(g) $ , $ {q}^h $ being increasing and concave in $ g $ , that is, $ {q}_g^h>0 $ , $ {q}_{gg}^h\le 0 $ , the subscripts denoting differentials. The preferences for the public policy outcome are continuously distributed with mean $ {\mu}_g $ and variance $ {\sigma}_g^2 $ . We denote the socially optimal level of the policy outcome by $ {g}^{\ast } $ .

An individual bureaucrat $ j\in \left\{1,\dots, J\right\} $ has a unique preference point for the public policy outcome, denoted by $ {g}^j\in G $ . We assume that the bureaucrat always prefers a higher level of the public policy outcome than the socially optimal one, that is, $ {g}^j>{g}^{\ast } $ .Footnote ⁶

A particular policy is given by $ p\in P $ , the policy being for example a carbon tax, an emissions trading system, a low carbon fuel standard, or a bonus-malus system for low-emission vehicles. The policy is perturbed by a random shock ( $ \epsilon $ ). The shock reflects society’s response to a given policy, which is not known in advance. For example, the price elasticity of fuel demand could be much lower (or higher) than believed. We assume that the shock is symmetrically distributed with a mean of zero, $ \overline{\epsilon}=0 $ , and variance of $ {\sigma}_{\epsilon}^2 $ , and that the influence of the shock is linear so that $ g=p+\epsilon $ . The outcome $ g $ is then a random variable that depends on policy so that the cumulative distribution function of $ g $ given policy $ p $ is $ F\left(g\left|p\right.\right) $ . $ g $ reflects the attainment of policy goals, for example, the Paris Agreement target of no more than 2 °C average global warming or the Swedish goal of reducing GHG emissions from the transport sector by 70 % compared with 2010 values by 2030.

Given uncertainty about the mapping from policies to outcomes, each bureaucrat has an optimal policy. The bureaucrat can, if they so choose, inform themselves about the mapping from policy to outcome, that is, they can learn the value of the shock. This requires effort, however. The cost of effort is denoted by $ {c}^j $ , with the cost increasing in effort. The greater the effort, the more likely the bureaucrat is to find the value of the shock, $ \epsilon $ . In practice, the cost of effort can be seen as the cost of performing a CBA or taking in and using information from one. Moreover, knowing the perturbation ( $ \epsilon $ ) permits complete “shock absorption”; if a decision-maker chooses a policy ( $ p $ ) after they observe the disturbance ( $ \epsilon $ ), they can always find a policy that, when combined with the shock, yields any desired (feasible) outcome ( $ g $ ) (Bendor & Meirowitz Reference Bendor and Meirowitz2004, p. 298). We can define the shock as a function of the cost of effort, $ \epsilon \left({c}^j\right) $ , with $ {\epsilon}_c\left({c}^j\right)<0 $ if $ \epsilon >0 $ ( $ {\epsilon}_c\left({c}^j\right)>0 $ if $ \epsilon <0 $ ), that is, the impact of the shock on the outcome approaches zero as the cost of effort increases. Furthermore, there exists some maximum level of effort, $ {c}^{\mathrm{max}} $ , such that $ \epsilon \left({c}^{\mathrm{max}}\right)=0 $ , that is, that given enough effort, the true value of the shock can be found.

The extensive form of the game is then as follows: A bureaucrat has the authority to propose policy $ p $ .Footnote ⁷ Before making their proposal, they decide whether to do/use information from a CBA or not. The bureaucrat consequently faces the following problem: They prefer an outcome $ {g}^j>{g}^{\ast } $ but cannot reach that outcome with certainty due to the shock. In the absence of a CBA, they choose policy to maximize their expected utility, given their expectations of the cumulative distribution of outcomes, $ F\left(g\left|p\right.\right) $ . Because the bureaucrat’s preferences are strictly single-peaked and because $ F\left(g\left|p\right.\right) $ is non-degenerate for all $ p $ , the expected utility of the bureaucrat is strictly less than the utility obtained by getting the ideal outcome with certainty.

Because utility varies continuously in outcomes, the bureaucrat has certainty equivalents (CE) in outcome space. The CE form a set of outcomes such that the bureaucrat would be indifferent between getting any such outcome for sure and taking a policy that maximized their expected utility.Footnote ⁸ This set of certainty equivalents defines the bureaucrat’s CBA set (the CBA set corresponds to the delegation set in Bendor & Meirowitz, Reference Bendor and Meirowitz2004, p. 299). It defines the bureaucrat’s propensity to do/use information from a CBA; the larger the CBA set, the more likely a bureaucrat is to carry out a CBA. We examine the determinants of the CE in the next proposition:

The result in Proposition 1 is ambiguous, however. What is the bureaucrat risk averse toward – the CBA outcome or the environmental outcome? This cannot be determined on theoretical grounds. Therefore, we formulate two alternative hypotheses that we can test empirically:

Turning to the question of the impact of attitudes on the choice of policy, consider a bureaucrat who must make a proposition for public policy, $ p $ , which together with the shock, $ \epsilon \left({c}^j\right) $ , determines the outcome of the public policy, $ g $ . Considering a discrete problem, the probability of $ \epsilon $ taking any given value is $ {\pi}_i,i\in \left\{1,\dots, n\right\} $ . The bureaucrat’s problem is to choose the effort cost to maximize the expected utility from $ g $ :

(1) $$ \underset{{\mathsf{g}}^{\mathsf{j}}}{\max}\sum_{\mathsf{i}=\mathsf{1}}^{\mathsf{n}}{\pi}_{\mathsf{i}}{\mathsf{q}}^{\mathsf{j}}\left(\mathsf{p}+\mathsf{\epsilon}\left({\mathsf{c}}^{\mathsf{j}}\right)\right),\mathrm{s}.\mathrm{t}.0\le {\mathsf{c}}^{\mathsf{j}}\le \mathrm{C}, $$

where $ C $ is some upper limit of the possible effort cost.

What happens to the effort put into finding the value of the shock as the bureaucrat’s policy preference, $ p $ , increases? We formulate the following proposition:

Proposition 2. If bureaucrat i has a higher preference for the policy outcome than bureaucrat j, that is, $ {g}^i>{g}^j $ , bureaucrat i will be less likely to do a CBA than bureaucrat j.

Proof: We prove Proposition 2 by totally differentiating (1) with respect to (w.r.t.) $ p $ and $ {c}^i $ and rearranging. We find

$$ \frac{d{c}^i}{dp}=-\frac{\sum_{k=1}^n{\pi}_k{q}_g^i\left(p+\epsilon \left({c}^i\right)\right)}{\sum_{k=1}^n{\pi}_k{q}_g^i\left(p+{\epsilon}_c\left({c}^i\right)\right)}<0.\qed $$

Thus, as $ p $ increases, the cost of effort, and consequently the effort put into finding $ \epsilon $ , falls. Therefore, a bureaucrat with higher preferences regarding the public policy outcome is less likely to put more effort into finding out the true value of $ \epsilon $ , that is, they are less likely to do a CBA than bureaucrats with less extreme preferences.

However, how the situation is perceived is also related to education and experience. With little experience with CBAs, a bureaucrat may perceive that the CBA does not reveal all necessary information about the shock. This case is analogous to that of informationally imperfect agents in the model of Bendor and Meirowitz (Reference Bendor and Meirowitz2004, p. 300). Thus, even if the bureaucrat does a CBA, it is possible that they will have to make an uninformed decision. This puts them in the worst of all possible worlds: There is policy bias but no compensating adaptation of the policy to the state of the world.

Thus, assume that with probability $ {\rho}^j $ the CBA reveals the true state of nature, the probability of the state of nature remaining unknown being $ \left(1-{\rho}^j\right) $ . Hence, the expected value of doing a CBA increases in the bureaucrat’s own competence, measured by $ {\rho}^j $ . As $ {\rho}^j $ falls, the CBA set contracts. We formulate the following proposition:

To illustrate Proposition 3, assume that with full certainty ( $ {\rho}^j=0\Big) $ , the expected CBA result equals the bureaucrat’s certainty equivalent, that is, the bureaucrat is indifferent between doing a CBA or not. Now introduce uncertainty about whether the CBA will reveal the true value of the shock or not. The value of the CBA is reduced by uncertainty. This means that the expected value of the CBA falls outside the bureaucrat’s CBA set and the bureaucrat under uncertainty is not willing to do a CBA.

We summarize the findings from Hypotheses 1 and 2, and Propositions 2 and 3 regarding the size of the CBA set in Table 1. A minus sign in Table 1 indicates that a bureaucrat with that particular property has a smaller CBA set and a plus sign that the CBA set is larger, the first sign referring to the individual’s risk aversion, the second to their policy preferences, and the third to the level of uncertainty about CBA results.

Table 1

The impact of risk preferences (Hypotheses 1 and 2, first sign), policy preferences (Proposition 2, second sign) and the uncertainty about CBA results (Proposition 3, third sign) on the CBA set.

A minus sign indicates a smaller CBA set; a plus sign a larger one.

Abbreviation: CBA, cost–benefit analysis.

Two unambiguous cases arise in Table 1 for Hypothesis 1 and Hypothesis 2, respectively. The bureaucrat who is risk averse with respect to the CBA outcome (Hypothesis 1), who has low policy preferences (Proposition 2), and low uncertainty about the CBA results (low cost of carrying out a CBA, Proposition 3) is the most likely to use information from a CBA, that is, they have the largest CBA set. The bureaucrat who is risk-neutral/loving with regard to the CBA outcome, with high policy preferences, and high uncertainty about the CBA results (high cost of carrying out a CBA) has the smallest CBA set.

On the other hand, a bureaucrat who is risk averse with respect to the environmental outcome (Hypothesis 2), who has high policy preferences (Proposition 2), and who experiences high uncertainty about the CBA results (Proposition 3) has the smallest CBA set while the corresponding risk-neutral/loving bureaucrat with low policy preferences and with low uncertainty about the CBA results has the largest set. All the other cases are of indeterminate size.

2.2 The influence of institutional norms on the use of CBA

In addition to the influence of individual characteristics, institutional norms may influence bureaucrats’ behavior and their search for information. An important norm of this type is the requirement that government agencies have to consider the implications of a regulation on the government’s budget. Hammes (Reference Hammes2017) argues that one difference between government agencies that regularly conduct CBAs and those that do not is that the former seems to be acting under a binding governmental budget constraint while the latter does not. A governmental budget constraint can, when binding, impose an upper limit on the supply of public goods. It may then be impossible for a bureaucrat to reach their ideal point. If, however, the bureaucrat, by doing a CBA, can get the government to relax the budget constraint, they may have an incentive to do so.

Consequently, if the governmental budget constraint limits the supply of the public good to a level below the bureaucrat’s optimal point, including the CBA set, the incentives to do a CBA depend on the bureaucrat’s expectations about the social optimum. An expected social optimum that corresponds to the budget constraint will not induce them to do a CBA since the CBA results would not make the government to relax the budget constraint. If, however, the CBA results likely lie closer to the bureaucrat’s ideal point, they will do a CBA since by doing one they can demonstrate the benefits of relaxing the budget constraint. The bureaucrat’s CBA set may consequently increase in this case. This is because the social optimum, though non-desirable for the bureaucrat, lies closer to their ideal point than the budget-constrained level of provision.

Proposition 4 is illustrated in Figure 1. The null of Proposition 4 is that there are no systematic differences between bureaucrats working for government agencies acting under a binding budget constraint and those not so constrained.

Figure 1

The impact of a binding governmental budget constraint ( $ {g}^B,\mathrm{the}\ \mathrm{thick}\ \mathrm{line} $ ) on the bureaucrat’s incentive to do a CBA. If a CBA demonstrating that $ {g}^{\ast } $ is preferable to $ {g}^B $ can move the budget constraint from $ {g}^B $ toward $ {g}^{\ast } $ , the bureaucrat’s CBA set expands from the original to the “expanded CBA set.”

To test the four propositions, we designed a choice experiment that was sent to civil servants at five government agencies. In the next section, we describe the questionnaire and data collection.

3.1 Description of the questionnaire and the data

The choice experiment tested the influence of CBA information on the participating civil servants’ response behavior. The questionnaire dealt with government policy to encourage the production of biofuels from forestry. The questions were based on a number of recent studies (Bosted et al., Reference Bosted, Mustonen and Gong2015; Henning, Reference Henning2015; Swedish Energy Agency, 2015, 2016; Apaydin, Reference Apaydin2016; Konjunkturinstitutet, 2016; Kungliga ingenjörsveternskapsakademien (IVA), 2016; Ministry of Enterprise and Innovation, 2016; Nationellt skogsprogram, 2016; Swedish Transport Administration, 2016). A pre-study was conducted on a reference group consisting of representatives from the Swedish Energy Agency (SEA), the Swedish Environmental Protection Agency (SEPA), the Swedish Transport Administration (STAdmin), the Swedish Transport Agency (STAgency), and Transport Analysis (Trafa), that is, the same agencies that answered the questionnaire.

As an introduction, the respondents were given information about the current policy landscape in Sweden with a political ambition of creating a fossil-free vehicle fleet by 2030. They were also informed that biofuels from forests could be part of the solution and about current energy use and CO₂ emissions from the car fleet in Sweden. Finally, they were told that the government in different ways encourages the transformation of the car fleet, for example, through government grants (see online Appendix).

The questions were divided into three sections. The first concerned the respondents’ education and previous work experience. The second part consisted of five choice experiment questions, four of a standard type evaluating the trade-offs between different attributes, and a fifth, which is the one analyzed in this study. Hence, when answering the fifth question, the respondents had already been forced to consider the possible trade-offs involved. The questionnaire ended with questions about individual characteristics (age, sex, income, etc.) and attitudes both to risk and more generally.

The aim of the fifth choice question was to create a scenario that resembles a real-life choice context. The assumption is that the choice situations that the bureaucrats encounter in their everyday work include a description of both the benefits and the costs of undertaking a certain action. To mimic this, the questionnaire contained information about the estimated societal benefit of reducing CO₂ emissions (0.8–1.2 SEK per kg CO₂ reduced) and the cost of four different action programs to reduce these emissions from the transport sector over the 10-year period 2020–2030; see Table 2.Footnote ⁹ The respondents also had the option to reject all alternatives, hence choosing the status quo. This we assumed to imply a zero willingness to pay.

Table 2

Action programs were presented to respondents in the questionnaire.

For Programs 1–3 (P1–P3), total benefits exceed the total costs. P1 is inefficient compared with P3, which results in approximately the same climate impact at a much lower cost. Which of the two alternatives, P2 or P3, is to be preferred depends on the choice of benefit estimate and decision rule (net present value (NPV) or benefit–cost ratio).Footnote ¹⁰ The upper benefit estimate, 1.2 SEK per kg CO₂, yields a higher NPV for P2 while the lower estimate of 0.8 SEK per kg CO₂, yields a higher estimate for P3. The average benefit estimate, 1 SEK per kg, yields the same NPV for P2 and P3. P2 makes a larger contribution to the reduction of CO₂ emissions, however. Using the benefit–cost ratio as the decision criterion, P3 is superior to P2. P4 is the only alternative that reaches the goal of reducing GHG emissions from the transport sector by 70 % by 2030 but at a very high cost.

In the third part of the questionnaire, we started by gauging the respondents’ risk preferences. The respondents were asked to choose between a series of small sums of money with certainty (starting from 10 SEK; about 1 EUR or 1.2 USD and running up to 200 SEK) or a lottery with a 50 % chance of winning 300 SEK (31 EUR or 35 USD), the game stopping when the respondent chose the sum of money with certainty over the lottery. The answers were used to construct the variable Risk neutral/loving (RN/RL), which takes the value of one if the respondent is risk neutral or loving and zero if risk averse.Footnote ¹¹

The third part of the questionnaire ended with a large number of questions examining the respondents’ attitudes. We used this information to construct two indexes of attitudes and aptitudes. Each index was constructed using the first components of principal component analysis. The first index (economic preferences) is composed of responses to questions about the importance of CBA, rationality, and cost-effectiveness. The variable is used as a proxy for the cost of doing a CBA (Proposition 3). The second index (environmental preferences) is composed of responses to questions about the importance of the natural environment, ecological sustainability, and reduced oil dependency. This variable is used as a proxy to test the intensity of preferences for the environment (Proposition 2). The responses to both these questions range from 1 (not at all important) to 6 (great importance). The correlation between economic and environmental preferences is 0.1032 (at the 5 % level of statistical significance).

Altogether there were $ 4\times 2=8 $ different versions of the questionnaire; four versions of the basic choice experiment questions, and two versions where half of the respondents received additional information about the policy context. The extra information contained, first, an explicit reminder of the CO₂ reduction goal for the transport sector to be achieved by 2030, that is, a reduction by 70 % compared with 2010 levels, and second, a reminder of the alternative cost of public funds. This is similar to an ordinary willingness to pay study where respondents are reminded about their own income and budget constraint. The binary variable Infotext captures this aspect of the questionnaire; it takes the value of one for those who got the additional information and zero otherwise. The variables used are summarized in Table 3.

Table 3

Descriptive statistics of the variables included in the regressions.

Abbreviations: SEPA, Swedish Environmental Protection Agency; STAdmin, Swedish Transport Administration; STAgency, Swedish Transport Agency; RL, Risk loving; RN, Risk neutral.

3.2 Data collection

The questionnaire was sent to 1,488 civil servants at five Swedish government agencies: the SEA, the SEPA, the STAdmin, the STAgency, and Trafa. These are the main agencies working with the transition to a fossil-free vehicle fleet by 2030. The data were collected in May 2017.Footnote ¹² As noted above, eight versions of the questionnaire were used, and 185–188 copies of each version were sent out.

At SEA, SEPA, and STAdmin, we could send the questionnaire to all staff involved in policy-related work, including heads of office. Only office clerks were excluded from the sample at these agencies. At the STAgency, we received permission to send the questionnaire to a random sample of 67 out of about 243 relevant workers, and at Trafa to the 11 employees working with the economic analysis. The results of the data collection are summarized in Table 4. As can be seen, some respondents were on leave of absence or were not relevant for other reasons. Hence, the valid sample consists of 1,383 respondents.Footnote ¹³

Table 4

Summary statistics for respondents from government agencies.

Abbreviations: SEA, Swedish Energy Agency; STAdmin, Swedish Transport Administration; STAgency, Swedish Transport Agency; SEPA, Swedish Environmental Protection Agency; Trafa, Transport Analysis.

The response rate varied considerably between the agencies. SEPA had the lowest response rate at 23 %, and Trafa the highest at 91 %. From comments sent by personal e-mail to the lead author as well as comments in the questionnaire, we found that there are several reasons for the variation in the response rate. An important one appears to be that bureaucrats are hesitant to answer questions outside their area of expertise. This may explain why the response rates from STAgency and Trafa are comparatively high while not all staff at SEA and SEPA work with transport-related questions. Moreover, the time available to answer the questionnaire appears to have been a limiting factor, and it seems the questions were considered complicated. Other comments along this line were that the choice questions do not reflect current management practices and/or that the questions do not fully reflect a more complex reality. Some respondents therefore deliberately chose not to answer the questionnaire. One person claimed that the reason for not replying was that bureaucrats only provide information to the decision-making process while the decisions and trade-offs are made by politicians. It is possible that only persons with a more favorable attitude to questionnaires of this type responded, which may above all bias the answers obtained from SEPA.

3.3 Description of respondents

Descriptive statistics of the respondents are provided in Table 5. A zero after the name of an agency indicates that the questionnaire version they received did not contain information about the emissions reduction goal and budget constraint (variable Infotext), and a 1 that the information was included.

Table 5

Summary statistics for respondents per agency and questionnaire version (minima and maxima in parentheses).

Abbreviations: SEA, Swedish Energy Agency; STAdmin, Swedish Transport Administration; STAgency, Swedish Transport Agency; SEPA, Swedish Environmental Protection Agency; Trafa, Transport Analysis.

^a 99 is used for those with a recorded time of over 1 h. In these cases, it is likely that they started to fill in the questionnaire, took a break and then came back later to complete.

The bureaucrats at STAdmin have worked the longest at the agency and also have the longest work experience. SEPA and STAdmin have the highest average ages of workers. Respondents at SEA, STAgency, and Trafa have relatively few years of work experience at their respective agency but especially employees at the STAgency have a lot of work experience from other jobs. One explanation for this is that STAgency is the result of a merger between previous regulatory agencies; hence, some of the employees have been working with these issues for a long time. A similar explanation applies for Trafa, which was created at the same time as STAdmin and STAgency. STAdmin together with STAgency have the lowest share of employees with a degree in economics, while Trafa has the highest. Regarding gender, there is some variation across agencies, with Trafa and STAgency having the lowest shares of women.

Summary statistics for risk preferences are shown in Table 6. The majority of individuals are risk averse, which is in line with standard economic theory. Breaking the numbers down by agency, we see that while risk aversion remains the biggest category in each case, there are marked differences in the degree of risk aversion, risk neutrality, and risk loving across agencies. These differences, however, are not statistically significant.

Table 6

Results from analysis of risk aversion.

Abbreviations: SEA, Swedish Energy Agency; STAdmin, Swedish Transport Administration; STAgency, Swedish Transport Agency; SEPA, Swedish Environmental Protection Agency; Trafa, Transport Analysis.

The differences between individuals who have a degree in economics and those with other degrees are statistically significant (see bottom boxes of Table 6). Thus, workers with economics degrees are significantly less risk averse than individuals with other degrees, and significantly more risk neutral and risk loving.

We analyze the responses to the choice question in two ways. First, we provide non-parametric information on the response behavior of the different agencies. Secondly, we add the influence of individual-specific characteristics with a multinomial logistic regression.