Performance Feedback Persistence: Comparative Effects of Historical Versus Peer Performance Feedback on Innovative Search

Performance Feedback Heterogeneity

A robust stream of behavioral-strategy research indicates that performance feedback (i.e., distance from aspiration level or attainment discrepancy) modulates firm search behaviors, including acquisition behavior (Iyer & Miller, 2008; Kim et al., 2015), alliance partner selection (Baum, Rowley, Shipilov, & Chuang, 2005), new-product development (Greve, 2007; Parker, Krause, & Covin, 2017), corporate lobbying (Rudy & Johnson, 2016), R&D investment (Chen & Miller, 2007; Lucas et al., 2018; Vissa, Greve, & Chen, 2010), and investment in absorptive capacity (Ben-Oz & Greve, 2015). Performance feedback can be either positive or negative depending on whether the firm’s performance exceeds or falls short of its aspiration. Extant research has concentrated on two primary referents for aspiration levels of financial performance: peer aspirations, where performance is compared with a reference group of other firms, and historical aspirations, where performance is compared with a firm’s own historical records (Cyert & March, 1963). Firms evaluate their current performance relative to these referents, interpret the resulting performance feedback, and use this information to modulate subsequent strategic decisions. Drawing on BTF and prospect theory, scholars have found that when firms are below aspiration levels, they tend to increase search behavior to address the performance shortfalls, and when they are above aspiration levels, they tend to be less motivated to make changes and even reduce risk-oriented behaviors (Cyert & March, 1963; Kahneman & Tversky, 1979; Shimizu, 2007).

Previous literature in this tradition has affirmed the importance of both peer and historical aspiration levels for organizational search decisions and has generally treated the two as theoretically and empirically consistent (Bromiley & Harris, 2014). However, while expecting the two types of performance feedback to have consistent effects on search behaviors, some empirical studies unexpectedly find meaningful discrepancies in the results. Greve (2003b) finds a clear dominance of the influence of historical aspiration levels over peer aspiration levels in affecting production growth in the Japanese shipbuilding industry. Greve (2003b) argues that this may be because managers in the shipbuilding industry view other firms as poor standards against which to evaluate their own performance. Baum and Dahlin (2007) show that peer performance feedback has a stronger effect than historical performance feedback in reducing accident cost in the U.S. freight railroad industry. The authors speculate that this result may be because the railroad industry promotes peer comparison of accident cost, causing this outcome to be highly scrutinized. In contrast, Iyer and Miller (2008) find that historical performance feedback has a stronger impact on acquisition behavior, especially for firms not threatened by bankruptcy.

Recent behavioral theory research has begun to grapple with these discrepancies by more directly examining the relationship between peer and historical performance feedback and allowing for misalignment. Joseph and Gaba (2015) demonstrate that the correlation between peer and historical performance feedback may be weak, negative, or positive and that the correlation influences a firm’s new product introductions. Lucas and colleagues (2018) focus on conflicts between performance feedback, for instance, when peer performance feedback is positive while historical performance feedback is negative. The authors contend that such inconsistency between performance feedback sends mixed signals that may foster ambiguity for organizational decision makers. Despite progress in recognizing discrepancies and allowing for misalignment between performance feedback, few studies have thoroughly grappled with theoretical and empirical differences in the nature of peer performance feedback compared with historical performance feedback.

A notable exception, and the most direct theoretical articulation of differences between peer and historical aspiration levels, is the paper by Kim et al. (2015) in which the authors theorize that performance feedback from peer aspiration is more ambiguous compared with historical aspiration. The authors argue that historical aspirations are “history dependent and reflect a firm’s capabilities and resources (Greve, 2003c: 42),” while peer aspirations rely on assumptions about the source of performance outcomes of other firms. Accordingly, they suggest that when performance is above peer aspiration levels, ambiguity regarding the cause of superior performance leads firms to attribute success to their own ability, engendering more confidence in strategic actions. In contrast, when performance is above historical aspiration levels, the less ambiguous nature of this performance feedback makes managers more cautious. ²

While the focus on ambiguity in different forms of performance feedback has advanced research, it remains a theoretical inference (Kim et al., 2015) that has not been explicitly tested. Further, the ambiguity aspect of performance feedback focuses primarily on the past experience of firms without paying sufficient attention to the forward-looking considerations of decision makers (Fiegenbaum et al., 1996; Nason et al., 2019). A manager’s perception of future prospects, such as expected over- or underperformance, can greatly influence firm R&D search (Chen, 2008). We contend that peer and historical performance feedback are likely to contain different information regarding future prospects, which is reflected in performance feedback persistence. As persistent performance feedback indicates that a firm’s current performance relative to a selected aspiration endures in subsequent periods, it affects the managerial interpretation of what performance feedback means for future performance, and thus, it affects subsequent organizational search.

The Persistence of Peer Performance Feedback

The concept of persistence comes from the profit persistence literature, which often defines performance persistence as the level of the convergence rate of a firm’s performance to the performance of its reference group (Chacar & Vissa, 2005; Mueller, 1986; Waring, 1996). More persistent performance indicates a slower convergence rate of performance to that of the reference group, while relatively less persistent performance indicates a faster convergence rate (Geroski, 1990; Mueller, 1986; Waring, 1996). In this body of work, it is well documented that the convergence of a firm’s profits with those of industry peers is remarkably slow (Geroski, 1990; Mueller, 1986; Waring, 1996). Although the degree of convergence, and thus the degree of persistence, varies across industries (Waring, 1996), empirical results confirm a high degree of persistence in a wide range of industries (e.g., Fisher & Hall, 1969; Mueller & Raunig, 1999). For example, General Motors had higher profits than Chrysler and Ford throughout the 1970s (Waring, 1996), and IBM consistently outperformed its competitors in computer manufacturing during the same era (Porter, 1979).

We draw on the concept of persistence and apply it to performance feedback. In particular, we suggest that, relative to historical feedback, peer performance feedback is likely to be especially persistent. This is due to both rather mechanistic empirical as well as deeper theoretical reasons. Historical performance feedback uses the firm’s own past performance as a referent, making the aspiration internally rather than externally determined. Thus, when a firm experiences poor historical performance for a given year, it adjusts its historical aspirations downward, making it more likely to match the historical aspiration level in the next period. In the case of positive historical performance feedback, however, overperformance shifts the subsequent aspiration level upward. Raising the necessary benchmark to achieve positive performance feedback reduces the likelihood that it will be achieved in the subsequent time period. However, this rather mechanistic feature of aspirations is bolstered by theoretical arguments for the persistence of peer performance feedback that can be derived from an understanding of interfirm mobility barriers and the constraining effects of a firm’s own resource trajectory.

Interfirm mobility barriers within an industry prevent firms from perfectly imitating or substituting their industry competitors’ resources and capabilities, thus constraining shifts in industry positions (Caves & Porter, 1977, Chacar & Vissa, 2005; Lippman & Rumelt, 1982; Waring, 1996). Specifically, interfirm mobility barriers come from two sources—the industrywide economic structure and the isolating mechanisms for firm resources and capabilities. In terms of the former, previous literature highlights such barriers as the degree of economies of scale, unionization, technological complexity (Lippman & Rumelt, 1982; Waring, 1996), and the existence of strategic groups within an industry (Porter, 1979) in reducing the speed of adjustment of a firm’s profits relative to its peers’. For example, point-to-point carriers in the airline industry would find it hard to compete directly with legacy carriers unless they invest in expensive regional hubs and build corresponding connections. Similarly, some industries have more complex knowledge structures and more legal protections, which impede the imitation of knowledge, thus safeguarding superior firm performance (Waring, 1996).

Further, the isolating mechanisms for firm-level resources affect a firm’s capability to sustain its performance position against industry peers. Research suggests that resources with valuable, rare, inimitable, and nonsubstitutable characteristics can lead to competitive advantage (Barney, 1991). Specifically, firms may have some unique resources, including their own and any partner’s experience (Madsen & Leiblein, 2015), good stakeholder relations (Choi & Wang, 2009), governance structure (Chacar & Vissa, 2005), and so on, enabling the firms that hold these resources to perform better than the others. However, because resources are heterogeneously distributed among firms, firms that do not have these kinds of resources need to acquire or develop them. The existence of isolating mechanisms, such as social complexity and causal ambiguity, makes it hard and time-consuming to acquire or develop these resources, thus generating sustainable competitive advantages for the firms that have them (Barney, 1991). To sum up, considering the two sources of interfirm mobility barriers, a firm’s performance relative to that of its industry peers is relatively stable.

Firms face fewer hurdles in updating and adapting their historical aspirations, making it faster to converge firm performance with historical aspiration levels. As historical aspirations stem from known past performance, they are more predictable, reliable, and reflective of a firm’s capabilities (Kim et al., 2015). Thus, firms are likely to understand their own sources of performance change better (Greve, 2003b; Kim et al., 2015), and accordingly, they can potentially make corresponding changes faster and implement changes relatively efficiently (compared with peer underperformance). This increases the firm’s chance of meeting its aspiration level in the next period when it is underperforming. Second, although a firm faces fewer isolating mechanisms to develop and acquire resources internally, its own resources and capability trajectory set a ceiling preventing it from attaining a similar rate of advantage in following periods. It is well established that firms have problems achieving continuous growth due to the “Penrose effect” (Penrose, 1959), which suggests that the firm’s continued growth rests on new, expanded managerial capabilities that need time to be acquired and integrated (Tan & Mahoney, 2007). This implies that there is a limit to a firm’s historical growth trajectory that affects the persistence of its historical performance feedback. In sum, we expect that a firm’s historical performance feedback is unlikely to persist.

While peer and historical performance are not fully independent and interfirm mobility barriers certainly can bound firm performance feedback fluctuation, we focus on differential levels of persistence in peer performance feedback relative to historical performance feedback. We contend that due to the reduced salience of interfirm mobility barriers, we expect that a firm’s historical performance feedback is less likely to persist than peer performance feedback. The case of Apple provides an illustrative example. Apple’s app ecosystem enables it to surpass peer performance for a relatively long period of time, and it would be hard for other manufacturing tech companies to quickly build a comparable ecosystem to challenge Apple’s position in the industry. However, Apple’s capability does not necessarily mean that it can continuously improve its performance in comparison to its own historical performance. Apple’s recent drop in iPhone sales is an example. Figure 1 illustrates this case more clearly by plotting annual peer feedback compared with historical performance feedback. While peer performance feedback remains consistently positive, there are sizable windows of time where historical performance feedback fluctuates more drastically from negative to positive and vice versa.

OPEN IN VIEWER

Figure 1

Apple’s Performance Feedback Comparison From 2006 to 2018

Apple may be a more extreme case, as its performance is at the top rank of the industry. It is possible that persistence is more salient for firms at the extreme ends of the industry, while firms in the middle face more fluctuation. Hence, we also draw performance trends for three firms performing in the middle of the same industry as Apple (see Appendix D in the online supplement). We find that in general, for the average firm in the industry, superior/inferior performance compared with historical aspirations is still more volatile or quicker to converge than peer performance feedback. ³

Overall, our examples demonstrate how superior/inferior performance, compared with historical aspirations, is more volatile or quicker to converge than performance related to peer aspirations. Together, the relative persistence of peer performance feedback and the comparatively fluctuating historical performance feedback lead us to predict a baseline hypothesis:

Feedback Persistence and Innovative Search

Behavioral strategy literature indicates that managers receive and interpret performance feedback, using it to regulate a firm’s search activities, especially its innovative search in terms of R&D expenditures (Gavetti et al., 2012). When a firm is below its aspiration level, it signals to managers that there is a problem and triggers a search for solutions (Cyert & March, 1963), which often include dramatic strategic change (Greve, 2002), risky decision making (Kahneman & Tversky, 1979), and long-term investments (Chrisman & Patel, 2012), to attempt to reach higher levels of performance in the future. Since innovative search is risky, with uncertain but potentially large future payoffs, firms will increase innovative search as a potential solution to their currently failing performance. Thus, the more the performance is below the firm’s aspiration level, the more a firm will invest in innovative search (Chen & Miller, 2007; Greve, 2003a; O’Brien & David, 2014).

When performance is above the aspiration level, however, managers interpret feedback and frame the same strategic decision very differently. Managers are generally satisfied with meeting expectations and thus are likely to maintain the status quo (Bromiley, Miller, & Rau, 2001). Prospect theory goes one step further by contending that individuals above their aspiration level immediately value (endow) their gains, which causes them to become loss averse (Tversky & Kahneman, 1991) and reticent to risk falling below the aspiration level. As a result, loss aversion leads to strategic conservatism, less risk taking, and a short-term orientation focused on protecting gains (Thaler, 1980). Since innovative search is a risky decision with a long-term time horizon, the greater the performance above aspiration level, the less a firm will invest in innovative search (Chen & Miller, 2007; Chrisman & Patel, 2012; Vissa et al., 2010).

As discussed previously, research in the behavioral strategy tradition generally treats peer and historical feedback as both salient and theoretically consistent influences on innovative search activities (Chen & Miller, 2007; Palmer & Wiseman, 1999). However, the contrasting persistence level of peer versus historical performance feedback is likely to play a critical role in how managers interpret performance feedback and thus calibrate strategic decision- making. While performance feedback certainly contains information on past performance, reflecting the need for the company to initiate a search for innovation (Chen & Miller, 2007), it also contains salient cues about future opportunities (Haleblian & Rajagopalan, 2005). The persistence characteristic of a given performance feedback is particularly relevant in providing information about the expected future state of the firm. Given the logic outlined in Hypothesis 1, peer performance feedback should be expected to be more likely to continue in the future in comparison to historical performance feedback. These contrasting temporal expectations are likely to factor into the cognitive processing of the feedback and bear on the manager’s strategic decisions.

Temporal expectations regarding performance feedback are particularly likely to influence how firms conduct innovative search. Innovative search has certain characteristics that naturally connect it with the persistence feature of performance feedback. Innovative search is an inherently long-term-oriented strategic decision (Rudy & Johnson, 2016). Investment in innovation faces technological and market uncertainty (Miller & Bromiley, 1990) and is unlikely to bring short-term gains (Bromiley & Washburn, 2011). However, despite (or perhaps because of) the long time frame until returns, unlike other types of organizational search, the return to innovation investment is likely to endure for a relatively long period (Dutta et al., 2005; Eberhart et al., 2004). Investments in innovation bolster absorptive capacity (Cohen & Levinthal, 1989) and innovative capabilities (Rothaermel & Deeds, 2006), which allow firms to generate more sustainable performance advantages over time (Dutta et al., 2005; Roberts, 2001).

When a firm faces positive historical performance feedback, the relatively nonpersistent nature of such performance implies that the positive performance is unlikely to last and performance shortfalls may ensue in the near future. The “anticipated failure in the immediate future” is expected to motivate managers of the firm to search for solutions to deal with the problem, even though it is doing well in the current period (Cyert & March, 1963: 121). Thus, even if firms are generally likely to reduce the search for innovative solutions in the face of overperformance (e.g., Audia, Locke, & Smith, 2000; Chen & Miller, 2007; Greve, 2003a, 2007; Miller & Chen, 1994; Vissa et al., 2010), when the historical aspiration is used as the reference point, managers will be cautious to cut innovative search significantly.

In contrast, when a firm receives positive peer performance feedback, the persistent nature of such performance signals to managers that superior performance is unlikely to disappear in the near future. Consequently, managers’ contentment with the status quo will buffer any sense of urgency to maintain innovative investments. This is particularly true considering the often persistent return from existing innovative capabilities and the lengthy harvest time frame of increasing innovative search (Rudy & Johnson, 2016), leading to the following hypothesis:

Managers’ interpretation of negative performance feedback is likely to be very different. Since negative historical performance feedback is expected to be less persistent, it signals to managers that firm performance may require only local solutions. Search triggered by underperformance starts locally in the neighborhood of the problem (Cyert & March, 1963). Since historical aspiration reflects a firm’s own capabilities, when firms fail to achieve what they can usually achieve, managers are likely to have a better understanding of what is wrong and how they could improve in a relatively short period of time. In such circumstances, managers are better positioned to execute local search directly in the neighborhood of the problem (e.g., Chen & Miller, 2007), rather than engaging in distant search, such as innovative investments with long-term returns (Yu et al., 2019). Further, such an approach is consistent with myopic demands from shareholders, particularly institutional investors, of publicly traded companies that focus on short-term incremental returns to firm performance rather than goals with long-term time horizons (Bushee, 1998; Bromiley & Washburn, 2011). Rather than significantly increasing innovative investments that require a long time to bring into fruition, managers may be more inclined to reduce discretionary expenses (Washburn & Bromiley, 2011) or enact search solutions that exploit existing knowledge and capabilities.

In contrast, when a firm faces negative peer performance feedback, its relatively persistent nature suggests to managers that the inferior performance is unlikely to improve in the near future. This signals that a local search may not be possible and induces a quicker transition to a more distant search. Indeed, discretionary cuts are unlikely to have any positive long-term impact on performance; rather, increasing innovative investments to develop long-term capabilities will be seen as critical to sustainably addressing underperformance. For example, Steenkamp and Fang (2011) find that in economic contractions, when a firm faces tight budget constraints, a strategy of increasing innovation but reducing advertising brings more profits than the opposite strategy. Considering the causal ambiguity in building innovative capability and the slow decay of this capability, innovative search is expected to bring returns that last for a relatively long period of time (Dutta et al., 2005; Eberhart et al., 2004). Altogether, we expect an organization will be more prone to increasing innovative search in the face of negative peer performance feedback compared with negative historical performance feedback.

Performance Feedback Persistence and Innovative Search by Industry

While the profit persistence literature argues that superior or inferior performance relative to peers is relatively enduring, there is also recognition that there is considerable heterogeneity in profit persistence across industries (Waring, 1996). For example, Hirsch and Gschwandtner (2013) find that relative profits in the food industry are less persistent than in other manufacturing sections because the food industry is very competitive and has a large concentration of retailers. Other studies confirm that different levels of profit persistence exist in some other industries, such as automobiles (Waring, 1996) and pharmaceuticals (Roberts, 1999). Heterogeneity in profit persistence across industries results from differences in the intensity of interfirm mobility barriers. Industries vary in attributes such as rivalry, switching cost, economies of scale, and information impediments to imitation. In high-persistence industries, these “mobility barriers” make it even harder for underperforming firms to catch up and ensure that overperforming firms enjoy superior performance for longer durations. Hence, peer performance feedback, reflecting the difference between performance and peer aspiration, is even more persistent in high-persistence industries, while it is less persistent in low-persistence industries.

Drawing from this evidence, we argue that the difference between peer and historical performance feedback is likely to be more pronounced in industries with high profit persistence. In high-profit-persistence industries, industry mobility barriers are harder to overcome. As a result, managers face even less pressure to engage in innovative search when they are overperforming compared with peers, but they sense an especially urgent need to correct performance shortfalls by conducting long-term and capacity-enhancing innovative search when they are underperforming relative to peers. It then follows that the differential effects between peer and historical performance feedback on innovative search will be stronger or amplified in high-persistence industries.

Data and Sample

We constructed our sample following previous BTF research on innovative search (e.g., Bromiley & Harris, 2014; Bromiley, Rau, & Zhang, 2017; Chen & Miller, 2007). The initial sample for this study comes from Standard and Poor’s Compustat database, covering the period from 1974 to 2018. We restrict our sample to firms in the manufacturing industry (Standard Industrial Classification [SIC] codes 2000–3999) as a large proportion of high-tech and innovative activities come from manufacturing industries (Hecker, 1999). This practice is also consistent with previous behavioral research examining innovative search (e.g., Bromiley & Washburn, 2011; Chen & Miller, 2007; Yu et al., 2019), which makes our study comparable to previous studies. To construct our sample, we undergo several data-cleaning procedures. First, we exclude industries with fewer than five firms to reduce biases that may arise from small industries (Chen & Miller, 2007; Gentry & Shen, 2013). Second, we exclude firms having R&D expenditures greater than sales because these firms may be R&D specialists or research firms with distinct search behaviors (Chen & Miller, 2007; Gentry & Shen, 2013). This step excludes 5,750 firm-year observations. Third, we drop observations with missing values in focal variables. The final sample includes 56,716 firm-year observations from 1974 to 2018. Following previous studies (e.g., Bromiley et al., 2017), unless specified otherwise, we winsorize variables used in the study to the 1st and 99th percentiles to reduce outlier influence.

Measures

Our hypotheses indicate different dependent variables. Thus, we separate variable descriptions for Hypothesis 1 and Hypothesis 2 through Hypothesis 4 for clarity’s sake.

Modeling Approach

We test our hypotheses using two related models. Our first model specification, persistence model, is used to test the persistence of peer and historical performance feedback (Hypothesis 1). Our second model specification, innovative search model, is used to test a firm’s innovative search behaviors in response to the peer and historical performance feedback (Hypotheses 2, 3, and 4). We control for firm and year fixed effects in all specifications to rule out the influence of unobservable firm-level stable factors, such as culture, managerial preference, and the influence of the business cycle.

Persistence models

The profit persistence literature typically measures persistence as a first-order autoregressive (AR [1]) difference process (Mueller, 1986; Waring, 1996). Specifically, we use the following equations:

P i , t − A h i , t = α + β 1 * ( P i , t − 1 − A h i , t − 1 ) + W i , t − 1 + N i , j , t − 1 + ϵ i , t

(1)

P i , t − A p i , t = α + β 1 * ( P i , t − 1 − A p i , t − 1 ) + W i , t − 1 + N i , j , t − 1 + ϵ i , t

(2)

where P_i represents firm i’s performance and A_hi and A_pi represent historical and peer aspirations, respectively. W_i,_t–1is a vector of firm-level controls, while N_i,_j,t–1 is a vector of industry-level controls. The slope coefficient β₁ for Equations (1) and (2) describes the persistence of profit, which is the proportion of a firm’s profits “in any period before period t and systematically remains in period t” (Waring, 1996: 1225). Generally, the higher β₁ is, the higher the profit persistence is.

Equations (1) and (2) are dynamic panels with lagged dependent variables. One potential problem is that the lagged dependent variables are likely to be correlated with the error term, which leads the ordinary least squares estimator to be biased. Another econometric issue is the “small T, large N” problem described by Nickell (1981). The two issues are related in our model, as we include lagged dependent variables and our time horizon (T = 47) is relatively small compared with the number of firms in the sample (N = 7,973). There are several ways to address these issues, such as using difference generalized method of moments (GMM) (Arellano & Bond, 1991), system GMM (Arellano & Bover, 1995; Blundell & Bond, 1998), bias correction (Nickell, 1981; McGahan & Porter, 1999), and QML estimation for a fixed-effect dynamic panel (Hsiao, Pesaran, & Tahmiscioglu, 2002). We choose QML because it offers potential efficiency gains and better finite-sample performance compared with GMM estimators (Hsiao et al., 2002). We use the Stata command “xtdpdqml” to perform QML for the fixed-effect dynamic panel (Kripfganz, 2016). We follow the same specification and estimate the persistence of the two forms of performance feedback in two separate equations and then compare the coefficients’ difference by performing the seemingly unrelated test (the “suest” command in Stata).

Before testing Hypothesis 4, the persistence model for peer performance feedback is also used to find which industries are more or less persistent. In particular, we add the interaction of four-digit industry dummy variables with lagged dependent variables in our persistence model for peer performance feedback using Equation (3). We measure industry performance persistence as the coefficients on the interaction, because these coefficients represent which industries preserve the larger or smaller proposition of previous profits. After that, we rank industry from high persistence to low persistence, as shown in Appendix A in the online supplement. We then assign the top half of industries to the high-persistence group and the bottom half to the low-persistence group. After that, we estimate Equation (4) in the two subsamples of high- and low-persistence industries.

P i , t − A pi , t = α + β 1 * ( P i , t − 1 − A pi , t − 1 ) + β 2 * ( P i , t − 1 − A pi , t − 1 ) * I n d i , j + W i , t − 1 + N i , j , t − 1 + ϵ i , t

(3)

Innovative search models

In innovative search models, we test how performance feedback affects the intensity of innovative search. Following the majority of previous BTF studies on innovative search (e.g., Bromiley et al., 2017; Bromiley & Washburn, 2011; Chen & Miller, 2007; Greve, 2003a), we used firm fixed-effect panel regression with robust standard errors to test our innovative search models.

Specifically, for Hypothesis 2 and Hypothesis 3, we use the following equation:

S i , t = β 0 + β 1 * I h i , t − 1 * ( P i , t − 1 − A hi , t − 1 ) + β 2 * ( 1 − I hi , t − 1 ) * ( P i , t − 1 − A hi , t − 1 ) + β 3 * I pi , t − 1 * ( P i , t − 1 − A pi , t − 1 ) + β 4 * ( 1 − I pi , t − 1 ) * ( P i , t − 1 − A pi , t − 1 ) + β 5 * W i , t − 1 + β 6 * N i , j , t − 1 + β i + β t + ϵ i , t

(4)

where S_i, _t is a vector of firm i’s innovative search intensity in period t. P_i,_t–1 represents firm i’s performance, and A_hi,_t–1 and A_pi,_t–1 represent historical and peer aspirations, respectively. I_hi,_t–1 measures whether firm i was underperforming relative to the historical aspiration level in period t – 1. I_pi,_t–1 measures whether firm i is underperforming relative to peer aspiration level in period t – 1. Wi,_t–1 is a vector of firm-level controls, which reflects a firm’s internal capability to support search activities, while N_i,_j,t–1 is a vector of industry-level controls, which implies the external capability to support organizational search activities (Chen & Miller, 2007). Following previous literature, we build industry-level controls by subtracting the corresponding firm’s own influence. β _i and β _t capture firm and year fixed effects, respectively, and ϵi, _t is the error term.

For Hypothesis 4, we split the sample into high persistence and low persistence, and we estimate Equation (4) in the two subsamples.

Industry Performance Persistence

Table 3 also shows the results comparing high-persistence industries and low-persistence industries. We test the difference between peer overperformance and historical overperformance, peer underperformance, and historical underperformance in the two samples. In the high-persistence industry group (Model 3), we find a significant differential effect between peer and historical overperformance (F statistic = 79.63, p = .000) and a significant differential effects between peer and historical underperformance (F statistic = 8.43, p = .004). In the low-persistence industry group (Model 4), we do not find a significantly differential effect between historical underperformance and peer underperformance (F statistic 2.65, p = .104). We find that there is a significantly differential effect between peer and historical overperformance (F statistic 5.37, p = .021), but that effect seems to be much smaller in the low-persistence industry group compared with what we find in the high-persistence industry group (coefficients of peer overperformance in high- vs. low-persistence industry: −0.10 vs. −0.04; coefficients of historical overperformance in high- vs. low-persistence industry: 0.02 vs. 0.02). Overall, our findings suggest that the differential effects between peer and historical performance feedback are most salient in high-persistence industries. Figure 3 provides a straightforward illustration of the estimated differential effects. From Figure 3a, we can see that the differential effect between peer and historical overperformance feedback is much larger in high-persistence industries than in low-persistence industries. Similarly, from Figure 3b, we can see that the differential effect between peer and historical underperformance feedback is larger in the high-persistence industries, while the differential effect is not significant in the low-persistence industries (i.e., overlap of confidence intervals). Taken together, these results provide support for Hypothesis 4 and also serve as evidence for persistence as the mechanism influencing the results of Hypothesis 2 and Hypothesis 3.

OPEN IN VIEWER

Figure 3

Innovative Search in Response to (a) Overperformance in and (b) Underperformance in High-/Low-Persistence Industries

Robustness Checks

We use the QML model in the baseline model for testing the persistence of performance feedback and present the results of alternative models in Appendix B in the online supplement. The system GMM approach has also been widely used to deal with the endogenous issue associated with lagged dependent variables in dynamic panel data (e.g., Girod & Whittington, 2017; Roodman, 2009). We present the results of the system GMM in Appendix B, Table B.1, in the online supplement. The system GMM enables us to build instruments using the lagged variables. We treat firm-level variables as endogenous, while we treat industry-level variables as exogenous. Using the lagged variables as instruments, we find that the results are broadly consistent with our baseline results using QML. The Arellano-Bond test for autocorrelation suggests our GMM specification is valid. We also use dynamic panel model using maximum likelihood as an alternative estimation. As this method usually does not work well when the time period is longer than 10 years (Williams, Allison, & Moral-Benito, 2018), we split our sample into 10-year periods and present the results in Appendix B, Table B.2, in the online supplement. The results from maximum likelihood estimation are consistent with our baseline results using QML.

We also test if our results are robust to different modeling approaches. First, we rerun the analysis using a two-stage process similar to the model used in Vissa et al. (2010). In the first stage, the firm decides whether it will engage in innovative search activity, and in the second stage, it decides how intensive the search effort will be. This approach enables us to correct the selection bias in our sample due to the fact that we observe that firms report R&D expenditure only when they decide to conduct innovative search. Apart from the two-stage process, we also rerun the analysis using the tobit model, which is a typical technique for dealing with censoring problems. Both the Heckman selection model (results in Appendix B, Tables B.5.1 and B.5.2, in the online supplement) and the tobit model (results in Appendix B, Tables B.3.1 and B.3.2, in the online supplement) give qualitatively similar patterns of results. We check other models, such as the two-part model (results in Appendix B, Tables B.4.1 and B.4.2, in the online supplement) following Belotti, Deb, Manning, and Norton (2015), and our results are consistent.

We further test if our results are robust to different measures and samples. Instead of using an aggregated slack index, we use disaggregated measures of available slack, recoverable slack, and potential slack, following Bromiley and Washburn (2011). ⁵ Our results still hold, and the magnitude of impacts is also comparable with our main results, as shown in Appendix C, Tables C.1.1 and C.1.2, in the online supplement. We also rerun analyses based on an updated version of Altman’s Z that was proposed by Altman (1983) for both private and public firms. As illustrated in Tables C.2.1 and C.2.2 in the online supplement, results are again robust. Finally, we reinclude those firms with R&D greater than sales and present the results in Tables C.3.1 and C.3.2 in the online supplement. The results are consistent with our predictions.

Moreover, we conduct several additional tests to rule out alternative explanations and verify that it is indeed the persistent nature of relative performance feedback that is driving our results. First, if persistence performance feedback indeed exists, temporal changes to macroeconomic conditions would create more variations in historical performance feedback than peer performance feedback does. We conducted a test that looks at changes in relative performance feedback before and during the 2007-to-2009 financial crisis. In particular, we look at average historical and relative peer performance for the years 2004 to 2006 as well as average historical and relative peer performance for the years 2007 to 2009. Then we calculate changes in relative performances and compare those changes. We find that compared with historical performance feedback (average change −0.05), peer performance feedback changes to a lesser extent (average change −0.03), and such difference is statistically significant (p = .03). This corroborates our arguments.

Second, an alternative explanation is that observations with inconsistent performance feedback (when peer performance feedback is positive and historical performance feedback is negative, or the other way around) are creating noise in our results. As Lucas et al. (2018) point out, inconsistent performance feedback may lead to confusion in managerial interpretation of the performance feedback. Thus, it may distort the findings on the comparison between historical and peer performance feedback. To control the effect of inconsistent performance feedback, we rerun our baseline regression (4) using observations with consistent performance feedback (when peer and historical performance feedback are aligned in the same direction). As shown in Appendix C, Tables C.4.1 and C.4.2 (in the online supplement), after controlling the effect of inconsistent performance feedback, we still observe a salient difference between peer and historical performance feedback in the full sample and in high-persistent industries as expected.

Apart from the aforementioned alternative explanation, we also check whether the difference between historical and peer performance feedback goes away in the nonpersistent group. If persistence is the key mechanism driving the different innovative search in the face of different performance feedback, we should see that the effects disappear in the nonpersistent group. We record observations for the nonpersistent group with current positive peer performance feedback and no consecutive positive peer performance in past years as well as observations with current negative peer performance feedback and no consecutive negative peer performance in past years. Both groups represent observations with nonpersistent performance feedback. We check whether our results are sensitive to how many consecutive years we use by including 6, 5, 4, 3, and 2 consecutive years. As shown in Appendix C, Tables C.5.1 and C.5.2 (in the online supplement), the difference between historical and peer performance feedback goes away in both nonpersistent groups. This further corroborates our theoretical development that persistence is the mechanism that explains the differential effect between historical and peer performance feedback on innovative search.

Finally, we check whether slack search drives the results. Slack is the availability of resources in excess of what is necessary to produce output for the organization (Bourgeois, 1981), and thus, it is more likely to be accumulated in previous positive historical performance. By restricting the sample to observations with nonprevious consecutive historical performance, we check whether our results are still held when we control for slack search. Again, we check whether our results are sensitive to how many consecutive years we use by including 6, 5, 4, 3, and 2 consecutive years. As shown in Appendix C, Table C.6.1 and C.6.2 (in the online supplement), we can see that our results still largely hold in the sample of observations with no previous consecutive historical overperformance (i.e., no slack accumulated). Our results confirm that after controlling for slack search, we can still observe the proposed difference between historical and peer performance feedback on innovative search.

Limitations

Our study is not without limitations, many of which provide promising avenues for future research. First, as with most organization-level behavioral theory research, we do not measure managers’ aspirations directly. More concerted investigation of how managers cognitively set and assess aspirations will be useful for future research (Short & Palmer, 2003). Second, and relatedly, we follow precedent and use median peer performance as the peer aspiration (Iyer & Miller, 2008). But firms may use nonmedian peer referents, including shifting their reference group downward to improve the perception of their performance (Jordan & Audia, 2012) or looking upward to benchmark industry leaders (Hu, Blettner, & Bettis, 2011). Third, we focus on the distinction between historical and peer referents. However, firms may also use other external referents, such as analyst predictions (Gentry & Shen, 2013) or stakeholder expectations (Nason, Bacq, & Gras, 2018). Future research should consider how firms handle multiple referents beyond historical and peer feedback. Fourth, we limit our empirical investigation to a theoretically salient context of public manufacturing firms in the United States. As a result, our findings may not be generalized to private firms, or firms in other geographic markets. Further investigation is necessary to reveal the generalizability of our findings across other industries, countries, and organizational forms.