ISSN 1403-2473 (Print)
Working Paper in Economics No. 687
Prices versus Standards: Evidence from an Artefactual Field Experiment on Managerial Investment Behavior
Magnus Hennlock, Åsa Löfgren, Conny Wollbrant
Department of Economics, Rev May 2021 (Nov 2017, Jan 2017)
Prices versus Standards: Evidence from an Artefactual Field Experiment on Managerial Investment Behavior
Magnus Hennlock
a, Åsa Löfgren
b, Conny Wollbrant
ca
Policy and Economics, IVL Swedish Environmental Research Institute, P.O. Box 530 21 S-400 14 Gothenburg, Sweden, tel. +46 31 708 65 08, e-mail: Magnus.Hennlock@ivl.se. (corresponding author)
b
Department of Economics, University of Gothenburg, P.O. Box 640, S-405 30 Gothenburg, Sweden, tel: +46 31 786 1375, e-mail: Asa.Lofgren@economics.gu.se
c*
Wollbrant, Economics Division, Stirling Management School, University of Stirling, Stirling, FK9 4LA, Scotland UK, tel: +44 1786 467276, e-mail: Conny.Wollbrant@stir.ac.uk
☆ Research funding from the Swedish Foundation for Strategic Environmental Research Mistra through the Mistra Carbon Exit research programme, the Swedish Research Council Formas through the project Policy Design for Correcting Market Failures Preventing a Circular Economy - Analysis and Evidence from Two Case Studies, and from the Swedish Environmental Protection Agency’s environmental research fund through the project Policies for Life-Cycles - an Integrated Assessment, is gratefully acknowledged.
Acknowledgments
The authors are grateful to Professor O. Johansson-Stenman, Department of Economics, University of Gothenburg, Professor M. Dufwenberg, Department of Economics, University of Arizona, and Senior Fellow D. Burtraw and K. McCormack at the Resources for the Future, Washington D.C. for their valuable comments and advice on a previous pilot design of the experiment. The research assistance from Lovisa Källmark, Fanny Isaksson Lantto, Louise Hwargård, Hannah Doherty, and Jonas Lind at the Swedish Environmental Research Institute is gratefully acknowledged.
Abstract
An artefactual field experiment is conducted using 166 experienced managers and senior advisors recruited from the chemical, pulp and paper, electricity, heating, and steel industries in Sweden. The experiment presented a strongly incentivized cost minimization task framed as an abatement investment decision in a hypothetical firm. Subjects were randomized into two treatments, and they made decisions in the presence of an emissions tax or an emissions standard. Treatments were calibrated to generate no treatment effects if subjects’ decisions are consistent with the predictions of rational choice theory. The results show that emissions standards reduce managers’ attentional focus on cost minimization, causing them to choose the most cost-effective alternative less often than is predicted by rational choice theory.
Although the emissions tax significantly increased managers’ focus on monetary information, managers
tended to minimize average abatement costs, resulting in abatement levels lower than optimal levels and
marginal costs lower than tax levels. The results demonstrate that the type of policy instruments per se
can induce decision-making bias via attentional focus, even when decision makers are experienced managers and senior advisors.
Keywords: artefactual field experiment, bounded rationality, attentional bias, pollution control, taxes, regulation
JEL-codes: C93, H32, L20.
1. Introduction
A common and important regulatory problem is how to optimally implement pollution control.
Neoclassical economic theory posits that profit-maximizing firms should equate the economic values of marginal products and marginal costs. This implies choosing the emissions level at the point at which the marginal abatement cost of reducing emissions equals the emissions tax level in the presence of environmental economic policy instruments, such as a Pigovian emissions tax, simultaneous with selecting the production level at which the price of the good equals the marginal cost of production. This result, which also guarantees cost-effective emissions reductions, is a cornerstone in the economics literature that argues for using economic policy instruments (such as taxes, subsidies, or tradable permits) over command and control instruments (such as emissions standards) to correct for externalities. However, that the optimal choice of policy instrument can be affected by other circumstances, including cost uncertainty, is well established (Weitzman 1974). Hence, understanding both the theory of pollution control and the potential limitations and inefficiencies that might arise when moving from theoretical predictions to the practical implementation of specific environmental policies is important from a regulator’s perspective.
Empirical evidence suggests that decisions in response to policy deviate in some circumstances
from predictions of neoclassical economic theory (for an overview, see, for example, Sorrell,
Mallett, and Nye (2011)). A prominent example of such a deviation is the so-called “energy
efficiency gap,” which refers to the observation that realized energy efficiency investments
diverge from the cost-minimizing level. Allcott and Greenstone (2012) comprehensively
reviewed the existence of the energy efficiency gap and pointed to substantial heterogeneity in
investment inefficiencies across different types of agents. Among the causes of these
investment inefficiencies are imperfect information and behavioral constraints (e.g., inattention
to information), although most of the discussion considers decision making from a consumer
perspective. In this paper, whether these types of investment inefficiencies are relevant for managerial decision making and whether the regulator’s choice of policy instrument (price versus quantity measures) affects the extent to which abatement investment decisions are cost effective are investigated.
Regarding firms’ decision making, neoclassical economic theory predicts that competition forces firms with long-run investment inefficiencies or low productivity to exit the market. Yet, well-documented and persistent empirical differences in productivity exist even within narrowly defined businesses and industries (for an overview, see Syverson 2011). Explaining empirical productivity differences has attracted strong interest from economists, and management has received particular attention. For instance, Bloom et al. (2013) showed that informational constraints and restrictions on competitive pressure can allow “badly run” firms to stay in business.
Armstrong and Huck (2010) and Cyert and March (1963) explained deviations from profit maximization in firms by introducing the concept of satisficing, first developed by Simon (1955). Satisficing is an instance of boundedly rational decision making and implies that, generally, a firm manager does not compare all universally possible investment choices;
instead, he or she chooses the first option deemed satisfactory.
1Alternatively, managerial decision making might be based on rules of thumb, such as imitating strategies of well- performing rivals or changing strategies only when profits decline to lower than an acceptable threshold rather than performing explicit calculations of optimal strategies (Armstrong and Huck 2010).
The psychology literature has suggested that decision makers might fail to engage in cost- benefit analyses and, instead, rely on decision-by-rules mechanisms learned through experience or social exchange (Simonson 1989; Prelec and Herrnstein 1991; Shafir, Simonson, and Tversky 1993). Amir and Ariely (2007) showed in an experiment that the need for monetary assessments might invoke previously learned rules, resulting in inconsistencies between monetary-based judgments and judgments based on other factors, such as effort and pleasure.
Similarly, Sunstein (2004) argued that the decision-making context could activate switching from preference maximization to, instead, applying a certain choice rule.
1 Decision-maker reliance on simplified decision rules for cognitively demanding tasks has received much attention in the literature on bounded rationality (Simon 1955, 1979). The term refers to the limited capacity for rationality that can arise when solving complex problems, processing large amounts of information, or making decisions in the presence of uncertainty and incomplete information.
This paper contributes to these areas of research by conducting an artefactual field experiment on investment choices among industrial managers and senior advisors in response to policy interventions (see Harrison and List (2004) for a taxonomy of experimental approaches). The overall aim is to test whether managers’ and senior advisors’ investment choices are consistent with predictions of neoclassical economic theory and whether different types of policy instruments induce systematic differences in investment choices. This research builds on empirical evidence that documented behavioral anomalies and the literature that emphasized the need to consider findings from behavioral economics and bounded rationality in the design and analysis of environmental policy (see, for example, Shogren (2002); Shogren and Taylor (2008); Gowdy (2008); Gsottbauer and van den Bergh (2010); Pollitt and Shaorshadze (2011);
Carlsson and Johansson-Stenman (2012); and Gsottbauer (2013)).
To this end, 166 firm managers and senior advisors from Swedish industry were recruited to participate in an incentivized experiment. Subjects were randomly assigned to one of two policy instrument treatments—either a price (framed as an emissions tax) or a quantity (framed as an emissions standard) instrument. In the experiment, managers and senior advisors were asked to make an abatement investment choice and were instructed to minimize total costs (with identical cost structures and incentives at the margin in both treatments). The experimental design allows for testing of whether the policy instrument type can i) increase the consistency of managers’ and senior advisors’ investment choices through the optimality prediction of neoclassical economic theory, ii) systematically shift investment choices toward other choice rules
2than the optimality rule, and iii) induce attentional biases in information use for decision making.
The literature on theoretical and experimental behavioral economics has revealed several judgment and attentional biases that depend on the information attribute that is more salient to decision makers. The dissociation between monetary assessments and predicted utility is a well- known judgment anomaly (see, for example, Thaler (1985); Hsee et al. (2003); and Amir, Ariely, and Carmon (2008)). Additionally, evidence suggests that monetary assessments make decision makers focus more on information variables related to transactions (for instance, prices, costs, and market norms) and less on variables related to the pleasure or utility of owning or consuming.
2 A choice rule is referred to as a mapping from a set of available information to a set of alternatives.
Amir, Ariely, and Carmon (2008) suggested that this phenomenon can be explained by focalism, meaning that different types of information and features of the evaluated stimuli inform different assessment tasks. They found experimental support for this conjecture by testing the impact of different types of information on subjects’ value assessments. They find a disparity between subjects’ willingness to pay and predicted utility by drawing subjects’
attention to either a monetary variable (production cost) or other attribute variables. Similarly, Schkade and Kahneman (1998) showed that decision-makers’ evaluations of changes are affected by the information emphasized. Finally, preference might be reversed when choices become more informed by the most prominent attribute of the evaluated options, known as the prominence effect (Tversky, Sattath, and Slovic 1988).
3In the context of our study, the results show that managers’ and senior advisors’ investment decisions differ from predictions of neoclassical theory of profit maximization. The results suggest that economic policy instruments and emissions standards naturally invoke two different assessment approaches—monetary-based and performance-based judgments—
leading to different attentional and judgment biases. Managers’ attentional focus on abatement levels when facing emissions standards seems to cause them to underweight cost information and to fail to select the alternative corresponding to less abatement, even when doing so is more cost effective. In contrast, when facing emissions taxes, managers’ focus on cost information increases toward consistency with the optimal rule of equalizing marginal abatement costs and the tax level but with an attentional bias toward minimizing average abatement costs. This experiment hence highlights important aspects of pollution control and policy choice, namely, that different types of policy instruments, ceteris paribus, induce differences in investment behavior.
The remainder of this paper is organized as follows. In section 2, the survey design and experimental treatments are described. In section 3, the results are presented, and the findings and their welfare and policy implications are discussed in section 4. Section 5 concludes.
3 Reversal of a preference occurs when a subject prefers one alternative in one response mode (e.g., choice) but exhibits the opposite preference order in another response mode (e.g., a rating over an attribute).
2. Survey Design and Experimental Treatments 2.1 Population and subjects
Experienced managers and senior advisors of large and medium-sized firms in the Swedish industry were recruited for the experiment. Their normal work duties involve decision making, analysis, preparation of technical and economic background information for investment decisions, and delivering decision recommendations to the firm’s CEO or board of directors.
The subject pool was identified from Bisnode’s database PARAD and the Swedish regulatory register of plants classified as engaging in environmentally hazardous activities (EHA) by the Swedish Environmental Code.
4Obtained from Bisnode’s database PARAD were personal contact information (phone number and e-mail address) of managers and board members of companies in the chemical, pulp and paper, and steel sectors, as well as of firms operating combustion plants to produce electricity and heat in Sweden. These search results included a total of 2,010 firms, each with more than 200 employees. The following profiles were used to search for subjects to be recruited from these firms to the experiment: CEO, financial manager, environmental manager, operative manager, regional manager, board member, and senior advisors, leaving us with a sample of 629 individuals belonging to at least one of these categories.
Both the Bisnode and EHA categories of the Swedish regulatory register of plants represent large and medium-size firms, including the largest firms in Sweden. Moreover, each regulated plant in the EHA category is generally regulated by various \policy instruments, including the command and control act, Swedish Environmental Code, and economic policy instruments, such as the Swedish NO
xcharge, the Swedish carbon tax, and the EU emissions trading system (EU ETS). Hence, this study’s subjects generally have broad experience with various types of economic policy instruments and standards.
A pilot study was conducted using 38 participants from the consulting industry who have previous experience as managers or senior advisors in the chemical, pulp and paper, and steel sectors in Sweden. After analyzing their responses, 10 subjects participated in two focus groups in which they were asked to explain how they understood the task and how they used the information available when making their choices. Subjects’ reported understanding of the task
4The Swedish regulatory register is Svenska miljörapporteringsportalen (SMP). The declaration data used are reviewed by the Swedish Environmental Emissions Data (Svenska Miljö Emissions Data, SMED) on behalf of the Swedish Environmental Protection Agency.
and their responses required only minor adjustments to the scenario text used in the final experiment.
During the recruiting process to the final experiment, the 629 managers and senior advisors in the sample were searched up to five times by phone. Phone contact was finally established with 374 (59.5%) of the 629 subjects. During the phone call, the subjects received a short introduction to the survey.
5Subjects were then asked whether they were willing to complete the survey. Out of the 374 subjects contacted, 268 (71.7%) were willing to complete the survey and were sent the survey link by e-mail directly after the phone call. After a maximum of three reminders by phone and then by e-mail, 166 subjects (110 managers and 56 senior advisors) completed the experiment, resulting in a 61.9% response rate.
Participants received the online survey with the stated aim of “better understanding the effects of environmental regulations on firms’ investments in cleaner technologies.” Participants were never informed in phone calls, e-mails, or through the online survey that they were participating in an experiment or that different treatments existed.
2.2 Information and choice sets
The purpose of this study is to investigate whether the type of policy instrument (the policy context) can:
i) cause managers’ and senior advisors’ investment choices to become more consistent with predictions of optimality of neoclassical economic theory;
ii) shift their investment choices toward other choice rules than the optimal response rule; and,
iii) cause different attentional biases in the revealed use of information for decision making.
A novel experimental design was developed to answer these questions. In an incentivized online experiment, subjects were presented with a scenario in which a hypothetical firm was about to invest in a new abatement technology to reduce its carbon dioxide emissions (see Appendix A.1 for the full survey). During installment of the technology, the firm has to choose the abatement level. Subjects were instructed to minimize compliance costs to the hypothetical
5 Call scripts are available from the authors on request.
firm
6and were then presented with a choice between two different abatement levels—A and B (henceforth referred to as a choice set)—with different marginal and average cost levels. The subjects were further informed that alternative A had a smaller abatement level than alternative B in all choice sets and that abatement costs were “continuously increasing”. This information was followed by the clarification that each unit of reduction in carbon dioxide emissions makes it more expensive to reduce emissions by another ton of carbon dioxide. As a result, the marginal cost of alternative B was always higher than that of alternative A.
7Each subject was presented with nine choice sets and, hence, made in total nine abatement decisions. Subjects were informed that succeeding in minimizing compliance costs for all nine investment decisions would earn them approximately USD 350 (after taxes), paid as a direct cash transfer to their bank account.
Table 1 presents an example of a choice set for the tax treatment. For a choice set, cost information for each alternative—A and B—is presented as marginal and average costs of one additional ton of carbon emissions reduction.
Table 1. Example of choice set in the tax treatment (choice set 1 in Appendix A.1)
Information in the tax treatment A B
Tax per ton of carbon dioxide (SEK/ton) 1 000 1 000
Marginal cost per ton of an additional unit of reduction (SEK/ton) 200 500 Average cost per ton of an additional unit of reduction (SEK/ton) 1 000 610
Because marginal costs are continuously increasing, average cost information and numerical abatement level values corresponding to each alternative are redundant for identifying the cost- minimizing alternative. The cost-minimizing alternative is always that for which the marginal cost equals the tax level (or is closest to the tax level for a corner solution). For a corner solution,
6 Abatement costs are included as a part of compliance costs. In the choice experiment task, subjects were instructed to maximize the cost savings, and that these costs included both abatement costs and any tax costs (in the tax treatment) or sanction fees (in the emission standard treatment) was explicitly stated. See Appendix A.1 for details about the survey design.
7 Thus, A and B are located along the same abatement cost function which limits the optimal problem to a problem consisting only of comparing marginal abatement cost levels between the two corner solution alternatives—A and B—and the Pigouvian tax level (or the Pigouvian sanction fee in the standards treatment), whereas the fixed cost is redundant for making the rational choice. Moreover, to simplify the choice set design there is no relationship between the production and emissions levels, implying that the change in abatement does not affect the optimal production level. The optimal production level is therefore constant during the decision on abatement levels.
as in Table 1, a subject first identifies the marginal cost as the relevant information and then identifies the alternatives for which their marginal cost equals (or is closest to) the marginal cost of 1,000 SEK/ton (alternative B in Table 1). (See Appendix A.1 for the full survey and details on the choice set design.)
Therefore, the experiment provides sufficient information for subjects to use formal reasoning to identify the cost-minimizing alternative. Nonetheless, identifying the cost-minimizing abatement option is still not a trivial task, which is a crucial feature of the design since the experiment aims to mimic the more complex decisions that managers and senior advisors must make. Because marginal costs are rarely available for investment decisions, industry managers might need to rely on choice rules depending on other information than that required by the optimality condition of neoclassical theory. For instance, a substitute for marginal costs could be average costs, which also account for part of the fixed costs in the decision. Hence, a possible choice rule could be “set average cost equal to the tax rate,” which would result in revealing choices with average costs close to the tax rate. Another possible choice rule could be
“minimize average cost,” which would result in revealed choices with average costs tending to be the lowest among alternatives.
As described above, each subject was presented with nine binary choice sets to identify the choice rules to which the subjects adhered. This design allows us to make statistical inferences about the choice rules revealed within subjects and, importantly, whether a subject’s revealed choice rule is consistent with the (incentivized) optimal rule of minimizing compliance costs.
For instance, as illustrated in Figure 1, choice set 5 contains the binary choice between alternatives A and B, which implies that alternative B is optimal (B being a corner solution), given the location on the marginal cost curve in relation to the tax level. In choice set 7, the optimal choice between A and B would be A (A also the unique interior solution).
8Figure 1. Examples of choice sets and their locations on the average and marginal costs of abatement curves (choice sets 5 and 7 in Appendix A.1).
8 To keep the marginal reasoning at the simplest level with marginal costs between alternatives A and B being compared with the tax level, choice sets comparing alternatives located at different sides of the interior solution were excluded. Otherwise, additional information about numerical values of the abatement levels would have been required to identify the optimal alternatives.
This experimental design therefore allows for testing the consistency between subjects’ choices and the following five choice rules: (i) average cost equals tax/sanction fee, (ii) marginal cost equals tax/sanction fee, (iii) minimize average cost, (iv) minimize marginal cost, and (v) minimize emissions (maximize abatement).
Table 2 illustrates the choice sequences of the nine choice sets corresponding to each of the five choice rules (i)–(v). For example, the optimal choice rule, MC = tax, implies choosing the abatement level for which the marginal abatement cost equals (or is closest to) the tax level. In Table 2, this implies choosing alternative B in choice sets 1–6 and alternative A in choice sets 7 to 9. In the experiment, the order of the nine choice sets presented to each subject was randomized to rule out order effects.
Wilcoxon matched-pairs signed-ranks tests were used to verify that the choice sequences of the five choice rules in Table 2 are significantly different from random sequences (Wilcoxon 1945).
The right-most column shows the test results for the choice sequence of each choice rule versus the sequence of expected values from a randomized sequence. The likelihood is 0.195% that a subject’s randomization among alternatives by chance should yield any of the choice rules in Table 2.
Table 2. Choice sequences of choice rules
The investment alternative with the largest emission reduction is placed intentionally as the B- alternative in all choice sets to reduce the complexity of the experiment. Information on the policy (and policy treatment) between the choice sets (the policy treatments are discussed in the next subsection) was also kept constant.
Subjects were informed that they would earn approximately USD 350 (after taxes) if they succeeded in minimizing the cost in all nine abatement choices, that is, following the choice sequence MC = tax in Table 2. The incentive scheme was tested in the pilot experiment and in focus groups. The payment was a significant incentive for the subject pool to focus on the experimental task. No time limits were imposed for completing the experiment and ex-post questionnaire, each subject needed on average 15 minutes to complete the full survey in both the pilot and the full experiment.
92.3 Policy treatments
Subjects were randomized to one of the two policy treatments. In the tax treatment, the aforementioned nine choice sets were presented as abatement decisions subject to a carbon emissions tax. In the standards treatment, the choice sets were presented as decisions subject to
9 Excluding outliers in response time exceeding 24 hours.
an emissions standard described in terms of an emission limit value. The two treatments started with identical descriptions of the investment decision and were then distinguished by a short information text of each policy as described in Table 3 (see Appendix A.1 for the full survey).
Table 3. Policy instrument treatments (translated from Swedish)
Investment decision (identical in both treatments)
We ask you to think about a situation in which you are involved in an investment decision in a firm that will invest in a new abatement technology to reduce their carbon dioxide emissions. Investments in technology impose fixed costs and marginal costs that are continuously increasing; that is, the more the firm reduces its carbon dioxide emissions, the more expensive it becomes to reduce another ton of carbon dioxide. With new technology, the firm is considering two different abatement levels: alternative A, which implies a smaller emission reduction, and alternative B, which implies a larger emission reduction. Both options have the same economic life.
Regulatory information in each treatment Emissions standard treatment Tax treatment The prerequisite for your decision is that the firm
produces in a country with an emissions standard that limits how much carbon dioxide emissions is allowed. The alternative with the lower abatement level (alternative A) results in an emissions level that exceeds the emissions limit value. If this alternative is chosen, the firm needs to pay SEK 1000 per ton in sanction fee for every additional ton that the emissions level exceeds the emissions limit value. a If the alternative with a higher abatement is chosen (alternative B), the emissions level does not exceed the standards, and no sanction fee needs to be paid.
The prerequisite for your decision is that the firm produces in a country with a carbon tax rate equal to SEK 1000 per ton carbon dioxide emission each year.a Both abatement alternatives A and B reduce emissions and, hence, the tax payments.
a SEK = Swedish krona
Although the regulatory information varied across treatments, it was identical for all choice sets within each treatment. The clarification for the standards treatment was that only alternative B would fulfill the standards. Choosing the lower abatement level, A, would lead to a Pigouvian sanction fee identical to the tax rate in the tax treatment.
10The subjects were explicitly asked to choose the abatement level, A or B, they believe gives the firm the greatest savings, including
10 In policy in practice, fixed or variable sanction fees are used. To obtain identical economic incentives at the margin between the two policy treatments, a “Pigouvian sanction fee” paid per unit emissions was used.
both abatement investment costs and any sanction fees (standard treatment) or tax payments (tax treatment).
The information provided in the choice sets in the two treatments was identical except for the policy information in the second row (see Table 3). The tax treatment reported the tax per ton, whereas the standards treatment reported the sanction fee per ton. In each choice set, the costs and the tax/sanction fee are identical across treatments, as observed in the examples in Tables 4a and 4b.
Table 4a. Example of choice set in the tax treatment
Information in the tax treatment A B
Tax per ton of carbon dioxide (SEK/ton) 1 000 1 000
Marginal cost per ton at attained level of reduction (SEK/ton) 200 500 Average cost per ton at attained level of reduction (SEK/ton) 1 000 610
Table 4b. Example of choice set in the standard treatment
Information in the standard treatment A B
Sanction fee per ton of carbon dioxide (SEK/ton) in excess of the standard 1 000 1 000 Marginal cost per ton at attained level of reduction (SEK/ton) 200 500 Average cost per ton at attained level of reduction (SEK/ton) 1 000 610
Although the two treatments had identical economic incentives at the margin, possible framing effects might result from paying a sanction fee for violating the regulation (emissions standards) rather than paying a tax. Framing effects might also exist because of informational biases resulting from the differences in information contexts between economic and regulatory instruments. This possibility is further discussed in the results section.
2.4 Descriptive statistics
Of the 166 subjects who completed the experiment, 90 were assigned to the tax treatment and 76 to the standards treatment. The average subject was 50 years old, 70% were male, and 80%
had a university degree. The subjects’ educational background was primarily in tech or
engineering (60%). Twenty-five percent had a background in science or environmental science, and 10% had a business administration background, and only 2% of the subjects had an economics background. The subjects had significant experience participating in environmental investment decisions at companies: 68% of the subjects stated that they had long (5–10 years) or very long (more than 10 years) experience. Only 2.5% of the subjects had no experience with making such decisions.
3. Results
This section discusses the analysis of the experimental data. First, whether the regulator’s choice of policy instrument (price versus emissions standards) affects the extent to which abatement investment decisions result in cost-effective choices is tested. More specifically, whether or not the type of policy instrument causes managers’ and senior advisors’ investment choices to become more or less consistent with the predicted optimality of neoclassical economic theory. Second, the choice rules that subjects use in their abatement decisions and whether they differ because the decisions were made subject to a tax or an emissions standard are examined. Finally, the effects of policy type on different attentional biases and revealed use of information for decision making are analyzed.
3.1 Rationality and treatment effects
Table 5 illustrates the frequencies and percentages of the observed choices across the two policy
treatments.
Table 5. Choice by treatment
Tax treatment Standards treatment
Abatement level A
Lower abatement
B Higher abatement
A Lower abatement
B Higher abatement
Observed choices* 382 428 219 465
Percentage observed choices*
47.2 52.8 32.0 68.0
Total number of choices made within each treatment*
810 684
*The larger number of choices in the tax treatment is the result of the larger number of subjects in the tax treatment (90) relative to the standards treatment (76). Each subject made nine choices.
Table 5 provides clear evidence of treatment effects on abatement investment levels (recall that
abatement increases from A to B; that is, for all choices, investment choice A corresponds to
the lowest abatement level and B to the highest). Alternative B is a more frequent choice for
the standards treatment than for the tax treatment: for the standards treatment, 68% of the
subjects chose alternative B (B complies with the emissions standards), and 32% selected
alternative A. For the tax treatment, 47% selected A and 53% selected B. Hence, although the
most commonly selected alternative in both treatments was alternative B (68% and 53% in the
standard and tax treatments, respectively), a larger share of the subjects chose abatement choice
A (lower abatement level) in the tax treatment relative to the standard treatment. (Table A.1 in
Appendix A.2 provides a quantitative overview of frequencies per choice set and treatment.)
Next, whether significant treatment effects exist on optimal abatement choices is tested. For
this purpose, a dummy variable is created that indicates for each choice whether it is consistent
with the optimal choice (dummy = 1) or not (dummy = 0). Wilcoxon matched-pairs signed-
ranks tests (Wilcoxon 1945) are performed on pairs of these dummy variables from each
treatment. Table 6 presents the Wilcoxon matched-pairs signed-ranks tests of equality of the
matched observation pairs. The null hypothesis is that no difference in distributions exists
between the two treatments. The total number of observations is 1,494 choices made by 166
subjects in nine choice sets. A statistical difference in the frequency of rational choices made
between treatment groups is found. The tax treatment seems to cause subjects to reveal a somewhat higher consistency with the optimal rule that equalizes marginal cost and the tax (sanction fee) level. For the tax treatment, 36% of all choices are consistent with the rational choice, whereas 32% of the choices are for the standards treatment. This effect is significant at the 10% level.
Table 6. Proportion of optimal choices (mean of dummy variables indicating consistency with the optimal rule)
Tax treatment Standards treatment Wilcoxon matched- pairs signed-rank test
Mean Mean Prob
Proportion of optimal choices
0.363 0.322 0.0565
Overall, these results suggest that the presence of a tax improves subjects’ consistency with the rational choice theory predictions relative to those facing an emissions standard, even though economic incentives and marginal costs are identical across the two types of policy treatments.
3.2 The effect of policy instrument on choices and attentional biases
This section discusses within-subject consistency with a specific choice rule. Table 7 shows that 25 subjects (27.8%) in the tax treatment and 24 subjects (31.6%) in the standards treatment are fully consistent with some specific choice rule for all nine choices. Notably, the most frequent choice sequences among subjects are consistent with the choice rules prescribing to either minimize emissions or minimize average abatement cost. Only the third most frequent rule consistent with the choices made by the subjects is to equalize marginal abatement cost with the tax (sanction fee) level, that is, the optimal cost-minimizing choice rule. The probability that subjects choosing randomly among alternatives should yield any of these choice rules consistently over the nine choices is a mere 0.195%.
11Statistically, therefore the claim can be made that if a subject’s choice sequence is found to be perfectly consistent with a given rule, his choices were based on some systematic belief about how the information should be used to minimize costs rather than by chance.
11 1/(29) = 0.001953
Table 7. Frequency and percentage of subjects per treatment consistent with a choice rule across all nine choices
Tax treatment Standards treatment Mann- Whitney U-test Choice rule Frequency Percentage Frequency Percentage Prob AC = tax level
AC = sanction fee level
0 0.0% 0 0.0% n/a
MC = tax level
MC = sanction fee level
5 5.6% 1 1.3% 0.1460
Min AC 10 11.1% 9 11.8% 0.8832
Min MC 0 0.0% 1 1.3% 0.2765
Min emissions 10 11.1% 13 17.1% 0.2668
It is noteworthy that a clear difference seems to exist between treatments when examining the number of subjects whose choice sequence exactly followed the optimal choice sequence (see Table 7). For the tax treatment, 5.6% of the subjects were fully consistent with the optimal rule—MC = tax level—across all of their choices, whereas 1.3% of the subjects were consistent with the optimal rule—MC = sanction fee level—in the standard treatment.
However, the Mann-Whitney U-test results in Table 7 show no significant treatment effects on the number of subjects fully consistent with a specific choice rule, supporting the conjecture that the subjects in table 7 made their choices on the basis of systematic beliefs about how to use the information that were strong enough to be unaffected by the policy treatments.
Next, whether the type of policy affects a subject’s total number of choices consistent with a
given rule is tested. Specifically, are there treatment effects that result in higher consistency
with a certain rule within subjects? For this purpose, Mann-Whitney U tests were performed on
the total number of choices for each subject, consistent with any of the choice rules. The results
in Table 8 indicate that the tax treatment significantly increases subjects’ consistency (higher
mean of the number of choices consistent with a given choice rule) with choice rules that minimize average and marginal abatement costs compared with the standard treatment.
12Table 8. Mean and standard deviation of total number of choices consistent with a given choice rule for each subject
Tax treatment Standards treatment Mann-
Whitney U- test
Choice rule Mean Std. Dev Mean Std. Dev Prob
AC = tax level
AC = sanction fee level
1.63 2.10 1.74 2.43 0.0155
MC = tax level
MC = sanction fee level
3.27 3.18 2.90 3.26 0.1087
Min AC 3.25 3.34 2.38 3.01 0.0155
Min MC 2.30 2.74 1.32 2.14 0.005
Min emissions 2.58 2.94 2.80 3.44 0.005
Moreover, the tax treatment makes subjects’ choices significantly less consistent with the rule of equalizing average cost and the tax (penalty) level and the rule of minimizing emissions relative to the standards treatment. An indication also exists that the tax treatment causes
subjects to be more consistent with the optimal rule that equalizes the marginal cost and the tax (penalty) rate (significant at the 10.1% level).
Finally, whether any significant treatment effects on single choices is tested. A dummy variable is created to indicate the choice rule with which each individual choice is consistent. Wilcoxon matched-pair signed-ranks tests (Wilcoxon 1945) are performed on the matched pairs of these dummy variables from each treatment. The total number of observations in each test is the 1,494 choices made by the 166 subjects in the nine choice sets. Table 9 shows that the tax treatment seems to increase subjects’ consistency with the optimal choice rule (treatment effect increase from 32.2% to 36.3% and p-value = 0.0565), the average cost-minimizing rule (the treatment
12 Tax versus performance standard: Mann-Whitney U-test, p = 0.0155 versus p = 0.005 for minimizing average and marginal abatement costs, respectively.
effect increases from 26.4% to 36.1% and p-value < 0.001), and the marginal cost-minimizing rule (treatment effect increases from 14.7% to 25.6% and p-value < 0.001).
The results in Tables 8 and 9 indicate that the tax treatment (compared with the standards treatment) significantly increases subjects’ consistency with choice rules that minimize average
abatement costs and marginal abatement costs, and equalizes the marginal cost and the tax (penalty) rate. No indication exists that the tax treatment significantly
affects subjects’ consistency with the equalize the average abatement cost and the tax level rule or a rule that minimizes emissions.
Table 9. Mean of dummy variables indicating proportion consistent with a given choice rule (dummy = 1 if consistent with a choice rule and dummy = 0 if not)
Tax treatment Performance standards treatment
Wilcoxon matched- pairs signed- rank test
Choice rule Mean Mean Prob
AC = tax level AC = sanction fee level
0.181 0.193 0.4216
MC = tax level MC = sanction fee level
0.363 0.322 0.0565
Min AC 0.361 0.264 0.0000
Min MC 0.256 0.147 0.0000
Min emissions 0.287 0.311 0.2157
A conditional logit model is applied to the full sample with the abatement choice as the dependent variable, cost variables as alternative-specific variables,
13and treatment and background variables as individual-specific controls to further investigate whether the use of
13Because cost information in the choice sets are generated along a non-linear cost function, the average and marginal cost levels are not linearly correlated. The correlation coefficient is 0.2314. A multicollinearity analysis by Belsley, Kuh, and Welsch (1980) showed no evidence of multicollinearity between the average and marginal cost levels.
information can explain observed abatement choices (McFadden and Train, 2000). Table A.2 in the Appendix presents predicted probabilities and marginal effects of the conditional logit model estimated for the full sample.
The conditional logit model shows that the policy treatments have the strongest effect on the choice probabilities. The tax treatment increases the probability of an average subject choosing A rather than B by 12.9 percentage points (p < 0.001). Average and marginal cost information represents alternative-specific attributes that are statistically significant at the 1% level for both investment choices (p ≤ 0.001). Overall, average cost information has a stronger effect on choices than marginal cost information. A one-unit increase in average cost increases the probability of choosing the lower abatement level A rather than the higher abatement level B by 0.060 percentage points. A one-unit increase in marginal cost increases the probability of choosing A by 0.018 percentage points.
Among the background variables in the survey only education level shows significant effects on the choice.
14A university degree increases the probability of choosing a lower abatement level (increasing the probability of investment level A by 17.4%). Furthermore, if the subjects stated that environment-related R&D is a prioritized area within the firm employing them, the probability that they choose a higher abatement level increases (increasing the probability of investment level B by 9.2%). The gender, age, type of degree, and years of working experience background variables do not have any significant effect on abatement choices.
4. Discussion
The optimal response to an environmental Pigouvian tax is to choose an abatement level for which the marginal abatement cost equals (or comes the closest) the tax level.
15The experiment in this study reveals that the number of choices that coincides with such an optimal response is higher when decisions are made subject to a tax than subject to an emissions standard with identical economic incentives as a tax. However, most importantly, other choice rules based on cost information are found to be more commonly used for the tax treatment than for the standards treatment. The choices that coincide with choice rules that minimize average and marginal costs are more frequent in the tax treatment than the choices corresponding to the
14 A multicollinearity analysis on alternatives- and case-specific variables in Table 3 was performed according to Belsley, Kuh, and Welsch (1980), resulting in a fairly moderate condition number = 39.62.
15 This holds as long as the tax level is higher than the minimum point at the average abatement cost curve, implying that the savings in environmental tax payments also cover the fixed portion of abatement investment costs.
optimal response of setting the tax equal to the marginal abatement cost. Hence, the experimental results indicate that managers’ and senior advisors’ use of information for their investment choices was significantly influenced by the type of policy instrument. In the standard treatment, the total number of choices within subjects tends to be more consistent with a rule to minimize emissions levels.
The different policy treatments seem to direct subjects’ attention to either a monetary variable (the tax treatment) or the emissions limit (the standard treatment). The experimental results show that these attentional biases induce differences between investment choices in the tax and standards treatments. In the standards treatment, subjects’ focus on abatement seems to have caused them to underweight cost information, failing to choose alternative A, which corresponds to less abatement, even when this is more cost effective.
However, regarding the tax instrument, managers’ and senior advisors’ revealed choice sequences are more consistent with the cost-effective choices that constitute the optimality condition of a marginal cost that equals the tax. Still, the strongest treatment effects were found for choice rules that minimize average or marginal abatement costs rather than the total cost to the firm (the sum of abatement and compliance costs). Consequently, subjects tended to minimize abatement costs across alternatives rather than the sum of the abatement and compliance costs. (In the experiment, the subjects were explicitly instructed to choose the investment that resulted in the largest cost savings considering compliance costs including abatement costs; see the instructions in the full survey in Appendix A.1.) Thus, subjects in the tax treatment overweighed the impact of their own abatement costs and failed to consider the full compliance costs. The results show that the attentional bias toward minimizing average and marginal abatement costs results in marginal abatement costs in the tax treatment that are lower than the marginal cost for the optimal choice sequence;
16853.7 SEK/kg compared with the expected means of the optimal choice of 900 SEK/kg.
The literature on behavioral economics suggests a number of explanations and results showing similar biases as in our findings. Schkade and Kahneman (1998) found that subjects overestimate the impact of events that are more salient to them than other events. Amir, Ariely, and Carmon (2008) found an additional explanation—that price assessments focus on the features of the transaction cues (prices, costs, market norms, and so on). Hsee et al. (2003) suggested that attentional bias with respect to the type of variable can lead to inconsistencies
16 The optimal choice sequence implies using corner solutions when interior solutions are not possible.
between choices and preferences. Further, the prominence effect might result in preference reversals because choices become more informed by the most prominent attribute of the evaluated alternatives (Tversky, Sattath, and Slovic 1988). Inattention to information is among the causes of the existence of the energy efficiency gap (Allcott and Greenstone, 2012). Even if this literature has so far focused mostly on consumers, the results indicate that managerial decision making is also subject to bias in their attention when responding to policies, which could cause investments in, for example, energy efficiency to diverge from the cost-minimizing level. If the problem is underinvestment, then the results indicate that information about the emissions or energy performance might need to be made more salient to the decision maker.
These results further suggest that appropriately balancing the importance of considerations that are the focus of attention with the importance of considerations that are currently in the background might be difficult. Our results show that this holds true even for experienced decision-makers.
4.1 Welfare implications
What general welfare conclusions can be reached from the experimental results? Theory shows that taxes result in an efficient outcome when a firm subject to a socially efficient tax bases its investment decisions on the optimal choice rule. However, minimizing the average abatement cost was the most frequent choice rule used by managers and senior advisors. This type of managerial investment behavior generally results in abatement levels lower than the socially efficient abatement level because the minimum point of the average abatement is associated with a lower abatement level than that of the socially efficient abatement level, resulting in a welfare loss.
17The efficiency of taxes versus standards depends on the choice rule used by firms. Taxes are weakly more efficient than standards when a firm’s investment decisions are based on the optimal choice rule. However, as our experimental result shows, minimizing average abatement cost was the most frequent choice rule, which introduces ambiguity in terms of the optimal policy choice for which standards can be more efficient than taxes.
17 For any tax level that is higher than the minimum point of the firm’s average cost curve, the optimal choice of the firm is a positive abatement level for which the marginal cost equals the tax level. Because the positively sloped marginal cost curve always cuts the minimum point of the U-shaped average abatement cost curve, any socially efficient abatement level that exceeds the minimum point of the average abatement cost implies a higher abatement level than that at the minimum point of the average abatement cost.
