Review of Accounting Studies

, Volume 15, Issue 3, pp 537–544

Discussion of: “Flattening the organization: the effect of organizational reporting structure on budgeting effectiveness”



DOI: 10.1007/s11142-010-9122-7

Cite this article as:
Young, R.A. Rev Account Stud (2010) 15: 537. doi:10.1007/s11142-010-9122-7


This paper is a discussion of Hannan et al. (Rev Account Stud, 2010), wherein subordinate-participants are endowed with private information, and superior-participants can potentially affect budget requests through their ability to reject them. Their findings are of interest to research on both the design of budgeting systems and span of control. I discuss the relevance of HRT to these literatures and comment on their implementation and experimental design. Also, I offer suggestions for future research, one of which is to explicitly introduce monitoring into experiments on span of control.


BudgetingExperimental economicsSpan of controlStrategy method

JEL Classification


1 Overview

This paper I have been invited to discuss, Hannan et al. (2010) (HRT), uses a laboratory experiment to investigate the effect of having superiors evaluate budget proposals from multiple subordinates simultaneously rather than individually. The budget protocols are referred to as High and Low Span of control, respectively. In the setting they investigate there would be no effect from increasing the span of control, unless superiors have preferences over something besides their own wealth, which I henceforth refer to as “other preferences.” HRT’s main contribution is its demonstration that superiors’ willingness to refuse to fund seemingly high-cost yet profitable projects, which has been well documented in similar settings, interacts with the span of control. In particular, the budgeting process becomes more efficient as the span of control increases. This is a well-designed, expertly administered, and carefully analyzed experiment.

2 Span of control and other relevant literatures

HRT relates to the span of control literature, in that it investigates the effect of having supervisors evaluate budget requests from three subordinates simultaneously versus processing them independently.1 It adopts a perspective that differs in many ways from previous work on span of control, emphasizing how supervisors’ “other preferences” influence the way they control the reporting behavior of employees which, in turn, affects the efficiency of the firm’s budgeting system.

The term “span of control,” which originates in military organization theory, has been studied further in the fields of management and economics. Span of control is often defined as the number of subordinates a manager can direct efficiently and effectively. Factors mentioned as affecting the optimal span of control include the geographical locations of the supervisor and employees, the similarity of the tasks performed by those under a supervisor, the capability of the employees, and the supervisor’s ability.

To put HRT’s contribution in perspective, I briefly reviewed some of the literature they cite. This literature’s view seems to be that the organization trades off the higher cost of paying supervisors against a “loss of control” that increases in the number of employees under a supervisor. This literature generally does not seem to model the details regarding the source of this loss of control. In particular, it does not directly address how informational issues affect span of control; clearly this would be an issue of first-order concern to accountants.

Williamson (1967) treats the span of control as exogenous and the number of hierarchies as endogenous. He investigates how the span of control affects the optimal size of the firm (number of employees). Calvo and Wellisz (1978) do not model span of control, but are concerned with the related loss of control. They model employees as maximizers and also focus on how characteristics of the monitoring environment affect the optimal size of the firm, although they do not directly address informational issues. Keren and Levhari (1979) make the span of control endogenous and assume the firm operates more efficiently if it reduces the time it takes to make decisions. This view of the firm is similar to that in Marschak and Radner (1972), in that the fundamental issue is more one of coordination than motivation.

Based on my cursory review, I believe one could make further contributions to this literature by taking a “first principles” approach in modeling both the decision and information environments. As far as theory goes, the existing principal-agent literature may already have much to say about optimal firm size and span of control.2 From the experimental side, accountants might explore how the decision and information environments affect the influence of “other preferences” on the span of control.

The HRT experiment makes a useful contribution in a number of ways. Consistent with the prior ultimatum game experiments and budgeting experiments in accounting, it illustrates within a laboratory setting that superiors react to manager’s reports, or budget requests, in ways different than what would be predicted by theories that rely solely on individual wealth maximization. It replicates the finding that “other preferences” induce superiors to punish high budget requests by rejecting them, even though it is personally costly so to do. Its unique contribution is supervisors react differently when they receive budget requests from more than one subordinate before they are permitted to make funding decision (High Span) than when they are required to consider them individually (Low Span). In particular, they become less tolerant of high budget requests when span of control is increased. Importantly, the subordinates anticipate this decreased tolerance and so build less slack into their budget requests. The two countervailing forces net out such that the firm operates more efficiently when span of control increases.

There are additional areas of the academic literature to which HRT makes contributions. Clearly, they contribute to the experimental economics literature and the experimental accounting literature on management control system design. More broadly, they make contributions to the general literature on the demand for organizing transactions through means other than a market. Demski (2004) presents the case for studying accounting in a setting where transactions are explicitly considered as endogenous. Further, his managerial accounting text (Demski 2008) is grounded in the idea that organizations exist to take advantage of interdependencies in the multiproduct cost function and can better combine multiple sources of information. Along these lines, HRT demonstrate how “other preferences” can be better-exploited if projects are evaluated simultaneously. This benefit would not be easily achieved in a market setting.

3 Experimental task, theory, and experimental design

The experiment uses a decision setting based on a version of Antle and Eppen (1985), with the major exception that superiors cannot commit as to the budget requests they will fund. This decision setting has also been used in Rankin et al. (2003, 2008) and Schwartz et al. (2009a). Lack of commitment, one might argue, activates preferences in addition to own-wealth maximization. Also, limitations on commitment may be a useful avenue to pursue, due to its applicability to situations in the world of affairs (Schwartz et al. 2009b).

HRT utilize this basic decision setting as a control, referred to as the Low Span treatment. The twist HRT incorporate is the introduction of the High Span treatment, where in each round of the experiment three subordinates submit budget requests to a single superior. After observing all three requests, the superior decides which to accept and which to reject. This setting has the advantage that under individual wealth maximization for superiors and subordinates a simple equilibrium results. Given the parameters, the superiors are weakly better off funding any possible request. Subordinates, in anticipation, request the maximum amount and extract all available profits from investment.

There is a multitude of evidence from similar experiments that superiors are, in fact, willing to reject high-budget requests. The widely accepted reasoning is superiors reject requests that would leave the subordinate with an inequitable share of the profits. This phenomenon could be consistent with rationality, if superiors derive utility from enforcing social norms, where the norm is superiors and subordinates should share in a way that is not too distant from a 50–50 split. This is the theory maintained by HRT. The sufficient condition HRT offer for why the High Span treatment might produce a different outcome is (within a single interaction) the superiors have (1) decreasing marginal utility for wealth and (2) constant marginal utility for punishing violations from social norms.

An additional feature of the experiment is the use of the strategy method for the last four rounds of the eight-round experiment. The first four rounds were implemented in the usual direct-response method. In the strategy method rounds superiors supply a threshold value for the maximum request they will fund to the experimenter, who then implements the superiors’ strategy. It is important that subordinates not be made aware of the superiors’ strategy, because this would be equivalent to superiors making binding commitments, as in the original theory paper by Antle and Eppen (1985). The strategy method has the advantage of providing a more efficient use of the data. Also, if experimental participants are concerned only with their own wealth, the two methods of extracting responses should not produce different outcomes.3 However, there is mixed evidence regarding whether experimental participants’ responses depend on which of the two methods is employed (Blount and Bazerman 1996; Brandts and Charness 2000, 2009).

Finally, I am unsure of the benefit of adding the Payoff-Adjusted Low Span treatment. I understand that, if one were primarily interested in the role of diminishing marginal utility in explaining the differences, some kind of additional treatment or analysis is warranted. If so, rather than add a treatment, one could choose three periods at random from the Low Span condition, which would equalize the superior’s potential payoff across the High and Low Span treatments. However, there may be issues related to differences in the perceived riskiness of the two conditions. Alternatively, I wonder if there is an additional way to analyze the data that might help to address the superior’s motivation for rejecting projects. Finally, it seems interesting to me if there exist control benefits from simply increasing the stakes in and of itself. This is accomplished in HRT by requiring superiors to delay reacting to budget proposals as they arrive, so they are forced to consider multiple projects simultaneously.4

4 Interpretation of results

The explanation offered by HRT for their results is, within a given interaction, superiors have decreasing marginal utility for money and constant marginal utility for enforcing social norms by turning down unacceptable budget requests. In my opinion, if this were the theory one wishes to test, one should not complicate the setting by assuming information asymmetry about the available surplus. In experimentation, just as in theoretical analysis, it is generally a good idea to make the setting no more complicated than necessary to study the forces of interest.

Taking the experiment as is, characterized by private information, several interesting issues come to mind. For example, as accountants, what might we learn about the role of private information and communication in similar budgeting experiments? An interesting paper that looks at such an issue is Coats et al. (2008), which demonstrates that the willingness to reject offers is decreased by the presence of information asymmetry. Given that a major purpose of budgeting is to extract and exploit subordinates’ private information, this finding is noteworthy. I wonder if there are other insights one could obtain from such experiments that would contribute more directly to accountants by concentrating on informational and reporting issues.

While HRT’s explanation relating to the utility functions of superiors is consistent with the data, there seem to be other valid explanations. For example, it seems plausible (to me) that superiors may be inclined to make relative comparisons, even though there is no reason for them so to do if they only have preferences for their own wealth. Further, although subordinates chose their requests without knowing other subordinates’ requests, they may anticipate this behavior and so be led to moderate their own budget requests. In addition, subordinates in the High Span setting may compete, even if they are not anticipating the superiors’ possible comparison of projects. As an aside, it is interesting how superiors in such experiments seem to be motivated by “other preferences,” yet subordinates generally seem to “best respond” as if they only care about their own wealth, as seems to be the case in HRT.5

There is, in fact, precedence for “comparison shopping” by superiors in similar economic experiments. For example, Bohnet and Zeckhauser (2004) conduct an experiment where responders (superiors) are informed about the average offer before they decide whether to accept or reject offers. This significantly increases both offers (decreases budget requests) and offer-specific (budget-contingent) rejection probabilities. Their findings are consistent with a distaste for deviations from the norm of equity but inconsistent with fairness theories. I wonder if a similar phenomenon might be taking place in the High Span treatment of HRT.

In another (somewhat) related experiment, Schwartz et al. (2009a) have a single subordinate make budget requests for three individual projects simultaneously (delayed). The superior can then compare the individual requests and choose freely which to accept and which to reject. They compare this treatment to two others, one in which superiors evaluate and make irrevocable funding decisions regarding each individual project iteratively (disaggregate) and—of greatest interest in their study—one where the subordinate makes a single funding request for the three projects combined (aggregated). In this latter treatment the superior must (of course) either accept all three projects or reject them all. Schwartz et al. (2009a) find differences between the aggregated treatment and the others, consistent with the idea that concealing information from the superiors about individual projects causes greater project acceptance, although it does not significantly improve the superiors’ welfare. However, they find no major differences between delayed (which is similar to the High Span treatment of HRT, except the funding requests are made by the same subordinate) and disaggregated (which is similar to HRT’s Low Span setting).

I am interested in how HRT use both the strategy and direct-response methods of extracting the superior’s decisions. The experimental economics literature suggests there will be a response method effect if participants have an emotional reaction to others’ choices that they do not anticipate when their decisions are obtained using the strategy method (Brosig et al. 2003). Once we admit to preferences for things other than own-wealth (a necessary condition for a difference in HRT’s treatments), I believe one cannot necessarily assume a game played using the strategy method and one played using the direct-response method will be played the same way. This issue is important because one of these games may be more representative than the other of the situation of primary interest for the researcher.6 While HRT state the responses in this particular setting are not affected significantly by the method of extracting superior decisions, they present no formal analysis. I do, however, understand and appreciate the advantage the strategy method has in terms of providing a useful dependent variable.

HRT implicitly acknowledge that this effect may be present in their footnote 19, which states that choices within the strategy method indicate superiors believed they would be tougher than they actually were ex post. Interestingly, this occurs even though they made their choices using the direct method in the first four rounds. I would guess this effect would be more pronounced if the strategy method were used first or if the response method were implemented as a between-subjects manipulation. We do not know whether the response method implemented in this fashion would interact with their treatment of interest, High versus Low Span. Also, the design does not allow for a clean comparison of the strategy and direct-response methods, as it would be confounded by a possible late- versus early-round effect due to, say, participant learning.

Finally, a few comments about Experiment 2 are in order. In an attempt to “isolate” how the superior and subordinate contribute individually to the outcomes in Experiment 1, HRT run an experiment where the subordinates are informed that superiors are pre-programmed to behave in an unspecified way. The thinking is that the behavior of subordinates can be compared between Experiments 1 and 2 and any differences can be attributed to their anticipation of the influence of the superior. I question the benefits of this exercise, given its stated purpose. While the difference between the two treatments was negligible, unlike in Experiment 1, I am reluctant to attribute the difference entirely to the effects within Experiment 1 of the subordinates’ anticipation of the superiors’ behavior. For one, subordinate behavior in both the High and Low Span treatments was much different than in Experiment 1, as indicated in Table 1. This makes me think subordinates may have viewed the experiments much differently for other reasons. More important, in my opinion it is not possible to attribute who is responsible for what behavior in Experiment 1; interactions are by definition not separable. This is the nature of game theoretic analysis—to study interactive settings. While Experiment 2 might be interesting in its own right, to me it is a completely different game from Experiment 1, because the endogeneity of the superiors’ decisions is removed.
Table 1

Subordinate cost reports in Experiments 1 and 2

Mean cost reports

Low span

High span

Experiment 1



Experiment 2



5 Summary

HRT examine issues related to both span of control and the design of budgeting mechanisms. Subordinates are endowed with private information, and superiors can implicitly control the subordinates’ budget requests through their ability to reject the requests. This is a simple setting yet useful way to begin addressing issues related to optimal firm size and span of control.

Possible directions for future research are worth considering. While the span of control literature generally refers to directing the employee and monitoring, these features are not built into HRT. Subordinates take no productive action. In addition, there are no informational efficiencies available to superiors from jointly considering projects, because project costs are not correlated, and superiors cannot commit as to how they will use information. Any of these features might be useful directions for future research in the areas of span of control and, more generally, management control system design.


The budget requests may be interpreted as cost reports, although these terms may induce a different decision frame for experimental participants.


In a similar decision and information environment, Arya et al. (1996) investigate a one-principal, two-agent setting. They provide conditions under which the optimal policy takes the form of relative project evaluation, even though there are no exogenous restrictions on available funding. Their setting differs from HRT in that it assumes (1) full commitment abilities for the principal, (2) a positive correlation between project costs, and (3) individual wealth maximization for all parties.


Of course, this assumption must be violated to make these sorts of experiments interesting.


This issue also arises in Schwarz et al. (2009a).


An exception can be found in Rankin et al. (2008), wherein subordinates seemed to care about being honest, but only if superiors were forced to accept the budget requests.


Schwartz et al. (2009b) study “unobservable commitments,” which are in the laboratory essentially the strategy method. They find that the method of extracting responses has a surprising and significant effect on the superior’s strategies in a setting very close to HRT.



The author benefited greatly from discussions with Anthony Meder, Steven Schwartz, and Eric Spires.

Copyright information

© Springer Science+Business Media, LLC 2010