TIMING FIRM INNOVATIONS
BASED ON ASPIRATIONS
How to make the innovation decision?
Abstract
This study adds new insights into the behavioral theory of the firm. In the behavioral theory of the firm the concept of organizational aspirations is central to an explanation of how organizations learn and make decisions. However, which aspiration might give the best performance is unclear. In this paper, four learning strategies are examined through computational simulation. We developed an agent-based model to formally examine the effects of learning strategies on innovation decisions and in turn firm performance. There are four learning strategies identified: the historical learning strategy, the social learning strategy, the simultaneous learning strategy and the mixed learning strategy. In the results, it became clear that the social learning strategy yields the best results, however, when market uncertainty increases, the performance difference between the strategies diminishes. Furthermore, we find that changing from one learning strategy to another can instigate direct short-term increase in performance. Finally, although more research is required in this direction, the findings suggest that industry selection has an inverted-u shape relation with industry performance and that industry selection moderates the effect of market uncertainty and technology path dependency on industry performance.
Keywords: behavioral theory of the firm, innovation decision-making, aspiration adaption theory,
Introduction
Goals and targets are widespread used by companies and there are many scholars that have studied the effects of goals and targets, especially in the strategic management literature (Hu, Blettner, & Bettis, 2011). The strategic management literature focuses, among other topics, on the organizational level decision making. As Cyert & March (1963) note in their behavioral theory of the firm, strategic decisions are made in an adaptive way through a process of organizational learning. There is, however, little consensus about a theory or model of organizational learning (Fiol & Lyles, 1985). Many researchers have studied the decision-making process of companies (Nicolas, 2004), especially in relation to organizational learning. The lack of consensus has led to multiple definitions of learning. In this paper, we study organizational learning as adaptions of the company to the new environments, which is similar to the definition used by Cyert & March (1963). The adaption to new environments occurs through innovation. Innovation decision making is central in this paper. The question when companies should innovate is not fully satisfied by the current literature. Innovation can be valuable, because it can bring creative solutions to problems in the firm or improve the overall performance of the firm. However, in the literature it is well understood that innovation also has a high potential to fail (Dong, 2017).
(Gaba & Bhattacharya, 2012). Others studies make contributions to the behavioral theory of the firm, such as the additions of a forward-looking approach to the aspiration adaption theory (Barreto, 2011), errors and biases in performance reporting (Fang, Kim, & Milliken, 2014), and a multiple reference point when it comes to R&D intensity (Chen & Miller, 2007). All these examples use or complement to the concept of aspirations, but there is only little research into how aspirations are measured and which basis of aspiration results in the best performance. Several papers do distinguish the differences between diverse strategies of aspiration foundation (Bromiley & Harris, 2014; Hu et al., 2011), however, this current paper differentiates itself as it uses computational simulations instead of empirical data to gain new insights into the subject. Moreover, this paper contributes not only which aspiration reference point results in the best performance, but also how organizations can switch from one learning strategy to another and how that affects firm performance. There are two questions central to this paper. First, how can firms change their
innovation-decision learning strategy to improve their performance? After that we try to answer
the question: how does industry selection affect the industry performance? To answer these questions, we use the agent-based model from Dong (2017) and improve that model to increase robustness and generalizability.
a better performing strategy. The results show that there is performance growth in the short-term due to the change in strategy. To validate the robustness of the model, various configurations are tested. Finally, in the discussion we incorporate the findings in the broader scope of the behavioral theory of the firm.
Theoretical Background
The literature in strategic decision making can be split up in four broad perspectives; the ‘individual decision perspective’, the ‘strategic or management choice perspective’, the ‘environmental determinism’ and a ‘firm characteristics and resource availability perspective’ (Papadakis et al., 1998). In this paper, we focus on the strategic or management choice perspective and the firm characteristics and resource availability perspective. The former emphasizes the role of decision makers at the corporate level, whereas the latter focuses on various internal factors (e.g. firm performance). First, a brief description of the theoretical concepts is given. After that, we explore and review the findings in the literature regarding this subject.
This study will complement to the behavioral theory of the firm (BTOF) that was introduced by Cyert & March (1963). Often BTOF is used to explain the actions of a firm (Blettner et al., 2014; Gavetti, Greve, Levinthal, & Ocasio, 2012; Rudy & Johnson, 2013). The BTOF emphasizes the decision-making process and introduces bounded rationality to organization behavior theory. Risk is a central component of innovation (Brown, 2010). Therefore, innovation often brings uncertainty in firms. The BTOF is especially useful when there is uncertainty involved in the decision-making. Thus, the BTOF is appropriate to develop theories around innovation decision making, as innovation brings high uncertainty.
term ‘aspirations’. Aspirations can be defined as the company’s desire to reach a certain performance level or goal (Levinthal & March, 1981). It can also be defined as the minimum level that is perceived as satisfactory by the decision maker (Xie et al., 2016).
In the aspiration adaption theory, which is part of the BTOF, the aspiration of the firm decides the level of risk it is willing to take. When the current performance of the company is below the level the company aspires, the company puts more emphasis on risky and non-routine moves in order to improve their performance. On the other hand, when performance is below the firm’s aspirations, the firm has no incentive to challenge the status quo (Dong, 2017; Hu et al., 2011; Rudy & Johnson, 2013). Although other literature suggests that the amount of difference between aspiration and performance results in different levels of risk seeking behavior (Hu et al., 2011; Xie et al., 2016), this is not taken into account in the model used in this paper, as it would unnecessarily complicate the model for the research question we aspire to answer.
Blettner et al. (2014) create a mathematical model that tries to calculate aspiration levels based on empirical data from companies in the German magazine industry. They state that it is plausible that firms pay varying attention to different reference points, as aspiration adaption is a dynamic process. Building upon that, there are two primary factors, or ‘reference points’, that influence firm’s aspirations (Xie et al., 2016): historical performance of the firm and the performance of competitors within the industry. With these two factors, four strategies can be identified: the historical learning strategy, the social learning strategy, the simultaneous learning strategy and the mixed learning strategy.
companies are inert and tend to attribute performance downfall to other environmental factors (Greve, 2002).
On the other hand, firms can base their aspirations on the performance of a reference group within their industry. Using the performance of the focal firm’s closest competitors instead of their own, the firm uses more information which makes it less prune to bias. It is likely, however, that it is less reliable data overall, because the firm has to acquire the information from its competitors. The simultaneous learning strategy takes information from both sides. It bases its own aspirations partly on the firm’s own performances and partly on the reference group within their industry. This strategy uses the most information and should be more reliable than the social learning strategy. (Dong, 2017) found, however, that this strategy results in lower performance than the social learning strategy.
Finally, the mixed learning strategy takes either of the two extremes. This strategy compares whether the focal firm’s own performance is higher than the average performance of the reference group. If this is the case, the firm uses the historical learning strategy. However, when the firm’s own performance is lower than the performance of the reference group, it uses the social learning strategy. Consequently, the firm bases its aspiration on the highest performance of either their own performance or the performance of the reference group.
These four strategies are compared in the paper of Dong (2017). This paper concluded that the social learning strategy and the simultaneous learning strategy performed optimally when only technological uncertainty was taken into account. This current paper tries to advance the research of Dong in several ways. First and foremost, this paper tries to answer the question what the effect is when some firm switches from a lower performing strategy to a higher performing strategy. Besides, this paper examines what the effects of industry selection on this model are. Finally, we perform extensive robustness checks to further validate the findings.
(Bromiley & Harris, 2014; Greve, 2003; Schimmer & Brauer, 2012). Different measures being patent applications (Gaba & Bhattacharya, 2012), new drug approval rates from the FDA (Tyler & Caner, 2016), and the number of magazines sold (Blettner et al., 2014). Other research methods such as interviews (Kirchoff et al., 2016) and simulations (Dong, 2017; Fang et al., 2014; Hu et al., 2011; Levinthal & March, 1981) are also used, though to a lesser extent.
The paper of Bromiley & Harris (2014) describes three types of models that are well established in prior research: the weighted average model, the separate social and self-measures model, and the switching model. Respectively, these models can be compared to simultaneous learning strategy, social learning strategy, historical learning strategy and the mixed learning strategy described earlier. In their findings, they present that all three models generally had significant parameters on finding the difference between performance and aspiration, although not all three models were equally good. Other research (Blettner et al., 2014) shows that a shift from organizational attention from own performances to others’ performances motivates more exploration. This leads to the statement that managers can change or redirect strategy by shifting the reference point of the organization. The research in this paper uses this implication as a basis in order to find out which strategy results in the best performance and what the effect is when firms change strategy.
Various scholars studied the effect of reference points in aspiration adaption theory. Almost all literature agrees that firms display more risk-taking behavior and innovativeness when the firm’s aspirations are below the actual performance. On the other hand, it is uncertain whether firms tend to be more innovative when performance exceeds aspirations due to slack resources or that organizational inertia stifles innovation at this point (Schimmer & Brauer, 2012).
This paper extends the agent-based model introduced by Dong (2017) and goes one step further. We will not only answer which strategy performs best, but also how firm performance is influenced in this model when organizations change from one strategy to another.
Methodology
Carley, 2007). Simulation can be defined as a method for using computer software to model the operation of “real-world” processes, systems or events (Davis et al., 2007).
Simulation is easily understood, relatively free of mathematics and often quite superior to mathematical methods that may be too complex or not available (Berends & Romme, 1999). A simulation approach is appropriate for this study, as it enables us to objectively study the relation between multiple variables. Furthermore, the paper of Bromiley & Harris (2014) states that there is no clear measure for firm aspiration. Most studies do not have a direct measure of aspiration, so the measure is calculated on accounting measures. Because multiple factors influence accounting measures, it is hard to generalize the results of these studies. Thus, in this case, the simulation approach is preferable to empirical or mathematical analysis, because objective data to study the social construct ‘aspiration’ is either unavailable or too complex to mathematically incorporate in a model.
This study is an extension and further validation of the research project started by Dr. John Dong. At first, the focus of this study was to reconstruct the results of the paper of Dong (2017). When similar results were found with the reconstructed code, we improved the model and deployed the code on a dedicated webserver. This enabled us to run and monitor 1000 iterations of one simulation and combine the results, which increases the robustness of the results. A webserver would ensure ease of use and accessibility, which is essential to rapid development of the model and enable the model to be used readily and quickly, with successive changes of input data (Bell, 1970). Therefore, the algorithms for these simulations are all written in the PHP programming language and follow an object-oriented programming structure. This programming structure is advantageous when programming an agent-based model.
Main parameters
The main parameters of this study are market uncertainty and technology path dependency. The four learning strategies of Hu et al. (2011) are primarily tested against these two independent parameters.
Technology path dependency
dependency results in minor changes in the companies’ technology status after each innovation, whereas low path dependency results in major shifts in technology status after each innovation.
Market uncertainty
The performance of a company is always influenced by market uncertainty, however, the amount of influence a market has on the company can differ. In this model, market uncertainty has a direct effect on company performance. A higher market uncertainty leads to lower and more fluctuating performance, whereas lower market uncertainty leads to a more stable performance levels.
The four stages of simulation
The structure of the simulations consists of four stages similar to the Howell Sequence (figure 1)(Bell, 1970). First, the simulation requires some input parameters to be set. After that, the simulation starts. The simulations generate the initial conditions for the sample. Then, it runs through the configured timespan and calculates new values for every company according to the model at that period of time. After these iterations, the average values of the complete sample add to the mean output data. When the data has been added, it iterates back to stage 2 (setting the initial conditions for the sample). When all iterations have finished, the simulations return the output values. Next, these four stages are explained in more detail.
Figure 1: Howell Sequence (Bell, 1970)
Input data: First, the sample size is determined. All simulations in this study are executed with a
sample set of 200 company objects. This is similar to other literature (Greve, 2002).
Parameters Symbol Description Default value(s)
Technology Path dependency 𝛼 The speed with which a new innovation
takes effect.
0.1 - 0.5 - 0.9
Market uncertainty 𝛾 The effect of the market on company
performance.
0.1 - 0.5 - 0.9
Aspiration path dependency 𝜑 The speed with which aspirations adapt to
circumstance.
0.5
Peer spread 𝑗 The number of close competitors it takes
into account.
10
History improvement 𝑞 The increased own performance the
company aspires.
0.05
Renewal rate 𝑑 Percentage of under-performing companies
that is replaced by new entry companies. 0
Table 1: The input parameters that were used during the simulations.
Beside those parameters, there is one parameters to differentiate between the learning strategies. This parameter (𝑤) indicates the balance between the social learning strategy and the historical learning strategy. This will be explained in more detail in the fourth action of the timespan iterations.
Initial configuration: Each company object gets an initial performance level (𝑃),+), aspiration level (𝐴),+) and technology level (𝑇),+). Initially, these values are randomly drawn from standard normal distribution.
Timespan iterations: Every time period the list of companies gets sorted based on their
performance. This way, the companies can easily check against the performance of their competitors. After that, it iterates over the list of companies and executes the following sequence for each company:
2. The simulation checks if the companies’ performance is below their aspiration, if so, the company innovates. When a company innovates, the technology level of the company (𝑇),/) changes. This change is calculated with the following formula:
𝑇),/ = 𝛼𝑇),/34∗ (1 − 𝛼)𝜇
Where 𝛼 is the path dependency and 𝜇 is the innovation. The outcome of the innovation is uncertain. So, we use an independent identically distributed standard normal distribution term. The term 𝑇),/34 is the current technology level.
3. The simulation calculates the new performance of the company based on their current technology level and market uncertainty, grasped by the following formula:
𝑃),/ = 𝑇),/+ 𝛾𝜇
Where 𝑃),/ is the performance for period 𝑡 and 𝑇),/ is the current technology level of the company. The parameter 𝜇 is a standard normal distribution that simulates the market uncertainty. This uncertainty is moderated by the market uncertainty parameter 𝛾.
4. Finally, the company recalculates their aspiration level. This calculation differs greatly depending on the values set for the parameters in table 1. In the formula 𝑤 determines the balance between the social and historical perspective. When the mixed strategy variable is set to 1, the 𝑤 is determined by comparing the companies’ performance with its competitors. The formula is as follows:
𝐴),/ = 𝜑𝐴),/34+ (1 − 𝜑) ∗ (𝑤 ∗ (𝑃),/ + (𝑞 ∗ |𝑃),/|)) + (1 − 𝑤)𝐶),>)
performance is better than competitor’s performance (𝑤 𝑖𝑠 𝑠𝑒𝑡 𝑡𝑜 1) or own performance is worse than competitor’s performance 𝑤 𝑖𝑠 𝑠𝑒𝑡 𝑡𝑜 0 .
Output and results: When the performance level for all companies over the time period have been
recorded, the average performance per time period for all companies is calculated. These values are added to the average performance levels of all iterations so far. These values are stored in a database that is connected to the web interface.
Figure 2 (Appendix A) shows the detailed flowchart of the simulation.
Model Robustness
There are two types of robustness checks to perform on the model in order to check the robustness of the results.
Control parameters
Different values for the control parameters should be tested to check if the model gives similar results. We test the benchmark model with variations in these parameters in the two most extreme cases; when market uncertainty and technology path dependency is low and when market uncertainty and technology path dependency is high.
This model uses four control parameters: aspiration path dependency, peer spread, history improvement and renewal rate.
Aspiration path dependency
Peer spread
The social, simultaneous and mixed learning strategy make use of the performance metrics of surrounding companies. The peer spread parameter indicates the number of companies the focal company uses to calculate the competitions’ performance. This study uses a peer spread of 10 companies. To check the effect of this parameter, the benchmark study is also executed in the two extreme cases with a peer spread of 50 companies.
History improvement
This study assumes that every company that uses its own performance for their aspirations, seeks to improve themselves. This parameter indicates the amount of self-improvement is used. In accordance with other studies (Bromiley, 1991), a five percent rate is used during the simulations. The extreme cases are tested with a 25% and a 50% self-improvement rate.
Renewal rate
The last control parameter is the renewal rate. This parameter introduces industry selection to the model. In the benchmark case, no industry selection is applied. This means that the companies that perform below the specified percentage, will be restarted. For example, when the renewal rate is 10%, the 10% lowest performing companies will be restarted. In the results, we do discuss the effect of industry selection on industry performance. Furthermore, we test all nine cases with a 10% renewal rate to see the effect of this parameter on the results found.
Multiple strategies
Results
Benchmark simulations
In the benchmark simulations, we study the four learning strategies of the behavior theory of the firm, namely the historical, social, mixed and simultaneous learning strategy. Each strategy is simulated in a matrix of three levels of market uncertainty and technology path dependency, resulting in total of 36 simulations. The three levels of market uncertainty and technology path dependency we use are low (0.1), medium (0.5) and high (0.9). Matrix 1 shows nine graphs each depicting the average performance of a sample. The performance of these samples is the result from the use of one of the four learning strategies used to decide whether to innovate.
As can be seen in matrix 1, the results of the simulations are similar to the findings of Dong (2017). With low market uncertainty and low technology path dependency, the social learning strategy and the simultaneous learning strategy perform best. However, when market uncertainty is increased, the effect of the learning strategy diminishes. By increasing the technology path dependency, the performance of the industry drops and curves do fluctuate more, but similar patterns are found for the learning strategies.
Matrix 1: Benchmark simulations
In all nine tiles, the social learning strategy results in the highest performance, except when both market uncertainty and technology path dependency are high. In that tile, no distinction can be made from the graph. The second-best strategy appears to be simultaneously learning from historical aspirations as from social aspirations. The mixed learning strategy performs marginally better than the history learning strategy.
Change of strategies
Matrix 2: Change of strategies
Matrix 2 shows an unexpected drop in the companies that initially use the historical learning strategy. The cause of this drop is unclear. In all other respects, the model initially behaves similarly to the benchmark case. When 500 time periods have passed, the strategy adapts to a new strategy. When market uncertainty is low and the firm quickly adapts to new innovations, the performance of the company increases rapidly when the new strategy takes effect. Similar to the benchmark case, as more market uncertainty is added, the effect of changing strategies declines.
Industry Selection
grows to 0.670. When the renewal rate is higher than 50%, the performance drops and at 70% the industry performance is on the same level as when there is no industry selection.
Matrix 3: The effect of industry selection to the model
All four learning strategies show a performance improvement when industry selection is introduced. The social and simultaneous strategy are affected the most. Industry selection does decrease the gap between the performance levels between the different learning strategies. When the renewal rate is 90%, there is no significant difference between the learning strategies.
Robustness of the model
Parameter robustness
There are eight parameter robustness checks performed to see whether these parameters had influence on the results. The results of these robustness checks can be found in figure 4A and 4B in Appendix B. The first thing that is notable, is that a 10% renewal rate leads to a clearer performance difference between the strategies when there is a high market uncertainty and a high technology path dependency. The results now show clearly that the social learning strategy performs optimal followed respectively by the simultaneous, mixed and historical strategy. To examine this effect further, the benchmark case is simulated again but with a 10% renewal rate. The results of that simulation can be found in Matrix 5 in Appendix B. Although the curves do grow closer when market uncertainty is increased, still a significant difference is visible between the curves. Furthermore, the performance drop from higher technological path dependency in the benchmark case is not visible when industry selection is included. Only when market uncertainty is added, the performance of the industry decreases significantly, but not drastically.
The other seven robustness checks show similar behavior as the benchmark case. In all cases the social learning strategy performs the best and the mixed and historical learning strategy perform the worst when there is low technology path dependency and low market uncertainty. Likewise, the results are indistinguishable from each other when market uncertainty is high and technology path dependency is high.
Multiple strategies
All cases we studied so far have used samples where all companies used the same strategy in that time period. In the real world, it is unlikely that all companies use the same strategy. Therefore, the model is tested with the benchmark case where all four learning strategies are mixed in one sample. To keep consistency in the total sample size, each strategy is represented with 50 companies in the sample. The results of this simulation can be found in appendix C.
All other cases show no clear distinction between the strategies. By mixing the strategies the average performance of the companies fluctuates more violently and it reduces the gap between the strategies.
Discussion
Theoretical Contributions
This paper was set out to contribute to the behavioral theory of the firm. Specifically, we tried to examine how organizations can improve their performance by changing their learning strategy regarding innovation decisions. Two contributions can be distinguished.
First, we take the work of Bromiley & Harris (2014) further as we did not only distinguish different learning strategies in the aspiration adaption theory, but also examined what the effects are when a firm changes from one strategy to another. The results show that shifting the aspiration reference point from your own performance to the average performance of the focal firm’s close competitors, the performance of the firm will increase. The work of Blettner et al. (2014) states that a shift from organizational attention from own performances to others’ performances motivates more exploration. This implies that in the behavioral theory of the firm the social learning strategy leads to more exploration, which in turn result in higher performance of the firm. This relation would be moderated by market uncertainty. With a high market uncertainty, the learning strategies result in only small to no differences in firm performance, whereas low market uncertainty results in significant performance differences in learning strategies. Consecutively, the effect of market uncertainty is moderated by industry selection, as a small amount of industry selection results in higher performances and a more noticeable difference between the strategies. As the inverted-u shape relation of industry selection indicates, when the competition is too harsh, it will hurt industry performance. Overall, when firms shift their attention from own performance to the average performance of others, the performance will increase to some extent, depending on the market uncertainty.
new innovations within the company have effect on the performance of the company. In other words, it defines the speed with which a company is able to adapt. Another change is that in all simulations the historical strategy sets future performance goals higher than the current performance. The study of Bromiley (1991) points out that firms tend to set higher targets than their current performance. Therefore, in all the simulations a 5% increase to the historical performance is added to determine the aspiration level. Finally, as it is unlikely that a decision-maker has information about all firms in the industry, the reference group that is used when calculating the social average is a subset of the industry. The focal firm only calculates the average performance of a subset of its closest competitors. All these changes are added to make the model more realistic and therefore more applicable for managers. Agent-based modelling is still in its infancy, but there are numerous possibilities how agent-based modelling can be advantageous for studying the innovation process (Garcia, 2005). This study shows that agent-based modelling can result in unexpected new insights in innovation decision making.
Practical Implications
A few practical implications can be derived from the results as well. First, a social learning strategy that is based on the performance of close competitors is more likely to result in higher performance for the organization. This is especially true for firms that compete in a stable market and when the firm quickly adopts to new environments. When more uncertainty is introduced in the model, then the effect of different learning strategies gradually disappears. This effect is reduced, however, when the industry replaces poorly performing organizations with new companies.
When decision-makers are currently using own historical performance or when decision-makers switch between own performance and others’ performance (mixed learning strategy), then the results do implicate that substantial performance improvements could be realized by paying more attention to close competitors in the industry. Even a 50% shift in attention would pay off significantly, as can be seen in the simultaneous learning strategy.
Limitations
Like other research methods, simulation does have its down sides. First, the results solely depend on the output of a computer program. It is very hard to reconstruct the code from a computer program for other scholars to reconstruct the study. Even if the codebase of the project is similar, different results could be attained (Harrison et al., 2007). This is due to the complete environment in which the computer program is executed. To remedy this to some extent, the most important parts of the source code of the simulations have been included in this paper (appendix D) and a detailed flowchart is used to depict the sequence of actions the computer program performs (appendix A). Moreover, small mistakes (bugs) can result in major result differences. This study has executed well over 275 simulations in order to correct mistakes or to add improvements to the model. This number is exclusive the (singular) test runs during development of the computer program. Finally, simulation models do tend to simplify reality to a large extend in order to be able to simulate them efficiently. Due to this simplification, more research is necessary to establish a useful theoretical founded framework.
Future Research Directions
The next step in this research project should be to delve into whether there actually is a mediating effect on the social learning strategy and firm performance through exploration and to what extend that effect explains the performance increase. This study found that industry selection plays a significant role in innovation-decision making, as it moderates the effects of technology path dependency and market uncertainty. However, it is still unclear how industry selection moderates these effects. Future research should try to find answers to these questions.
References
Argote, L., & Greve, H. R. (2007). A Behavioral Theory of the Firm—40 Years and Counting: Introduction and Impact. Organization Science, 18(3), 337–349.
Barreto, I. (2011). A Behavioral Theory of Market Expansion Based on the Opportunity Prospects Rule. Organization Science, 23(4), 1008–1023.
Bell, J. (1970). Production strategy decisions a simulation model approach. Long Range Planning,
2(4), 62–73.
Berends, P., & Romme, G. (1999). Simulation as a research tool in management studies. European
Management Journal, 17(6), 576–583.
Blettner, D. P., He, Z., Hu, S., & Bettis, R. A. (2014). Adaptive aspirations and performance heterogeneity: Attention allocation among multiple reference points. Strategic Management
Journal, 36(7), 987–1005.
Bromiley, P. (1991). Testing a causal model of corporate risk taking and performance. Academy of
Management Journal, 34(1), 37–59.
Bromiley, P., & Harris, J. D. (2014). A comparison of alternative measures of organizational aspirations. Strategic Management Journal, 35(3), 338–357.
Brown, L. (2010). Balancing Risk and Innovation to Improve Social Work Practice. The British
Journal of Social Work, 40(4), 1211–1228.
Chen, W.-R., & Miller, K. D. (2007). Situational and Institutional Determinants of Firms’ R&d Search Intensity. Strategic Management Journal, 28(4), 369–381.
Cyert, R. M., & March, J. G. (1963). A behavioral theory of the firm. In Organizational Behavior
2: Essential theories of process and structure (Vol. 2, pp. 169–187). NJ: Englewood Cliffs.
Davis, J. P., Eisenhardt, K. M., & Bingham, C. B. (2007). Developing theory through simulation methods. Academy of Management Review, 32(2), 480–499.
Under Technological and Market Uncertainty, (Under peer review).
Fang, C., Kim, J.-H. (Jason), & Milliken, F. J. (2014). When bad news is sugarcoated: Information distortion, organizational search and the behavioral theory of the firm. Strategic Management
Journal, 35(8), 1186–1201.
Fiol, C. M., & Lyles, M. A. (1985). Organizational learning. Academy of Management Review,
10(4), 803–813.
Gaba, V., & Bhattacharya, S. (2012). Aspirations, innovation, and corporate venture capital: A behavioral perspective. Strategic Entrepreneurship Journal, 6(2), 178–199.
Garcia, R. (2005). Uses of agent-based modeling in innovation/new product development research.
Journal of Product Innovation Management, 22(5), 380–398.
Gavetti, G., Greve, H. R., Levinthal, D. A., & Ocasio, W. (2012). The behavioral theory of the firm: Assessment and prospects. Academy of Management Annals, 6(1), 1–40.
Greve, H. R. (2002). Sticky aspirations: Organizational time perspective and competitiveness.
Organization Science, 13(1), 1–17.
Greve, H. R. (2003). A Behavioral Theory of R&D Expenditures and Innovations: Evidence from Shipbuilding. The Academy of Management Journal, 46(6), 685–702.
Harrison, J. R., Lin, Z., Carroll, G. R., & Carley, K. M. (2007). Simulation modeling in organizational and management research. Academy of Management Review, 32(4), 1229–1245. Hu, S., Blettner, D., & Bettis, R. A. (2011). Adaptive aspirations: performance consequences of risk preferences at extremes and alternative reference groups. Strategic Management Journal,
32(13), 1426–1436.
Kirchoff, J. F., Omar, A., & Fugate, B. S. (2016). A Behavioral Theory of Sustainable Supply Chain Management Decision Making in Non-exemplar Firms. Journal of Supply Chain
Management, 52(1), 41–65.
Levinthal, D., & March, J. G. (1981). A model of adaptive organizational search. Journal of
Nicolas, R. (2004). Knowledge management impacts on decision making process. Journal of
Knowledge Management, 8(1), 20–31.
O’Brien, J. P., & David, P. (2014). Reciprocity and R&D search: Applying the behavioral theory of the firm to a communitarian context. Strategic Management Journal, 35(4), 550–565.
Papadakis, V. M., Lioukas, S., & Chambers, D. (1998). Strategic decision-making processes: the role of management and context. Strategic Management Journal, 19(2), 115–147.
Rudy, B. C., & Johnson, A. F. (2013). Performance, aspirations, and market versus nonmarket investment. Journal of Management, 42(4), 936–959.
Schimmer, M., & Brauer, M. (2012). Firm performance and aspiration levels as determinants of a firm’s strategic repositioning within strategic group structures. Strategic Organization, 10(4), 406– 435.
Selten, R. (1999). Bounded rationality. Journal of Institutional and Theoretical Economics
(JITE)/Zeitschrift Für Die Gesamte Staatswissenschaft, 146(4), 649–658.
Tyler, B. B., & Caner, T. (2016). New product introductions below aspirations, slack and R&D alliances: A behavioral perspective. Strategic Management Journal, 37(5), 896–910.
Appendices
Appendix A: Flow chart of the simulation
Sample parameters are send to server
Sample data stored in database Array with the size of the sample size is
created and filled with sample subjects
Sample array is sorted based on performance
Grabs closest competitors and calculates their average performance
Company restarts with new start values
Company innovates
𝑇),/= 𝛼𝑇),/34∗ (1 − 𝛼)𝜇
Company calculates its new performance
𝑃),/= 𝑇),/+ 𝛾𝜇
Company derives new aspiration from current position
𝐴),/= 𝜑𝐴),/34+ (1 − 𝜑) ∗ (𝑤 ∗ (𝑃),/+ (𝑞 ∗ 𝑃),/)) + (1 − 𝑤)𝐶),>)
Calculate and save mean of sample set Add sample averages of iteration to overall mean
Store overall means in the database
Appendix B: Parameter Robustness
Appendix C: Mixing the samples
Matrix 6: This matrix shows the effect of combining all the strategies in one sample.
Appendix D: Source code for the simulation
The simulations make use of the PHP Laravel Framework (https://laravel.com/, version 5.4.36). This framework enables the use of programmable models that contain the information for the simulation and their connection to the database. These three files are the core of the simulations run in this study.
SimulationController.php: <?php namespace App\Http\Controllers; use Illuminate\Http\Request; use Illuminate\Support\Facades\Input; use Illuminate\Support\Facades\Storage; use App\Simulation;
class SimulatorController extends Controller {
private $bResponse = false;
public function index()
{
return view('simulator/content');
}
public function simulate()
{
$debug_array = array();
ignore_user_abort(true);
$simulation = request()->all();
/**
* Store the simulation in the database */
$sim = new Simulation;
$sim->Handle = $simulation['id'];
$sim->Samplename = $simulation['name'];
$sim->Timespan = intval($simulation['timespan']);
$sim->Iterations = intval($simulation['iterations']);
$sim->Samples = json_encode($simulation['samples']);
$sim->Current = 0;
$sim->Time = 1;
$sim->Status = "Pending";
$sim->Result = json_encode([]);
$sim->save();
$this->flush(null, false, ['Simulation received: '.$simulation['name']]);
Simulation::where('Status', '=', 'Cancelled')->delete();
while (Simulation::where('Status', '=', 'Simulating')->get()->count() >= 4) { sleep((60*2));
set_time_limit(60);
$sim->fresh();
if ($sim->Status === 'Cancelled') { $sim->delete();
return;
} }
$sim->Status = "Simulating";
$sim->save();
$iterations = $sim->Iterations;
for ($j = 0; $j < $iterations; $j++) { $mean = array();
$subjects = array();
/**
* Create an array with new subjects based on the given "Object". */
$simulSamples = json_decode($sim->Samples);
for($k=0; $k < count($simulSamples); $k++) { for ($i=0; $i < $simulSamples[$k]->size; $i++) {
$sampleObject = "App\Mockobjects\\".$simulSamples[$k]->object;
} }
for($t=0; $t < $sim->Timespan; $t++) { set_time_limit(60);
$sim->Time++;
$sim->save();
/**
* Sort the objects from low performance to high performance. */
usort($subjects, function ($a, $b) {
if ($a->getPerformance() > $b->getPerformance()) { return 1;
} elseif ($a->getPerformance() < $b->getPerformance()) { return -1; } else { return 0; } }); /**
* Loop over all the subjects and iterate the following actions: * 1: Set time, so it will grab the correct parameterset
* 1.5 Check if company is below the dropOut rate, if so, restart the company * 2: Grab the peerset from the sampleset
* 3: Calculate and set the peer average for the subject * 4: Check if performance is lower than aspirations * 4.5: If so, innovate
* 5: Update the performance based on techstatus and market uncertainty. * 6: Adapt aspirations based on new tech status and current time *
*/
foreach ($subjects as $i => $company) { $company->setTime($t);
// To win speed, it does not calculate the peer average if it is not used in the calculation.
if ($company->get("historySocialBalance") != 1 || $company->get("mixedLearning") == 1) { $peers = $this->getPeers($subjects, $i,$company->get('peerSpread'), $company);
$avg_new = $this->averages($peers, false);
$company->setPeerAverage($avg_new['performance']);
}
if ($i < round($company->get('dropOutrate') * count($subjects))) { $company->restart();
} else {
if ($company->getPerformance() < $company->getAspirationlevel()) { $company->innovate(); } } $company->calcPerformance(); $company->adaptAspirations(); }
$means = $this->averages($subjects, true);
foreach ($means as $groupId => $metricList) { if (!isset($mean[$groupId])) {
$mean[$groupId] = array();
}
foreach ($metricList as $metric => $value) { if (!isset($mean[$groupId][$metric])) { $mean[$groupId][$metric] = array();
}
array_push($mean[$groupId][$metric], $value);
} } } /**
* Add the mean results from this simulation to the overall average resultset. */
foreach ($mean as $groupId => $metricList) { $result[$groupId] = $metricList;
} } else {
foreach ($result as $groupId => $metricList) { foreach ($metricList as $metric => &$valueArray) { for ($l=0;$l<count($valueArray);$l++) {
$valueArray[$l] = $valueArray[$l] + (($mean[$groupId][$metric][$l] - $valueArray[$l])
/ (count($valueArray)+1)); } } unset($valueArray); } } ksort($result);
$result = array_values($result);
/**
* Reset to start values for next simulation and store progress */
$sim->Time = 1;
$sim->Current++;
$sim->Result = json_encode($result);
$sim->save();
$sim->fresh();
/**
* If status of simulation has changed to 'cancelled', stop the simulation. */
if (strtolower($sim->Status) === 'cancelled') { return;
} }
$sim->Status = "Ended";
$sim->save();
}
private function getPeers($subjects, $current, $spread, $self = null)
{
// Get total number of subjects
$countsubjects = count($subjects);
// Ensure that the spread is not bigger than the number of subjects
$spread = min($spread, $countsubjects);
// Split the spread in half to get bottom and top half
$spreadhalf = floor($spread / 2);
$min = max(0, $current-$spreadhalf - (($current + $spreadhalf > $countsubjects) ? $current + $spreadhalf -
$countsubjects + 1 : 0));
$subset = array_slice($subjects, $min, ($spread+1));
// Remove the company itself from the subset
foreach ($subset as $key => $peer) { if ($peer->id() === $self->id()) { unset($subset[$key]); break; } } return $subset; }
private function averages($subjects, $groupById = false)
{
$totalMetrics = [];
$averages = [];
$groupCount = [];
foreach ($subjects as $i => $subject) { $subjectMetrics = $subject->toMetrics();
$groupId = ($groupById) ? $subject->getGroupId() : 0;
$subjectMetrics = (is_array($subjectMetrics)) ? $subjectMetrics: [$subjectMetrics];
if (!isset($totalMetrics[$groupId])) { $totalMetrics[$groupId] = array();
}
if (!isset($totalMetrics[$groupId][$metric])) { $totalMetrics[$groupId][$metric] = 0; } $totalMetrics[$groupId][$metric] += $value; } if (!isset($groupCount[$groupId])) { $groupCount[$groupId] = 0; } $groupCount[$groupId]++; }
foreach ($totalMetrics as $groupId => $metricList) { $averages[$groupId] = array();
foreach ($metricList as $metric => $totalValue) {
$averages[$groupId][$metric] = $totalValue / $groupCount[$groupId];
} }
return ($groupById) ? $averages : $averages[0];
}
private function average($subjects, $method, $groupBy = false)
{
if (is_array($method)) { $totals = [];
$result = [];
foreach ($subjects as $i => $subject) { foreach ($method as $m) {
if ($groupBy) {
if (!isset($totals[$subject->getGroupId()])) { $totals[$subject->getGroupId()] = array();
}
if (!isset($totals[$subject->getGroupId()][$m])) { $totals[$subject->getGroupId()][$m] = 0;
}
$totals[$subject->getGroupId()][$m]+= $subject->$m();
} else { if (!isset($totals[$m])) { $totals[$m] = 0; } $totals[$m]+= $subject->$m(); } } } foreach ($method as $m) {
$result[$m] = $totals[$m] / count($subjects);
}
return $result;
} else { $total = 0;
foreach ($subjects as $i => $subject) { $total+= $subject->$method();
}
return $total / count($subjects);
} }
private function flush($sim = null, $quit = false, ...$args)
{
if (!$this->bResponse) {
$response = json_encode($args);
header("Connection: close");
header("Content-Length: " . mb_strlen($response));
echo $response;
flush();
$this->bResponse = true;
if (is_object($sim)) { $sim->forceDelete();
}
} } } }
The Mockobject is the abstract class that contains the base logic for every object is studied in this paper (just one: the company object) or that might be studied in the future.
Mockobject.php:
<?php
namespace App\Mockobjects;
abstract class MockObject {
protected $parameters;
protected $id;
protected $time;
public function __construct(array $parameters = array(), $id = null, $groupId = 0)
{
$timeperiods = [];
$this->parameters = [];
$this->id = $id;
$this->groupId = $groupId;
foreach ($parameters as $i => $parameterset) { $this->timeperiods[] = $parameterset->_startAt;
$this->parameters[$parameterset->_startAt] = $parameterset;
} }
abstract public function toMetrics();
public function id()
{
return $this->id;
}
public function getGroupId()
{
return intval($this->groupId);
}
public function get($parameter)
{
return floatval($this->parameters[$this->timeperiod]->{$parameter});
}
public function randomNormalDistribution()
{ $u = 0; $v = 0; while ($u===0) { $u = mt_rand() / mt_getrandmax(); } while ($v===0) { $v = mt_rand() / mt_getrandmax(); }
return sqrt(-2.0*log($u)) * cos(2.0 * pi() * $v);
}
public function setTime($t)
{
$this->time = $t;
foreach ($this->timeperiods as $tp) { if ($t >= $tp) {
}
return $this;
} }
Finally, the company object contains the metrics and calculations that each company makes during the simulation.
companyObject.php:
<?php
namespace App\Mockobjects;
use App\Mockobjects\Mockobject;
class Company extends Mockobject
{
private $fTechstatus;
private $fAspiration;
private $fPerformance;
private $fPeeraverage;
public function __construct($arguments, $id = null, $groupId = 0)
{
parent::__construct($arguments, $id, $groupId);
$this->fTechstatus = [$this->randomNormalDistribution()];
$this->fAspiration = [$this->randomNormalDistribution()];
$this->fPerformance = [$this->randomNormalDistribution()];
$this->fPeeraverage = [];
}
public function innovate()
{
$inv = $this->get("innovationAdaptionSpeed");
array_push($this->fTechstatus, $inv * $this->getTechstatus() + (1-$inv) * $this ->randomNormalDistribution());
}
public function getTechstatus(int $t = null)
{
return ($t === null) ? end($this->fTechstatus) : $this->fTechstatus[$t];
}
public function getPerformance(int $t = null)
{
return ($t === null) ? end($this->fPerformance) : $this->fPerformance[$t];
}
public function setPeerAverage($fPeeraverage)
{
array_push($this->fPeeraverage, $fPeeraverage);
return $this;
}
public function getPeerAverage(int $t=null)
{
return ($t === null) ? end($this->fPeeraverage) : $this->fPeeraverage[$t];
}
public function calcPerformance()
{
$performance = ($this->getTechstatus() + ($this->get("marketUncertainty") * $this ->marketuncertainty()));
array_push($this->fPerformance, $performance);
public function adaptAspirations()
{
$a = $this->get("aspirationAdaptionSpeed");
$w = $this->get("historySocialBalance");
$q = $this->get("historyImprovement");
$peeraverage = $this->getPeerAverage();
if ($this->get("mixedLearning") != 0) {
$w = ($this->getPerformance() > $peeraverage) ? 1 : 0;
}
$asp = $a * $this->getAspirationlevel() + (1 - $a) * (($w * ($this->getPerformance() + ($q *
abs($this->getPerformance())))) + (1 - $w) * $peeraverage);
array_push($this->fAspiration, $asp);
}
public function getAspirationlevel(int $t = null)
{
return ($t === null) ? end($this->fAspiration) : $this->fAspiration[$t];
}
public function marketuncertainty()
{
return $this->randomNormalDistribution();
}
public function checkRank($rank, $samplesize)
{
return $rank > floor($this->get('dropOutrate') * $samplesize);
}
public function toMetrics(int $t = null)
{
return [
'performance'=> $this->getPerformance($t),
'techstatus' => $this->getTechstatus($t),
'aspiration' => $this->getAspirationlevel($t),
'peerperformance' => $this->getPeerAverage($t)
];
}
public function restart()
{
array_push($this->fTechstatus, $this->randomNormalDistribution());
array_push($this->fAspiration, $this->randomNormalDistribution());
