FORMALIZING MANAGERIAL BEHAVIOR: EVALUATION APPREHENSION AND LACK OF CONTROL

Hierarchy and Information Flow 1995 administrative costs is that they reflect the time

Table 2. Experimental data: number of ideas submitted the participant has to spend to deal with further

per period as a function of experimental treatment evaluation requests by her superior(s), regardless

of whether she correctly sends up a good idea or Model 3a Model 3b mistakenly passes up a bad one.

Prediction1

0.135 0.135 The parameters for the three mechanisms were

Hierarchy

( 0.099) ( 0.189) chosen to ensure that each mechanism allows

Predicition 2

for the predicted pattern to emerge; that is, for − 0.385 ( 0.323)

Evaluation apprehension

information economics, participants’ expected

− 0.235 * − 0.235 * earnings increase if they pass more ideas up in the

Evaluation

( 0.123) ( 0.066) three-layer hierarchy as compared to the two-layer

apprehension × hierarchy

Prediction 3

0.790 apprehension and lack of control, the opposite

hierarchy, all else being equal. For evaluation

Lack of control

− 2.150 ** − 2.150 picture emerges: participants’ expected earnings ** ( 0.285)

Lack of control × hierarchy

( 0.000) increase if they pass fewer ideas on in the steeper

5.480 ** 5.615 ** hierarchy relative to the less steep hierarchy, all else

Constant

( 0.267) ( 0.000) being equal. Thus, if participants exhibit the pre-

1,200 1,200 dicted sensitivity in terms of their idea-submission

Observations

60 R behavior as a function of hierarchy, this provides 2 (total)

Groups

conclusive evidence about the existence of the 0.48 mechanism in question. Furthermore, the chosen

R 2 (between)

Model 3a: Cross-sectional data, standard errors clustered by

parameters allow us to make inferences about the

subject.

relative strength of the various mechanisms.

Model 3b: Multilevel data, fixed effects for subjects, standard errors clustered by subject and adjusted for degrees of freedom. *Significant at 10 percent; **significant at 1 percent.

Results and discussion

Given the experimental nature of our data, few and hierarchy exceeds the one for evaluation observations appear noteworthy in terms of purely

apprehension and hierarchy by almost an order descriptive statistics. The average number of ideas

of magnitude. Not surprisingly, subjects generally passed up by a subject in the different treatments

pass up significantly fewer ideas in both treatments was 5.68 (information economics), 5.43 (evaluation

than in the information economics base case (for apprehension), and 4.40 (lack of control). Individ-

Model 3a: Evaluation apprehension + Evaluation uals earned, on average, about US$31 per session,

apprehension × Hierarchy = −0.62, p = 0.00; Lack with the variable, performance-based component of

of control + Lack of control × Hierarchy = −1.36, their payoff amounting to 90 percent of their total

p= 0.00). Finally, we do not find support for revenue—thus indicating that subjects engaged in

Prediction 1; however, directionally our results are the experimental task and seemed to understand it

consistent with the predicted pattern. Participants well overall.

passed on more ideas the steeper the hierarchy Table 2 presents the results pertaining to the

that surrounds them in the information economics core relationships we sought to unravel in the

treatment. In any case, even if our subjects suffered experiment. Model 3a provides experimental

from an omission bias (Baron and Ritov, 1994) support for both Prediction 2 and Prediction 3,

and over-appreciated the relative benefits from suggesting that participants—when being sanc-

avoiding commission errors, this omission bias tioned for commission errors or when incurring

would be too weak to account for a negative net administrative costs for forwarding ideas—pass

effect of hierarchy on agents’ propensity to pass up along fewer ideas the steeper the hierarchy that

information to their superiors. surrounds them. The results remain robust in Model

3b, in which we also control for time-invariant

CONCLUSIONS AND FURTHER

subject-specific effects. Notably, while both

RESEARCH

treatments—evaluation apprehension and lack of control—interact negatively with the degree of

The current paper shows that ultra-parsimonious hierarchy, the interaction effect for lack of control

nonbehavioral

models—as suggested by

1996 M. Reitzig and B. Maciejovsky information economists—insufficiently account

for mid-level managers’ information transmission behaviors in real-world organizations. Contrary to the model’s predictions, mid-level managers in the field pass on fewer ideas the steeper the hierarchy, suggesting the existence of more complex underpin- ning behavioral mechanisms. In our specific case, introducing another manager at the second highest level within the business unit would lead to roughly

10 percent fewer ideas being passed up from below, all else being equal. Whereas this figure is likely context-specific, it provides an idea of the order of magnitude of effect sizes. Complementary exper- imental results indicate that agents pass up fewer ideas the steeper the hierarchy that surrounds them once they fear negative feedback for commission errors they make or once they incur administrative costs for passing on ideas irrespective of their quality. Of the two behavioral mechanisms, the latter seems to be the most powerful by far.

Our findings appear to be relevant to differ- ent communities of scholars in the field of strate- gic management and organizational studies more broadly. Both behavioral strategists more broadly and colleagues examining the evolution of strate- gies within firms more specifically may find it inter- esting that employees’ unwillingness to transfer information within the firm need not necessarily originate from the relationship between the sender and receiver of such information (Hansen, 1999; Szulanski, 1996). A middle manager, irrespective of her liking of a subordinate, may decline to pass a proposal by the latter on to top-level management if the firm’s hierarchy either leads to the manager’s detachment from corporate goals or increases the likelihood of her receiving negative feedback. Put differently, motivational barriers to idea diffusion within the firm may well be exacerbated by struc- tural organizational features such as hierarchy. This finding also adds to a more refined picture of the true role of middle managers within the strategy for- mation process—a picture that other scholars have recently begun drawing (Huy, 2011; Reitzig and Sorenson, 2013), and which takes account of the particular behaviors middle managers may display because of their position within the actual corpo- rate line of command. Notably, these prior works as well as the findings we report here highlight the challenges of using middle managers to counter- balance the actions by visionary CEOs who pursue (overly) narrow strategic trajectories in their firms,

as scholars before us have suggested (Rotemberg and Saloner, 2000).

Organizational economists may find it insightful that hierarchies in which agents are connected by a line of command can alter agents’ screening func- tions in such a way that endogenous screening pro- duces even more omission errors than exogenous screening. With increasing psychological detach- ment from corporate omission errors, the rela- tive benefits of having decisions rechecked by

superiors—to the extent that they exist 25 —appear to pale in comparison to the costs of endorsing potentially bad proposals or spending time on ini- tiatives that are of (perceived) little value to the agent. Arguably, the most interesting practical con- sequence is that managers who wish to “weed out” overoptimistic decisions by their subordinates by funneling these decisions through hierarchies (Christensen and Knudsen, 2010) may want to con- sider implementing fewer hierarchical layers than prior contributions would suggest. Colleagues from psychology may consider our (formal) theoretical integration of evaluation apprehension and lack of control with a rational model of information pro- cessing insightful. Also, we hope they see value in our empirical test that originally adds to the largely conceptual literature on employee voice. Finally, given the empirical context of our field data, our findings also speak to scholars of corporate inno- vation. In their literature, there has also long been debate about how organizational structure affects the type of innovation a firm can solicit (Teece, 1996); much less attention has so far been ded- icated, however, to the effects of organizational structure on the actual number of employee ideas reaching corporate management.

As usual, many intriguing questions remain unanswered, and addressing them in future work would appear worthwhile. At this point, we conclude by touching briefly on three different categories of remaining questions, the first of which results from the imperfections of our current design.

25 In an earlier pilot study, which we ran with analytically trained engineering and science undergraduates from one of the finest

engineering colleges worldwide, we do get traction on the infor- mation economics base case; i.e., these highly formally trained students do realize the value of using superiors as rechecking devices, and they pass up more ideas the steeper the hierarchy—as per the predictions of Sah and Stiglitz (1986). However, even these subjects are easily conditioned to succumb to the mechanisms of evaluation apprehension and lack of control. More information is available from the authors upon request.

Hierarchy and Information Flow 1997 Despite best efforts, we must not exclude the pos-

sibility that omitted variables in our field data may bias our existing findings. In an ideal world, we particularly would like to control even better for an idea’s fit with global business in the first stage of our regressions. We take some comfort in the fact, however, that our specifications are largely robust to the inclusion of a variable that measures whether the idea was simultaneously exploited at the local level—serving as a crude control for an idea’s potential lack of fit for global business. Also, instead of relying on word counts to capture idea quality, we would optimally like to draw on multi-respondent idea reevaluations. While these data could not be feasibly obtained in this research project, future researchers may find ways to generate field data that contain these controls. Finally, to rule out that omission biases in the field might cause a negative net effect of corporate hierarchy on mid-level man- agers’ propensities to pass up information on their own, future researchers would ideally rerun our study and compare results across organizations that differ in their emphasis on pursuing opportunities as opposed to avoiding threats.

The second category of open questions speaks to the link between organizational structure and agent behavior. Apart from the results of this first paper gained in a particular setting, what do we know about how an employee’s actions are affected by the structural organizational environment we expose her to? Could an omission bias in the field ever be strong enough to account for a negative net effect of hierarchy on agents’ propensity to pass up informa- tion on its own? Are our operationalizations of eval- uation apprehension and lack of control optimally suited to capture what is happening in real-world organizations? And if they are, are the costs for employees in organizations similar in magnitude to our experimental test? Can the effects of hierarchy on information transfer be counteracted through incentive schemes? Do different types of managers react differently to the same environment? These and other issues appear to be critical when seri- ously thinking about designing organizations that employ human actors. Their examination, however, necessitates a type of data that may be difficult to find in the field, in turn stressing the need to devise intelligent laboratory experiments of sufficient validity to be meaningful for scholars interested in real-world organizations. The second kind of remaining questions evolve around the discussion about agent behavior, organizational structure, and

organizational performance. Clearly, increasing or reducing vertical information flow by affecting agents’ willingness to pass up data is not good or bad per se, but its optimal degree depends on its link to corporate performance. Extending prior works in this domain (Csaszar, 2012) by allowing for more complex behavior of managers in organizations appears to be an issue worth investigating.