Amsterdam Business School
The impact of the Controllability and the possibility to create
Slack in presented information on the Target Setting Process
Name: Robbert de Jager Student number: 10851178 Thesis supervisor: dr. Peter Kroos Date: 20th of June 2016
Word count: 12976
Amsterdam Business School MSc Accountancy and Control
Statement of Originality
This document is written by student Robbert de Jager who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Acknowledgements
First of all, I would like to thank my supervisor dr. Peter Kroos for his supervision and guidance during the research. His expertise in the field of target setting helped me with the theoretical constructs and designing the experiment. Furthermore I would especially thank my family supporting me through all these years of studying. Also I would like to thank my friends who supported me through the difficult times writing this thesis. At last I would like to thank three people in particular with helping me with the statistical analysis: Daan Willers, Patrick Doornekamp and Frieda Hartoog. They helped me structuring my thoughts about the analysis.
Abstract
The development of setting a target is affected by several factors. This research examines the impact of the controllability of a performance measure and the possibility to create slack in the presented information. In the literature was described that business unit managers face a higher target when they have the possibility to slack the information and that they get a higher revised target when there is a high controllability of a performance measure (for example no noise). These findings are tested in this research in a two by two scenario-based survey experiment. The aforementioned findings were formulated in two hypotheses. The results were contradictory to what was expected based on the literature. Based on the results, the controllability of a measure doesn’t seem to impact the revised targets, while the place where the sales are registered gives opposing results to what was expected. This means that the revised targets are higher when the sales are registered centrally rather than when they are registered locally.
Contents
1 Introduction ... 8
1.1 Background ... 8
1.2 Research question ... 9
1.3 Relevance of the research question ... 10
1.4 Structure of this research ... 10
2 Literature review and hypothesis ... 11
2.1 Target setting and performance evaluation ... 11
2.2 Development of targets ... 12
2.3 Difficulty in target setting ... 12
2.4 Target ratcheting concept ... 13
2.4.1 Target ratcheting ... 13
2.4.2 Ratchet effect ... 14
2.4.3 Using knowledge of employees lower in the organization ... 14
2.5 Commitment of managers ... 15
2.6 Hypothesis development ... 16
2.6.1 Budgeting slack ... 16
2.6.2 Controllability of a performance measure ... 18
3 Research design ... 20
3.1 Conceptual research model ... 20
3.1.1 Scenario-based survey experiment ... 20
3.1.2 Manipulations ... 21
3.2 Procedures ... 22
3.3 Post manipulation checks ... 22
3.4 Sample characteristics ... 23
4.1 Descriptive statistics ... 25
4.2 Univariate analysis of variance ... 25
4.3 Correlations ... 27
4.4 Regression analysis ... 28
4.4.1 Test assumptions ... 28
4.4.2 Outcomes regression analysis ... 30
4.5 Additional analysis ... 32
4.5.1 Additional test; Free of distribution ... 32
4.5.2 Additional test; regression analysis with full sample ... 33
5 Discussion and conclusion ... 35
5.1 Findings ... 35
5.2 Discussion and limitations ... 35
5.3 Conclusions ... 37
References ... 38
Appendices ... 42
Appendix I: Case descriptions ... 42
Instructions ... 42
Background ... 42
Dealer-specific circumstances case 1 ... 42
Dealer-specific circumstances case 2 ... 43
Dealer-specific circumstances case 3 ... 43
Dealer-specific circumstances case 4 ... 44
Questionnaire (For each case exactly the same) ... 44
Appendix II: Flowchart working sample ... 46
Appendix III: Interaction effects ANOVA ... 47
Appendix VI: Graphs for testing linearity ... 50
Appendix VII: Homoscedasticity test ... 51
List of tables ... 52
1 Introduction
1.1 Background
Many firms use performance measurement and compensation systems as a crucial vehicle for the daily decision making and control activities within their firms. Performance evaluation and compensation systems are much researched. Less research has been devoted to the topic target setting, despite the fact that target setting is an important component of performance evaluation and compensation. Much research on target setting has focused on target ratcheting. This refers to a practice where next-period targets are revised on the basis of the actual performance vis-à-vis a target in the current period (Indjejikian and Nanda, 2002). That is, targets are revised upward when the current performance exceeds the target and vice versa.
Research on target setting has also devoted attention to the adverse consequences of target ratcheting. That is, when managers exceed the performance in the current period, they are penalized by means of higher targets in the subsequent period (Indjejikian, Matĕjka and Sloetzer, 2014; Leone and Rock, 2002). Murphy (2001) finds that employees are incentivized to alter their performance when new targets are based on prior performances. Bouwens and Kroos (2011) show that managers that have a favourable year-to-date performance are more likely to slow down their performance at the end of the year to lower their target increase in the next period. Bouwens and Kroos show that managers do this through real economic decisions. For example, a manager slows down his sales performance in order to influence the next year’s target. The financial incentive for quarterly bonus pay-outs are not strong enough to prevent the ratchet effect. However, Indjejikian et al. (2014, p. 1260) find that contractual commitments to disregard or deemphasize information about past performance in target setting may alleviate the ratchet effect. Indjejikian and Matĕjka (2006) show how firms may use less of the current performance information in revising the current period target in an attempt to reduce this so-called ratchet effect. In this way, managers who report a favourable performance are rewarded by smaller subsequent target increases.
This thesis aims to examine to what extent current performance information relative to the target is used to determine the next-period target update. For example, firms can strongly increase the target following a favourable target deviation or firms can decide to only partially increase the target after a favourable deviation. My specific goal is to examine the determinants underlying that decision.
First, targets must be challenging but attainable. That is, they must be effort-inducing. So, targets that are set too high are not motivating as the managers involved will perceive them as
‘out-of-the-money’. An important factor in this case is the controllability of the measure. Specifically, when factors outside the control of a manager (e.g., the economic climate) take a bad turn, this will negatively affect the performance and can render the target as unachievable. So, one might expect that managers are given less challenging targets when they operate in a more volatile environment. This volatile environment is referred to as noise. Baker and Jorgensen (2003) describe noise as the uncertainty to which the manager should not react by changing his actions. Aside from the economic climate, there are other factors that are outside the control of a manager, for example, decisions taken by other managers in the company or division and decisions taken by other seniors in the rank (Giraud, Langevin and Mendoza, 2008).
Second, the information underlying the performance measures can be generated under the supervision of the firm or are generated at the lower-level. The latter case implies that lower-level managers may influence the numbers that are subsequently used for target setting. That is, lower-level managers might build slack into the reported performance numbers that subsequently may lead to easier next-period targets for those managers. Anderson, Dekker and Sedatole (2010) stated that although the most of the real knowledge about the company’s processes and performances is lower in the organization, employees take their chances to create slack in the target setting process. That is, when slack is successfully created and it is not corrected for in target setting, this will lead to lower targets in the subsequent period.
1.2 Research question
For this research, I will focus on two determinants in the target setting process. The first determinant will be the controllability of a performance measure. The controllability of a performance measure is described by Bol and Smith (2011) as the informativeness of a measure about the actions and effort of an employee. Hölmstrom (1979) found that controllable (objective) measures are preferred form a contracting perspective because uncontrollability reduces the informativeness of a measure. Also, managers need to be compensated for the additional risk of uncontrollable factors. However, in practice is shown that completely controllable factors are very rare (Bol and Smith, 2011). This means that there always will be some noise in target setting and performance evaluation. Measures are therefore never completely controllable.
The second determinant which fulfil a central role in this research is how sensitive a performance measure is to manipulation. For example, the business unit manager delivers information about the performance to the corporate headquarters. However, Hartman and Maas (2010) find that managers have incentives to bias the information to maximize the expected value of the payoffs from the target setting process. When this information is biased, corporate managers
and owners cannot make an accurate estimation of the performance potential of the business unit. This means that the business unit manager has the possibility to create slack in the information and in the target setting process. This research will seek to find how much the influence of biased information has on the target setting process.
Both determinants will be examined in the controlled setting of an experiment. Based on aforementioned description, the research question is formulated as follows:
‘’How does the controllability of a measure and the susceptibility to manipulation of a measure affect the degree in which the current performance on that measure is incorporated in the subsequent target revision on that measure?’’
1.3 Relevance of the research question
This research is interesting from an academic perspective as well from the social perspective. It contributes in several ways. First, much research looked at performance measurement, performance evaluation and incentive compensation systems. Less research focused on target setting.
Much on the literature on target ratcheting has focused on documenting the adverse consequences of target ratcheting (e.g. Bouwens and Kroos, 2011). Some authors propose that firms use less of the current performance information as a way to mitigate the tendency of managers to slow down performance. I contribute by examining which considerations companies may face when they decide to use more or less performance information in setting next period targets. Little research up to now has explored this.
Given the prevalence of performance evaluation and target setting in organizations, this study features a societal contribution as it shows how targets are updated in the presence of high uncontrollability and high susceptibility to manipulation of the underlying performance measure.
1.4 Structure of this research
For the remainder of this research the structure of this research will be as follows. Chapter two contains the literature review which displays the theoretical constructs, concepts and the hypothesis development. Chapter three contains the research methodology which explains the requirements for the participants, the case design and the procedure during the experiment. Chapter four will review the results of the experiment. Chapter five is the conclusion and there will be discussion of limitations and potential follow-up research.
2 Literature review and hypothesis
2.1 Target setting and performance evaluation
Target setting and performance evaluation are two crucial components for the daily operations of a company which cannot be seen separately from each other. There are several ways to measure the performance of, for example, an employee, a division or a product. According to Murphy (2000), in most cases the performance is evaluated based on a comparison with the target. Therefore, setting a (accurate) target is of great importance to evaluate a performance in a proper way.
A target has two main functions within a company: a motivational function and a coordinating function (Macintosh and Daft, 1987). First there are the coordinating activities. It is for coordinating purposes important that the financial records and information are realistic and accurate when making the budgets for the purchasing, selling and production cycles. When those budgets are not accurate, the actual results will differ from the budgets. Budgets are according to Réka, Ştefan and Daniel (2014) and Wetherbe and Dickson (1979) rigid, costly, time consuming and not suitable for setting targets. Though, many companies are still using the budget for the coordinating functions in the company.
Second there is the motivational aspect of target setting. A target is for many people an incentive to work harder and obtain a bonus with the achievement of a target. Gómez-Miñambres (2012) find that targets do not only incentivize individuals, but also when targets are set for a group. Gómez-Miñambres (2012) find evidence that performance increase significantly when targets are set for groups. Biggs (1978) and McConnel, Sherman and Hamilton (1997) find that target setting in a group works as peer pressure and the group performance will increase. Though, they state that this is partly dependent of the personal qualities and institutional environment.
Targets are set by entities based on information they collect. This information can be collected from independent organisations (for example sector analysis by Rabobank) or inside the corporation. Sometimes this information is available at lower levels of the organization. That is, entities are dependent of the information that is delivered by the lower level managers. Specifically, this means that lower level managers have some degree of control over the information. This is where an agency conflict could arise.
2.2 Development of targets
Not only the type of target is important, but also the development of the target is important. Companies are often using internal resources to set the targets. They often use one of the three following ways: I) use of earlier obtained results, II) using the performance of peers (comparable business units, stores or employees) or III) using knowledge that resides with employees lower in the organization. The way of target setting is chosen not only by those three factors, but also depends on the availability, costs, reliability and comparability of the objective/subjective targets. Most important in setting the target is that the performance of the employee best shows according to Bol and Lill (2012).
Bouwens and Kroos (2016) predict that forward-looking information will be used in addition to set next-period targets1. This implies that not only historical performance on the
respective measure can be used, but that also the prior performance on other measures may influence the target revision on a specific measure. They found evidence that, after controlling for current sales performance, business unit managers with higher current nonfinancial performance scores report higher future sales. They also find that forward-looking information is used with the objective sales information to set subsequent sales targets.
2.3 Difficulty in target setting
Latham and Locke (1991) researched goal setting theory. According to them is the goal setting theory based on the simplest of introspective observation, the conscious human behaviour is purposeful. This is mainly shaped by the goals of individuals. However, they state that goal-directed action is not restricted to conscious action.
One finding of Meyer and Gellatly (1988) is that with regard to goal content is that specific and challenging or difficult goals lead to a higher level of performance than vague but challenging goals. The consistent superiority of the former is, according to Latham and Locke, 1991) that vague goals are compatible with many different outcomes. Kernan and Lord (1989) find for example that individuals with no specific goal generally evaluated their performance more positively than those with specific and hard goals in response to varying degrees of negative feedback. They find that maximum effort is not aroused under a do best goal. Furthermore, Latham and Locke (1991) find that goal specificity as such affects the variability of performance.
When controllability is assumed, people with very specific goals show less variation in performance that people with vague goals.
Therefore, it is difficult for every organization to set optimal targets. Choi, Kim and Merchant (2013) find that targets should be challenging, but obtainable. When targets are challenging, but obtainable, employees are incentivized to put more effort in achieving the target. Specifically, employees put more effort in achieving the target when there is a bonus attached to achieve the target. A bonus does not necessarily need to be a financial incentive.
When a target is too difficult to achieve, this can cause lack of interest by the employee (Delfgaauw, Dur, Non and Verbeke, 2014). The employee will not maximize their effort which could harm the company value in the long run. According to Delfgaauw et al. (2014) this works also the other way around. When targets are set too low, the employee will lack effort because they know that they will achieve the target. This means that it is for an organisation very difficult to find the balance in setting targets that are challenging and obtainable. Merchant and Mazoni (1992) find that more than 70 percent of the managers achieves the targets that are set. Therefore they concluded that companies choose to set more achievable targets instead of setting more challenging targets. Webb, Williamson and Zhang (2013) argue that while challenging targets and target-based pay can hinder the discovery of production efficiencies, they can motivate high productive effort whereby individuals work harder and more productively using either the conventional task approach or more efficient task approaches when discovered.
2.4 Target ratcheting concept
2.4.1 Target ratcheting
A common way to set targets for the coming period is using earlier obtained results according to Murphy (2000). The results of the past year(s) are analysed to set the new targets for the coming year. The target setter is always looking at why the performance was good or bad. Bol and Lill (2012) find that a good performance normally is caused by the effort of the manager, noise or a target that is too low. Bol and Lill (2012) states that it is difficult to determine which of those three factors is of the greatest influence in the performance. Normally, the target should fluctuate every year with good or less good performances.
Leone and Rock (2002) document how business units’ targets are ratcheted. That is, current financial performance relative to the target is used to set the subsequent period target. Likewise, Bouwens and Kroos (2011) show how sales targets for retail stores are updated based on prior sales performance. Both studies also show how targets are ratcheted asymmetrically. This
means that targets are increased more after favourable performance compared to the decrease in targets following unfavourable performance. Targets are typically ratcheted consistent with the intuitive nature of setting targets in this way. However, ratcheting targets may also give rise to the ratchet effect.
2.4.2 Ratchet effect
It was therefore described that many firms use prior performance to set next-period targets, commonly described as target ratcheting. However, this also gives rise to the ratchet effect where managers trade-off the benefits of current performance against the costs of higher future targets. It is argued that to prevent higher targets in the future, managers may withhold performance. Consistent with this, Bouwens and Kroos (2011) show how store managers with a favourable year-to-date performance report a lower end-of-year performance. In this way, it is argued do managers still cash their bonuses for the current year, but reduce the target increase for the next year. This increases their odds for making the targets and bonuses in the subsequent year. It is argued that the ratchet effect represent a cost when firms decide to ratchet their targets.
2.4.3 Using knowledge of employees lower in the organization
Anderson, Dekker and Sedatole (2010) find that not every company can use the participation of employees in target setting. This is, employees take their chances to create slack in the target setting process which will lead to lower target. This will lead to the achievement but not outperformance of targets. Anderson et al. (2010) find though that the most of the real knowledge about processes and performances is lower in the organisation. Specifically, this means that not all of the information will be shared with the corporate headquarters.
Charness, Kuhn and Villeval (2010) do not find it surprising that employees try to influence targets in a negative way. They found in their research that people always will choose the easier target. Specifically, individuals need to be incentivized to choose the more difficult target. This creates a dual role for business unit managers. On the one side they want to deliver unbiased information to get as accurate targets which are challenging, but still achievable. On the other side this also creates incentives to lower their targets so they can easier obtain the bonus and maybe make a bigger bonus. The business unit manager will therefore always keep essential information for themselves. As long as these targets are a part of the bonus system, the business unit manager will give biased information to the target setters (Leone and Rock, 2002).
2.5 Commitment of managers
The agency theory suggests that an agent is capable of engaging in dysfunctional behaviours known as adverse selection and moral hazard (Chong and Eggleton, 2007). This behaviour arises when the interests of the owners do not align with the interests of the employee. A clear example of such behaviour is the ratchet effect where agents hide the true performance potential of their units by withholding performance at the end of the year. The agency theory predicts that the principal can minimize the moral hazard effect by having an incentive-based compensation scheme which aligns the interests of principal and agent.
Indjejikian et al. (2014) find that when moral hazard and adverse selection are present, the standard contracting theory prescription calls for arrangements where a high-profitability managers earn informational rents in return for high effort and vice versa. Indjejikian et al. (2014, p 1230) state the following: ‘’Informational rents are a necessary ingredient in such separating contracts because otherwise managers in a high-profitability environment would be reluctant to truthfully reveal this information’’. This shows that the commitment of a manager is dependent of the way the information they delivered is used. This important information is often held privately, because managers can shirk and misrepresent the profitability as perceived low. Firms are better off if they commit to reward managers that truthfully report their profitability and provide high effort.
When compensation contract are not specified until the start of the period, this is described as no-commitment contracts (Indjejikian et al., 2014). No-commitment contracts are sequentially optimal from a company’s perspective in the way that they incorporate all relevant information. This may lead to an incentive conflict because managers anticipate that being successful today, may lead to lower compensation in a subsequent period. Therefore, companies find it difficult to distinguish between high-profitability and low-profitability managers and may therefore opt for pooling arrangements, where both managers are offered a comparable compensation contract.
On the other hand there are full commitment contracts. Full commitment implies that a company and its managers enter into a long-term employment contract with a commitment not to renege on any of its terms during the contract. Full-commitment contracts should increase the range of incentives because any no-commitment contract is still achievable. Specifically, this means that companies guarantees that information about profitability in the current period is not going to be exploited in future periods. This has as consequent that full-commitment contracts are ‘‘separating’’ in the sense that high-profitability managers work hard and earn informational rents that persist through time, while low-profitability managers work less and earn little or no rents.
Indjejikian and Nanda (2002) find that companies increase the likelihood that business unit managers provide high effort and faithfully represents their performance potential. These companies should not exploit all available information in the next-period target update and reward high performers this way (Bouwens and Kroos, 2016). The commitment to make limited use of available information in target setting should decrease the likelihood that managers misrepresent their capacity. Aranda, Arellano and Davilla (2014) find consistent with their expectations that business units that perform well relative to peers, targets are less likely to be revised upward after better performance than targeted. Bol and Lill (2015) show that past performance is partially incorporated in future targets as part of an implicit agreement where the principal allows the agent to receive economic rents from positive target deviations that result from superior effort.
The way of evaluating performances and target setting gives the target setter more information about the economical capacity of the business unit or store. Another advantage with using peer performance is that good performing business units are ‘rewarded’ in comparison with business units that are performing below expectations. Their targets will be less strongly revised based on the information they get from peers. In contrast, business units that are performing below expectations will get a more strongly revised target based on the information of peers. Since the effort of the manager can be seen separately from the market circumstances, a good performing manager will be rewarded based on its effort. It becomes clear for the company which managers are performing well under the current market circumstances. A ‘reward’ is that their target is less strongly revised so the manager has a greater chance in achieving their bonus or other (financial) benefits. This should also (partially) address the earlier described ratchet effect.
2.6 Hypothesis development
2.6.1 Budgeting slack
The role of business unit managers in target setting is one which cannot be underestimated. Hartmann and Maas (2010) find in their research that there are two sets of responsibilities for business unit managers. The first role is to contribute to corporate control by reporting about the activities and the performance of their unit to higher-level management. The second role is that business unit managers are responsible for strategic and operational decision making with regard to their store.
The main problem with involving business unit managers in target setting is that managers gets incentives to create slack. Slack is described by Webb (2002) as intentional biasing of performance targets below the expected levels. A common concern in slacking the performance
is that business unit managers try to maximize the expected value of the payoffs from target achievements (Hartmann and Maas, 2010; Nouri and Parker, 1996; Webb, 2002). This happens despite corporate managers and owners of a company want an as accurate performance as possible (Indjejikian and Matĕjka, 2006). Common way to create slack in the targets is through budgets, since the budget fulfil a central role in the management control system (Nouri and Parker, 1996).
Indjejikian and Matĕjka (2011) researched how the delegation of accounting decision rights affected the way and how performance measures are used. Typically it is argued in prior literature that when local managers have superior information about critical success factors in their local markets, they are delegated decision rights to act on this knowledge. However, this also means that now the need arises to guide the decisions of the local managers to assure that their decisions will not only benefit themselves, but will also benefit the firm (Abernethy, Bouwens and Van Lent, 2004; Bouwens and Van Lent, 2007). So, they find that operational decentralization and incentive strength are complementary organizational design choices. However, Indjejikian and Matejka (2011) argue that operational decentralization and accounting decentralization should go together. But they also note that more incentives tied to financial measures makes those local managers more incentivized to abuse their accounting discretion to make the financial numbers look more appealing than warranted by the underlying economics. So, this is consistent with the earlier argument that when target setting relies on information from lower-level managers, they could distort if to serve their own interests (build slack and therefore lower targets)
In another research, Indjejikian and Matĕjka (2006) find that business unit managers who have considerable authority to design their own accounting system, enjoy informational rents in form of greater slack in the targets. This suggests that authority to design their own accounting system is associated with private information. Accounting systems are characterized as centralized when business unit (financial) reports are standardized by corporate accounting rules.
Nouri and Parker (1996) find that there are contradictory results with regard to the involvement of business unit managers. They find that there is an agency problem when a business unit manager has specific information about the local conditions which are not known by the corporate managers. The study of Lukka (1998) observed that the participation of sales managers led to the introduction of slack into the sales forecast as managers tried to improve their chances of attaining the budget (Nouri and Parker, 1996). The other side of the story is that participation can bring goal congruence within the organization.
The first hypothesis is built on the involvement and influence of business unit managers on the target setting process. The literature has proven that business unit managers will bias
information about their performance intentionally to maximize the expected value of the payoffs from target achievements (Hartmann and Maas, 2010; Nouri and Parker, 1996; Webb, 2002). As this form of manipulation by lower level managers represents a cost to firms (i.e., lower targets are not optimally effort-inducing), I expect that when principals are aware of the susceptibility to manipulation of the target they will adjust for this by providing them with higher targets. Based on this expectation, the following hypothesis is developed.
Hypothesis 1: ‘’When a business unit managers has the opportunity to create slack in delivered information to corporate headquarters, there will be higher revised targets’’
2.6.2 Controllability of a performance measure
In this section the controllability of a performance measure is examined and how this could influence the (revised) target setting. Hölmstrom (1979) find that controllable objective measures are preferred from a contracting perspective because uncontrollability reduces the informativeness of a measure about employee actions and effort and because agents need to be compensated for the additional risk of uncontrollable factors. However, in practice is shown that completely controllable measures are very rare (Bol and Smith, 2011). This means that there always will be some noise which can’t be filtered out in the objective performance measures. As consequence this means that objective performance measures are never completely controllable.
Employees are more satisfied when there is a better controllability of performance measures. Specifically, uncontrollable items in their performance assessment are neutralized and this brings fairness in the evaluation of an employee. Giraud, Langevin and Mendoza (2008) find that there are three factors which seems to be uncontrollable and needs to be neutralized. They assumed that the three factors are: external factors, decisions taken by other managers in the company or within the division and decisions taken by other seniors in the rank. The findings of Giraud et al. (2008) showed that it is not desired that the controllability principle is totally applied. Managers want the controllability principle applied when their performance is affected by internal uncontrollable factors, mostly when their performance is affected by decisions from others, but not when the results are affected by external factors. Managers accept that there is a certain level of risk in their job. Therefore they consider that they have to manage these outside risks because everyone in the market has to deal with market conditions. Fairness is concerned as a very important characteristic for business unit managers in evaluating their performance. Vincil (1979) find that it is considered unfair when profit centre managers are evaluated based on results which are affected by uncontrollable factors.
The agency theory focuses on situations with outcome uncertainty, information asymmetry between the principal and the agent, and compensation contracts based on performance measures that are not perfect in providing information about the agent’s effort. Therefore, linking the payment to the agent’s effort, some risk is transferred from the principal to the agent. The optimal compensation contract trades off the costs and benefits of imposing compensation risk on the agent. That is, if a lot of variable compensation is at stake for a manager and the performance measure has a high degree of uncontrollability, there is a lot of risk imposed on the manager for which he or she wants to be compensated. This makes incentive contracting very costly for firms. This is referred to by Bol and Smith (2011). They find that if an employee perceives the compensation outcome to be unfair, the employee will have a negative behavioural reaction toward the compensation system and will as a result not be motivated. Supervisors therefore need to take fairness concerns into account when managing a compensation plan and setting targets.
The second hypothesis is built on effects of the environment on the target setting process. Prior literature has argued that firms can use subjectivity to afterwards correct for the impact of uncontrollable events on the performance measure outcome. Likewise, firms can compare the performance with that of peers to filter out the impact of uncontrollable events that impact both the manager’s performance as well as that of peers. Here, I focus on the impact of controllability on the degree in which targets are revised. Bol et al. (2010) show how firms set easier targets when units face greater environmental volatility. That is, when the uncontrollability of the environment in which units operate is greater, units are provided insurance against negative external events by granting them lower targets.
Given that the literature shows that the sales target will be adjusted downward when there is a great influence of market conditions, the following hypothesis is developed.
3 Research design
3.1 Conceptual research model
In figure one is schematically shown how the independent variables are operationalized and how these will possibly affect the dependent variable. The independent variables are the controllability of a performance measure and the possibility to create slack in the sales information. The dependent variable is the revised target.
Figure 1: Conceptual research model
3.1.1 Scenario-based survey experiment
To test whether the independent variables have an influence on the targets for the subsequent period, an experiment is conducted. This relationship is tested through a simulation of the target setting process with manipulated factors taken into account in different scenarios. This study is a controlled experiment, as the variables will be manipulated. Therefore, a scenario-based survey experiment will be conducted. Scenario-based experiments have advantages and disadvantages. An experiment can provide a good internal validity, but may lack external validity and generalizability. The scenarios described a defined context, but it is not guaranteed that there will be an effect in the target setting process.
In this two by two experimental case study, two characteristics will vary systematically: (1) the economic climate is or is not of influence on the performance of last year, and (2) the registration of the sales will be executed at either the local dealership which makes the performance
susceptible to manipulation or at the company’s headquarter. Each scenario described the performance of last year and based on that performance the participants needed to set the new target for coming year.
The four scenarios are presented in table one. Each scenario has other influences. This should lead to differences in the targets set by the participants.
Table 1: Expectations revised target 3.1.2 Manipulations
The cases were designed as shown in table one. Each cases started with a general introduction to the experiment and a background of the company. Hereafter, the participant was directed to one of the four cases.
The performance of the manager was in two of the four cases influenced by the economic climate (i.e. low controllability). The economy was stagnating for the coming year, so this should affect the performance. In the other two cases was no influence of the economic climate (i.e. high controllability). The second manipulation was that in two cases the sales registration took place at the local dealership, so there should be a correction for possible sensitivity to manipulation (i.e. high susceptibility to manipulation). In the other two cases the registration of the sales took place at the company’s headquarter (i.e. low susceptibility to manipulation). The case descriptions are attached in appendix I.
The participants faced two of the four described situations. They should make the decision as owner if they correct for the variables when setting the target for the subsequent period.
Both manipulations were derived from earlier researches. The manipulation for controllability is for example derived from the Bol et al. (2010) paper, whilst the budget slack manipulation was derived from the Hartmann and Maas (2010) research. The effects of the
Independent
variable Impact Economic Climate low (controllability high) Impact Economic Climate high (controllability low) Sales Registration at dealership (susceptibility to manipulation high) Case 4 Case 1 Sales Registration at headquarter (susceptibility to manipulation high) Case 3 Case 2
variables on the target setting process were derived from these studies, the manipulations itself were designed by me.
3.2 Procedures
The survey was conducted through the online application of Qualtrics. This is, all the respondents were recruited through the internet. At first, it was intended to recruit only business students of the University of Amsterdam. Since it was difficult to gain enough participants, the choice was made to amplify the recruitment of participants. Participants were approached through my personal network (i.e. Facebook and LinkedIn). Both networks made it possible to recruit a sufficient number of respondents for each case.
After the participants clicked on the Qualtrics link, they were randomly assigned to a scenario description in which they need to set a target for a car dealership. The independent variables are manipulated as showed in table one and can be high or low in one situation and the sales are registered at either the local dealership or the company’s headquarter. The dependant variable is the target set for the coming year. The target can be set on a quantitative scale between a restricted range of 100 and 130. This is, a target will probably not be lowered in a subsequent period. Since the target for the past year was 100, this was the lower bound. The upper bound of 130 was determined because the performance of last year was 125. Typically, target ratcheting parameters do not exceed 1. Since there should be some room for the participant to increase the target beyond the realization of the current year, 130 was chosen as upper bound.
3.3 Post manipulation checks
At the end of the questionnaire, the participant was asked four control questions to see if they had read and if they understood the case. Those four questions were based on the case they read and on the different manipulations.
The goal of those questions was twofold. On the one hand luck needed to be excluded. It is unlikely that the participants answered all four questions right based on luck. On the other hand were the four questions used to check whether the participants understood the questions.
The first question was about the sales of the unit in the last year and was used to see whether the participant had read the case. The second question was for the first manipulation, namely if the economic climate was of impact of the performance of last year. The third question continued on this effect by asking what the dealership sold. The dealership sells only new cars when the economic climate was of impact on the performance and did also sell second hand cars
when the economic climate was not of impact. The last question was based on the second manipulation, where the sales are registered. These checks were used to exclude participants which made a mistake.
3.4 Sample characteristics
A total of 164 respondents started in this research (N=164). There were 21 persons in total who dropped out during the survey (incomplete data/missing values N=21). This resulted in a drop out percentage of approximately 13 percent. This means that the total of respondents which did complete the questionnaire was 143 (N=143). It was not possible to proceed to the next page when not all questions were answered (i.e. due to obligatory answering settings in Qualtrics), respondents might have left the survey without completing it. Respondents were excluded when they questioned one of the control questions wrong. This resulted in a decrease in the sample of 60 respondents (N=60). Therefore the final sample consists 83 respondents (N=83).
After eliminating participants for making a mistake in the control question (i.e. for the reliability of the data), the sample size useful for analysis was N = 83 (scenario 1: N = 21; scenario 2: N = 20; scenario 3: N = 21; scenario 4: N = 21). Hereafter the data was checked for missing’s (i.e. participants who did not complete the whole questionnaire). After eliminating the participants which made a mistake in the control questions, there were no participants left that which didn’t complete the questionnaire. A flowchart of the calculation of the working sample is attached in appendix II.
In this research is controlled for three characteristics; (1) age, (2) gender and (3) education level (Dutch style). The age was ranging from 18 to 62 years old, with an average of 29.5 years. The controlled characteristics are displayed in table two.
Characteristics sample
Age N
Mean [SD] 29.46 [1.138] 83
Median [IQR] 24 [23; 33] 83
Education-level Frequency Valid Percentage Cumulative percentage
High 22 26.51 26.51 Middle 42 50.60 77.11 Low 11 13.25 90.35 Other 8 9.65 100.00 Gender Female 21 25.3 25.3 Male 62 74.7 100.0
* Dutch education system. High level consists of VWO and university students, middle consists of HAVO and HBO and low consists of VMBO-TL and MBO. Other is a collective group for non-graduates or foreign people.
4 Results
4.1 Descriptive statistics
For each of the four cases is analysed how many participants there are, what the estimates means, medians and standard deviations are. In table three is shown how the final sample is divided.
Descriptive statistics
Case 1 N
Mean new target [SD] 114.52 (7.09) 21
Median new target [IQR] 115 (110;120) 21
Case 2 N
Mean new target [SD] 118.65 (8.86) 20
Median new target [IQR] 120 (115; 125) 20
Case 3 N
Mean new target [SD] 118.38 (8.31) 21
Median new target [IQR] 120 (110; 125) 21
Case 4 N
Mean new target [SD] 114.20 (5.66) 21
Median new target [IQR] 115.00 (110;119) 21
Table 3: Descriptive statistics per case
4.2 Univariate analysis of variance
The ANOVA allows to check for combined influences of the independent variables on the dependent variable. In this instance, the univariate analysis of variance (UNIANOVA, furthermore the ANOVA) has been executed. The one-way ANOVA was not possible to run due to the significant outcome of the Levene’s test for homogeneity of variances. A multivariate analysis of variance (MANOVA) is not possible since there are two independent variables and a MANOVA needs at least three.
The Levene’s test of equality of error variance in the ANOVA does not show a significant statistic (F-statistic = 1.836 and p = 0.147). The F-statistic is greater than one, which means that the experimental manipulation had some effect above and beyond the effect of individual
differences in performance (Field, 2013, p. 441). When the Levene’s test does not show a significant p-value, this means that the variances are not significantly different.
The ANOVA tested the influence of both the economic climate and the sales registration and the interaction effect on the revised targets. The ANOVA show that for the independent variable economic climate, there is no significant effect on the revised targets (F-statistic = 0.044, p-value = 0.834). This shows that there is no effect of the controllability (i.e. economic climate) on the revised target. This is not according hypothesis two, where was expected that when there is no influence of the economic climate on the outcome of the performance measure (i.e. high controllability), the revised target will be higher.
For the sales registration, the ANOVA show that there is a significant effect of the sales registration (e.g. the possibility to create slack in information) on the revised targets (F-statistic = 6.412, p-value = 0.013). This means that it does matter where the sales are registered for the target for the subsequent period. There is a negative relation between where sales are registered and the target. The independent variable is compared with the registration of sales at the company’s headquarter. This means that when the sales are registered at the local dealership, the revised target will be lower than when they are registered at the company’s headquarter. This is not according to the first hypothesis which expects that when sales are registered at the local dealership, the revised target will be higher (i.e. high susceptibility to manipulation). This means that an alternative hypothesis must be recognized that when the sales are registered at the local dealership, the revised target for the subsequent period will be lower than when sales are registered at the company’s headquarter.
The ANOVA also tests for an interaction effect of the economic climate and the sales registration. There is no interaction effect between the two independent variables (F-statistic = 0.002, p-value = 0.962). The graphs to support this finding are attached in appendix III.
The explained variance by the two independent variables in the ANOVA (R-squared) is 7.5 percent of the total variance. This is an indication that the data is not fitted with the regression line. The ANOVA table is shown in in table four below.
Levene’s test for homogeneity
F-statistic Df1 Df2 P-value
1.836 3 79 0.147
Test of between-subjects effects
Source Df F-statistic Partial η2 P-value
Economic Climate 1 0.044 0.001 0.834 Sales Registration 1 6.412 0.075 0.013* Economic Climate * Sales Registration 1 0.002 0.000 0.962 Error (within groups) 79 * Significant at p < 0.05
Table 4: Outcomes ANOVA
4.3 Correlations
The correlation research, as exhibited in appendix IV, is conducted to see if there is a univariate relation between the dependent, independent and control variables.
In this correlation matrix is chosen for the Pearson-r correlation statistic rather than Spearman correlation statistic. The assumption for the correlation matrix is made that the data is normally distributed in which case the Pearson statistic is used.
The correlations show a significant effect between the independent variable sales registration and the dependent variable revised target (r = -0.274, p < 0.05). The direction of this specific correlation is not in the expected direction with the hypothesis. The independent variable economic climate does not show any significant effect with regard to the revised target (r = -0.019, p = 0.861). This means that there probably is no effect of the economic climate on the revised target. This is also not in line with the expectation. This will be further elaborated in the next paragraphs.
Of the control variables, age appears to have a positive significant correlation with the revised target (r = 0.251, p < 0.05). It appears to be that age does influence the revised target in some way.
On the independent variables are two control variables significant. Both age and gender have a (slightly) significant influence on sales registration. Age appears to have a slightly negative significant influence on sales registration (r = -0.206, p < 0.10). Gender, on the other hand, has a positive and significant influence on sales registration (r = 0.256, p < 0.05).
It appears to be that cases were not completely randomized (i.e. hence the significant effect of control variables). Therefore is in the next paragraph a regression analysis is performed, after the data is tested for the assumptions for the regression analysis.
4.4 Regression analysis
4.4.1 Test assumptions
Before the regression analysis is run, the data is tested for three assumptions; (1) normality of the data, (2) linearity of the data and (3) multicollinearity. If any of these assumptions is violated (i.e. nonlinear relationships, multicollinearity etcetera) then the regression model may be inefficient or biased / misleading (Field, 2013).
The normality of the data is tested on the dependent variable. For the dependent variable is tested on the kurtosis and the skewness. When both values are between -1 and 1, there is a normal distribution. Both measures indicate that there is a normal distribution as kurtosis has a value of -0.815 and the skewness has a value of 0.090. The normal P-P plot for the revised target is shown in appendix V. This is also shown in table 5 and in the histogram which is presented in the appendix V.
Statistics dependent variable New Target (total sales)
N Valid 83 Missing 0 Mean [7.72] 116.39 Median 115 Kurtosis -0.815 Skewness 0.090
Table 5: Statistics for normality
Secondly, the data is tested for linearity. All of the control variables and the dependent variables are tested if there is no exponential relation in the data. The scatter dot plots of the control variables are in appendix VI. All variables (revised target, age, education level and gender) show a linear relationship.
The third assumption that needed to be tested is multicollinearity. Multicollinearity is tested through the measures VIF and tolerance. The VIF may not be substantially larger than 1, while tolerance may not be lower than 0.2 to prevent potential problems. The multicollinearity is tested on the independent variables and on the control variables. Also multicollinearity is no problem for a linear regression as is shown in table six.
Multicollinearity matrix
Variables Tolerance VIF
Age 0.918 1.090
Economic Climate 0.956 1.046
Education Level 0.925 1.081
Gender 0.929 1.077
Sales Registration 0.859 1.164
Table 6: Multicollinearity matrix independent and control variables
The last assumption to test whether the data is useful for a linear regression, is the homoscedasticity. The variance of the residuals should be homogeneous across levels of the predicted values. If the variance of the residuals is non-constant then the residual variance is said to be heteroscedastic. The scatterplot for homoscedasticity is in appendix VII. In this scatterplot is shown that the outcomes are homogeneous across the levels of the predicted values. This means that the variables are suitable for a linear regression analysis.
4.4.2 Outcomes regression analysis
In the last paragraph is shown that the data is suitable for a linear regression analysis. The regression is first run on the dependent variables and the independent variables. In the second model is the regression run with the control variables. Both regression models are shown in table seven.
The regression shows the same outcomes as the correlation matrix and the ANOVA at a 95 percent confidence interval (95% CI). The regression shows that for economic climate there is no significant influence (B = -0.023, p-value = 0.833 (CI: -3.63; 2.94)). Though the economic climate does not have a significant effect, the direction of the standardized beta is negative where a positive influence was expected. This means that the revised targets will be higher when there is an influence of the economic climate. However, the p-value is not significant thus hypothesis two is not supported.
The regression shows for the sales registration that there is a significant influence at a 95 percent confidence interval (B = -0.274, p-value = 0.013 (CI -7.49; -0.92)). The standardized beta has a negative direction. This is opposed to what was expected. This means that the revised targets are corrected when they take place at the local dealership instead of the company’s headquarter. This means that the first hypothesis is not supported. Since the sales registration has a significant influence, an alternative hypothesis needs to be adopted.
4.4.2.1 Regression model
A regression analysis is run with the independent variables and the control variables. This allows us to check if the control variables have any influence on the independent variables and on the dependent variable. This regression is also run on a 95 percent confidence interval. The ANOVA for the second model shows that there is a lower significant impact of the total model on the dependent variable and is only significant a p-value < 0.10 (F-statistic = 2.095, p-value = 0.075). The fit of the model increases to 12 percent.
With the control variables included, the direction of the standardized beta for the economic climate becomes positive. However, the influence of economic climate is still not significant (B = 0.014, p-value = 0.901 (CI: -3.13; 3.55)). There is no statistical support for the second hypothesis.
For the other independent variable, sales registration, the influence on the revised targets is lower. The standardized beta for the sales registration had a negative coefficient and the p-value was significant (B = -0.216, p-value = 0.065 (CI: -6.83; 0.22)). This means that the effect of the
location of the sales registration lower when the control variables are included in the regression model. The effect is further elaborated in the next paragraph.
With regard to the control variables is only age of significant influence on the dependent variable (B = 0.210, p-value = 0.064 (CI: -0.01; 0.32)). Also for the control variable age appears to be that the effect on the dependent variable less strong when the independent variables are included.
The interaction effect is included in the second model. This is to test what effect the control variables have on the combined effect of the independent variables. This shows that the effect of age becomes significant at p < 0.05 level. The interaction effect has a p-value of 0.354, which is not significant. This means that the control variables don’t have such impact on the interaction effect that it becomes significant. This regression model is included in the model below.
* Significant at p < 0.10, ** significant at p < 0.05
Table 7: Linear Regression analysis.
4.5 Additional analysis
4.5.1 Additional test; Free of distribution
Since the results are contradictory with the expectations, another test has been performed. The free of distribution test is chosen to test whether all assumptions are right and the outcomes are the same as with the ANOVA and the regression,
For the first hypothesis is chosen for the independent-samples Mann-Whitney U Test. This allows to test for two independent conditions. The Mann-Whitney U test works by looking at differences in the ranked positions of scores in different groups. The test relies on scores being ranked from lowest to highest. The significance level of the Mann-Whitney U test shows then that
Linear Regression model coefficients Model Unstandardized coefficients Standardized coefficients Confidence interval (CI 95%) B Std. Error Beta t-statistic p-value Lower bound Upper bound 1 Economic climate 0.210 1.679 0.014 0.125 0.901 -6.834 0.217 Sales Registration -3.309 1.771 -0.216 -1.868 0.065* -0.820 0.694 Age 0.156 0.083 0.210 1.881 0.064* -3.132 3.553 Education level -0.063 0.380 -0.018 -0.166 0.869 -0.009 0.322 Gender -1.311 1.959 -0.074 -0.669 0.505 -5.211 2.589 2 Sales registration * economic climate -1.859 1.995 -0.105 -0.932 0.354 -5.829 2.112 Age 0.169 0.083 0.227 2.044 0.044** 0.004 0.333 Education level 0.060 0.375 0.018 0.160 0.873 -0.686 0.806 Gender -1.825 1.962 -0.103 -0.930 0.355 -5.731 2.082
there is a significant effect (p-value < 0.05) between sales registration and the revised target. This means that the earlier analysis was correct.
For the second hypothesis is also chosen for the independent-samples Mann-Whitney U Test. Also for the influence of the economic climate on the revised target is analysed if the earlier analysis are correct. Also in this case was the earlier analysis correct. The economic climate does not have a significant influence on the revised targets since the p-value = 0.807.
* Significant at p < 0.10, ** significant at p < 0.05
Table 8: Free of distribution
4.5.2 Additional test; regression analysis with full sample
The second additional analysis is done with the full sample of 143 (N = 143). The participants that didn’t finish were eliminated in this analysis. This regression analysis is done to see if there were any effects in the full sample on the independent variables. In the first model are only the independent variables included, in the second model are also the control variables included.
In both models do the independent variables and the control variables not have any influence on the dependent variables. This means that with eliminating the participants that made an error in the control questions, sales registration becomes significant. The outcomes of the full sample regression is reported below.
Free of Distribution test
Hypothesis Test p-value
(2-sided)
1 Mann-Whitney U 0.807
Table 9: Regression analysis with full sample
Linear Regression model coefficients Model Unstandardized coefficients Standardized coefficients B Std. Error Beta t-statistic p-value 1 Economic climate 1.383 1.373 0.084 1.007 0.316 Sales registration 1.824 1.377 0.111 1.325 0.187 Economic climate 1.482 1.380 0.091 1.073 0.285 Sales registration 1.454 1.408 0.089 1.033 0.304 Age 0.100 0.064 0.135 1.579 0.117 Education level -0.214 0.310 -0.058 -0.690 0.491 Gender 1.896 1.494 0.107 1.269 0.207
5 Discussion and conclusion
5.1 Findings
The aim of this research was to examine whether the controllability of a measure and the possibility to create slack in the presented information to the company’s headquarter are of influence in the target setting for the subsequent period. Specifically, the research question was:
‘’How does the controllability of a measure and the susceptibility to manipulation of a measure affect the degree in which the current performance on that measure is incorporated in the subsequent target revision on that measure?’’
This research question was tested with two hypotheses and how the effects were of each of the separate factors on the target setting process. The test was constructed in a manipulated situation in which the participants faced two conditions. The participants needed to set a new target for the next period based on the manipulations.
For the first hypothesis was expected that when the business unit manager has the possibility to create slack, the target will be revised upwards in the subsequent period. The ANOVA and the regression analysis show that there is a significant effect of the location of the information, but that the direction of the effect is opposed to what was expected. This means that there is no statistical support for the first hypothesis, but that an alternative hypothesis is supported and needs to be adopted. The alternative hypothesis describes the situation that when the information is registered at the company’s headquarter, the revised targets will be higher that when the information is registered at the local dealerships. However, this effect is weakened when including the control variables. Specifically, when gender and age are included.
With the second hypothesis was expected that a higher controllability of a performance measure (i.e. no impact of the economic climate on the performance measure) will lead to higher revised target. As well the ANOVA as the regression analysis show that there is no significant effect when there is influence of the economic climate impacting the performance or not. This means that there is no statistical support for the hypothesis and therefore the hypothesis is not adopted.
5.2 Discussion and limitations
This experiment has shown that the results are contradictory with what was expected. The opposed significant effect of the registration of sales was not in the line of expectations. The data shows that the revised targets are higher at the company’s headquarter compared with the local dealership. A possible explanation for this effect could be that the participant thought that there is a higher
workload at local dealership than when the sales are registered at the company’s headquarter. The workload for administrative tasks was not mentioned in the case. Hartmann and Maas (2010) find a connection between social pressures of a business unit controller and the personality of the controller. However, Hartmann and Maas find also that when the sales are registered, the targets should be revised upwards since the business unit manager has the chance to create slack in the performance. Since this is not mentioned in the experimental case, this is an effect what needed to be further researched.
The type of respondents could also be an influence in the contradictory results. The first intent was to only select business students. Since it was difficult to gain enough participants, the choice was made to widen the range of participants. This was unforeseen and therefore not controlled for. Since this effect cannot be investigated in this research, this could be interesting for future research to investigate. Also a research only under business students or people with a financial background could be interesting.
Another possible explanation for these results could be that the questionnaire is not completely understood by the participants. Possibly, the questions were too financial for the participants that did not have the background to answer the questions properly. This connects with the more selective sample.
The results of this research are interesting on scientific and on practical level, because these are opposed of what was expected in the literature. Further research is needed to see if these results can be repeated in a more selective research sample. The sample is not completely with participants with a financial background, which could have an influence on the results. On the other hand, a sample with mixed backgrounds could give more insight in how people set targets in different fields, but this need to be controlled in future research. However, there are also some practical implications on this research. First of all, when sales are registered at the company’s headquarter, this means that they have full insight in the information. Therefore, it could be possible that this is taken into account by the participants. Second practical implication is the workload when sales are registered locally. This increases because the business unit manager has more tasks to do and could therefore be corrected for having less time to do other things.
Note to the found results is that the fit of the regression model of the independent variables on the dependent variable is only 7.5 percent. This is low and indicates that there are more factors of influence on the target setting process. When the control variables are included, the fit of the regression model increases to 12.5 percent, which still is low for the explained variance.
5.3 Conclusions
The main conclusion of this research is that the results are opposed to what was expected based on literature. This research could give more insights in the target setting process and what the influences are of (1) the controllability of a measure and (2) the possibility to create slack in the presented information to the company’s headquarter.
In the experiment were simulations in the economic climate and the sales registration based on studies of Hartmann and Maas (2010) and Bol and Smith (2011). Also the expectations for the directions of the hypotheses were based on those studies. From the outcomes of the statistical analysis can be concluded that the impact of the economic climate does not affect the revised targets (i.e. no statistical support that there is any influence). For the sales registration was a significant effect found, but opposed to what was expected from the theory. Therefore there was no support for the hypothesis (i.e. no statistical support for this hypothesis), but an alternative hypothesis is adopted. From the outcomes can be furthermore concluded that there are more factors that influence the target setting. 7.5 percent of the variance in the model was explained by the two independent variables.
This research has shown that the outcomes are opposed to what was initially expected from the literature. Therefore the target setting process is something that needs to be further explored. It seems to be that though these factors influence the (reported) performance of a business unit, these factors have different effects on the target setting process.
