Oppgrader til nyeste versjon av Internet eksplorer for best mulig visning av siden. Klikk her for for å skjule denne meldingen
Ikke pålogget
{{session.user.firstName}} {{session.user.lastName}}
Du har tilgang til Idunn gjennom , & {{sessionPartyGroup.name}}

Chapter 14: Seasoned parliamentarians perform worse than students in a lobbying experiment 1

Gjermund holds an MSc in Political Economy from BI Norwegian Business School. He is currently a Deputy Director of Government Relations at the Confederation of Norwegian Enterprise (NHO). Gjermund has previously worked with government relations in Energy Norway, and was a political advisor to the Minister for Church and Culture in the second Bondevik government.

Leif Helland holds a Dr. Polit. degree from the University of Oslo and is Professor of Economics at BI Norwegian Business School. He is also director of BI’s Centre for Experimental Studies and Research (CESAR). He works in applied game theory, experimental economics, and political economy. Helland has published in top journals within economics and political science, such as the Journal of Economic Theory, Journal of Theoretical Politics, Journal of Economic Behavior and Organization, Electoral Studies, and Journal of Conflict Resolution.

Lars holds a Dr. Econ. degree from BI Norwegian Business School. He currently works as a researcher at OsloMet. His research falls mainly within health economics, political economy, and local government studies. Monkerud has published his work in top journals such as Journal of Theoretical Politics, European Journal of Political Research, Health Economics, and Human Reproduction.

We present results of a laboratory experiment on costly lobbying, comparing the behavior of elite politicians and students. Our main finding is that members of the Norwegian national assembly deviate more from equilibrium predictions than students. This is in opposition to earlier experimental findings comparing the behavior of students and experienced public relations officers. Our finding is somewhat troubling, given that the underlying model addresses experienced real-world, decision makers. Ours is the first systematic study using members of a national parliament as subjects in a lobbying experiment.

Keywords:: Decision-making, Laboratory experiment, Lobbying

14.1 Introduction

In this chapter, we depart from a stylized game theoretic model of the interaction between a lobbyist and a decision maker. Thus, we are addressing an issue in the interface between business and politics. The core implications from the model are tested in a highly controlled environment: a laboratory experiment. Laboratory experiments have become an important and established tool in business studies over the last couple of decades.2 In the experiment we tap into a highly unusual pool of subjects, namely seasoned members of the Norwegian National Assembly (the Storting). Their behavior in the lobbying experiment is compared to the behavior of master students at BI Norwegian Business School.

Special interests are an integral part of democratic decision making, and they expend significant resources in their attempts to impact public policy.3 Frequently, influence is sought by strategically transmitting private information to policy makers; or, in short, by lobbying.4

Recent decades has witnessed advances in the understanding of lobbying.5 A core insight is that lobbying costs can increase the informative content of lobbying messages in equilibrium, and thereby improve democratic decision making.

We investigate experimentally the canonical model of costly lobbying (Potters and van Winden 1992).6 In the model there is only one lobby. Nature is in one of two possible states, and payoffs are state dependent. The lobby is privately informed about the true state. The lobby moves first. It chooses whether to send a costly signal (to lobby), or not to send a signal at all. The decision maker moves last. She confronts a binary choice; either preserve the status quo, or implement a fixed alternative policy. Independent of the true state of nature, the lobby prefers the alternative policy. In the absence of new information, the decision maker’s prior belief favors the status quo. For an interesting range of cost–payoff parameters, a strategically rich signalling game results in which semi-separating and pooling equilibria coexist. Various refinements can be employed in order to select between the equilibria of the model. The stylized environment of the model captures core trade-offs in the interaction between lobbyists and decision-makers.

This model has been subjected to experimental tests (Potters and van Winden 1996; 2000). One such test (Potters and van Winden 2000) compares behavior in two distinct subject groups: students and experienced public affairs officials (professionals).7 They find that professionals behave more in accordance with the model than students, thereby achieving a higher degree of separation and higher earnings. They conclude that professionals outperform students in playing the lobby game in a controlled environment.

Comparing inexperienced students and subjects with (model) relevant real-life experiences provides an external validity check on laboratory findings (Ball and Cech 1996). The approach is fairly common.8 Confidence in the empirical content of the underlying model is strengthened (weakened) if relevant professionals outperform (are outperformed by) students.

We subject Potters and van Winden’s (1992) costly lobbying model to an additional test of external validity: by comparing the behavior of students and representatives of the Norwegian National Assembly. To the best of our knowledge, ours is the first study that systematically uses elite politicians as subjects in a controlled laboratory experiment on decision-making.9 Our main finding is that elite politicians behave less in accordance with the costly lobbying model than students do. Elite politicians, having extensive real-life experience with lobbying, achieve lower degrees of separation and lower expected gains than inexperienced students. In our opinion, these results challenge the external validity of the costly lobbying model.

The remainder of the paper is organized as follows. In the next section our design, hypotheses, and procedures are outlined. Thereafter we present our result, and end the paper with a brief conclusion.

14.2 Design, hypothesis and procedures

We start by presenting the parametrized version of the game played by our subjects, the equilibria of this game, and the ensuing behavioral predictions that are tested. Subsequently, we outline our experimental procedures. Our game parameters are identical to the ones used by Potters and van Winden (2000). Apart from some (innocuous) economizing on language in the instructions used, and the employment of a different payoff procedure, we seek to replicate the experimental protocol of Potters and van Winden (2000) faithfully.

14.2.1 The game

The signaling game under study has two players, a sender (S) and a receiver (R). The time line is as follows; first nature draws a black ball with probability 1/3 or a white ball with complementary probability 2/3; the outcome of the draw is observed by S, but not by R; S decides on whether to send a costly signal (c > 0) (which may either have the content “the ball is black” or the content “the ball is white”), or to send no signal at all (c = 0); R observes the signal, updates his prior probability assessment on the color of the ball, and takes the decision B1 or B2, finally payoffs are distributed and the game ends.

Implementing the game with neutral language, such as white ball and black ball instead of for instance bad state and good state; sender and receiver instead of lobbyist and politician; and signal instead of lobbying was done in order to minimize the effect of context on behavior.

The parameters of the game are reproduced as a state-decision matrix in Table 14.1 (cost of signaling and a priori probability of state in parenthesis). In the table, the payoffs of S are provided first, then the payoffs of R. The strategic tension lies in the fact that R prefers B2 only if the state is black ball (the good state), while S always prefers B2. Evidently, signals cannot be trusted at face value. In deriving the equilibria of the model, the rules of the game and the information contained in Table 14.1 is assumed to be common knowledge.

Table 14.1:

Parameters of the experiment.

R’s choice
B1 B2
Low cost White (2/3) 2, 34, 1
(c = 0.5) Black (1/3) 1, 07, 1
High cost White (2/3) 1.5, 33.5, 1
(c = 1.5) Black (1/3) 1.5, 05.5, 1

The game has two equilibria for the parameters considered.10 The first one is a pooling equilibrium in which S sends no message in either state, and R always chooses B1. Since all sender types send the same signal, it reveals no new information. The prior then favors the decision B1. If no receiver chooses B2, incurring a cost by sending a signal is futile.

The second equilibrium is separating. In this equilibrium, S sends a costly message for sure if the ball is black, and with probability 1/4 if the ball is white. R responds to no message by choosing B1 for sure. If costs are high, R responds to a costly signal by choosing B2 with probability 3/4. If costs are low, R responds to a costly signal by choosing B2 with probability 1/4.

The logic in this equilibrium is as follows. After observing a costly message, R’s update (according to Bayes rule) is Pr(Black ball | c > 0) = 2/3. Given the update, R is indifferent in her choice. A best reply is then to randomize over her choice so as to make S indifferent between sending a costly message or no message at all. If costs are high, this is achieved for a probability of choosing B1 equal to 3/4. If costs are low, this is achieved for a probability of choosing B1 equal to 1/4. Given this response by R, sending a costly message for sure if the ball is black, and with probability 1/4 if the ball is white, is a best reply for S. Note that the content of the message does not matter in equilibrium. What matters is whether or not S incurs a cost in sending the message.

It has been shown that the separating equilibrium, but not the pooling equilibrium, passes the “universally divine” refinement criteria (for out of equilibrium beliefs), suggested by Jeffrey Banks and Joel Sobel in 1987. In Potters and van Winden’s (1992, 2000) opinion, this favours the separating equilibrium as a behavioral prediction.11

14.2.2 Hypotheses

We identify six testable hypotheses. Departing from the separating equilibrium, the following three follow immediately.

Hypothesis 1 (Signaling): a) the probability of a costly signal is higher following a black ball than a white ball; b) the probability of decision B2 is higher following a costly signal than no signal.

Hypothesis 2 (Treatment effect, receivers): a) the probability of decision B2 following a costly signal is higher in the high cost treatment than in the low cost treatment); b) the probability of decision B2 following no signal is independent of treatment.

Hypothesis 3 (Treatment effect, senders): a) the probability of a costly signal following a black ball is independent of treatment; b) the probability of a costly signal following a white ball is independent of treatment.

The next three hypotheses stipulate that elite politicians are better at playing the lobby equilibrium than students, both in the roles of senders and receivers. Successful candidates in elite politics are selected in highly competitive environments. One trait they are selected on is their ability to effectively signal their types in political campaigns. Such signalling can be modelled in ways that are structurally identical to the lobby game sketched above, albeit with the candidate as sender and the voter as receiver (Sloof 1998:55–8, and chapter 4).

Successful candidates enter elite politics, and become exposed to lobbyists on a regular basis in the role of receivers. At times, the efficient handling of such relationships can be consequential for political survival.

Empirics on the regularity of exposure can be found in a survey of members of the Norwegian Storting in 2001. Representatives were asked: How common is it to be contacted by professional lobbyists? By professional lobbyists in this context is meant hired guns, that is, lobbyists working for a customer on a contract-by-contract basis (as opposed to identifiable special interests that lobby in established corporatist channels). Norwegian institutions do not require professional lobbyists to disclose the identity of customers, and professionals are careful to protect their identity. Thus, the survey question addresses a relationship where precise prior information on preferences is largely unavailable, and signaling ought to be of particular importance.

Responses to the survey question were conditioned on the issue considered most important to the respondents in the current session (presumably an issue of ultimate consequence for political survival), and other issues. Above 20% of the representatives indicated that some or all of the organizations active in the issue considered most important used professional lobbyists. Close to 90% responded likewise with respect to the other issues category (Gullberg and Helland 2003).

Students, of course, are not selected on their ability to act effectively as senders in signalling games, and do not acquire substantial experience as receivers in signalling games through their studies.

The fourth hypothesis stipulates that the behavior of parliamentarians is closer to equilibrium behavior of senders than is the behavior of students. We use the two measures of signal error suggested by Potters and van Winden (2000) to check on this. The first one is an unweighted measure: 1/2|σ b S b | + 1/2|σ w S w |, where σ b  = 1 is the equilibrium frequency of costly signals after a black ball, σ w = 1/4 is ditto after a white ball, and S b , S w are actual frequencies of costly signals after a black ball and a white ball, respectively. The second measure uses the prior probability of drawing a black and white ball respectively as weights: 1/3|σ b S b | + 2/3|σ w S w |, with notation as in the first measure.

Hypothesis 4 (Gamesmanship senders): politicians have lower signal errors than students.

The fifth hypothesis relates to the level of gamesmanship achieved by politicians in the role they presumably master best; as receivers. Decision errors are defined in the following way 1/2|ρ 0β 0| + 1/2|ρ k β k |. In the formula, ρ 0 is the equilibrium frequency of B2 decisions after observing no-signal; ρ k is the equilibrium frequency of B2 decisions after observing a costly signal, which is conditioned on the cost treatment k = {C L ,C H }; β 0 is the actual frequency of B2 decisions given no signal; and β k is the actual frequency of B2 decisions given a costly signal. We believe that politicians will make more correct decisions (i.e. be closer to equilibrium) than students.

Hypothesis 5 (Gamesmanship receivers): politicians have lower decision errors than students.

The last hypothesis is more ad hoc, and stipulates that politicians—who need a measure of trust in order to get (re)elected—tend to follow the dictums of honesty more closely than students. An implication is that politicians are less likely than students to send costly signals unless a black ball has in fact been drawn. A similar hypothesis is used for the professionals in Potters and van Winden (2000). Their measure of separation is: , i = (Students,Politicians), with S b and S w defined as above.

Hypothesis 6 (Separation): politicians achieve a higher degree of separation than students.

14.2.3 Experimental procedures

In Potters and van Winden (2000), some sessions are conducted as computerized experiments, others as pen and paper experiments. Results are robust to this variation. We replicate with computerized sessions only. Potters and van Winden (2000) programmed their experiment with software no longer in use.12 Our sessions were programmed in zTree (Fischbacher 2007). We were not able to obtain screenshots from the original computerized sessions. Screenshots were therefore replicated based on information available in written instructions, working papers and final reports.

The written instructions for the computerized sessions in Potters and van Winden (2000) were in Dutch. These were translated to Norwegian with the help of two independent translators. One translated from Dutch to Norwegian, the other compared the translated Norwegian text with the original Dutch text. We had extensive communication with the translators in this process.

Instructions concerning payoffs were re-written to match the payoff structure in our lottery sessions (see below). In addition, some minor adjustment were made in order to economize on language and align the instructions with the typical format used in present day experiments.

Students were recruited from among first-year master students at BI Norwegian Business School via student e-mail and during lecture breaks. None of the recruited students had prior knowledge of the experiment, or experience as subjects in previous economics experiments at the school. No subject was used in more than one session.13

To attract parliamentarians for the experiment, a seminar on taxation and redistribution (including lunch) was arranged at the school.14 The idea was for this seminar to function as a showup fee in kind. Effort was made to pick a day that avoided clashes with essential business in the Storting; to announce the seminar and experiment widely in relevant networks; as well as to secure standbys in the (highly likely) case of last minute drop outs. Forty-seven out of the 169 elected representatives to the Storting responded positively to the invitation. Due to a busy schedule, 12 present and two former parliamentarians eventually participated in the experiment. In addition, six (non-elected) political advisors from the parliamentary party groups were recruited. Only elected representatives played in the role as responders (R) in the sessions with politicians.

In Potters and van Winden (2000), performance based monetary payoffs are used in all sessions (adjusted for differences in going wages of students and professionals). In some of our student sessions, we used performance based monetary payoffs, while in other student sessions we used the binary lottery procedure. In all sessions with politicians, we used the binary lottery procedure (see details in Table 14.2).

Why the binary lottery procedure? Norwegian parliamentarians are likely to consider monetary rewards for participation in research illegitimate, and are generally careful to accept monetary payments for services considered part of their duties as elected representatives. For example, most representatives do not accept money for lectures provided at business schools or universities, and many do not accept monetary rewards for articles published in journals or newspapers.

A potentially attractive alternative to monetary payments is the binary lottery procedure (Roth and Malouf, 1979; Berg et al., 1986). In this procedure, subjects earn points, and these points are lottery tickets in a random draw of a fixed price. Since a von Neuman-Morgenstern utility function is linear in the probabilities of outcomes, this payoff procedure renders individual utility linear in points. Thus, the binary lottery procedure controls for variation in risk preferences over subjects. From a purely theoretical point of view, this is the proper way to reward subjects in experiments, if controlling for risk preferences is desirable. On the flip side, results on how the procedure works in practice are mixed (Camerer, 2002:40–41; Berg et al., 2009). Whenever the binary lottery procedure was used in our experiment, the fixed prize consisted of two bottles of good wine. The procedure was accepted by the politicians as a legitimate form of reward.

Students participating in our lottery procedure sessions were paid 100 NOK as a showup fee. In student sessions with performance pay in NOK, there was no showup fee. On average, a session lasted 45 minutes. The exchange rate of experimental currency to NOK in sessions with monetary payoffs produced an expected payoff (in equilibrium) of 200 NOK per subject, which is above the typical optional wage for a student.

The lottery procedure was implemented in the following way. At the end of the session, the sum total of points earned by a subject in that session was divided by a fixed and publicly announced denominator common to all subjects of that session, resulting in the assignment of a number between 0 and 1 to each subject. A draw was then made from a random uniform variable in the interval 0 to 1. If the assigned number of a subject was larger (smaller) than the randomly drawn number, this subject would win (not win) a prize consisting of two bottles of good wine. The fixed denominator was calibrated in such a way that half of the subjects in a treatment could be expected to win a prize.

In all student sessions, the experiment was conducted twice. First we ran a session with one cost treatment, then a session with the other cost treatment. The participants were not informed about the second session before the first session had been concluded. Subjects played in the same role in both sessions. Because of time constraints, only one treatment was conducted in each of the two sessions with politicians. Since data from second sessions are likely to be biased (due to dependencies of observations over time), we only use data from first sessions in our analysis.

After a short introduction in a reception room, the participants were asked to draw roles (S or R) from an envelope, and were subsequently randomly assigned a number indicating their placement in the lab. After seating the subjects, instructions were read aloud (to achieve public knowledge of rules). Subjects where then allowed some time to read through the instructions independently and ask questions. Lastly, subjects where assigned a few simple questions, to make sure they understood the state-payoff matrixes. Subjects (in all sessions) first participated in a non-paying test round (in order to familiarize themselves with the screens), and subsequently played 10 paying rounds of the game.

The same matching protocol as in Potters and van Winden (2000) was employed. In this protocol, no subject is matched with the same subject for two or more consecutive periods, and no subject meets the same subject more than twice during the experiment. All interactions in the experiment preserved the anonymity of the subjects.

Table 14.2:

Session characteristics (c L : low cost treatment; c H : high cost treatment)

Session N Subjects Payoff Treatment(s)
Pol1 10 Politicians Lottery c L
Pol2 10 Politicians Lottery c H
Stud1 30 Students Lottery c L
Stud2 30 Students Lottery c H
Stud3 10 Students Money c L
Stud4 10 Students Money c H

14.2.4 The sample of politicians

The 14 (current and former) parliamentarians were on average 47.2 years old (standard deviation 10.9 years). Males and females where equally represented. On average they had served as elected representatives in parliament for 1609 days at the date of the experiment (standard deviation 1093.8 days; maximum of 3769 days; and minimum 889 days).15 Participating parliamentarians were distributed quite evenly between parties,16 while members on the finance committee where clearly overrepresented.17 Of the 12 currently serving parliamentarians participating in the experiment, 67% where reelected into the new Storting in the general election of 2009. The two former parliamentarians used in the experiment both left the Storting after the general election of 2005.

Of the six political advisors used in the experiment, four were males. Their average age was 32.2 years (maximum of 36 and minimum 27 years). The left wing side of the political spectrum was underrepresented in the group of advisors,18 while advisors to the parties’ finance fractions in parliament were overrepresented.19

Clearly the sample is not representative of the Storting at large. For the purpose of this study, however, we believe the distribution of party affiliation, age, gender and committee assignments is not likely to have a significant impact on the result, the reason being that the experiment was framed in neutral language without references to gender, age, party ideology or committee tasks.

14.3 Results

We present the results in two sections. First, we assess behavioral patterns within the lottery payoff procedure. That is, we test for certain predictions following from the model in terms of signalling patterns and treatment effects (hypotheses 1 through 3) and we check for subject pool effects (hypotheses 4 through 6) using only the sessions in which the wine lottery payoff structure was applied (Pol1,2 and Stud1,2). In particular, this allows for a controlled evaluation of subject pool effects in terms of the expectation that politicians are better decision makers than students. In a subsequent section, we briefly compare results from the lottery and money sessions in order to evaluate the impact of payoff structure on observed patterns. To this end, we evaluate and compare results using only the different sessions among the student population (Stud1,2,3,4).

Before presenting the results, we note the following points pertaining to some quite important analytic choices. First, we use non-parametric tests with p < 0.10 (one-tailed) as the threshold for significance: Wilcoxon’s signed rank test (WSR) for related samples and the Mann-Whitney U-test (MWU) for independent samples.20 Second, what counts as an observation? According to Potters and van Winden (2000:509), “[t]here are three possibilities: each play of the game, each individual subject or each session”. Potters and van Winden themselves end up using sessions as observations, of which there were 15 in their design. In our study we have 4 comparable sessions.21 However, only analyzing at the session level risks neglecting differences that are significant at the subject level. While the former strategy is the more conservative, we choose to present patterns at the subject level, noting that analysis at the session level produces no significant results for any of the tests presented below, save for tests of hypotheses 1a and 1b (to which we return shortly).

14.3.1 Signalling, treatment and subject pool effects

Table 14.3 presents the average proportion of costly signals (as averaged over subject means), conditioned on cost treatment (c L , c H ), and contingent on whether the draw was a white ball (S w ) or a black ball (S b ).22 Standard deviations over subject means are presented in parentheses. For convenience, the lower rows in these tables also presents the equivalent averages for the student-money sessions (Stud3,4).23

Table 14.3:

Costly signals by treatment (c L , c H ), contingent on draw of a white ball (S w ) or a black ball (S b ). Average subject level proportions (standard deviations)

Treatment: c L c H Average
S w S b S w S b S w S b
Politicians .66 (.26) .93 (.15) .20 (.25) .44 (.33) .43 (.34) .69 (.35)
Students Lottery .39 (.29) .85 (.17) .49 (.37) .85 (.19) .44 (.33) .85 (.18)
Average .46 (.30) .87 (.16) .42 (.36) .75 (.29) .44 (.33) .81 (.24)
[Students Money .63 (.37) .93 (.15) .45 (.19) .70 (.45) .54 (.29) .82 (.34)]

In a similar fashion, Table 14.4 displays average proportions of B2 decisions contingent on whether the decision maker is playing in the high or low cost treatment and whether he or she has received a costly signal (S) or not (0).

Table 14.4:

B2 decisions by treatment (c L , c H ), contingent on no signal (β 0) or costly signal (β S ). Average subject level proportions (standard deviations)

Treatment: c L c H Average
β 0 β S β 0 β S β 0 β S
Politicians .37 (.42) .33 (.19) .27 (.12) .37 (.44) .33 (.29) .35 (.32)
Students Lottery .12 (.20) .51 (.30) .16 (.22) .36 (.24) .14 (.14) .43 (.28)
Average .18 (.28) .46 (.29) .19 (.21) .37 (.29) .18 (.24) .41 (.29)
[Students Money .00 (.00) .43 (.22) .15 (.22) .68 (.24) .08 (.17) .55 (.26)]

Since our focus is on predicted differences between student and politician subjects pools, we concentrate our discussion of results on hypotheses 4–6. First, however, we summarize results for general model predictions as laid out in hypotheses 1a–3b.

First of all, costly signalling is significantly more frequent after a black ball has been drawn (hypothesis 1a). As can be seen from Table 14.3 the overall difference in relative frequency is Δ = 0.33, which is highly significant in a WSR test (p = 0.000).24 Moreover, this result holds in both the student and politician subject pools. The relative frequency of B2 decisions is also significantly higher after a costly signal has been sent (hypothesis 1b), with an overall Δ = 0.23 (p = 0.00). However, this result is driven entirely by student subjects and is clearly not attributable to politicians’ behavior, where the difference of only Δ = 0.02 is not near significant (p = 0.47).

As can be seen from Table 14.4, there is little support for hypothesis 2a, i.e. the proposition that B2 decisions following a costly signal is more frequent in the high cost treatment. The overall difference in relative frequencies, Δ = –0.10, is in fact the reverse of the expectation. This anomaly is even marginally significant among students (Δ = –0.15, p = 0.09), while the small positive difference (Δ = 0.05) for politicians is insignificant (p = 0.50).

As for the remaining hypotheses, overall results are in line with expectations, since neither B2 decisions following a no-message signal (hypothesis 2b), nor costly signals following the draw of a black ball (hypothesis 3a) or a white ball (hypothesis 3b) are significantly more frequent in either the high cost or low cost treatment (Δ = –0.01, Δ = –0.13, Δ = –0.04 and p = 0.26, p = 0.11, p = 0.26 respectively). However, whereas the results for the latter two hypotheses hold in the student population, there are significant anomalies in politicians’ behavior. Here, the differences in relative frequencies of costly signalling (Δ = –0.49 and Δ = –0.46 respectively) are significant (p = 0.02 and p = 0.03 respectively). We return to these general patterns and anomalies in the the two subject pools below. For now we move on to a more detailed discussion of behavior amongst politicians and student populations (hypotheses 4–6).

Hypothesis 4 (gamesmanship, senders). We apply the same measure for prediction errors (for players in the role of senders) as is used in Potters and van Winden (2000:511). It turns out that both the unweighted and the weighted versions of this measure of out-of-equlibrium behavior is substantially larger for politicians (0.31 and 0.30 respectively) than for students (0.21 and 0.23 respectively). Moreover, the differences are statistically significant in MWU tests (p = 0.05 for the unweighted difference and p = 0.07 for the weighted difference). This finding is in opposition to expectations. One would expect politicians to perform better than students. At the very least they should have some knowledge of the doings of “[p]ublic affairs managers [who] are professionally skilled to transmit information and to influence the beliefs and behavior of policy makers” (ibid. 505), and some experience in having played a structurally similar signalling game as senders in their electoral campaigns.

Hypothesis 5 (gamesmanship, receivers). The perhaps most interesting difference between the two subject pools concerns behavior in the role of receivers, i.e. the role to which politicians should arguably be especially accustomed. Applying our measure of out-of-equlibrium behavior ( ) reveals that politicians, even in the role of receivers, have larger decision errors (0.33) than students (0.25). However, this difference is not statistically significant (p = 0.13). Nevertheless, the data speak against the general hypothesis that professional experts (here: politicians in the role of receivers) do (significantly) better than laymen.

Hypothesis 6 (separation). We use an equivalent test to the one in Potters and van Winden (2000:505–506). Specifically, the measure in question (S b S w ) is meant to tap the ability of achieving full separation (S b = 1– and S w = 0), a result that is off-equilibrium (S b = 1 and S w = 1/4) but that will nevertheless leave subjects with higher earnings. Since politicians are presumably more experienced in lobbying situations, we expect them to achieve a higher degree of separation than students.25 The results, however, go in the opposite direction. While politicians on average achieve a separation measure of 0.26, students achieve a separation measure of 0.41. Moreover, this difference of Δ = –0.15 is significant (p = 0.09).

Finally, one may wonder if there are signs of convergence on (perhaps higher) levels of separation over the course of the ten rounds. Since we have only two independent draws per period per subject pool, a same-color draw in the two renders the separation measure undefined. A complete round-by-round mapping of separation is therefore not possible. However, we may compare separation in the first five rounds with that for the last five rounds. Figure 14.1 displays the pattern for the two subject pools. As can be seen from the figure, separation seems to be reasonably stable between early and late parts of the experiment26, with levels for students consistently higher than that for politicians.

Figure 14.1

Separation of students and politicians

14.3.2 Payoff structure effects

We now look specifically at differences in behavioral patterns between student subjects playing under the lottery and money payoff structures, respectively. We focus on tendencies for equilibrium (or sensible off-equilibrium) behavior in the two types of games. In other words we parallel the above tests for hypotheses 4 through 6, but now with payoff structure substituting for subject pool categories as the group effect.

First, however, we note that play in the experimental rounds with payment in the form of money (and with students as players) is by and large in accordance with the expectations of hypotheses 1 through 3. With the relevant differences (Δ) calculated from terms in the lower rows of Tables 14.3 and 14.4, the appropriate statistical tests (WSR or MWU tests) reveal that costly signals are sent more frequently after a draw of a black ball (Δ = 0.28, p = 0.07, see hypothesis 1a); B2 decisions are more frequent in the face of a costly signal (Δ = 0.48, p = 0.00, see hypothesis 1b); B2 decisions given a costly signal are more frequent in the high cost treatment (Δ = 0.26, p = 0.10, see hypothesis 2a); B2 decisions in situations of no signal are somewhat more frequent in the high cost treatment (Δ = 0.15, p = 0.09, see hypothesis 2b). The last finding is in breach of expectations; the relative frequency of costly signals after a draw of a black ball is independent of whether the game is played in the high or low cost treatment (Δ = –23, p = 0.22, see hypothesis 3a); the same goes for the sending of costly signals in the face of a white draw dependent on treatment (Δ = –0.18, p = 0.34, see hypothesis 3b).

Turning finally to differences in behavior between the money and the lottery sessions, it turns out that sender gamesmanship is not much different from the one version to the next. Whereas students in the money game have decision error measures of 0.25 (unweighted measure) and 0.27 (weighted measure), students in the lottery game can show for quite similar measures of 0.21 (unweighted) and 0.23 (weighted) (as reported earlier). Moreover, the differences are not statistically significant (p = 0.43 for unweighted measures; p = 0.40 for weighted measures). It seems, however, that receiver behavior may be influenced by the payoff structure of the game. While students in the money game, on average, have receiver decision errors of 0.14, students in the lottery game display a somewhat higher error rate of 0.25. Moreover, the difference between the two is statistically significant (p = 0.04). Lastly, it turns out that levels of separation do not seem to vary between the two types of games, with students in the money games achieving a slightly inferior separation measure of 0.28 compared to the reported 0.41 for students in the lottery games (the difference not statistically significant at p = 0.29).

14.3.3 Summing up the results

Statistical tests show that politicians behave less in accordance with predictions from the model than do students. The behavior of the former falls short of expectations on four (hypotheses 1b, 2a, 3a, 3b) out of six counts (hypotheses 1a through 3b). Students perform more in accordance with model expectations, failing to conform to one expectation (hypothesis 2a). Moreover, in terms of behavior politicians perform significantly worse than students in the role of senders and, importantly, do no better than students in the role of receivers. Finally, students achieve significantly higher rates of separation than politicians. We stress that these results pertain to patterns in games where the lottery payoff scheme is applied. However, in supplementary tests of effects from the applied payoff structure, it is found that students by and large conform to model expectations, regardless of payoff scheme.

14.4 Conclusion

We have replicated a costly signaling game, comparing the behavior of students and elite politicians. Both groups deviate from equilibrium predictions. However, elite politicians are substantially more off-mark than students. We cannot entirely rule out that our choice of payoff procedure is partially responsible for the results obtained, though the robustness of student behavior to payoff procedure indicates that the lottery procedure might not be culpable for the results. Caution has to be taken due to the low number of observations in our experiment. This said, the main pattern of more equilibrium deviations by elite politicians seems fairly robust. Most surprisingly, perhaps: elite politicians are no better than students in the role of receivers (interpreting and acting on lobby signals).

In our opinion, and with the appropriate methodological qualifications, our results raise questions pertaining to the external validity of the costly signaling model.

Why do elite politicians deviate more from equilibrium than students? There may be a number of reasons for this. The experiment was conducted without contextualizing it as a lobby problem. A speculation is that a less abstract context could have primed the experience of the politicians more effectively.

Furthermore, any one–or any combination–of the highly stylized assumptions underpinning the lobby model may constitute a poor approximation to the kind of lobby relations experienced by elite politicians in their daily dealings. For instance, real-life interactions are usually repeated under an open horizon, making reputations salient, and expanding the set of equilibria. Also, politicians in real life situations typically encounter more than a single signal. Multiple signals are usually more informative than single signals, and should induce more honesty. If such experiences prime the lab behavior of politicians, it may produce excessive trust (compared to the equilibrium of the single signal model). Our experiment, however, was not designed to isolate reputational concerns and multiple signals.


Abbink, Klaus and Bettina Rockenbach (2006): Option Pricing by Students and Professional Traders: A Behavioral Investigation. Managerial and Decision Economics 27:497–510.

Alevy, Jonathan, Michael Haigh and John List (2007): Information Cascades: Evidence from a Field Experiment with Financial Market Professionals. The Journal of Finance LXII(1):151–180.

Ansolabehere, Stephen, John de Figuerido and James Snyder Jr. (2003): Why Is There so Little Money in U.S. Politics. The Journal of Economic Perspectives 17(1):105–130.

Atlas, Vivi, Lisa Cameron, Ananish Chaudhuri, Nisvan Erkal and Lata Gangadharan (2006): Subject Pool Effects in a Corruption Experiment: A Comparison of Indonesian Public Servants and Indonesian Students. Research Paper No.975: Department of Economics/University of Melbourne.

Austen-Smith, David and John Wright (1992): Competitive Lobbying for a Legislators Vote. Social Choice and Welfare 9:229–257.

Austen-Smith, David and JohnWright (1996): Theory and Evidence for Counteractive Lobbying. American Journal of Political Science 40(2):543–564.

Ball, Sheryl and Paula-Ann Cech (1996): Subject Pool Choice and Treatment Effects in Economic. Laboratory Research. In: R. Mark Issac (ed.) Research in Experimental Economics, Vol. 6. Creenwich: JAI-Press.

Banks, Jeffrey and Joel Sobel (1987): Equilibrium selection in signaling games. Econometrica 55:647–661.

Bhattacharyya, Gouri and Richard Johnson (1977): Statistical Condepts and Methods. New York: John Wiley and Sons.

Belot, Michèle, Raymond Duch and Luis Miller (2010): Who should be called to the lab? A comprehensive comparison of students and non-students in classic experimental games. Discussion Papers No. 2010:001. University of Oxford: Nuffield College.

Berg, Joyce, Lane Daley, John Dickhaut and John O’Brian (1986): Controlling preferences for gambles on units of experimental exchange. Quarterly Journal of Economics 101:281–306.

Berg, Joyce, Thomas Rietz and John Dickhaut (2009): On the performance of the lottery procedure for controlling risk preferences. In: Charles Plott and Vernon Smith (eds.) Handbook of Experimental Economic Results: Volume 1. Amsterdam: North-Holland.

Camerer, Colin (2002): Behavioral Game Theory: Experiments in Strategic Interaction. Princeton: Princeton University Press.

Cleave, Blair L., Nikos Nikiforakis and Robert Slonim (2013): Is there selection bias in laboratory experiments? The case of social and risk preferences. Experimental Economics 16(3):372–382.

Cooper, David (2006): Are Experienced Managers Experts at Overcoming Coordination Failure? The B.E. Journal of Economic Analysis and Policy, Berkeley Electronic Press, 0(2).

Cooper, David, John Kagel, Wei Lo and Qing Liang Gu (1999): Gaming Against Managers in Incentive Systems: Experimental Results with Chinese Students and Chinese Managers. American Economic Review 89(4):781–804.

Dejong, Douglas and Robert Forsythe (1988): A note on the use of businessmen as subjects in sealed offer markets. Journal of Economic Behavior and Organization 9:87–100.

Dohmen, Thomas , Armin Falk, David Huffmann and Uwe Sunde (2008): Representative trust and reciprocity: Prevalence and determinants. Economic Inquiry 46(1):84–90.

Donohue, Karen, Elena Katok and Stephen Leider (2018): The Handbook of Behavioral Operations. Hoboken, New Jersey: John Wiely and Sons.

Dyer, Douglas, John Kagel and Dan Levin (1989): A Comparison of Naive and Experienced Bidders in Common Value Offer Auctions: A Laboratory Analysis. The Economic Journal 99(394):108–15.

Egas, Martijn and Arno Reidel (2008): The economics of altruistic punishment and the maintenance of cooperation. Proceedings of the Royal Society of London B: Biological Sciences 275(1637):871–878.

Fatas, E., T. Neugebauer and P. Tamborero (2007): How politicians make decisions: a political choice experiment. Journal of Economics 92(2): 167–196.

Falk, Armin, Stephan Meier and Christian Zehnder (2013): Do lab experiments misrepresent social preferences? The case of self-selected student samples. Journal of the European Economic Association 11(4):839–852.

Gullberg, Anne Therese and Leif Helland (2003): Profesjonell lobbyisme. Norske eliters bruk og holdninger (“Professional lobbyism. Use and attitudes among Norwegian elites”). Oslo: Makt- og demokratiutredningen. Rapport 70/03.

Løyning, Gjermund (2008): Lobbyism, a waste of time? Parliamentarians and students in a lobbying experiment. MSc Thesis (Political Economy). Oslo: Norwegian School of Management.

Mahoney, Christine (2004): The power of institutions: State and interest group activity in the European Union. European Union Politics 5: 441–466.

Potters, Jan and Frans van Winden (2000): Professionals and students in a lobbying experiment. Professional rules of conduct and subject surrogacy. Journal of Economic Behavior and Organization 43(4):499–522.

Potters, Jand and Frans van Winden (1996): Comparative statics of a signaling game. An experimental study. International Journal of Game Theory 25:329–354.

Potters, Jan and Frans van Winden (1992): Lobbying and asymmetric information. Public Choice 74(3):269–292.

Roth, Alvin and Michael Malouf (1979): Game-theoretic models and the role of information in bargaining. Psychological Review 86:574–894.

Samuelson, Larry (1997): Evolutionary Games and Equilibrium Selection. Cambridge Mass.: The MIT-Press.

Sloof, Randolph (1998): Game Theoretic Models of the Political Influence of Interest Groups. Dordrecht: Kluwer Academic Publishers.

Wooders, John (2010): Does Experience teach? Professionals and Minimax Play in the Lab. Econometrica 78(3):1143–1154.

1Financial assistance from BI Norwegian Business School is greatly appreciated. Thanks to Jan Potters and Frans van Winden for generously sharing material underlying their 2000 experiment. We are indebted to Rolf Aaberge and Alexander Cappelen for constructive comments on an earlier draft.
2An excellent overview of laboratory experiments in operations management is provided by Donohue et al., (2018). To assess the breadth of the experimental program in current business studies, it is a good idea to consult recent issues of Management Journal, the top journal in the field.
3Estimates of expenses and activities for special interests in the federal US process are provided by Grossman & Helpman (2001:chapter 1). For a critical interpretation of such numbers, see Ansolabere et al., (2003). For interest group activity in the EU, see the survey of Mahoney (2004) with references.
4Grossman and Helpman (2001:4–13), and Sloof (1998:18–20).
5See for instance Grossman and Helpman (2001, part I)I; Persson and Tabellini (2000, ch.7).
6The model has been extended in various directions. An overview of many extensions, and a thorough analysis of a few more, is provided by Sloof (1998). For a model that departs from slightly different assumptions (two lobbies), but reach qualitatively comparable conclusions for certain parameters, see Austen-Smith and Wright (1992). Austen-Smith and Wright (1996) provide tests of the model on field data from the US Congress.
7Professionals are described as executives who subscribed to two different conferences on public affairs (one in Amsterdam, the other one in The Hague). In particular, they held positions as public affairs and public relations officers, in the public as well as the private sector.
8For instance: Cooper (2006) and Cooper et al., (1999) compare managers and students in coordination experiments; Dyer et al., (1989) compare students with experienced contractors in experiments on common value auctions; Dejong and Forsythe (1986) compare businessmen and students in sealed offer experimental markets; Alantas et al., (2006) compare (Indonesian) students and public servants in a corruption experiment; Abbik and Rockenbach (2006) compare professional traders with students in an options-pricing experiment; Wooders (2010) compares students and soccer professionals’ ability to play minimax in an experiment on a zero-sum game. Alevy et al., (2007) compare data collected in a field experiment on information cascades using professional traders from the Chicago Board of Trade, with laboratory data using students.
9Fatas et al., (2007) explore framing effects in a survey experiment including Spanish elite politicians. They find weak evidence of less deviation from the rational-choice model among elite politicians.
10A general treatment of the equilibria in the costly signalling game is provided in Potters and van Winden (1992), and in Sloof (1998).
11Using this (and other sophisticated) refinement criteria to underpin behavioral predictions is not uncontroversial. Se, for instance, the discussion in Samuelson 1997:5–12.
12The computerized sessions in Potters and van Winden was programmed in EASEL for the OS 2 operating system of IBM.
13Subjects used in economics experiments at the school are entered in a historical data base. For an evaluation of the representativeness of student populations used in lab experiments relative to the population at large see Egas and Reidel (2008), Dohmen et al., (2008), and Belot et al., (2010). For an evaluation of the representativeness of students participating in lab experiments relative to students that do not, see Cleave et al., (2011), Falk et al., (2013).
14Two leading Norwegian experts presented their views on the need for a tax reform: professor in tax legislation at BI Norwegian Business School, Ole Gjems Onstad, and senior researcher Rolf Aaberge at Statistics Norway.
15Though the participants must be said to have substantial experience as MPs, it is, at the end of the day, a matter of taste whether one feels that their experience merits the label seasoned or not.
16Progress Party (Fremskrittspartiet) 2; Conservative Party (Høyre) 3; Christian Democratic Party (Kristelig Folkeparti) 3; Liberal Party (Venstre) 2; Labor Party (Arbeiderpartiet) 2; and Socialist Left party (Sosialistisk Venstreparti) 2.
17Finance committee 6; Justice committee 1; Labour and Social services committee 1; Church, Education and Research committee 1; Energy and Environment committee 2; Industry committee 1; Foreign Affairs committee 1; Local administration committee 1.
18Conservative party (“Høyre”) 1; Christian Democratic party (“Kristelig Folkeparti”) 4; and Center party (“Senterpartiet”) 1.
19Advisors to the finance fraction of a party in parliament 4; Advisors to the local administration fraction of a party in parliament 1; Advisor to the prime minsters’ office 1.
20The WSR test is a non-parametric test for within-subject differences, while the MWU test is a non-parametric test for between-subject differences. A good exposition is provided by Bhattacharyya and Johnson (1977, chapter 15).
21Potters and van Winden (2000:note 11) are able to demonstrate significant differences between sessions with the same experimental design, indicating dependencies within sessions. Obviously, we are not able to (statstically) test for such differences in our material.
22As for the content of messages; politicians lie somewhat less than students (13% vs. 20% average lie rate over subjects), and refrain from sending a signal somewhat more often (47% vs. 40% for students), but none of these differences are significant (p=0.29 and p=0.52 respectively).
23The next section deals with payoff structure effects.
24Consulting Table 14.3, the average difference of 0.33 in costly signaling (c H ) between subjects drawing a black and a white ball is Δ = π(c H |S b ) – π(c H |S w ) = 0.75 – 0.42 = 0.33 (in which π stands for proportion).
25In line with Potters and van Winden (2000:505), we may argue that behavior to this end in particular implies that senders should avoid sending deceitful messages when the color is white. Also, one may argue that the existence of professional rules of conduct such as a never cheat or misinform rule is conducive in that respect (ibid.). Again, we may argue that politicians’ experience with real-life lobbying situations may also result in behavioral influence from what originates more specifically from calculations on the part of lobbyists. Or one may assume that strong, but nevertheless general norms of honesty harbored by politicians themselves also influences behavior in situations that are not defining features of the policy-maker role.
26There are no significant differences between separation measures in early and late periods, neither for students nor for politicians. This is similar to patterns found in Potters and van Winden’s (2000:512) study. The authors find no particular pattern of convergence, but they do, conversely, find that professionals (lobbyists) consistently achieve higher separation than non-professionals (students).

Idunn bruker informasjonskapsler (cookies). Ved å fortsette å bruke nettsiden godtar du dette. Klikk her for mer informasjon