Case Study using Metaphors

Administrative Science Quarterly 2022, Vol. 67(4)1012–1048 � The Author(s) 2022 Article reuse guidelines: sagepub.com/journals-permissions DOI: 10.1177/00018392221117996 journals.sagepub.com/home/asq

The Two Blades of the Scissors: Performance Feedback and Intrinsic Attributes in Organizational Risk Taking

Xavier Sobrepere i Profitós,1 Thomas Keil,2

and Pasi Kuusela3

Abstract

We draw on the behavioral theory of the firm and prospect theory to examine how performance feedback (decision context) and the characteristics of the alternatives (decision content) that decision makers face jointly determine orga- nizational risk-taking choices. While the behavioral theory of the firm has identi- fied performance feedback’s important role in driving organizational risk-taking decisions, it has not considered the intrinsic attributes of alternatives, specifi- cally the magnitude and likelihood of their outcomes, which have been the focus of prospect theory. We argue that these two attributes play a key role in decision makers’ assessment of alternatives, but because achieving organiza- tional goals is the prime objective in organizations, performance feedback drives how decision makers process information regarding these attributes. Analyzing 23,895 fourth-down decisions from the U.S. National Football League, we find that decision makers weigh attainment discrepancy and the magnitude and likelihood of outcomes in their choices, depending on deadline proximity. Furthermore, the size and valence of attainment discrepancy modify the weight of the magnitude and likelihood of outcomes in risky choices. Our arguments and findings suggest extensions to the behavioral theory of the firm and imply modifications to prospect theory when applied to the organizational context.

Keywords: risk taking, behavioral theory of the firm, prospect theory, aspirations, performance feedback

1 UPF Barcelona School of Management, Spain 2 University of Zurich, Switzerland 3 University of Groningen, Netherlands

Corresponding Author:

Xavier Sobrepere i Profitós, UPF Barcelona School of Management, Balmes 132, Barcelona,

Catalonia 08008, Spain. Email: xavier.sobrepere@bsm.upf.edu

How do organizational decision makers make risky choices? Using the meta- phor of the two blades of scissors, Simon (1947, 1990) suggested that decisions are shaped by the decision context and how decision makers process information regarding the decision content. Yet prior research has typically focused on only one blade of the scissors. Emphasizing the organizational deci- sion context, research drawing on the behavioral theory of the firm (Cyert and March, 1963) has focused on performance feedback relative to an organiza- tional goal (e.g., Bromiley, 1991; Greve, 1998; Lehman and Hahn, 2013) and has thus not considered the attributes of the alternatives that decision makers face: the decision content. In contrast, research on prospect theory (Kahneman and Tversky, 1979; Tversky and Kahneman, 1986, 1992) has focused on how decision makers process information regarding the decision content, specifi- cally the magnitude and likelihood of outcomes, which constitute the intrinsic attributes of an alternative, yet it has not considered how the decision context may modify information processing in organizations. It is difficult to imagine that managers would make any important decision without considering both the potential consequences of the alternatives and the performance relative to the organizational goal. To progress toward better understanding of organiza- tional risk taking, we therefore integrate and modify arguments from these two theoretical perspectives to theorize how performance feedback (decision con- text) and intrinsic attributes of alternatives (decision content) jointly affect orga- nizational risk taking.

We argue that in the organizational context, decision makers draw on infor- mation regarding both intrinsic attributes and performance feedback, but per- formance feedback drives information processing. From the organizational perspective, decision makers’ prime objective is to achieve organizational goals (Cyert and March, 1963). We expect organizational decision makers to use information that has high diagnostic value (Greve, 2003)—i.e., is informative, useful, and important—for assessing the potential to achieve the organizational goal. As decision makers do so, contextual factors may influence decision mak- ing directly and indirectly, as they modify the diagnostic value of other information.

For example, consider a team in the National Football League (NFL) facing the high-risk fourth-down decision, our empirical context. As we explain in detail below, the fourth down is a team’s last attempt to advance a total of 10 yards, which allows them to retain possession of the ball. The risky choice in the fourth down is to attempt to win these yards, i.e., to ‘‘go for it.’’ Before deciding which play to choose, the team will consider the possible outcomes of each alternative and their likelihoods, the score, and the remaining time in the game. To provide some intuition, Figure 1 depicts heatmaps of risk taking in fourth-down decisions as a function of these parameters. The four heatmaps depict the propensity to go for it on fourth downs, which captures risk-taking intensity. Each heatmap’s x-axis shows the difference in game score, which captures the attainment discrepancy; in the y-axis, the yard line position on the field captures the magnitude of outcomes because the expected points scored/received when starting a new play are a direct function of the field posi- tion (see Romer, 2006). Heatmaps in the left column capture scenarios with low likelihood of losses, and those on the right capture scenarios with high

Sobrepere i Profitós et al. 1013

likelihood of losses. Loss likelihood is captured by the remaining yards to com- plete the down, as the likelihood of succeeding in the attempt to go for it decreases as the yards left increase. Finally, the heatmaps in the top row are for the first half of the game, far from the deadline, and the heatmaps on the bottom are for the second half, close to the deadline. This figure shows strik- ingly different risk-taking patterns, and in the following sections, we explain the theory and empirically examine the decision making that leads to such differences.

Figure 1. Heatmaps of Risk Taking Based on the Raw Data

1014 Administrative Science Quarterly 67 (2022)

First, we argue that deadline proximity modifies the diagnostic value of the intrinsic attributes of alternatives and of attainment discrepancies, that is, the size and valence of the difference between organizational performance and an aspiration level regarding the organizational goal (Lant, 1992). Deadline proxim- ity captures the time to performance evaluation and is a dimension of within- period performance feedback.1 We know from prior research (Lehman et al., 2011) that decision makers weigh attainment discrepancies more strongly the closer the deadline is because closing performance gaps or avoiding jeopardiz- ing performance advantages becomes more important for reaching the organi- zational goal. In contrast, intrinsic attributes of an alternative, specifically the magnitudes and likelihoods of outcomes, which capture the extent to which the alternative is expected to make the decision maker better off (Kahneman and Tversky, 1979), have more diagnostic value when the deadline is distal, as any performance improvement is valuable at that time. This is not the case when the deadline is near. Close to the deadline, decision makers may diverge from the better-off perspective by pursuing unattractive opportunities or rejecting attractive ones if the choice allows them to possibly reach the goal. For instance, at the beginning of the game, a team will likely choose a play that improves the team’s score; close to the end of a game, the team may pursue a highly risky play that they would not have considered at the beginning of the game but may now be the only way to win the game.

Second, we argue that the size and valence of an attainment discrepancy also modify the diagnostic value of intrinsic attributes in risky choices. Because of the primacy of achieving the organizational goal, the larger the attainment discrepancy, the higher the diagnostic value of the magnitude attribute becomes. This occurs because with a large attainment discrepancy, achieving or failing to achieve the goal becomes more dependent on the magnitude of the outcome. Furthermore, because of the shape of the subjective value func- tion (Kahneman and Tversky, 1979), decision makers exhibit loss aversion and are expected to become less sensitive to changes in performance the larger the positive attainment discrepancy is, which reduces the diagnostic value of both intrinsic attributes: magnitude and likelihood of outcomes. Combining these arguments, we posit that (1) when performance is below aspirations, the effect of the magnitude of outcomes increases with the size of the attainment discrepancy; (2) when performance is close to the aspiration level, the likeli- hood of outcomes drives choices, and its effect is stronger above than below aspirations; and (3) when performance is above aspirations, the effect of the likelihood of outcomes decreases as the attainment discrepancy increases. In our NFL example, when the team is losing by many points, the team’s choice

1 Research on performance feedback has taken different approaches to temporal structure of per-

formance feedback. Some studies have focused on performance feedback across performance

periods such that performance in past periods shapes current levels of risk taking (e.g., Palmer and

Wiseman, 1999; Miller and Chen, 2004; Kacperczyk, Beckman, and Moliterno, 2015). Other studies

use forward-looking performance feedback, either by measuring performance expectations relative

to aspirations (e.g., Bromiley, 1991; Wiseman and Bromiley, 1996; Chen, 2008) or by relying on

within-period performance feedback (Lehman et al., 2011; Lehman and Hahn, 2013). In within-

period studies, performance feedback from actions during a performance period affects subsequent

behavior within the same performance period; this feedback signals whether the goal of achieving

a satisfactory performance level at the end of the period (when final performance evaluation takes

place) is at risk. Our study belongs to the within-period performance feedback stream of research.

Sobrepere i Profitós et al. 1015

will focus more on the fact that a play will have the potential to score 6–8 points, whereas close to the aspiration, the likelihood of success will matter more. When the team is winning by many points, neither factor will play a strong role.

We test and find support for these predictions by analyzing NFL teams’ risk- taking decisions. This setting has been used previously to study risk taking (e.g., Lehman et al., 2011; Lehman and Hahn, 2013; Gonzales, Mishra, and Camp, 2017; To et al., 2018) given that it provides systematic data, good measures of key constructs, and a structure of decision making resembling that of other business contexts.

Our study makes important contributions to both the behavioral theory of the firm and prospect theory. We extend the behavioral theory of the firm by incorporating the intrinsic attributes of alternatives into explanations of risk tak- ing and by explaining how these attributes, jointly with performance feedback, affect risky choices. Incorporating intrinsic attributes into the theory is impor- tant given that in real-world decisions, decision makers typically face a choice among specific alternatives, and a theory of risk-taking decisions that ignores the attributes of these alternatives would seem incomplete.

Our study also has important implications for the application of prospect theory to organizational contexts. We argue for skepticism of directly applying prospect theory arguments alongside performance feedback to explain organi- zational risk taking. Given that predictions based on these theories are often similar on the surface, prior research has at times ignored their important differences (Bromiley, 2010; Bromiley and Rau, 2022). Our results suggest, however, that because decision makers in organizations try to reach organiza- tional goals, important modifications to prospect theory arguments are needed to theorize the effects of intrinsic attributes in the organizational context. In other words, the organizational context modifies information processing regard- ing risky choices, and the behavioral theory of the firm provides the framework to explain how this occurs.

THEORY DEVELOPMENT

In his seminal analysis of decision making in organizations, Simon (1947: 241) highlighted that a theory of organizational decision making ‘‘must be concerned with the limits of rationality, and the manner in which organizations affect these limits for the person making a decision’’ (emphasis added). For our analysis of organizational risk-taking choices, this implies that our theory must clarify how decision makers process information regarding the decision content and how the organizational decision context affects this information processing. We therefore draw on two theories that scholars have commonly used to theorize the organizational context and the decision content: the behavioral theory of the firm and prospect theory.

Risky Choices in the Organizational Context

Formulated by Cyert and March in 1963, the behavioral theory of the firm has become the central theory explaining risk taking in the organizational context. In their initial specification, Cyert and March (1963) focused on explaining orga- nizational search (and, to a lesser extent, change) as an organizational response

1016 Administrative Science Quarterly 67 (2022)

to performance feedback and remained silent on risk taking. Later research in the 1980s extended these arguments to risk taking (e.g., Singh, 1986; March and Shapira, 1987; Bromiley, 1991), following the same logic originally devel- oped for search and change (Greve, 2003).

The central idea is that in the organizational context, decision makers focus on a specific organizational goal and then regulate behavior by comparing per- formance feedback with an aspiration level regarding that goal. When perfor- mance feedback deviates from the aspired level—that is, when there is an attainment discrepancy (Lant, 1992)—decision makers respond by adjusting behavior. When an organization performs below aspirations, risk taking should increase with the attainment discrepancy because to close the aspiration– performance gap, the organization needs to take risky actions (Bromiley, 1991; Greve, 1998).2 When an organization is performing above aspirations, two alternative predictions exist. Some studies argue that, above aspirations, organizations reduce risk taking with larger attainment discrepancies because the organizations perceive increasingly less need to take risky actions (Greve, 1998; Arrfelt, Wiseman, and Hult, 2013; Joseph, Klingebiel, and Wilson, 2016; Smulowitz, Rousseau, and Bromiley, 2020). Other studies argue that, above aspirations, risk taking could increase because with a larger attainment discrep- ancy, decision makers are less concerned with falling below aspirations in the event of losses and therefore relax controls (March and Shapira, 1992; Chen and Miller, 2007).

Given the focus on performance feedback, empirical studies, with rare exceptions (March and Shapira, 1987), have not considered how decision makers process information regarding the intrinsic attributes of the alternatives an organization faces. This omission is not surprising given that performance feedback theory was originally not designed to explain individual risky choices. Prior empirical research has therefore focused mostly on the question of whether to search, adjust risk levels, or change, thereby emphasizing the acti- vation and intensity of aggregate responses to performance feedback (Greve, 2018; Posen et al., 2018). Underlying this focus has also been the difficulty of observing the steps preceding risky choices (Posen et al., 2018) and the assumption that organizations have too little information about alternatives’ outcomes and their likelihood to consider them (Knudsen and Levinthal, 2007). This latter assumption may not hold in many organizational contexts. Rather, managers at least have ‘‘concrete . . . if not necessarily accurate’’ (Cyert and March, 1963: 99) estimates about the intrinsic attributes of the alternatives they consider; therefore it would appear at odds with the behavioral realism of the Carnegie tradition to assume that decision makers make no use of this information.

The Decision Content of Risky Choices

Prospect theory has focused on consideration of decision content, particularly how decision makers process information regarding the intrinsic attributes of

2 This logic mostly applies in the relative vicinity of the aspiration level, and some studies suggest

that very large shortfalls may threaten survival and lead to rigidity in behavior rather than risk taking

(e.g., Staw, Sandelands, and Dutton, 1981; March and Shapira, 1992; Audia and Greve, 2021). In

this study, we therefore focus on the vicinity of the relative aspiration level.

Sobrepere i Profitós et al. 1017

risky choices (Kahneman and Tversky, 1979; Tversky and Kahneman, 1992). The theory was originally developed to explain one-shot decisions regarding risky choices of individuals seeking to improve their performance by choosing an option if it makes them better off than alternative choices do (Kahneman and Tversky, 1979). In other words, decision makers will choose the course of action with the highest overall subjective value given the decision makers’ estimates of the magnitude and likelihood of outcomes. Thus the higher the magnitude of gains compared to losses and the higher the likelihood of a posi- tive outcome, the more prone decision makers are to choose an alternative.

In the process of assessing the subjective value of outcomes, decision makers’ estimates of magnitude and likelihood of outcomes are thought to be biased (Kahneman and Tversky, 1979). In estimating the value of outcomes, decision makers set a reference point and classify outcomes as either gains or losses depending on that reference point. They consider the value of outcomes in decreasing returns, but the decreasing returns are more pronounced above than below aspirations; the value function is more concave above aspirations than it is convex below aspirations (Kühberger, 1998; DellaVigna, 2009; Ruggeri et al., 2020). Additionally, when considering the likelihood of outcomes, deci- sion makers do not use exact likelihoods but, rather, biased and cognitively sim- plified estimates (Tversky and Kahneman, 1992; Prelec, 1998; Gonzalez and Wu, 1999). As a consequence of their biased estimates, they exhibit loss aver- sion and risk aversion in the domain of gains and risk-seeking behavior in the domain of losses (Kahneman and Tversky, 1979).

Prospect theory has also been frequently applied in the organizational con- text alongside arguments derived from the behavioral theory of the firm (e.g., Miller and Leiblein, 1996; Palmer and Wiseman, 1999; Gomez-Mejia and Wiseman, 2007; Shimizu, 2007). Research integrating both theories has often assumed that organizational aspirations provide a natural reference point and that otherwise, no other important modifications need to be considered. Yet this practice has been criticized for ignoring the fact that risky choices in the organizational context may differ in important ways from the assumptions of prospect theory (e.g., Bromiley, 2010; Bromiley and Rau, 2022). Specifically, Bromiley and Rau (2022: 125) highlighted that the ‘‘belief that the [prospect] theory leads to some relatively straightforward hypotheses regarding the relations between firm or individual performance and risk-taking . . . stems from an oversimplification or incomplete application of the core ideas in the theory.’’ Building on this critique, we argue first that research integrating arguments from prospect theory into the organizational context should consider the intrin- sic attributes of alternatives; second, we argue that while decision makers may use aspiration levels as a reference point, performance feedback may modify how they process information regarding the intrinsic attributes. We therefore focus our theory on how intrinsic attributes and performance feedback interact to jointly influence risk-taking choices.

Organizational Risk-Taking Choices: Integrating Decision Context and Decision Content

In building our theory of risk taking, we argue that in the organizational context, decision makers draw on information regarding both the intrinsic attributes of the alternatives and performance feedback. As Simon’s (1947, 1990) notion of

1018 Administrative Science Quarterly 67 (2022)

scissors suggests, the organizational context shapes risk-taking decisions (Simon, 1947; Kacperczyk, Beckman, and Moliterno, 2015) and leads to decisions that deviate in important ways from the better-off logic and more closely resemble the satisficing logic. This occurs because organizations hold goals, and achieving an aspiration level regarding these goals is the primary objective of organizational decision makers (Cyert and March, 1963; Greve, 2003). Decision makers satisfice by aiming to surpass the minimum aspired performance level regarding the goal, rather than maximizing performance (Simon, 1955).

To achieve the organizational goal, decision makers often need to make a sequence of choices (March, 1996, 2010) and relate each choice in that sequence to the overall goal of surpassing the performance aspiration, rather than viewing each choice separately from a better-off perspective. As a result, in some situations decision makers may choose an alternative that, viewed indi- vidually, they do not expect to make them better off but that could help achieve the organizational goal. For instance, when an organization is experiencing a large underperformance, its decision makers may choose an option that offers low likelihood of large gain and high likelihood of loss because it would allow them to close the gap with the aspiration level, even if the subjective expected value is negative; or the same decision makers in an organization that is overperforming by a narrow margin might refuse an alternative with a clear posi- tive subjective expected value but a small risk of losses, to ensure that the potential loss does not shift performance below aspirations. This behavior is possible because organizations absorb the cost of individual choices, and the decision maker is primarily evaluated on performance relative to the perfor- mance aspiration that arises from the cumulative effect of all choices.

Organizational contexts are characterized by the availability of performance feedback that gives decision makers information regarding attainment discrepancy (Cyert and March, 1963; Greve, 2003). Decision makers directly incorporate this information into their choices. If we view decision makers as mindful in their infor- mation processing (Levinthal and Rerup, 2006)—that is, if they can focus time, energy, and effort in a controlled manner on selected information (Ocasio, 2011)— we can expect them not only to directly incorporate information about intrinsic attributes and performance feedback but also to be strategic in their information processing and to use in their decisions information that has diagnostic value for assessing an alternative’s contribution to achieving the organizational goal.

Given the primacy of organizational goals in the organizational context, we argue that the diagnostic value of intrinsic attributes depends on the informa- tion the decision maker has about the potential to reach the organizational goal and, therefore, that the attributes’ weight in risky choices will be modified by different dimensions of performance feedback. Next, we offer hypotheses regarding this interplay of intrinsic attributes and dimensions of performance feedback in risky choices.

HYPOTHESES

Deadline Proximity and the Diagnostic Value of Intrinsic Attributes and Attainment Discrepancy

In our first set of predictions, we focus on how the temporal proximity of per- formance evaluation, also called deadline proximity, shapes the diagnostic value

Sobrepere i Profitós et al. 1019

of each intrinsic attribute and attainment discrepancy (Lehman et al., 2011). While deadline proximity has sometimes been viewed as distinct from perfor- mance feedback, we can think of the former as a dimension of within-period performance feedback. Deadline proximity is important for several reasons. Organizations typically set their goals and evaluate performance for clearly defined periods of time, i.e., performance periods (Greve, Rudi, and Walvekar, 2021), with performance evaluation events at the end of those periods. For instance, organizations may set and measure weekly sales goals, quarterly earnings goals (Chen, 2008), and yearly employee assessment or profitability goals (Mezias, Chen, and Murphy, 2002). Performance periods may play an even larger role in discrete activities such as change task forces, new product development projects (Sethi and Iqbal, 2008), funding rounds in new ventures, or sports games (Lehman et al., 2011; Greve, Rudi, and Walvekar, 2021). While decision makers tend to monitor goal achievement throughout the performance period (Eisenhardt, 1989; Sutcliffe, 1994; Lehman et al., 2011) and adjust their behavior during this period (Simon, 1947: 62; Cording, Christmann, and King, 2008; Hohnisch et al., 2016), the final evaluation occurs at the end of the period, and organizational rewards and punishments are linked to achieving goals at that point. As a result, deadline proximity should affect the weight of information regarding both attainment discrepancies and intrinsic attributes in risky choices.

Specifically, we expect the weight of intrinsic attributes in risk-taking decisions to decrease with proximity to the deadline and, following Lehman et al. (2011), we expect the weight of the attainment discrepancy to increase. In other words, deadline proximity modifies the diagnostic value of the intrinsic attributes and attainment discrepancy, but in opposite directions. Intrinsic attributes allow the decision maker to evaluate whether an alternative is making the organization better off, and early in the performance period, being better off is the best contribution to reaching the organizational goal. In contrast, early in the performance period, information on attainment discrepancy can be consid- ered noisy and therefore offering little information regarding ability to reach the goal. As the performance evaluation nears, being better off may not be enough to reach the goal; therefore the diagnostic value of intrinsic attributes is reduced. In contrast, the closer the deadline comes, the attainment discrepancy becomes more predictive of the organization’s ability to reach the organizational goal. When the end of the performance period is very close, performing below the aspiration level strongly suggests the need to take risks to achieve the goal, even if a decision may be expected to generate negative outcomes on average, and performing above the aspiration level strongly suggests avoiding risks even if a decision is expected to generate positive outcomes on average.

Relatedly, time can be viewed as a resource that decision makers can use to reach the goal (Svenson and Maule, 1993). The more the organization has of that resource, the lower the pressure (Busemeyer, 1985) to deviate from the better-off perspective and to respond to attainment discrepancies. Finally, the temporal proximity of an event influences how information related to that event will be processed (Loewenstein and Elster, 1992; Liberman and Trope, 1998; Liberman, Sagristano, and Trope, 2002). In particular, decision makers ascribe higher importance to information regarding events that will occur soon than to information regarding events in the distant future (McElroy and Mascari, 2007; Peetz, Wilson, and Strahan, 2009; Nadkarni, Pan, and Chen, 2019); therefore

1020 Administrative Science Quarterly 67 (2022)

the closer the time to the performance evaluation, attainment discrepancies have increasing weight in choices, which may justify deviating from the better- off assessment. These observations suggest the following hypotheses:

Hypothesis 0 (H0): The farther away the performance evaluation is, the weaker the effect of attainment discrepancy on risk taking.

Hypothesis 1 (H1): The farther away the performance evaluation is, the stronger the effect of the magnitude of potential losses/gains on risk taking.

Hypothesis 2 (H2): The farther away the performance evaluation is, the stronger the effect of the likelihood of adverse/positive outcomes on risk taking.

Attainment Discrepancy and the Diagnostic Value of Intrinsic Attributes

In our second set of predictions, we argue that in the organizational context, the size and valence of the attainment discrepancy also influence the diagnostic value of the intrinsic attributes. According to performance feedback theory, the aspiration level marks the threshold between success and failure (Simon, 1955) and acts as a master switch for behavioral changes (Greve, 2003): below aspirations, decision makers exhibit risk-seeking behavior to restore perfor- mance; above aspirations, they are mainly concerned with avoiding losses to secure performance. In prospect theory, the reference point marks the thresh- old between losses and gains, which also triggers similar risk-seeking and loss- avoiding behaviors (Kahneman and Tversky, 1979; Tversky and Kahneman, 1992; Barberis, 2013). Based on these similarities, prior research integrating the theories (e.g., Audia and Greve, 2006) has suggested that in the organiza- tional context, the aspiration level regarding the organizational goal may be viewed as a natural reference point for decision makers. But previous research has not considered how intrinsic attributes may be processed differently as a function of attainment discrepancy in the organizational context.

Negative attainment discrepancy. From an organizational perspective, when performance is below aspirations decision makers focus on achieving gains to amend the shortfall, and the more so the larger the attainment discrep- ancy. Thus the size of the attainment discrepancy should modify the diagnostic value of the magnitude of outcomes because with increasing attainment discrepancies, restoring performance above aspirations can be achieved only through increasingly larger performance improvements. In comparison, close to the aspiration level, almost any improvement will suffice to restore perfor- mance above aspirations. Performance improvements closing the performance gap are highly valued, but further performance improvements beyond closing the gap are considerably less relevant for organizational decision makers since they satisfice (Simon, 1947, 1955). For instance, when the organization is underperforming by two units, improvements of three versus six units are val- ued similarly, given that both are sufficient to restore performance above the aspiration level; but these same improvements of three and six units are valued differently when the organization underperforms by four or even ten units because their contributions to amending the shortfall are different. Thus the larger the negative attainment discrepancy is, the higher the diagnostic value of the magnitude of the outcome attribute and, therefore, the stronger its impact

Sobrepere i Profitós et al. 1021

on risky choices. An additional argument relates to the smart use of resources. Generally, pursuing risky alternatives requires the use of resources. Thus with larger attainment discrepancies, only alternatives with large potential gains are worth considering to avoid wasting resources, such as time (Greve, Rudi, and Walvekar, 2021). These observations suggest the following hypothesis:

Hypothesis 3 (H3): For performance below the aspiration level, the larger the attainment discrepancy is, the stronger the effect of the magnitude of outcomes on risk taking.

Performance at the aspiration level. Based on the argument above, in the vicinity of the aspiration level, the likelihood of outcomes will receive weight in risk-taking choices, whereas the magnitude of outcomes will not. Sufficiently close to the aspiration level, even small losses (gains) can shift the perfor- mance from success (failure) to failure (success), reducing the diagnostic value of the magnitude attribute, whereas the likelihood attribute maintains its impor- tance for assessing the degree to which a risky choice contributes to achieving the organizational goal. Because below aspirations decision makers focus on gains whereas above aspirations they focus on losses, and because decision makers respond more strongly to losses than to gains (Kahneman and Tversky, 1979), it follows that the effect of the likelihood of outcomes is stronger when performance is slightly above the aspiration level (and decision makers focus on the likelihood of losses) than when it is slightly below (and decision makers focus on the likelihood of gains):

Hypothesis 4 (H4): Close to the aspiration level, the effect of the likelihood of outcomes is stronger when performance is above rather than below aspirations.

Positive attainment discrepancy. As argued, when organizations are slightly overperforming, decision makers focus on avoiding losses and turn to the likelihood of losses as the key diagnostic attribute. Only alternatives with very small likelihood of potential losses will be considered, while alternatives with substantial likelihood of potential losses will be dismissed. However, with a larger positive attainment discrepancy, some losses are affordable, and thus the effect of the likelihoods of outcomes should weaken.

Above the reference point, decision makers exhibit loss aversion (Kahneman and Tversky, 1979; Tversky and Kahneman, 1992; Barberis, 2013) but less so the further performance is above the reference point (Bromiley, 2009). This behavior is based on the strongly concave value function in the domain of gains: performance improvements above the reference point are perceived as less valuable the further away performance is from the reference point (Kahneman and Tversky, 1979; Tversky and Kahneman, 1992). As a conse- quence, in the organizational context, the strong effect of the likelihood of outcomes when performance is just above the aspiration level should weaken as performance increases further.

Similarly, organizational research has suggested that when organizations per- form above aspirations, decision makers focus on avoiding losses that could jeopardize their over-performance (Wiseman and Bromiley, 1996; Miller and Chen, 2004). However, larger attainment discrepancies above aspirations cre- ate a performance buffer that ensures achievement of the goal even in the case of losses. The larger the performance buffer is, the more managers

1022 Administrative Science Quarterly 67 (2022)

become confident in their ability to maintain performance above aspirations (Xu, Zhou, and Du, 2019) and thus relax controls regarding risk taking (March and Shapira, 1992; Chen and Miller, 2007). As a consequence, the strict dis- criminatory behavior based on the likelihoods of outcomes attribute (that we predict just above aspirations) should weaken as the positive attainment dis- crepancy increases:

Hypothesis 5 (H5): For performance above the aspiration level, the larger the positive attainment discrepancy is, the weaker the effect of the likelihood on risk taking.

We do not make predictions for potential moderation of the magnitude attri- bute and positive attainment discrepancies or of the likelihood attribute and nega- tive attainment discrepancies. For the magnitude attribute, two conflicting mechanisms are at play above aspirations. On the one hand, the decreasing sen- sitivity to changes in performance above aspirations may also weaken the effect of the magnitude of outcomes. On the other hand, also above aspirations, magnitudes may hold higher diagnostic value with a larger attainment discrep- ancy because the potential to jeopardize the performance surplus becomes more dependent on the magnitude attribute. For the likelihood attribute, the prediction for the decreasing effect above aspirations is based predominantly on the flatten- ing of the value function above aspirations. Yet below aspirations the flattening of the value curve is much weaker than it is above aspirations (Kühberger, 1998; DellaVigna, 2009; Ruggeri et al., 2020). As we focus on attainment discrepancy in the relative vicinity of the aspiration level, we assume that the flattening of the value curve below aspirations does not play a significant role, and thus we do not expect a moderation effect. While we make no predictions regarding these potential additional interaction effects, we explore them in our models for robust- ness purposes. Figure 2 provides a summary of our hypotheses.

DATA AND METHODS

Research Setting

We test our hypotheses by analyzing fourth-down decisions from 2,304 regular season NFL games during the 2009–2016 seasons. In an NFL game, teams have up to four attempts, called ‘‘downs,’’ to advance a total of at least 10 yards on the field to receive a new first down and continue their attack; other- wise the right to attack shifts to the opponent. Teams face a risky choice dur- ing fourth downs, which is their last attempt to achieve these yards. On the fourth down, the conservative choice is to punt the ball so that the opposing team begins its attack from as far away as possible, while the risky choice is to go for it, that is, to attempt to win the necessary yards for a new first down. If the team goes for it and succeeds, it continues its attack. If the team fails, the opponent begins its attack from where the play ended. (For a more detailed description of the setting, see the appendix in Romer, 2006.)

The NFL offers an ideal setting for testing our predictions given the con- trolled nature of the game, a common theme in studies using sports data (Day, Gordon, and Fink, 2012; Moliterno et al., 2014; Fonti and Maoret, 2016), in which organizations regularly face risky choices. The decision to go for it involves multiple individuals with different responsibilities and positions in the hierarchy (head coach as the final decision maker, offensive coordinator,

Sobrepere i Profitós et al. 1023

quarterback, and the rest of the players). This hierarchical structure is similar to decision making in many top management teams discussing strategic decisions: the CEO may hold final decision-making power, the CFO may have a particularly strong role in advising on financial aspects of a decision, and multiple executives participate in the discussion.

The NFL setting also allows external observers to capture relevant proxies for all variables of interest in the study. In this setting, key data regarding every play are documented in detail by teams, fans, and pundits. Outside the sports

Figure 2. Hypothesized Effects of the Influence of Intrinsic Attributes on Risk Taking

H3

H3

performance evaluation far

performance evaluation near

R el

at iv

e in

fl u

en ce

o f

m ag

n it

u d

es o

f o

u tc

o m

es o

n r

is k

ta ki

n g

Attainment discrepancy

abovebelow 0

H1

Magnitudes of Outcomes

R el

at iv

e in

fl u

en ce

o f

lik el

ih o

o d

o f

lo ss

es o

n r

is k

ta ki

n g

Attainment discrepancy

abovebelow 0

H5

H5

performance evaluation far

performance evaluation near

H2 H4

Likelihoods of Outcomes

H4

1024 Administrative Science Quarterly 67 (2022)

setting, such detailed proxies are often available only to the decision maker, making it difficult for researchers to create variables that resemble the decision makers’ information. In the NFL setting, the organizational goal is to win the game, the aspiration level of each team is to score more than the opponent, and performance feedback is readily available in the form of the score (Lehman and Hahn, 2013). Furthermore, games last 60 minutes; thus deadline proximity is clearly defined, and decision alternatives are linked with relevant information on the magnitudes (yard lines) and likelihoods (yards left) of outcomes.

Sample

The initial sample size consisted of 30,526 fourth-down decisions from the 2,304 games played in the 2009–2016 regular NFL seasons. We exclude 225 fourth downs played in overtime from the sample, as very different rules apply to this situation (Lehman et al., 2011). We also exclude those fourth downs in which attainment discrepancy was very large (bottom/top 5 percent) because prior research has suggested that extremely low and high levels of attainment discrepancy may switch decision makers’ attention from the aspiration level to survival and slack focus, respectively (March and Shapira, 1987; Lehman et al., 2011; Ref and Shapira, 2017). Results are consistent when attention to survival and slack are considered (see Online Appendix Part 5). The final sample includes 23,895 fourth-down decisions.

Variables

Dependent variable. The dependent variable is Risk taking, which is a dummy variable taking a value of 1 when the team chooses to go for it on the fourth down and 0 otherwise. As in previous studies (Lehman et al., 2011; Lehman and Hahn, 2013; To et al., 2018), we treat not going for it equally, regardless of whether the team chooses to punt the ball or to attempt a field goal conversion. We do so because these alternatives to going for it depend predominantly on the position on the field (Romer, 2006; Lehman et al., 2011).3

Independent variables. We measure Attainment discrepancy as the differ- ence in points between the focal team and the opponent.4 Following previous

3 To control for the possibility that the focal team may attempt a field goal instead of simply punting

the ball, we add a control variable (see the Control variable section). We also test an alternative

(multinomial logit) model that allows for the possibility of three alternative decisions (go for it,

punting the ball, and field-goal alternative). We report the results of this alternative specification in

the Online Appendix Part 3. The results are qualitatively similar to the main results. 4 While it is reasonable in sports settings to assume that the goal of a team is always to win the

game, some teams may have additional socially defined aspirations such as winning with a margin

or avoiding losing by many points or losing to specific teams. Such socially defined aspirations

either could affect decisions separately, or the decision maker could form an integrated aspiration

that combines socially defined aspirations with the team’s aspiration of winning the game.

However, research on the behavioral theory of the firm is not very clear on how social aspirations

and historical aspirations are to be integrated (Bromiley and Harris, 2014), and how to define social

aspirations is both theoretically and empirically contested (Moliterno et al., 2014; Kuusela, Keil, and

Maula, 2017). We therefore focus on the simple aspiration of winning the game and add relative

team quality and rivalry as control variables (see the Control variable section) to capture instances

that may lead to differing social aspirations.

Sobrepere i Profitós et al. 1025

conventions (e.g., Greve, 1998; Lehman et al., 2011), we create a spline func- tion capturing Negative attainment discrepancy and Positive attainment dis- crepancy. The former is captured by the difference in points of the focal and the opposing teams when the points of the opposing team exceed those of the focal team and 0 otherwise; symmetrically, the latter is captured by the same difference in points when the points of the focal team exceed those of the opposing team and 0 otherwise. Because we use the aspiration level of performance as a natural reference point (Audia and Greve, 2006) and as a ‘‘master switch’’ (Greve, 2003: 76) that changes decision makers’ behavior, our model allows for discontinuity at the aspired level of performance by including an additional variable, Valence indicator, which takes a value of 1 when performance is above the aspiration level and 0 otherwise (Lehman and Hahn, 2013).

To capture the intrinsic attributes of alternatives, we create separate measures for the magnitude of outcomes and their likelihood.5 We do not expect decision makers to systematically calculate exact magnitudes or likelihoods of outcomes during the game but, rather, to use easily available proxies about which they can form beliefs based on experience.

To capture the magnitude of outcomes, we use the variable Yard line, which reflects the position of the attacking team on the field. When the team is in a position close to the opponent’s end zone, the magnitude of potential gains is high because scoring becomes easier on successive plays. At the same time, the magnitude of potential losses is low because in the case of failing to achieve a new first down, the opponent would start the next attack far away from the focal team’s end zone, making it difficult to score. As a result, although the team receives the same number of points when scoring, the expected value of the play differs greatly with the yard line because several additional plays may be needed to score (Romer, 2006). We center the Yard line at the middle of the field, such that value 0 represents being in the middle and values –49 and 49 represent being one yard from one’s own end zone and one yard from the opponent’s end zone, respectively.

We capture the likelihood of outcomes with Yards to go for first down, that is, the number of yards needed to complete the 10-yard advance necessary for a first down. When teams attempt to go for it on a fourth down, the fewer the yards left for a first down, the more likely they are to succeed in their attempt. We measure Yards to go for first down on a logarithmic scale because the increase in difficulty is concave rather than linear.

Moderator variables. To test our hypotheses, we rely on several modera- tor variables. For Hypotheses 0, 1, and 2, we focus on the interaction between deadline proximity and the independent variables just described. Specifically, we create a set of dummy variables for quarters 2, 3, and 4 (quarter 1 serves as a baseline), denoted Quarter 2, Quarter 3, and Quarter 4. We use quarter dummies instead of continuous time to facilitate the interpretation of the

5 This operationalization assumes that decision makers view choices as drawn from a Bernoulli dis-

tribution and is aligned with empirical research in prospect theory (Kahneman and Tversky, 1979)

and recent theoretical work on the behavioral theory of the firm (Posen and Levinthal, 2012;

Stieglitz, Knudsen, and Becker, 2016).

1026 Administrative Science Quarterly 67 (2022)

results.6 The dummy variables take the value of 1 if the play is in the respective quarter and 0 otherwise. We then create interaction terms of the dummy variables with our independent variables: Negative attainment discrepancy, Positive attainment discrepancy, Valence indicator, Yard line, and Yards to go for first down.

To test the changing effect of magnitudes and likelihoods at different sizes and valences of attainment discrepancy as predicted in Hypotheses 3, 4, and 5, we create interaction variables of the attainment discrepancy variables (Negative attainment discrepancy, Positive attainment discrepancy, and Valence indicator) with the magnitude of gains and losses (Yard line) and their likelihood (Yards to go for first down).

Control variables. We introduce several control variables into our models. First, to account for the possibility that the relative benefit of risk taking depends on other options available, we control for the Field goal alternative to capture instances when a field goal attempt could be an alternative to punting the ball, depending on the yard line. We build this variable by following Romer (2006). We first estimate the average points gained when not going for it at each Yard line and then average the value at each Yard line with its five nearest neighbors, to reduce noise in the event of few observations at a specific Yard line value. The return of not going for it is approximately 0 when Yard line is below 10; it then steeply increases to 3 as a field goal conversion becomes possible and stabilizes close to 3 when Yard line is above 35. We also include as additional controls the moderation between the Field goal alternative and the moderation variables discussed above.

Second, we control for two distinct periods of the game when the rules are slightly different. We include the variables Last minutes quarter 2 and Last minutes quarter 4, which take the value of 1 if the play is taking place during the last two minutes of quarters 2 and 4, respectively, and 0 otherwise.

Third, we control for several factors potentially influencing the team’s pro- pensity to go for it. Home-field advantage is a dummy variable taking the value of 1 when the team plays at home and 0 when the team is the visitor (Lehman et al., 2011; To et al., 2018). We also control for the Difference in team quality, operationalized as the difference in the percentage of wins between the two teams for that season. We further include the focal team’s Within-game momentum (positive and negative) and Across-game momentum (positive and negative) to control for the existence of performance momentum. Following Lehman and Hahn (2013), Positive (negative) within-game momentum starts after two consecutive instances of scoring (being scored against) in the game, and Positive (negative) across-game momentum begins after two consecutively won (lost) games. The four momentum variables are initially set to 0 and then take the number of cumulative positive (negative) points within the game for

6 In Hypotheses 0, 1, and 2, we argue for a moderating effect of temporal proximity to performance

evaluation, but we do not specify whether the change in the effect of the intrinsic attributes with

time is linear; thus we relax the linear moderation assumption by using categorical variables.

Splitting the sample into quarters helps us to discuss the effects of our variables of interest in four

meaningful parts of the performance period. We also discuss the results when we operationalize

temporal proximity to performance evaluation continuously in a set of robustness analyses. Our

results remain qualitatively similar.

Sobrepere i Profitós et al. 1027

positive (negative) Within-game momentum and the cumulative number of won (lost) games for positive (negative) Across-game momentum once momentum has started. Once the performance momentum is broken, the vari- able is set back to zero. Finally, to account for previous success records on the decision to go for it, we create the variable Memory two last attempts, which captures the percentage of success in the last two attempts of going for it (with the baseline, set at the beginning of the game, being 1).

Fourth, we control for two factors that may affect the team’s motivation to win the game and therefore risk-taking behavior. First, we create a Rivalry con- trol variable based on historical rivalry between the two teams at play identified by sports experts (To et al., 2018). We also control for Attainment discrepancy for playoff classification, which captures the across-period attainment discrep- ancy with the social aspiration of qualifying for the playoffs at the end of the season. It is measured as the difference in wins compared with the closest team currently qualifying for playoffs, and it takes the value of 0 if the team qualifies.

Fifth, we control for the characteristics of the most relevant decision makers in the decision-making process. We account for the Quarterback’s quality using his passing rating, as established by the NFL, because the quality of the quar- terback may affect the decision to go for it. We also control for the Coach’s tenure on the team because it partly captures his power and knowledge of the team, as well as for the Quarterback’s tenure on the team, following the same logic. Finally, we also include two sets of dummy variables controlling for the week of the game and the opposing team.7

Analytical Procedure

Because the dependent variable is dichotomous, we use a conditional fixed- effects logit model with team and season fixed effects to account for time- invariant unobserved heterogeneity. We test moderating hypotheses and sup- port the interpretation with a graphic presentation of marginal effects (Hoetker, 2007; Greene, 2010; Mize, 2019). We show the average marginal effects, expressed as semielasticities (dy/dx x 1/y), of the two variables measuring intrinsic attributes—Yard line and Yards to go for first down—as a function of attainment discrepancy.8 These marginal effects reflect the impact of the unit change in the intrinsic attribute on the proportional change in risk taking.9 To examine the interactions, we use contrasts (difference) in average marginal

7 In a set of separate robustness analyses that we report in the Online Appendix, we also control

for the total number of points held by the offensive team, the number of players injured during the

match, playoff eligibility, intradivision games, and years of overlap between the quarterback and

head coach. None of these additional controls had a meaningful impact on our results, and all were

therefore dropped from the main analysis presented here to facilitate interpretation. 8 Even though the models in Table 2 are based on a conditional fixed-effects logit model (Stata

xtlogit, fe), the marginal effects graphs are based on the corresponding logit model (Stata logit) with

cluster robust standard errors, where the fixed effects are inserted as a set of dummy variables.

This is done to circumvent the problem that the conditional fixed-effects logit model does not esti-

mate the intercept, which is needed to calculate the marginal effects properly (Wooldridge, 2010). 9 In calculating the marginal effects, we keep all continuous variables at their mean and all categori-

cal variables at their most common value in the data. As marginal effects are not constant in nonlin-

ear models but depend on values of explanatory variables, we test the validity of our results at

different levels of our explanatory variables. This is reported in the Online Appendix Part 1.

1028 Administrative Science Quarterly 67 (2022)

effects between the base reference level, the average marginal effect when attainment discrepancy is zero (the points labeled R in Figures 4 and 5, intro- duced below), and the attainment discrepancy at the third quintile (the points labeled M in Figures 4 and 5). To further understand interaction effects, we also test regression coefficients and the significance of their difference, using the Wald test, and display the significance levels of the interactions (Greene, 2010), which leads to the same interpretation. We report these analyses in Online Appendix Part 1.

RESULTS

Table 1 summarizes the descriptive statistics and correlations. Table 2 reports the results of the logit models using odds ratios. Model 1 in Table 2 contains the control and independent variables, without their interactions, and Model 2 adds the interaction terms to test our hypotheses. To facilitate comparison, the results for the interaction of the independent variable with deadline proximity (measured with Quarter 2, Quarter 3, and Quarter 4 dummies) are presented in four columns, although they were included simultaneously in Model 2. The first column shows the results for baseline quarter 1, and the other three columns indicate the results for quarters 2, 3, and 4, displaying the interaction term between the respective variable (row) and the quarter in question (column). The effects of the variables in Model 2 that are not interacted with deadline proximity are presented in a single mid-centered column below these four columns. Both models in Table 2 include the dummy variables for weeks and opposing teams, although to conserve space we do not report these here.

To support the interpretation of our moderating hypotheses, we first present three figures. Figure 3 presents the average marginal effects of attainment dis- crepancy in two time periods: quarter 1, when the deadline is far, and quarter 4, when the deadline is near. Figures 4 and 5 present the average marginal effects of magnitude (Yard line) and likelihood (Yards to go for first down) of outcomes, respectively, as a function of attainment discrepancies at the same time points. The shaded area shows 95 percent confidence intervals, and the labels above the x-axes show the percentile distribution of the data. Figures 4 and 5 follow the conceptual top and bottom of Figure 2, respectively, and the narrow-dashed arrows illustrate the effects expected according to the hypothe- ses. However, because the values on the y-axis in Figure 5 are always nega- tive, the size of the effect increases downward, and the dashed arrows are therefore horizontally mirrored compared to the bottom image of Figure 2. The marginal effect in Figure 5 is negative because of our measure of likelihood of outcome; an increase in Yards to go for first down corresponds to a decrease in the likelihood of gains.

Baseline Effects

Model 1 in Table 2 suggests that direct baseline effects for our independent variables are in line with our expectations and prior research. A higher potential magnitude of gains compared to potential losses (Yard line) increases risk tak- ing (p < 0.001), whereas a higher likelihood of losses (Yards left) decreases risk taking (p < 0.001). Furthermore, in line with prior research suggesting that performance shortfalls trigger risk taking, our results show a negative relation

Sobrepere i Profitós et al. 1029

between Attainment discrepancy and Risk taking for Negative attainment dis- crepancy (p < 0.001). In contrast, we do not find a statistically significant rela- tionship between Positive attainment discrepancy and Risk taking (p = 0.158), which also aligns with previous mixed findings for above-aspiration effects

Table 1. Descriptive Statistics and Correlations for Dependent, Independent, and Control

Variables*

Variable Mean S.D. Min Max 1 2 3 4 5 6 7 8 9

1 Risk taking 0.108 0.311 0 1

2 Yards to go for

first down (log)

1.726 0.871 0 3.871 –0.2965

3 Yard line –1.021 25.206 –49 49 0.1946 –0.2214

4 Negative attainment

discrepancy

–3.491 4.956 –19 0 –0.2042 –0.0292 0.0169

5 Positive attainment

discrepancy

2.385 4.037 0 16 –0.0891 –0.0169 0.0334 0.4196

6 Valence indicator 0.340 0.474 0 1 –0.1107 –0.0157 0.0312 0.5096 0.8234

7 Field goal alternative 0.723 1.185 –0.404 3 0.147 –0.1629 0.881 0.0023 0.0257 0.0212

8 Last minutes quarter 2 0.070 0.255 0 1 –0.0206 –0.0005 0.0643 –0.0368 0.0208 0.0272 0.0634

9 Last minutes quarter 4 0.051 0.221 0 1 0.2822 –0.0112 0.07 –0.0131 0.0196 0.0164 0.0577 –0.0634

10 Home-field advantage 0.491 0.500 0 1 0.0049 –0.0062 0.026 0.0884 0.0867 0.0859 0.0236 –0.0043 0.0103

11 Difference in

team quality

–0.012 0.377 –1 1 –0.0166 –0.0047 0.0278 0.1145 0.1114 0.1106 0.0213 0.0047 0.0001

12 Within-game momentum (> 0) 0.691 2.319 0 24 –0.0178 –0.0176 0.0291 0.1705 0.4268 0.3125 0.0266 0.0106 0.0342

13 Within-game momentum (< 0) –1.312 3.211 –28 0 –0.085 –0.0197 0.0222 0.4965 0.1883 0.2064 0.0097 –0.0326 –0.0393

14 Across-game momentum (> 0) 0.373 0.965 0 9 –0.0064 –0.0063 0.0018 0.0229 0.0302 0.0276 0.0046 –0.0067 0.006

15 Across-game momentum (< 0) 0.434 1.078 0 10 0.0128 0.0077 –0.0138 –0.0585 –0.0429 –0.0455 –0.0098 0.0009 –0.0117

16 Memory two last attempts 0.919 0.265 0 1 –0.0712 0.0111 –0.0149 0.1013 –0.0027 –0.0018 –0.0136 0.0037 –0.1013

17 At. discr. for playoff (<Asp) 1.235 1.740 0 9 0.0178 0.0095 –0.0333 –0.0827 –0.0744 –0.0755 –0.026 –0.008 –0.0014

18 Rivalry 0.061 0.240 0 1 0.0022 0.0044 0.0074 –0.004 0.0026 0.0006 0.0066 0.0134 0.0031

19 Quarterback’s quality 85.285 13.357 5.9 124.8 –0.0102 –0.0151 0.0539 0.1377 0.1413 0.1322 0.0518 0.0081 0.004

20 Coach’s tenure on the team 3.840 3.831 0 16 0.0052 –0.0063 0.0133 0.0499 0.0499 0.0376 0.0141 –0.0069 0.0057

21 Quarterback’s

tenure on the team

3.283 3.555 0 16 0.0032 –0.0104 0.0302 0.0669 0.0802 0.0669 0.0284 0.0008 0.0133

22 Quarter 2 0.291 0.454 0 1 –0.0713 0.0062 0.035 –0.0375 0.0169 0.0351 0.0342 0.427 –0.1486

23 Quarter 3 0.207 0.406 0 1 –0.0706 0.0172 –0.0412 –0.0829 0.0817 0.083 –0.0384 –0.1399 –0.1176

24 Quarter 4 0.252 0.434 0 1 0.2325 –0.0067 0.0638 –0.071 0.124 0.1161 0.0587 –0.1586 0.4001

Variable 10 11 12 13 14 15 16 17 18 19 20 21 22 23

11 Difference in

team quality

–0.0866

12 Within-game

momentum (> 0)

0.0211 0.0461

13 Within-game

momentum (< 0)

0.0261 0.0551 0.1213

14 Across-game

momentum (> 0)

–0.0282 0.2788 0.0055 –0.0035

15 Across-game

momentum (< 0)

0.0476 –0.2857 –0.0146 –0.0336 –0.1684

16 Memory two

last attempts

–0.0123 –0.0048 –0.0018 0.06 0.0107 0.0077

17 At. discr. for

playoff (<Asp)

0.0219 –0.4258 –0.0367 –0.0486 –0.2359 0.496 –0.0047

18 Rivalry 0.0062 –0.0041 –0.0127 –0.0003 0.0442 0.0028 0.0114 0.0051

19 Quarterback’s quality –0.0108 0.2698 0.041 0.0666 0.2241 –0.2517 0.0221 –0.3742 0.0241

20 Coach’s tenure

on the team

0.0084 0.0809 0.0195 0.0188 0.0686 –0.079 –0.0009 –0.1426 0.0083 0.1904

21 Quarterback’s tenure

on the team

–0.0105 0.1229 0.0216 0.0374 0.1026 –0.1184 0.0088 –0.2222 0.0983 0.3962 0.4345

22 Quarter 2 –0.0073 –0.0013 –0.0065 –0.0129 –0.0036 –0.0003 0.0653 0.0007 0.0076 0.0064 0.0064 0.0051

23 Quarter 3 0.0005 0.005 0.0591 –0.0806 0.0011 0.0056 –0.0563 0.008 –0.0072 –0.0169 0.002 –0.0042 –0.3277

24 Quarter 4 0.0099 0.0008 0.1048 –0.1003 –0.0036 –0.0066 –0.1674 –0.0092 –0.0016 0.0044 –0.0092 0.0005 –0.3714 –0.2938

*n = 23,895; correlations above .012 are significant at p < .05; correlations above .016 are significant at p < .01.

1030 Administrative Science Quarterly 67 (2022)

Table 2. Logistic Regression Analysis for the Likelihood of a Fourth-Down Attempt

Model 2

Model 1 Baseline ×Quarter 2 ×Quarter 3 ×Quarter 4

Negative attainment discrepancy 0.868••• 0.971 0.976 0.915•• 0.811•••

(0.006) (0.024) (0.026) (0.026) (0.022)

Positive attainment discrepancy 0.982 0.978 0.957 0.991 0.970

(0.014) (0.066) (0.067) (0.071) (0.067)

Valence indicator 0.582••• 1.028 1.277 1.271 0.484

(0.073) (0.486) (0.639) (0.692) (0.244)

Yards to go for first down (log) 0.273••• 0.127••• 1.781••• 1.467•• 3.4•••

(0.009) (0.014) (0.234) (0.22) (0.414)

Yard line 1.054••• 1.069••• 1.004 1.011 0.954•••

(0.003) (0.009) (0.011) (0.012) (0.009)

Field goal alternative 0.513••• 0.536••• 0.668• 0.645• 1.144

(0.025) (0.08) (0.125) (0.137) (0.197)

Yards to go for first down (log) × Negative attainment discrepancy

1.009

(0.008)

Yards to go for first down (log) × Positive attainment discrepancy

1.066•••

(0.019)

Yards to go for first down (log) × Valence indicator

0.424•••

(0.073)

Yard line × Negative attainment

discrepancy

0.998•••

(0.001)

Yard line × Positive attainment

discrepancy

0.999

(0.002)

Yard line × Valence indicator 1.008

(0.014)

Field goal alternative × Negative

attainment discrepancy

1.028•

(0.012)

Field goal alternative × Positive

attainment discrepancy

1.007

(0.027)

Field goal alternative × Valence

indicator

1.231

(0.306)

Last minutes quarter 2 1.089

(0.132)

1.096

(0.138)

Last minutes quarter 4 8.81•••

(0.794)

11.347•••

(1.147)

Home-field advantage 1.185••

(0.063)

1.178••

(0.066)

Difference in team quality 1.066

(0.103)

0.996

(0.101)

Within-game momentum (>0) 0.995

(0.013)

0.998

(0.014)

Within-game momentum (<0) 1.028••

(0.008)

1.01

(0.008)

Across-game momentum (> 0) 0.975

(0.032)

0.965

(0.033)

Across-game momentum (< 0) 1.027

(0.008)

1.026

(0.033)

Memory two last attempts 1.222•

(0.111)

1.296••

(0.125)

Attainment discrepancy for playoff

classification (<Aspiration)

1.021

(0.03)

1.002

(0.032)

Rivalry 1.128

(0.139)

1.169

(0.151)

Quarterback’s quality 1.006

(0.004)

1.003

(0.005)

(continued)

Sobrepere i Profitós et al. 1031

(Lant, Milliken, and Batra, 1992; Miller and Chen, 2004; Lehman and Hahn, 2013; Posen et al., 2018).

Results for Deadline Proximity, Performance Feedback, and Intrinsic Attributes

Hypothesis 0 predicts that the effect of attainment discrepancy on risk taking strengthens as the deadline approaches. Figure 3 strongly supports this for below-aspiration performance, as the marginal effect of attainment discrepancy is significantly larger during Q4 (deadline near) than Q1 (deadline far), and this difference is statistically significant (p < 0.05) over the whole range of data below aspirations, except the two data points where the 95 percent confidence intervals overlap (approximately 2 percent of the data). The marginal effect of Negative attainment discrepancy decreases as attainment discrepancy increases in quarter 4 because Risk taking is upper-censored such that it can- not be higher than 1. In contrast, above aspirations, there is no moderating effect, as Figure 3 shows, given that the null baseline effect of Positive attain- ment discrepancy remains for all quarters.

Hypotheses 1 and 2 predict that the effects of the magnitude of outcomes (captured by Yard line) and their likelihood (captured by Yards to go for first down) weaken as the deadline approaches. Figures 4 and 5 strongly support both hypotheses, showing that the marginal effects are clearly weaker during quarter 4 (deadline near) compared to quarter 1 (deadline far) on all levels of attainment discrepancy. The difference in the average marginal effects between Q1 and Q4 is also statistically significant (p < 0.001) at all data points where the 95 percent confidence intervals overlap in Figures 4 and 5.

Table 2. (continued)

Model 2

Model 1 Baseline ×Quarter 2 ×Quarter 3 ×Quarter 4

Coach’s tenure on the team 1.436+

(0.266)

1.481•

(0.295)

Quarterback’s tenure on the team 1.03

(0.024)

1.033

(0.024)

Quarter 2 0.969

(0.088)

1.153

(0.228)

Quarter 3 0.863

(0.085)

0.847

(0.199)

Quarter 4 3.472•••

(0.303)

0.818

(0.168)

Week control Included Included

Opposing team control Included Included

n 23,895 23,895

LR w2 5,526.51••• 6,617.9•••

+ p < .10; • p < .05; •• p < .01; ••• p < .001 (two-tailed tests).

1032 Administrative Science Quarterly 67 (2022)

Results for Attainment Discrepancy and Intrinsic Attributes

Hypothesis 3 predicts that the larger the size of the attainment discrepancy below aspirations, the stronger the effect of the magnitude of outcomes. Figure 4 shows that H3 is clearly supported in quarter 1 but not in quarter 4 for most levels of attainment discrepancy. The 0.020 difference in the marginal effect between MQ1b and RQ1b is statistically significant (χ2 = 13.98, p <

0.001), whereas the 0.003 difference between MQ4b and RQ4b is not (χ2 = 0.45, p < 0.501). Below aspirations, quarter 2 and quarter 3 behave similarly to quarter 1, suggesting that H3 is supported in the vast majority of the data. The non-support to H3 for large attainment discrepancy levels at quarter 4 arises because risk taking approaches 1, thereby censoring the increasing effect of Yard line. For additional details, see our analysis in Online Appendix Part 1.

Hypothesis 4 predicts that the effect of likelihoods in the neighborhood of the aspiration level is stronger above than below the aspiration level. Figure 5 shows support for H4 for both quarters 1 and 4, as the average marginal effect becomes stronger (more negative) as attainment discrepancy shifts from below to above the aspiration level. The differences in marginal effects between points RQ1b and RQ1a, as well as between RQ4b and RQ1a, are statistically signifi- cant (difference 0.83, χ2 = 13.69, p < 0.001 and 0.83, χ2 = 17.86, p < 0.001, respectively). As detailed in Online Appendix Part 1, H4 is supported through- out the data.

Hypothesis 5 predicts that the effect of the likelihood of outcomes becomes weaker the larger the size of the attainment discrepancy above the aspiration level. Figure 5 shows that H5 is strongly supported: above aspirations, the mar- ginal effect of the likelihood attribute becomes less negative when the attain- ment discrepancy increases in quarter 1 and quarter 4. The differences in marginal effects between MQ1a and RQ1a, as well as between MQ4a and RQ4a,

Figure 3. Average Marginal Effect (Semielasticity ey/dx) of Attainment Discrepancy at Q1 and

Q4

Sobrepere i Profitós et al. 1033

are 0.583 and 0.595, respectively, which are both statistically significant (χ2 = 6.00, p < 0.05 and χ2 = 7.39, p < 0.01). As detailed in Online Appendix Part 1, H5 is supported throughout the data.

We do not formalize a prediction for the moderation between attainment dis- crepancy above the aspiration level and the magnitudes of outcomes or for the moderation between attainment discrepancy below the aspiration level and the

Figure 4. Average Marginal Effect (Semielasticity ey/dx) of Yard Line (Magnitude of Outcome)

at Q1 and Q4

Figure 5. Average Marginal Effect (Semielasticity ey/dx) of Yards Left (Likelihood of Outcome)

at Q1 and Q4

1034 Administrative Science Quarterly 67 (2022)

likelihood attribute. As displayed in Figures 4 and 5, no clear effect can be con- cluded regarding these moderations.

Additional Analysis and Robustness Tests

We conduct many robustness tests, which we report in the Online Appendix. Here we focus on a specific analysis that provides further insight: continuous spec- ification of deadline proximity. In the analysis reported in Model 3 in Table 3, we replicate Model 2, replacing the variables Quarter 2, Quarter 3, and Quarter 4 with the variable Remaining time, a continuous measure that captures the remaining minutes for performance evaluation. We include both linear and quadratic terms of Remaining time to avoid imposing strictly linear effects because the main analysis suggests that the moderation is not linear. Support for all hypotheses is robust to this alternative specification, and results from Model 3 suggest that Remaining time moderation is, indeed, not linear but in decreasing returns the further the deadline is (for more detail, see Online Appendix Parts 1 and 2). In addition to test- ing the robustness of our findings to the alternative specification, we use Model 3 to illustrate effect sizes that the results above imply.10

For the effect of Yard line, we proposed that the further the deadline (H1) and the larger the attainment discrepancy are (H3), the stronger the effect is. We observe that when the deadline is 40 minutes away (far from the deadline) and the team is trailing by 14 points (experiencing a large shortfall), changing Yard line from –15 to +15 increases Risk taking 26-fold (from 1.5 to 39 per- cent); the same change in Yard line when the deadline is only 5 minutes away and the game is tied increases Risk taking only 1.7-fold (from 13 to 22 percent).

For the effect of Yards left, we proposed that the further the deadline is (H2), the stronger the effect is; that the effect is strongest with an Attainment discrepancy just above aspirations (H4); and that it weakens as Attainment dis- crepancy increases above aspirations (H5). We observe that when the deadline is 40 minutes away and the team is ahead by only 1 point, changing Yards left from 5 to 1 increases Risk taking 30-fold (from 1.4 to 42 percent); the same change in Yards left when the deadline is only 5 minutes away and the team is ahead by 14 points increases Risk taking only 3-fold (from 5 to 15 percent).

In addition, for the effect of Attainment discrepancy we predicted that the closer the deadline is, the stronger the effect is (H0). We observe that changing Attainment discrepancy from 0 to –14 increases Risk taking 5.4-fold (from 17 to 92 percent) when the deadline is 5 minutes away, while the effect is only 1.3- fold (from 7 to 9 percent) when the deadline is still 40 minutes away.

DISCUSSION

In this study, we set out to explain organizational risk taking, considering the two blades of Simon’s (1990) scissors: the decision context and how decision makers process information regarding decision content. We show that deadline

10 The effect sizes are based on predictive margins, which requires the model’s intercept to be esti-

mated. As conditional maximum likelihood logit models do not estimate the intercept (Wooldridge,

2010), we use the corresponding unconditional maximum likelihood logit model, following the same

logic as with the figures presented earlier in the paper. We set Yards left at 3 for the Yard line and

Attainment discrepancy examples and Yard line at 0 for the Yards left and Attainment discrepancy

examples. All other variables are set at mean or at their most frequent value for the case of categor-

ical variables.

Sobrepere i Profitós et al. 1035

Table 3. Logistic Regression Analysis for the Likelihood of Fourth-Down Attempt (Proximity to

Performance Evaluation in Linear and Quadratic Form)

Model 3

Baseline ×Remaining Time ×Remaining Time^2

Negative attainment discrepancy 0.672•••

(0.018)

1.02•••

(0.002)

0.9998•••

(0.00003)

Positive attainment discrepancy 1.044

(0.042)

0.9896••

(0.003)

1.002••

(0.00007)

Valence indicator 0.196•••

(0.077)

1.143•••

(0.034)

0.998•••

(0.001)

Yards to go for first down (log) 0.624•••

(0.055)

0.926•••

(0.007)

1.001•••

(0.0001)

Yard line 0.997

(0.007)

1.005•••

(0.001)

0.9999•••

(0.0001)

Field goal alternative 0.74•

(0.098)

0.938•••

(0.011)

1.001•••

(0.002)

Yards to go for first down (log) × Negative attainment discrepancy 1.013

(0.009)

Yards to go for first down (log) × Positive attainment discrepancy 1.044•

(0.019)

Yards to go for first down (log) × Valence indicator 0.583••

(0.1)

Yard line × Negative attainment discrepancy 0.997•••

(0.001)

Yard line × Positive attainment discrepancy 0.9998

(0.002)

Yard line × Valence indicator 0.999

(0.014)

Field goal alternative × Negative attainment discrepancy 1.033•

(0.013)

Field goal alternative × Positive attainment discrepancy 1.008

(0.027)

Field goal alternative × Valence indicator 1.372

(0.338)

Last minutes quarter 2 1.296•

(0.164)

Last minutes quarter 4 4.63•••

(0.592)

Home-field advantage 1.176••

(0.067)

Difference in team quality 1.073

(0.1)

Within-game momentum (>0) 0.992

(0.013)

Within-game momentum (<0) 1.005

(0.009)

Across-game momentum (> 0) 0.949

(0.031)

Across-game momentum (< 0) 1.021

(0.031)

Memory two last attempts 1.243•

(0.123)

Attainment discrepancy for playoff classification (<Aspiration) 1.002

(0.024)

Rivalry 1.145

(0.151)

(continued)

1036 Administrative Science Quarterly 67 (2022)

proximity modifies the diagnostic value and therefore the weight of an attain- ment discrepancy regarding the organizational goal and of the intrinsic attributes of alternatives in risky choices. We further show that information processing regarding the intrinsic attributes also depends on the size and valence of the attainment discrepancy. These findings have important implications for research on organizational risk taking, extend the behavioral theory of the firm, and suggest modifications to prospect theory when applied in the organizational context.

Implications for the Behavioral Theory of the Firm

In its original formulation, the behavioral theory of the firm focused on search and change as the main organizational responses to performance feedback, and later research extended this theory to organizational risk taking (Singh, 1986; March and Shapira, 1987; Bromiley, 1991; March and Shapira, 1992; Wiseman and Bromiley, 1996). This research has mostly stayed true to the core idea that attainment discrepancies regarding an organizational goal drive organizational responses. While this focus on performance feedback as the sole driver of organizational response has provided a useful simplification for explaining aggregate risk taking, it is insufficient for explaining specific choices. Our study extends the theory by incorporating the intrinsic attributes of alternatives and explaining how these attributes, jointly with performance feed- back, affect risky choices.

Our core argument has been that decision makers use intrinsic attributes in their risky choices based on the attributes’ diagnostic value for assessing the potential to achieve an organizational goal, which is shaped by performance feedback. In particular, we focused on the moderating effects of deadline prox- imity and the size and valence of attainment discrepancy. But other

Table 3. (continued)

Model 3

Baseline ×Remaining Time ×Remaining Time^2

Quarterback’s quality 1.002

(0.003)

Coach’s tenure on the team 1.003

(0.011)

Quarterback’s tenure on the team 1.007

(0.013)

Remaining time 1.036•

(0.016)

Remaining time^2 0.999•

(0.0003)

Week control Included

Opposing team control Included

n 23,895

LR w2 7,398.07•••

+ p < .10; •p < .05; ••p < .01. •••p < .001 (two-tailed tests).

Sobrepere i Profitós et al. 1037

characteristics of performance feedback may also affect the diagnostic value of intrinsic attributes. For instance, we observe that organizations do not clearly react to performance feedback that is ambiguous because of inconsistent feed- back (Joseph and Gaba, 2015), when multiple goals exist (Audia and Greve, 2021; Levinthal and Rerup, 2021), when decision makers are prone to self- enhancement (Jordan and Audia, 2012), or when the feedback is highly noisy or may be systematically distorted (Fang, Kim, and Milliken, 2014). In those instances, and parallel to our findings regarding the moderating role of deadline proximity, we might expect decision makers to rely more on the intrinsic attributes of the alternatives they encounter when performance feedback does not trigger organizational reactions. As our study shows that organizational decision making depends on decision makers’ perception of the diagnostic value of information to assess its contribution to achieving the organizational goal, future research should strive to identify additional factors that underlie decision makers’ perception of such diagnostic value.

Integrating intrinsic attributes of alternatives with organizational performance feedback has broader implications for performance feedback theory’s applica- tion beyond risk taking. Because of the focus on aggregate responses, problems involving choice among a limited number of alternatives have typically not been theorized by employing performance feedback theory (for notable exceptions see, for instance, Greve, 1998; Kuusela, Keil, and Maula, 2017). For example, when organizations choose among different entry modes, alternative technologies, or different organizational forms, performance feedback alone is insufficient to explain the choice. Our findings suggest that we can extend per- formance feedback theory to these choice problems by incorporating some attributes of the alternatives that organizations face. Future research should therefore identify additional attributes that allow our arguments to extend to a broader set of choice problems.

Finally, our results indicate a decision maker who is informationally far more sophisticated and strategic than the behavioral theory of the firm has previously considered. Previous research has perhaps taken an extreme view of bounded rationality and has therefore underplayed decision makers’ capacity to consider the information most useful and important for making choices regarding risky alternatives and to flexibly process this information. Future research needs to consider decision makers who are smart and strategic in their information use within the boundaries of their cognitive abilities and biases.

Implications for Prospect Theory

Previous research applying arguments from prospect theory in the organiza- tional context has often done so in conjunction with theorizing about perfor- mance feedback. Yet this research has typically ignored differences between the two theories and potential boundary conditions to their applicability, proba- bly because arguments in both theories appear similar on the surface (Bromiley, 2010; Kacperczyk, Beckman, and Moliterno, 2015; Bromiley and Rau, 2022). Our arguments and results suggest that caution is warranted in combining these two theories given their important differences in focus and assumptions. While prospect theory has proven robust in the context for which it was originally developed, applications in the organizational context require important modifications.

1038 Administrative Science Quarterly 67 (2022)

Specifically, prospect theory assumes that a decision maker seeks to maxi- mize outcomes based on their biased assessments of the alternatives. In con- trast, in the organizational context, decision makers aim to achieve an aspired level of performance—that is, they satisfice (Simon, 1947, 1955). Prospect the- ory further assumes that decisions are evaluated separately and are made from a strict better-off perspective. These assumptions typically do not hold in organizations. Rather, decision makers in organizations may care more about whether a decision contributes to reaching the organizational goal when com- bined with other decisions than about whether they expect each decision to make the decision maker better off. Furthermore, unlike most individual deci- sion makers, for organizational decision makers, the organization tends to bear the cost of each decision. Organizational decision makers’ personal perfor- mance evaluations are based mainly on achieving the organizational goal at the end of the performance period, not on the results of individual choices. As a result of these important differences, decision makers weigh information as a function of its diagnostic value for assessing the extent to which taking each specific alternative may contribute to achieving the organizational goal; there- fore decision makers process information regarding intrinsic attributes condi- tionally upon performance feedback.

By identifying important modifications in the organizational context, our research adds to studies that have identified microcontextual factors, such as presentation format or target of the task, that can modify the effect theorized by prospect theory (e.g., Levin and Chapman, 1990; Takemura, 1994; Wang, Simons, and Brédart, 2001; Imas, 2016). While we acknowledge the useful- ness and importance of the theory to inform organizational choices and thus incorporate it into our theory, our arguments suggest that the organizational context modifies information processing regarding the intrinsic attributes in important ways. When scholars apply prospect theory to the organizational con- text without taking these modifications into account, the resulting predictions may be misleading.

Despite our emphasis on modifying the theory for the organizational context, our arguments nonetheless align in spirit with the logic of deviations from expected utility that is central to prospect theory. By accounting for decision makers’ cognitive biases in interpreting the intrinsic attributes of risky choices, prospect theory’s key insight has been that a decision maker will not use the expected utility (i.e., the probability-weighted mean of the magnitude of outcomes) of an alternative to make decisions but, rather, will deviate from this logic—for instance, depending on whether the outcome is in the domain of gains or losses or whether likelihoods are very small or very large (Kahneman and Tversky, 1979; Tversky and Kahneman, 1992; Barberis, 2013; Ruggeri et al., 2020). Our arguments suggest further deviations from expected utility in the organizational context.

Practical Implications

Our results suggest that deadline proximity reduces the diagnostic value of intrinsic attributes because reacting to attainment discrepancy becomes urgent and thus justifies deviating from better-off assessments. In our context, tempo- ral proximity of performance evaluation was exogenous, but in many organiza- tional contexts it may be under the control of the organizational designer. This

Sobrepere i Profitós et al. 1039

implies that an organizational designer could guide choices toward organiza- tional goals by changing the length of performance periods. For instance, an organization may move from yearly to quarterly performance periods or increase the number of evaluation steps in the product development process to tie choices more strongly to performance feedback rather than to intrinsic attributes. In contrast, by lengthening the performance period, organizational designers may shift the focus in risky choices from the choice’s impact on attaining the organizational goal toward whether it makes the organization bet- ter off. When performance feedback is highly noisy or otherwise distorted, such behavior may be advisable.

We also showed that decision makers’ focus on achieving the organizational goal may lead them to diverge from the better-off logic as a function of attain- ment discrepancy. Such behavior may not be desirable in all circumstances. For instance, if a change in the market environment makes an aspiration no lon- ger attainable, a large negative attainment discrepancy may lead to risk-taking behavior that is likely to make the organization worse off if decision makers choose high-risk alternatives to close an attainment discrepancy that can no longer be closed. Additionally, due to the concavity of the value function above the reference point, large positive attainment discrepancies may lead to behav- ior that undervalues changes in current performance. While in our context the performance aspiration was exogenous and not modifiable, in many contexts organizational designers may address this by adjusting the performance aspira- tion when such behavior is not desired.11

Boundary Conditions, Limitations, and Future Research

A theoretical boundary condition of our study is that we focused on explaining risk taking as an organizational response to performance feedback. Organizations also respond through search and change, and future research should explore the implications of our theory for these responses. For instance, performance feedback theory suggests that organizations stop searching once they identify a solution that restores performance above aspirations (Cyert and March, 1963; Posen et al., 2018), but we have proposed that such satisficing behavior is contingent on the diagnostic value of the attainment discrepancy. Thus an implication of our theory to search behavior may be that when attain- ment discrepancy holds low diagnostic value, decision makers might not stop search behavior but may explore additional solutions, aiming to improve current performance. Similarly, given the importance of the intrinsic attributes of alternatives that organizations face, organizations might be motivated to change if they encounter very attractive alternatives, even in cases when per- formance is already above aspirations.

11 While we kept the practical implications generic for organizations, our findings also have specific

implications for NFL members. For instance, if the offensive team is on third down with still 10

yards to go and winning the game by a narrow margin, the defensive coordinator can safely assume

that anything resulting in more than 3 yards left will be sufficient for the offensive team to not go

for it on the fourth down. In that case, the defensive coordinator could frame the defense strategy

to make sure the offensive team is unable to advance more than 7 yards (for instance provoking a

run decision or even a short pass) rather than just defending aggressively, making the possibility of

advancing more than 7 yards more likely.

1040 Administrative Science Quarterly 67 (2022)

Instrumental for our study’s theoretical development has been the combina- tion of a limited number of choice alternatives and a largely binary goal (whether to win or not). These boundary conditions of our theory should each be relaxed to test the generalizability of our arguments to a broader set of contexts. One relevant research opportunity relates to the question of how decision makers will process information regarding the intrinsic attributes of alternatives when the number of choice alternatives increases. Given decision makers’ bounded rationality, we may expect that information regarding the alternatives is being processed differently from the processes discussed in this paper when a very large number of alternatives exists. Another relevant research opportunity relates to relaxing the binary goal. In our study we assume that winning by a large margin is not important for decision makers. In other contexts, the level of under- or over-performance may have higher impor- tance for decision makers, such as when performance incentives are tied to the degree of under- or over-performance.

Scope limitations also exist in our study that provide opportunities for further research. We chose to focus on the first-order moderating effects of deadline proximity and the size and valence of attainment discrepancy. One may expect additional higher-order effects of combining these attributes, and while a sys- tematic theoretical and empirical treatment of these effects was outside the scope of this study, our ex-post explorations suggest that such effects are likely to exist. Future research should extend our study to these higher-order effects. In particular, the three-way interaction between attainment discrep- ancy, the intrinsic attributes, and deadline proximity is of theoretical interest: are the interactions between attainment discrepancy and intrinsic attributes enhanced by deadline proximity, due to higher diagnostic value of attainment discrepancy, or is such an effect counterbalanced by intrinsic attributes losing diagnostic value with time? Additional higher-order effects could relate to the interaction between the two intrinsic attributes conditional on attainment dis- crepancy or on deadline proximity.

Finally, the NFL context of our study has specific empirical limitations. First, in the decision we examine, baseline risk taking is low (the organization chooses to go for it in only approximately 10 percent of cases), which may affect the nature of information processing. Future research could investigate settings in which risk taking is the norm and decision makers may have the capacity to be more selective about alternatives and might therefore engage in different types of information processing. Second, decisions during NFL games are highly emotionally charged, and affective responses may influence decision making. For instance, the position in the field, particularly when the attacking team approaches the end zone, may create excitement, and such emotions may also affect how decision makers respond to intrinsic attributes. Similarly, several consecutive, successful plays may create positive emotions that could affect how decision makers respond to the magnitude of the potential reward.

More generally, while the NFL context provided a highly suitable context for testing our theoretical arguments, future research would need to test and extend our arguments to other industries. Such research may begin with other sports industries that provide a context with similarly clear rules. But to exam- ine the limits to generalizability of our findings, other industries should also be investigated, despite the difficulty of deriving similarly clear measures.

Sobrepere i Profitós et al. 1041

Conclusion

Theories of how performance feedback shapes organizational risk-taking decisions and how decision makers draw on intrinsic attributes when making risky choices have developed largely in separation. Drawing on Simon’s (1947, 1990) notion of scissors, our arguments and results suggest that such a single- sided approach is inadequate and may lead to misleading or at least incomplete predictions. In organizational risk-taking choices, decision makers attempt to make decisions that are ‘‘organizationally’’ rational, that is, decisions that are ‘‘oriented to the organization’s goals’’ (Simon, 1947: 85). In doing so, they con- sider information regarding the intrinsic attributes of choices; yet how they pro- cess information regarding these attributes is conditional on performance feedback and subject to biases. We hope that future research will build on our arguments, which are based on Simon’s original view, to develop a richer behavioral theory of organizational decision making by accounting for both the organizational context and the cognitive approach of decision makers.

Acknowledgments

This article would not have been what it is without the thoughtful guidance and insights of the editor Henrich Greve. We would also like to acknowledge the insightful comments by three excellent reviewers. The project started during the first author’s dis- sertation at IESE Business School, and the first author would like to thank the participants in IESE’s brown bag seminars, the members of his dissertation committee, and particularly his advisor, Africa Ariño, as well as Nadim Elayan, who provided detailed feedback during the data collection. We also would like to acknowledge the thoughtful comments we received on earlier versions of the paper from Pere Arqué-Castells, Pino Audia, Dovev Lavie, Johannes Müller-Trede, and Ohad Ref. Earlier versions of the paper were presented at seminars at Bocconi and ESSEC as well as at various conferences, and we thank the participants for their comments and feedback.

ORCID iDs

Xavier Sobrepere i Profitós https://orcid.org/0000-0003-2203-1118 Thomas Keil https://orcid.org/0000-0001-6124-0655 Pasi Kuusela https://orcid.org/0000-0003-0254-598X

Supplementary Material

Find the Online Appendix at https://journals.sagepub.com/doi/full/10.1177/0001839222 1117996#supplementary-materials.

REFERENCES

Arrfelt, M., R. M. Wiseman, and G. T. M. Hult 2013 ‘‘Looking backward instead of forward: Aspiration-driven influences on the effi- ciency of the capital allocation process.’’ Academy of Management Journal, 56:

1081–1103. Audia, P. G., and H. R. Greve

2006 ‘‘Less likely to fail: Low performance, firm size, and factory expansion in the shipbuilding industry.’’ Management Science, 52: 83–94.

Audia, P. G., and H. R. Greve 2021 Organizational Learning from Performance Feedback: A Behavioral Perspective

on Multiple Goals. Cambridge, UK: Cambridge University Press.

1042 Administrative Science Quarterly 67 (2022)

Barberis, N. C. 2013 ‘‘Thirty years of prospect theory in economics: A review and assessment.’’ Journal of Economic Perspectives, 27: 173–196.

Bromiley, P. 1991 ‘‘Testing a causal model of corporate risk taking and performance.’’ Academy

of Management Journal, 34: 37–59. Bromiley, P.

2009 ‘‘A prospect theory model of resource allocation.’’ Decision Analysis, 6:

124–138. Bromiley, P.

2010 ‘‘Looking at prospect theory.’’ Strategic Management Journal, 31: 1357–1370. Bromiley, P., and J. Harris

2014 ‘‘A comparison of alternative measures of organizational aspirations.’’ Strategic

Management Journal, 35: 338–357. Bromiley, P., and D. Rau

2022 ‘‘Some problems in using prospect theory to explain strategic management issues.’’ Academy of Management Perspectives, 36: 125–141.

Busemeyer, J. R. 1985 ‘‘Decision making under uncertainty: A comparison of simple scalability, fixed-

sample, and sequential-sampling models.’’ Journal of Experimental Psychology: Learning, Memory, and Cognition, 11: 538–564.

Chen, W.-R. 2008 ‘‘Determinants of firms’ backward- and forward-looking R&D search behavior.’’

Organization Science, 19: 609–622. Chen, W.-R., and K. D. Miller

2007 ‘‘Situational and institutional determinants of firms’ R&D search intensity.’’ Strategic Management Journal, 28: 369–381.

Cording, M., P. Christmann, and D. R. King 2008 ‘‘Reducing causal ambiguity in acquisition integration: Intermediate goals as

mediators of integration decisions and acquisition performance.’’ Academy of Management Journal, 51: 744–767.

Cyert, R. M., and J. G. March 1963 A Behavioral Theory of the Firm. Englewood Cliffs, NJ: Prentice Hall.

Day, D. V., S. Gordon, and C. Fink 2012 ‘‘The sporting life: Exploring organizations through the lens of sport.’’ Academy of Management Annals, 6: 397–433.

DellaVigna, S. 2009 ‘‘Psychology and economics: Evidence from the field.’’ Journal of Economic

Literature, 47: 315–372. Eisenhardt, K. M.

1989 ‘‘Making fast strategic decisions in high-velocity environments.’’ Academy of

Management Journal, 32: 543–576. Fang, C., J.-Y. Kim, and F. J. Milliken

2014 ‘‘When bad news is sugarcoated: Information distortion, organizational search and the behavioral theory of the firm.’’ Strategic Management Journal, 35:

1186–1201. Fonti, F., and M. Maoret

2016 ‘‘The direct and indirect effects of core and peripheral social capital on organiza- tional performance.’’ Strategic Management Journal, 37: 1765–1786.

Gomez-Mejia, L. R., and R. M. Wiseman 2007 ‘‘Does agency theory have universal relevance? A reply to Lubatkin, Lane,

Collin, and Very.’’ Journal of Organizational Behavior, 28: 81–88.

Sobrepere i Profitós et al. 1043

Gonzales, J., S. Mishra, and R. D. Camp 2017 ‘‘For the win: Risk-sensitive decision-making in teams.’’ Journal of Behavioral Decision Making, 30: 462–472.

Gonzalez, R., and G. Wu 1999 ‘‘On the shape of the probability weighting function.’’ Cognitive Psychology, 38:

129–166. Greene, W.

2010 ‘‘Testing hypotheses about interaction terms in nonlinear models.’’ Economics

Letters, 107: 291–296. Greve, H. R.

1998 ‘‘Performance, aspirations, and risky organizational change.’’ Administrative Science Quarterly, 43: 58–86.

Greve, H. R. 2003 Organizational Learning from Performance Feedback: A Behavioral Perspective

on Innovation and Change. Cambridge UK: Cambridge University Press. Greve, H. R.

2018 ‘‘Where to search?’’ In M. Augier, C. Fang, and V. P. Rindova (eds.), Behavioral

Strategy in Perspective, vol. 39: 91–100. Bingley, UK: Emerald Publishing. Greve, H. R., N. Rudi, and A. Walvekar

2021 ‘‘Rational fouls? Loss aversion on organizational and individual goals influence decision quality.’’ Organization Studies, 42: 1031–1051.

Hoetker, G. 2007 ‘‘The use of logit and probit models in strategic management research: Critical

issues.’’ Strategic Management Journal, 28: 331–343. Hohnisch, M., S. Pittnauer, R. Selten, and A. Pfingsten

2016 ‘‘Designing for deliberative goal-based decision making in environments with rare adverse events—An experimental study.’’ Organization Science, 27: 1417–1434.

Imas, A. 2016 ‘‘The realization effect: Risk-taking after realized versus paper losses.’’

American Economic Review, 106: 2086–2109. Jordan, A. H., and P. G. Audia

2012 ‘‘Self-enhancement and learning from performance feedback.’’ Academy of

Management Review, 37: 211–231. Joseph, J., and V. Gaba

2015 ‘‘The fog of feedback: Ambiguity and firm responses to multiple aspiration levels.’’ Strategic Management Journal, 36: 1960–1978.

Joseph, J., R. Klingebiel, and A. J. Wilson 2016 ‘‘Organizational structure and performance feedback: Centralization, aspirations,

and termination decisions.’’ Organization Science, 27: 1065–1083. Kacperczyk, A., C. M. Beckman, and T. P. Moliterno

2015 ‘‘Disentangling risk and change: Internal and external social comparison in the

mutual fund industry.’’ Administrative Science Quarterly, 60: 228–262. Kahneman, D., and A. Tversky

1979 ‘‘Prospect theory: An analysis of decision under risk.’’ Econometrica, 47: 263–291.

Knudsen, T., and D. A. Levinthal 2007 ‘‘Two faces of search: Alternative generation and alternative evaluation.’’

Organization Science, 18: 39–54. Kühberger, A.

1998 ‘‘The influence of framing on risky decisions: A meta-analysis.’’ Organizational

Behavior and Human Decision Processes, 75: 23–55.

1044 Administrative Science Quarterly 67 (2022)

Kuusela, P., T. Keil, and M. Maula 2017 ‘‘Driven by aspirations, but in what direction? Performance shortfalls, slack

resources and resource-consuming vs. resource-freeing organizational change.’’ Stra- tegic Management Journal, 38: 1101–1120.

Lant, T. K. 1992 ‘‘Aspiration level adaptation: An empirical exploration.’’ Management Science, 38: 623–644.

Lant, T. K., F. J. Milliken, and B. Batra 1992 ‘‘The role of managerial learning and interpretation in strategic persistence and reorientation: An empirical exploration.’’ Strategic Management Journal, 13: 585–608.

Lehman, D. W., and J. Hahn 2013 ‘‘Momentum and organizational risk taking: Evidence from the National Football League.’’ Management Science, 59: 852–868.

Lehman, D. W., J. Hahn, R. Ramanujam, and B. J. Alge 2011 ‘‘The dynamics of the performance–risk relationship within a performance period: The moderating role of deadline proximity.’’ Organization Science, 22:

1613–1630. Levin, I. P., and D. P. Chapman

1990 ‘‘Risk taking, frame of reference, and characterization of victim groups in AIDS

treatment decisions.’’ Journal of Experimental Social Psychology, 26: 421–434. Levinthal, D., and C. Rerup

2006 ‘‘Crossing an apparent chasm: Bridging mindful and less-mindful perspectives

on organizational learning.’’ Organization Science, 17: 502–513. Levinthal, D., and C. Rerup

2021 ‘‘The plural of goal: Learning in a world of ambiguity.’’ Organization Science, 32:

527–543. Liberman, N., M. D. Sagristano, and Y. Trope

2002 ‘‘The effect of temporal distance on level of mental construal.’’ Journal of

Experimental Social Psychology, 38: 523–534. Liberman, N., and Y. Trope

1998 ‘‘The role of feasibility and desirability considerations in near and distant future

decisions: A test of temporal construal theory.’’ Journal of Personality and Social Psychology, 75: 5–18.

Loewenstein, G., and J. Elster 1992 Choice Over Time. New York: Russell Sage Foundation.

March, J. G. 1996 ‘‘Learning to be risk averse.’’ Psychological Review, 103: 309–319.

March, J. G. 2010 The Ambiguities of Experience. Ithaca, NY: Cornell University Press.

March, J. G., and Z. Shapira 1987 ‘‘Managerial perspectives on risk and risk taking.’’ Management Science, 33:

1404–1418. March, J. G., and Z. Shapira

1992 ‘‘Variable risk preferences and the focus of attention.’’ Psychological Review,

99: 172–183. McElroy, T., and D. Mascari

2007 ‘‘When is it going to happen? How temporal distance influences processing for

risky-choice framing tasks.’’ Social Cognition, 25: 495–517. Mezias, S. J., Y.-R. Chen, and P. R. Murphy

2002 ‘‘Aspiration-level adaptation in an American financial services organization: A

field study.’’ Management Science, 48: 1285–1300. Miller, K. D., and W.-R. Chen

2004 ‘‘Variable organizational risk preferences: Tests of the March–Shapira model.’’

Academy of Management Journal, 47: 105–115.

Sobrepere i Profitós et al. 1045

Miller, K. D., and M. J. Leiblein 1996 ‘‘Corporate risk–return relations: Returns variability versus downside risk.’’ Academy of Management Journal, 39: 91–122.

Mize, T. D. 2019 ‘‘Best practices for estimating, interpreting, and presenting nonlinear interaction

effects.’’ Sociological Science, 6: 81–117. Moliterno, T. P., N. Beck, C. M. Beckman, and M. Meyer

2014 ‘‘Knowing your place: Social performance feedback in good times & bad times.’’

Organization Science, 25: 1684–1702. Nadkarni, S., L. Pan, and T. Chen

2019 ‘‘Only timeline will tell: Temporal framing of competitive announcements and rivals’ responses.’’ Academy of Management Journal, 62: 117–143.

Ocasio, W. 2011 ‘‘Attention to attention.’’ Organization Science, 22: 1286–1296.

Palmer, D., and R. M. Wiseman 1999 ‘‘Decoupling risk taking from income stream uncertainty: A holistic model of risk.’’ Strategic Management Journal, 20: 1037–1062.

Peetz, J., A. E. Wilson, and E. J. Strahan 2009 ‘‘So far away: The role of subjective temporal distance to future goals in motiva-

tion and behavior.’’ Social Cognition, 27: 475–495. Posen, H. E., T. Keil, S. Kim, and F. Meissner

2018 ‘‘Renewing research on problemistic search—A review and research agenda.’’ Academy of Management Annals, 12: 208–251.

Posen, H. E., and D. A. Levinthal 2012 ‘‘Chasing a moving target: Exploitation and exploration in dynamic

environments.’’ Management Science, 58: 587–601. Prelec, D.

1998 ‘‘The probability weighting function.’’ Econometrica, 66: 497–527. Ref, O., and Z. Shapira

2017 ‘‘Entering new markets: The effect of performance feedback near aspiration

and well below and above it.’’ Strategic Management Journal, 38: 1416–1434. Romer, D.

2006 ‘‘Do firms maximize? Evidence from professional football.’’ Journal of Political

Economy, 114: 340–365. Ruggeri, K., S. Alı́, M. L. Berge, G. Bertoldo, L. D. Bjørndal, A. Cortijos-Bernabeu,

and C. Davison et al. 2020 ‘‘Replicating patterns of prospect theory for decision under risk.’’ Nature Human

Behaviour, 4: 622–633. Sethi, R., and Z. Iqbal

2008 ‘‘Stage-gate controls, learning failure, and adverse effect on novel new products.’’ Journal of Marketing, 72: 118–134.

Shimizu, K. 2007 ‘‘Prospect theory, behavioral theory, and the threat–rigidity thesis: Combinative

effects on organizational decisions to divest formerly acquired units.’’ Academy of Management Journal, 50: 1495–1514.

Simon, H. A. 1947 Administrative Behavior: A Study of Decision-Making Processes in Administra-

tive Organization, 4th ed. (1997). New York: Free Press. Simon, H. A.

1955 ‘‘A behavioral model of rational choice.’’ Quarterly Journal of Economics, 69:

99–118. Simon, H. A.

1990 ‘‘Invariants of human behavior.’’ Annual Review of Psychology, 41: 1–20.

1046 Administrative Science Quarterly 67 (2022)

Singh, J. V. 1986 ‘‘Performance, slack, and risk taking in organizational decision making.’’ Academy of Management Journal, 29: 562–585.

Smulowitz, S. J., H. E. Rousseau, and P. Bromiley 2020 ‘‘The behavioral theory of the (community-oriented) firm: The differing response of community-oriented firms to performance relative to aspirations.’’ Strategic Management Journal, 41: 1023–1053.

Staw, B. M., L. E. Sandelands, and J. E. Dutton 1981 ‘‘Threat rigidity effects in organizational behavior: A multilevel analysis.’’ Administrative Science Quarterly, 26: 501–524.

Stieglitz, N., T. Knudsen, and M. C. Becker 2016 ‘‘Adaptation and inertia in dynamic environments.’’ Strategic Management Journal, 37: 1854–1864.

Sutcliffe, K. M. 1994 ‘‘What executives notice: Accurate perceptions in top management teams.’’ Academy of Management Journal, 37: 1360–1378.

Svenson, O., and A. J. Maule 1993 Time Pressure and Stress in Human Judgment and Decision Making. New York: Springer Science & Business Media.

Takemura, K. 1994 ‘‘Influence of elaboration on the framing of decision.’’ Journal of Psychology, 128: 33–39.

To, C., G. J. Kilduff, L. Ordoñez, and M. E. Schweitzer 2018 ‘‘Going for it on fourth down: Rivalry increases risk taking, physiological arousal, and promotion focus.’’ Academy of Management Journal, 61: 1281–1306.

Tversky, A., and D. Kahneman 1986 ‘‘Rational choice and the framing of decisions.’’ The Journal of Business, 59: S251–S278.

Tversky, A., and D. Kahneman 1992 ‘‘Advances in prospect theory: Cumulative representation of uncertainty.’’ Journal of Risk and Uncertainty, 5: 297–323.

Wang, X. T., F. Simons, and S. Brédart 2001 ‘‘Social cues and verbal framing in risky choice.’’ Journal of Behavioral Decision Making, 14: 1–15.

Wiseman, R. M., and P. Bromiley 1996 ‘‘Toward a model of risk in declining organizations: An empirical examination of risk, performance and decline.’’ Organization Science, 7: 524–543.

Wooldridge, J. M. 2010 Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press.

Xu, D., K. Z. Zhou, and F. Du 2019 ‘‘Deviant versus aspirational risk taking: The effects of performance feedback on bribery expenditure and R&D intensity.’’ Academy of Management Journal, 62: 1226–1251.

Authors’ Biographies

Xavier Sobrepere i Profitós is the Director of the Academic Department of Business and Management Strategy at the Universitat Pompeu Fabra, Barcelona School of Management. Xavier received his Ph.D. at IESE Business School. His research interests lie within the Behavioral Strategy field, with particular interest in the Behavioral Theory of the Firm, Upper Echelons Theory, and Rational Ecology.

Sobrepere i Profitós et al. 1047

Thomas Keil holds the Chair in International Management at the University of Zurich, Switzerland. Thomas received his D.Sc. (Tech.) at Helsinki University of Technology (today Aalto University), Finland. In addition to the Behavioral Theory of the Firm, his research focuses on mergers and acquisitions, corporate entrepreneurship, and corpo- rate governance.

Pasi Kuusela is an Assistant Professor in the Faculty of Economics and Business at the University of Groningen, the Netherlands. He holds a D.Sc. (Tech.) in technology strat- egy and venturing from Aalto University, Finland. His research interests fall within the Behavioral Theory of the Firm, mergers and acquisitions, and innovation.

1048 Administrative Science Quarterly 67 (2022)

Copyright of Administrative Science Quarterly is the property of Administrative Science Quarterly and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.