Will a minimum effort requirement crowd out intrinsic motivation in
a real world setting?
Jeffrey Gerhard Dinham
10826440
Supervisor: Jeroen van de Ven
July 2015
MSc Business Economics: Organizational Economics
1
TABLE OF CONTENTS
SECTION I
Introduction 2
Research Question and Contribution 2
Brief Summary of Findings 3
Setup of Thesis 3
SECTION II
Related Literature 4
Related Literature – Empirical Evidence 6 Critical Evaluation of Existing Studies 7
Contribution of this study 8
Summary of Identified Crowding Out Causes 9 SECTION III
Description of Field Study 11
Formulation of Hypothesis 13
SECTION IV
Summary Statistics 13
Defining Groups 15
Initial Results 16
The September Effect 19
SECTION V
Identifying Crowding Out 21
Second Set of Results: Survey Results 24
Econometric Analysis 29
Possible Limitations 31
SECTION VI
Discussion 34
Conclusion 35
2 SECTION I
Introduction
“Incentives are the essence of Economics”
(Prendergast, 2011)
In the last decade, persistent anomalies with the standard disciplining effect of incentives and controls1 have led to the acceptance that under specific circumstances certain individuals may show a crowding out of intrinsic motivation (See Upton, 1973). To date this has phenomena has been accepted theoretically, however its empirical effects have not yet been well proven. Numerous recent studies have been done, finding mixed empirical evidence in support of and against crowding out. This creates a level of uncertainty for organizations wishing to use control and incentives to elicit a particular response from agents (See Frey & Oberholzer-Gee, 1997). When we consider the scale of these interventions in modern organizations, the importance of accurate outcome predictions becomes paramount.
Research Question and Contribution
Many of the studies done to determine crowding out of intrinsic motivation are experimental and the variables which may lead to crowding out are limited by re-creation through proxy. This paper looks at a natural field experiment run on a well-established real world organization which implemented a minimum performance requirement (a measure of control) on its agents. Control was placed on the number of outbound sales calls required to be made each day. This environment is unique for two reasons; firstly it is a situation of control possibly crowding out intrinsic motivation rather than the more frequently analysed incentive crowding out situation. Secondly control was implemented on an agent‘s effort measurable, not on the agents actual output (sales revenue). There are fewer studies analysing this aspect of the principal agent relationship and as such it adds further field data to the rather slim body of literature that currently exists. The research question will be ―Will a minimum effort
requirement crowd out intrinsic motivation in a real world setting?” Brief Summary of Findings
1
3
Data was analysed for thirty individuals in the sales division of a large software company over seven months: three months pre introduction of control and four months post introduction of the control, giving a total of 3942 observations. Results show that, as expected, the introduction of a minimum effort requirement (MER) did have a significant effect on the targeted variable, namely number of outbound calls per day (henceforth called average daily calls (ADC)). This effect was overwhelmingly positive, in support of the disciplining theory of control.
Nevertheless a definite crowding out effect was identified. This occurred predominantly in August, one period post introduction of the MER, and affected in total 11 individuals or close to 43% of the full sample group. Using only significant crowding out values brings this proportion to 15.4% of the full sample group. Finally the use of a survey provides information on various characteristics of this group; however results of this survey were not completely in line with those expected from theory. Overall our field study shows that the crowding out of intrinsic motivation is a measurable effect that complicates the traditional disciplining effect of control, but not enough to create a net negative effect.
Setup of Thesis
This paper will be structured as follows: Section I deals with the introduction and overview. Section II analyses related literature on the crowding out phenomenon and identifies possible variables that will lead to crowding out. The table constructed in this section is to the best of my knowledge the first such summary of possible contribution variables done. Section III presents formal hypotheses and describes the field study in more detail. Section IV illustrates the broad results from the experiment across different groups for greater clarity. Section V takes an individual-level look at the data and searches for evidence of crowding out. This section continues by analysing the results of a survey testing individual measurements of the characteristics identified in Section II. Finally Section VI concludes with a summary of all findings.
4 SECTION II
Related Literature
The traditional relationship between incentives and agent response is well established and built on the foundation of neoclassical relative price effect theory. By incentivizing (or increasing the price an agent receives for) an action you increase its supply. By dis-incentivizing through punishment or control you increase the cost of an action for an agent and reduce its supply. Economically this straightforward relationship enjoyed a lot of support (see Bandiera et al, 2009). Anecdotally when presented with an anomaly neither of the Nobel Prize winning economists Solow and Arrow could ―detect any reasons why increasing monetary incentives, or equivalently, the price paid… should not increase the quantity supplied‖ (Frey, Jegen 2001). To use the terminology of Falk and Kosfeld (2006) this classical relationship predicts the ―disciplining‖ effect of incentives.
Nevertheless, evidence that suggested traditional incentive theory was not encompassing of all agents responses began to filter in, most strongly from the field of psychology. Psychologists had long believed that external incentives and control could actually lead to a reduction in the intrinsic motivation of an individual. Deci and Ryan (1985) published an influential experimental result in which subjects were given interesting tasks during three phases. In phase I and III no reward was given; in phase II a reward was given based on performance. It was shown that effort rose from phase I to phase II, however in phase III effort fell below phase I levels. The authors explained this as an incentive ―crowding out‖ intrinsic motivation, with intrinsic motivation defined as ―the innate natural propensity to engage ones interests without the aid of extrinsic rewards or controls2‖ (Deci and Ryan, 1985). If this reduction in intrinsic motivation was not compensated for by a corresponding increase in external motivation, output and performance could suffer: a reverse result to the traditional discipline effect. Frey and Jegen (2001) illustrate this effect graphically as shown below.
2
A good example of intrinsic motivation is the desire to learn a new language; according to psychologists this would engage our intrinsic desire for competence. This is just one area of three more detailed definitions of intrinsic motivation, generally used more frequently by psychologists than economists. See Deci & Ryan (1990)
5
In isolation the disciplining effect of incentives are represented by the slope S. An increased reward (R) shifts the agents effort from A to A‘ however the crowding out effect induces a leftward shift of S to S‘. If the crowding out effect is larger than the discipline effect, work effort may reduce from A to A‘‘ resulting in a net negative effect on effort of increasing incentives.
Until this point in the paper the crowding out effect has only been identified in positive incentive situations. However there is no reason to expect any difference in punishment or control situations. One of the leading experiments in this regard comes from Falk and Kosfeld (2006) who analyse how agents see the principals‘ decision to control, and how this affects the agents‘ behaviour. Their results show that the decision to control significantly reduces the agents‘ performance in response (See Appendix I). Gneezy and Rustichini (2000) use an earlier field experiment to introduce a fine to late coming parents at a nursery school. They show that the fine actually induces more late coming (decreased performance) from the parents.
Figure 1. Net outcome of the disciplining and crowding out effect Frey & Jegen (2001)
6 Related Literature – Empirical Evidence
We have seen then that the crowding out effect is a measurable real world counter argument to the traditional incentive disciplining effect. The traditional disciplining effect is well established; both Jenkins et al (1998) and Ford et al (2014) provide conclusive evidence to this effect looking at meta-analysis on incentive driven performance over more than forty years. Both show that incentives and control are significantly correlated (positively and negatively respectively) with performance quantity; although they do point out that performance quality is better explained through intrinsic motivation. This position is well summarized by Condly et al (2003) who show ―the overall average effect of all incentive programs in all work settings was a 22% gain in performance‖. In light of these strong results, how prevalent is the crowding out effect and perhaps more importantly, under what conditions does it arise?
The incentive side of crowding out has a fairly well documented history; meta-analysis has been done by psychologists Deci, Ryan and Koestner (1999) and economists Frey and Jegen (2001). Both of these meta-analyses do find evidence of crowding out: ―Careful consideration of reward effects reported in 128 experiments leads to the conclusion that tangible rewards tend to have a substantially negative effect on intrinsic motivation [within the conditions specified]‖ (Deci, Ryan and Koestner, 1999). Unfortunately there exists no meta-analysis for punishment and control crowding out intrinsic motivation. This is an important gap in the literature as the threat of punishment is a widely used tool in many organizational structures, being both easier and cheaper to implement than a positive incentive.
Looking more specifically at crowding out under control rather than incentives Frey (1993) provides a strong theoretical reasoning for crowding out under principal control and uses empirical evidence from Barkema (1995) looking at the response of 116 Dutch managers in medium sized firms to monitoring. Barkema shows that if monitoring is done by a fellow executive the crowding out effect makes such monitoring detrimental. Monitoring is only effective if done by a 3rd party or a more senior figure, such as a parent company. This leads to Frey‘s hypothesis that crowding out dominates the disciplining effect the more personal the relationship between principal and agent. (See Table 1).
7 Critical Evaluation of Existing Studies
As discussed in the previous section, empirical evidence supporting the crowding out effect under the control/punishment condition is lacking. While compelling theoretical evidence exists, compelling empirical evidence does not. To date the strongest empirical evidence comes from Falk and Kosfeld (2006) discussed above.
Many studies find weak support for the crowding out theory at best, see Dickinson and Villeval (2006) for a laboratory experiment , Nagin et al (2002) for a field experiment, Boly (2011) for both a laboratory and field experiment and Landry et al (2011) for a field experiment. Schnedler and Vadovic (2011) duplicate Falk and Kosfeld‘s (2006) experimental design with slightly different treatments and find a small crowding out effect; nevertheless the disciplining effect far outweighs the crowding out effect in all treatments. Finally Ziegelmeyer, Schmelz and Ploner (2011) conduct four repetitions of Falk and Kosfeld‘s (2006) most significant experimental designs (Low and Medium, C5 and C10 in Appendix I). Again they find statistically significant evidence of the crowding out effect but ―not substantial enough to undermine the effectiveness of economic control‖.
The problem of external validity of laboratory experiments aside, field studies and experiments to date have their own issues. Many of the more recognized papers looking at field data on the crowding out effect use situations that do not represent the large majority of organizational principal agent relationships in the real economy. Barkema (1992) looks at top level managers with a board seat, a class of employee who definitely does not represent the average in a firm. Fehr and Rockenbach (2003) show convincingly that sanctions can crowd out altruism; while this is applicable in a public good setting, altruism does not equate well to intrinsic motivation in a principal agent setting. Finally the oft quoted paper by Gneezy and Rustichini (2000) may simply have set the fine too low for late coming parents resulting in no real disciplining effect being produced, a situation easily remedied by a stronger control.
Finally establishing a size of the crowding out effect is one of the largest difficulties related to previous literature. The large majority of studies and meta-analysis look at significant differences between means, or regression analysis coefficients on specific variables across treatments (Deci, Ryan, Koestner., 1999; Dickinson and Villeval., 2006; Boly., 2011). While these methods can show the crowding out effect, they do little to give a sense of size of the effect, a weakness that this paper hopes to improve upon. Schendler et al (2011), Falk &
8
Kosfeld (2006) and Ziegelmeyer (2011) all indicate relative proportions of crowded out individuals. Unfortunately their findings vary greatly across sample group and treatment from 64.29% to 9%. Looking only at the more conservative, and by extension restrictive, treatments we see a slightly smaller range of values between of 9-37% but still too large to form the basis of meaningful predictions.
Contribution of this Study
This paper seeks to add to the body of literature searching for the crowding out effect in field data in a real world organization. The use of this field data brings two distinct advantages to previous studies. Firstly there is no need to try and simulate variables as experimental laboratory data needs to do; subjects in the study already have a rich history of interpersonal relationships. These relationships would need to be artificially induced in a laboratory setting, perhaps by placing subjects in groups to try and create interpersonal relationships. This creates far more externally valid results, should any be found, and allows identification of various contributing factors that may lead to crowding out in specific individuals. As Boly (2011) puts succinctly ―most of the existing experiments on the crowding out effects of monitoring use numeric effort tasks for which it may be difficult to argue that agents are intrinsically motivated to complete‖. In contrast, working with agents in an existing employment contract, it is reasonable to assume that some level of intrinsic motivation does already exist.
Secondly the sample and nature of the principal agent relationship is far more representative of the average principal agent relationship in the majority of organizations. The sample uses employees reporting to a manager, a structure that is used throughout the hierarchy of most modern firms. Ideally results from this study will have high external validity in terms of actual recommendations for principals contemplating the introduction of a control. The first way to add to previous research is to create a summary of literature identified crowding out causes under control.
9 Summary of Identified Crowding Out Causes
A review of the literature makes it clear that any crowding out effect that appears is dependent on a number of variables, sometimes under very specific situations. In order to properly identify and predict crowding out in this paper‘s sample group, a summary table of the theoretical and experimental conditions for crowding out to occur has been extracted from previous literature (see table 1 below). To the best of my knowledge this is the first summary table of its type looking at the conditions for crowding out under control rather than positive incentives.
While accurate measurement of many of these variables is difficult by virtue of their nature, a good approximation can be made through the use of a survey administered to all individuals within the sample group. These survey results will be linked to individual response to MER control implementation and results analysed for significance and agreement or disagreement with existing literature.
10 Table 1. Conditions identified for the emergence of the crowding out effect under control
Condition Direction Identified By
Intrinsic motivation Intrinsic motivation must exist for crowding out to occur, prerequisite condition.
Frey and Jegen (2001), Boly (2010)
Legitimacy of control Illegitimate perception of control, more crowding out. Control imposed unilaterally (agents have no say), more crowding out.
Kessler and Leider (2013), Kessler and Leider (2014), Schnedler and Vadovik (2011)
Principal agent relationship More personal relationships, more crowding out
Frey (1993), Barkema (1995), Dickinson and Villeval (2008), Falk & Kosfeld (2006)
Preference for control Individuals averse to control (and the framing of this control) will exhibit more crowding out
Fehr and Gachter (2002) Social environment, team
cohesion
More an individual feels part of a team, less likely crowding out will occur.
Enforcement of punishment Punishment that is noticeable and enforced increases extrinsic motivation, less crowding out will occur.
Cerasoli, Ford, Nicklin (2014), Kessler and Leider (2013)
Principal‘s payoff directly and continuously increases with the agent‘s output
The more the Principal benefits from the agents effort, the larger the crowding out effect
Dickinson and Villeval (2008)
Penalties change agents information
If control signals to agents that non-performance is a widespread behaviour, crowding out may increase.
Gneezy and Rustichini (2000), Sliwka (2003) Fairness Control perceived as fair reduces crowding out Fehr and Rockenbach (2003),
Falk et al (2008), Kessler and Leider (2014)
Hierarchy of monitoring Being monitored by an individual in the same perceived 'level' as you increases crowding out.
Barkema (1995)
Playing hard to get Individuals may act strategically to force a reduction in the control, in this case increased crowding out will occur.
Schnedler and Vanberg (2014)
Additional Conditions that may affect agent response to control Risk aversion and loss
aversion
Greater risk and loss aversion is associated with increased effort to avoid punishment Outside option A larger outside option reduces the likelihood
of increased effort, increases crowding out
Nagin et al (2002) Job satisfaction Lower job satisfaction, less response to
control, more crowding out
Nagin et al (2002) Pre-existing norm There is a pre-existing norm of agents
providing high effort, less crowding out
11 SECTION III
Description of Field Study
Data is gathered from a sales division of a multinational software company over the period April 2013 to October 2013. The sales division is typical of the vast majority of principal agent relationships in business organizations; 30 individual consultant‘s report to one manager in a pyramidal hierarchy.
The individuals in the sample are exclusively telephone sales consultants; they receive a fixed wage and in addition are incentivized on a commission basis, based on the revenue they bring into the firm. Typically they have target revenue, under which they earn zero, and over which their commission increases exponentially at multiples of the target value. This incentive scheme does not change for the duration of the sample period.
In the first half of the sample period April to June, sales consultants are only exposed to the revenue incentive; the number of telephone calls they wish to make to clients is entirely a personal decision. However in July, consultants were required to make a minimum number of outbound calls per day (MER). This intervention forms the basis of the experiment analysed in this paper, with a target initially set at 50. However due to leniency the effective hard floor requirement was 35 outbound calls per day (see Appendix II for description and punishment terms).
This form of control is unique amongst experiments in the literature. In this study control is placed on one of the effort functions of the agent‘s3 output function, rather than on the output itself as more common in previous studies. To understand this concept we can imagine a simple agent production function (where output is revenue brought into the firm) as
Where S = Sales skill, N = Number of outbound calls and P = Product knowledge. More straightforward experimental control such as that used by Falk & Kosfeld (2006) would place a minimum effort requirement on directly. In a real world sales environment it is
3
Throughout this paper, the terms Consultants, agents and individuals are used interchangeably to mean the same group of people. ‗Consultant‘ for detail from a company perspective, ‗agent‘ when the context is focussed on principal agent relationships and ‗individual‘ in all other cases.
12
acceptable to incentivize this output (as is done throughout the sample), however it is very contentious to place a control on output. By placing a control (through the MER) on the number of outbound calls, management is indirectly stimulating output without placing a control on output itself.
The analysis of this experiment will be broken up into two broad sections; initially the results of the MER on agent effort as measured by the average daily calls (ADC) will be examined in broad overview to provide the reader with a general understanding of group response. To enhance this understanding, various groups of individuals are created to illustrate how the MER control condition affects individuals differently. Building on these broad brush strokes, a more detailed analysis will be done looking for individual level crowding out effects. Being the crux of this paper, these results will be tested econometrically and integrated with the results of a survey sent out to all individuals analysed for the effect of the variables identified in Table 1 of this paper.
Before continuing, two important assumptions need to be made with regards to the data and a final working definition of crowding out needs to be established.
Assumption 1: Agents have chosen their optimum call level based on their own utility function. Based on standard neoclassical utility theory, it is not unreasonable to assume that
in the period pre-MER agents choose their daily call volume based on the weightings of their own utility functions. Utility of expected revenue per call, less disutility of effort may represent a simple version of such a utility function. This assumption is important as it allows us to assume that pre-MER agents are operating at their intrinsic effort level, a vital assumption in the identification of crowding out.
Assumption 2: All other aspects of the individual’s environment remain the same. At this
point we use this assumption to ensure that all changes in an individual‘s ADC result purely from the interaction between the introduction of the MER and their own personal characteristic preferences. Ceteris paribus is one of the hardest conditions to validate in any field study; limitations to this assumption are addressed more thoroughly in Section IV.
Working definition of crowding out: If an agent’s performance decreases at some point following the introduction of the MER control, this effect will be called crowding out. Of importance: This decrease must be sufficient to bring agents performance below the intrinsic level established pre-MER introduction.
13 Formulation of Hypothesis
Based on previous theoretical and empirical research, the following hypotheses are proposed:
I.) The introduction of MER will have a positive effect on ADC.
II.) For a select subset of individuals the introduction of the MER will result in a decrease in ADC due to the crowding out effect.
III.) In aggregate the positive effects on ADC will outweigh the negative crowding out effects.
IV.) Characteristics of the crowded out subset will fit those characteristics identified via previous literature in Table 1.
SECTION IV
Summary Statistics
In this section our aim is to provide the reader with a broad overview of individual‘s response to the introduction of the MER. This will form the foundation of our analysis which will focus specifically on identifying crowding out. Our main variable of interest is the average number of daily outgoing calls per consultant per month (ADC) and how this value is affected by the introduction of the minimum effort performance requirement (MER) on the 1st July 2013.
The variable ADC was chosen to represent an individual‘s response primarily because it is intuitive to understand. The MER is a control placed on the number of daily calls made by a consultant, evaluated at the end of each month. Therefore any agent with an ADC over 35 would be exempt from punishment (as shown in Appendix II).
An alternative measure is the number of days in the month that the individual reached the 35 call target. This measure is used to identify one group of individuals in Table 3; however it is only used for identification and not analysis as this becomes more difficult to interpret and subsequently a less robust measure of performance. Conceivably an individual may make the majority of calls in the last week of the month, enough to avoid punishment but not provide a high measure of days made target.
14 Table 2.Summary Statistics of Main Variable - Average Daily Calls per individual
Full Sample Pre Control
(MER)
Post Control (MER)
Sample Time period 7 months 148 days 3 months 61 days 4 months 87 days Individuals 26 30 Observations 3819 1431 2388 Minimum Effort Requirement (MER)
NA NA 35 calls per day
Average Daily outbound calls (ADC) Mean 34.83 32.96 36.13 Std Dev. 6.62 6.00 6.75 Median 32 31 32 Minimum 1 1 Maximum 143 131
Table 2 above summarizes the variable ADC; of interest is the range between minimum and maximum daily calls and how far this maximum is from the mean. Looking at this variable on a daily basis pre and post introduction of the MER illustrates how the distribution changes – these histograms can be found in Appendix IIA; they illustrate a fairly normal distribution based on measure of skewness and kurtosis, although it should be noted that post-MER introduction the upper bound tail (50-100 calls per day) becomes denser than pre-MER. Table 3 below, continues investigating the same variable over the full seven months of the sample period.
Table 3. Summary Statistics of main variable – Average Daily Calls per individual - continued
April May June July Aug Sept Oct
Mean 34.23 32.76 31.89 38.60 32.50 39.18 34.35
Std Dev: all daily observations 17.26 19.32 17.81 23.65 21.66 24.76 20.76 Std Dev: daily average 5.81 5.42 6.40 6.32 5.42 5.27 7.08
Cursory examination indicates a general increase in ADC following the introduction of the MER. In addition it can be seen that the standard deviation of outgoing calls during these months is high, which suggests that despite the use of ADC as a measure of performance; standard deviation will still be affected by aggregate effects such as the Monday blues or quiet Fridays. Of interest standard deviation is higher in all months post-MER implementation when looking at daily observations, perhaps due to individuals pushing to make the MER in the last few days of the month.
15 -25 -20 -15 -10 -5 0 5 10 w p l n d s e v m ac ab r u h j z aa i g t f c o y a b x ad q k Cha nge in ADC
Individuals Unique Identifier
Figure 2B: July - August ADC change
-20.00 -10.00 0.00 10.00 20.00 30.00 40.00 t e a y z n q d x i k r o s h f g m w b v c l u p j Cha nge in ADC
Individuals Unique Identifier
Figure 2A: Pre July - July ADC change
Next we turn to individual‘s responses in the two main periods of interest. Figure 2A looks at how the relative change in ADC for each individual overwhelmingly increases in July from pre-July periods (each letter along the X – Axis represents a unique individual). 20 of the 26 individuals demonstrate an increase in performance. Figure 2B illustrates a reverse effect on the individual‘s ADC going from July to August, with the majority of individuals decreasing performance.
Defining Groups
The above summary statistics provide a broad overview of the experiment results. Looking at Table 3 it is clear that the ADC increase with the introduction of the MER in July, decrease in August and then increase again in September. In order to further breakdown the effect of the MER it is helpful to define groups of individuals. These groups of individuals are designated to demonstrate how varied the response to the MER is, this variation makes it clear that introduction of a control is definitely not a blanket effect across the full sample.
16 Full Sample: All individuals tracked pre and post-MER.
Top MER Performers: the 5 top performing individual‘s pre-MER. Top performing in this
instance refers to the five individuals who most often reached the MER target before it was introduced. This is roughly 20% of the full pre-MER sample group.
Target Achievers: All individuals identified pre-MER who would have reached the monthly
target of 35 calls per day had it been in place. This gave a total of 10 individuals which includes all 5 of the top MER performers group. This is roughly 38% of the full pre-MER sample group.
Lowest ADC Performers: The 5 individuals with lowest ADC identified pre-MER, again
representing roughly 20% of the full pre-MER group.
Target Non-Achievers: Essentially the reverse of the Target Achievers grouping: All
individuals pre-MER who would not have reached the monthly target MER had it been in place. This gives 16 individuals, roughly 62% of the pre-MER sample group.
It is important to realize that ‗performer‘ in the below context relates only to how well an individual acts with respect to the performance variable mentioned, a high performer in the above groups does not necessarily correspond to a high revenue sales individual, or any other measure of performance.
Initial Results
The groups identified above were tracked in their response to the introduction of the MER in the summary table, Table 4. As the results show, all groups except lowest ADC performers indicate a significant improvement in July, the month of the MER introduction. (The full significance table between months can be found in Appendix IV). Significance was tested using a two-sided t-test, to use this test the most important assumption that needs to be made is that of independence. In this sample we can make the assumption that each new day‘s data point is independent of the last, because the occurrence of one day‘s data does not affect the
17
others. The null hypothesis in all these t-tests is that the difference between means is zero. Test statistics t-stats and p-values between the various months can be found below Graph 4.
Secondly it appears that the increase in ADC decreases back to non-significant levels within a month, however it is also clear to see that September shows yet another large significant increase in individual ADC across nearly all groups, illustrated in the following graph. Full graphs of all group responses can be found in Appendix IV. This full table shows significant differences from any pre-MER month to July for the same groups as found significant in Table 4.
18 35 40 45 50 55 60
April May June July Aug Sept Oct
ADC p er con su lta nt
Figure 3A. Target Achievers ADC Table 4. Summary Table - All groups ADC
Full Sample Top MER Performers Target Achievers Lowest ADC Performers Target Non-Achievers April 34.23 50.89 46.25 26.36 27.52 (5.81) (7.75) (6.52) (7.2) (6.31) May 32.76 47.52 44.9 24.85 25.61 (5.42) (8.09) (7.75) (4.43) (4.55) June 31.89 44.46 42.71 23.03 24.69 (6.4) (7.89) (8.25) (6.08) (5.68) July 38.6 58.76 54.51 26.82 30.81 (6.32) (13.11) (9.11) (8.25) (7.01) Aug 32.5 46.65 44.17 24.17 26.62 (5.42) (10.13) (8.74) (5.36) (4.71) Sept 39.18 58.86 54.66 25.69 30.19 (5.27) (11.99) (10.31) (5.38) (4.4) Oct 34.35 47.61 45.39 26.42 28.41 (7.08) (14.09) (10.45) (8.39) (6.15) T-Test July - June ** *** *** - ** T-stat -3.307 -4.257 -4.295 -1.670 -3.049 P-value 0.002 0.000 0.000 0.102 0.004 Aug - June - - - - - T-stat 0.319 0.745 0.528 0.612 1.130 P-value 0.751 0.461 0.600 0.544 0.266
19
With regards to the research question, a few notable effects are already apparent. There is clear support for Hypothesis 1: That the introduction of MER will have a positive effect on ADC we see this in summary in Table 3 and statistically across nearly all groups in Table 4. With respect to the full sample we can reject the null-hypothesis that there is no effect of MER on ADC: the ADC in July is 6.71 calls above the ADC in June and this difference is statistically significant from zero. (two-sided t-test, t-stat= -3.307 p=0.002). Appendix IV provides further support that July ADC are statistically significant to any other month except September for the full sample.
This section also provides support for Hypothesis 3: In aggregate the positive effects on ADC will outweigh the negative crowding out effects as seen above for the full sample group. Only the lowest ADC performers do not show a clear increase. Using the same two-sided t-test, we cannot reject the null-hypothesis that there is no effect of MER on ADC for this grouping: the ADC in July is 3.79 calls above the ADC in June and this difference is not statistically significant from zero. (two-sided t-test, t-stat= -1.670 p=0.102). In fact we have no evidence of crowding out at this point. Findings from this broad overview are very much in line with traditional theory: the introduction of a control raises the cost of an action for an agent (in this case the action is providing lower than MER level average daily calls) and reduces its supply.
The September Effect
September represents an anomaly in our results. To understand the sudden significant increase in individuals ADC, further information was gathered from the company. This information revealed that the financial year of the company runs from October to September. In 2013 the sales division was behind its forecasted budget and as a result, there was a lot of increased pressure from management for consultants to increase their daily outgoing calls in an attempt to increase sales in this last month of the financial year.
This represents an artificial incentive mechanism that disrupts the underlying effect of the MER introduced in July, and has been termed the September Effect. Looking at the data in more detail, we find that the middle two weeks of September are the most prominent in terms of increased ADC. This disruption can be corrected by smoothing out individuals‘ ADC over these two weeks. Average consultant data for the same two weeks from August and October was used to replace the two problem weeks in September. This corrects the September effect
20
quite well as can be seen in the adjusted figure below (Adjusted figures for all groups can be seen in Appendix VI).
Naturally the more days the smoothing effect is applied to, the more September would reduce to the mean of August and October. This exercise is simply to highlight the main results of the MER introduction and reduce the unwanted effects of another incentive placed on the sample group.
Once September is smoothed our sample period naturally falls into three distinct periods upon which the remainder of this paper will focus.
April – June: This period will be termed pre-MER introduction and forms the intrinsic motivation level of the Agent. Due to the fact that all groups in Table 4 display some downward trend from April to July, intrinsic motivation will be formed from the average of these three months.
July: With the introduction of the MER this month becomes very important and will form the candidate period for identification of crowding out despite there appearing to be an almost unanimous significant increase in ADC.
August: Once September has been excluded or smoothed, all individual groups decrease their ADC significantly in August and October. August therefore naturally presents itself as the last distinct period in the sample and will also form a candidate for identification of crowding out.
35 40 45 50 55 60
April May June July Aug Sept Oct
ADC p er con su lta nt
Figure 3B. Target Achievers ADC - September Effect smoothed
21 SECTION V
Identifying Crowding Out
Results thus far have been strongly in support of the traditional disciplining effect of control, in line with Hypothesis 1 and traditional incentive theory. As Table 4 and Appendix IV show, a positive significant effect is seen across all groups (except the target non-achievers) following the introduction of the MER. There seems to be little sign of the crowding out effect in July; August however sees a sharp significant drop in the ADC across the same groups. Adjusting for the September effect, this drop in ADC remains until the end of the sample period. It appears that the majority of individuals exhibit a disciplining effect initially; however this effect wears off quickly.
Since crowding out is not immediately apparent in the larger groupings of the previous section, we need to look at individual responses in both July and August in an attempt to identify crowding out through a decrease in performance. As stated in the opening section of this paper, for a performance decrease to be correctly identified as crowding out, the decrease needs to take the individuals performance level below the intrinsic level established prior to MER introduction. Since this requires a baseline of intrinsic motivation the four July new hires were dropped from the sample group and will not be included in the results of this section.
Using the above definition we find eleven individuals exhibit crowding out either in July with the introduction of the MER or in August, one period post introduction of the MER. We term these eleven; crowded out individuals (COI) in contrast to the remainder of the group, non-crowded out individuals (non-COI). Table 5 shows these eleven individuals displaying crowding out (two individuals display crowding out in July and August) however only four individuals‘ decreases are found to be significant when compared to intrinsic levels. Again significance was tested using a two-sided t-test, with the assumption that each new day‘s data point is independent of the last, because the occurrence of one day‘s data does not affect the others. The null hypothesis in all these t-tests is that the difference between means is zero, in other words that there is no effect of MER on ADC in the potential period of crowding out. We reject the null hypothesis for individuals T, D, N and E. These individuals do display significant differences in their ADC measurements between the potential period of crowding
22
out and their intrinsic levels. The test statistics (t-stat) for August are (2.188), (5.770), (4.857) and (6.788) respectively.
Table 5: Individuals Exhibiting Crowding Out
Individual Potential Period Pre-July (Intrinsic ADC) Potential Period ADC Significant difference A July 38.3 36.7 - T July 36.4 26.1 ** E July 23.2 17.3 - H August 44.0 41.3 - W August 49.1 42.6 - S August 41.2 33.0 - Z August 33.5 27.1 - T August 36.4 22.4 * R August 31.0 28.8 - D August 27.1 11.3 *** I August 24.9 22.7 - N August 27.7 10.9 *** E August 23.2 4.5 ***
*** indicates p-value ≤ 0.001. ** indicates p-value ≤ 0.01. * indicates p-value ≤ 0.05
As a proportion of the full sample group eleven individuals represent close to 43% of the sample when new July hires are excluded. This proportion is substantially higher than predicted by theory. Using only significant crowding out values decreases the proportion to 15.4%.
In order to gain a clearer understanding of this subgroup, we look at their simple correlation coefficients in response to the introduction of the MER across two measures of performance in Table 6.
‗Made target‘ (MT) is a binary variable indicating if an individual i‘s calls on the particular day t were at or above the MER level. This was extended to include pre-MER call levels per day.
‗Daily calls‘ (DC) is a continuous variable representing the actual number of calls made by an individual. These correlations are for data across the full sample period on a daily basis.
23
Table 6: Correlation Coefficients: correlation with MER
Sample Full COI Only Full Sample less
COI (Non-COI)
Variable DC MT DC MT DC MT
MER 0.072 0.031 -0.054 -0.115 0.132 0.126
It is clear from these correlation coefficients that the COI group of individuals display crowding out and show a negative response to the introduction of a control. It is important to remember that these individuals exhibited crowding out either in July or August; as such they may have increased performance in July and then subsequently decreased. Table 7 below gives a summary of the crowded out individual performance over the three key periods to examine their attributes within the scope of the sample data (personal characteristics are examined in more detail through a survey in the following section)4.
Table 7: Performance characteristics of crowded out individuals
MER Performance Relative performance Individual Pre-July July Aug Pre July
-July July - Aug A 1 1 1 D i H 1 1 1 i D W 1 1 1 i D S 1 1 - i D Z 1 1 - s D T 1 - - D D R - 1 - i D D - - - i D I - - - i D N - - - s D E - - - D D
Individual reaching MPR over the month. Yes =1 i= Increase, D = Decrease, s = Same (No Change)
Immediately apparent is that crowding out does not equate to under performance. Six of the COI‘s reach the MER post introduction and the same number are already above the MER level before it is introduced. Three of the eleven or close to 30% remain above the MER in August. However the larger majority of individuals exhibiting crowding out behaviour are
4
This Table is an extract from Appendix IX displaying the same information for all individuals.
24
underperformers, with four of the eleven (37%) never reaching MER and six of the eleven (55%) never reaching it more than once.
This compares favourably to the non-COI group, where six of the fifteen (40%) never reach MER and eight of the fifteen (53%) never reach it more than once. These results demonstrate that the COI do not differ from the non-COI in their performance ability to reach or not reach MER.
In summary of this section and with reference to the research question at hand we find support for Hypothesis 2: For a select subset of individuals the introduction of the MER will result in a decrease in ADC due to the crowding out effect. For individuals T, D, N and E we can reject the null-hypothesis that there is no effect of MER on ADC in August: their ADC in July is lower than their ADC in June and this difference is statistically significant from zero. Two-sided t-test, t-stats= (2.188), (5.770), (4.857) and (6.788) respectively and p = (0.036), (0.000), (0.000), (0.000) respectively - where 0.000 simply means the p-value is too small to be represented within the number of decimals. Identified crowding out is small and significantly identified crowding out is even smaller. Although previous experiments offer a wide range of crowding out proportions, our significant 15.3% proportion of crowded out individuals offers another point of data to add to the literature, and from which to base future predictions. Looking at performance of the group with reference to the MER in Table 7 and Appendix IX, we see no large differences in the composition of the COI‘s and the non-COIs.
Second Set of Results: Survey Results
We now turn our attention to identifying differences in individual preferences with regards to characteristics that may increase or decrease the crowding out effect and test these preferences against those predicted by theory. To this end a survey was sent to all consultants. This survey used questions designed to test the conditions identified in Table 1 using a rating system in response to statements. For example, the statement: ―My Manager and I get on well together‖ with 1 being disagree and 5 being agree. Overall there were 27 questions, grouped together as one of eleven characteristic variables, the above question forms part of the characteristic ―principal agent relationship‖.
A small number of conditions identified could not be satisfactorily transferred to a survey setup and were therefore not tested. The structure of this survey and a summary of questions
25
can be seen in Appendix VII. The Survey response was 22 individuals, 85% of the sample, with most non-respondents having left the company. A control question was included to ensure that respondents knew what the value of the MER was supposed to be; all respondents answered this question correctly.
Summary statistics of the survey are laid out below in Table 8. Risk aversion and loss aversion were tested using two ‗games‘ devised by Holt and Laury (2002) and Gachter et al (2007). This allowed the results to be compared to the standard population as shown in Appendix VIII. Most of the answers for the other variables tend closer to the higher rating of ―strongly agree‖; in fact no mean is below the neutral point of 3. This issue is covered in more detail in the section on possible limitations.
26 Table 8: Survey Summary Data
Measure Mean Min Max Notes
Risk aversion 4.77 0 9 Scale of 0-10 safe choices; higher values denote higher risk aversion
Loss aversion 3.90 2 6 Scale of 0-6 accepted gambles; higher values denote lower loss aversion
Intrinsic motivation 4.30 2.8 5 Scale of 1-5: Higher values denote higher motivation for career, position, quality of work
Legitimacy of control 3.45 1 5 Scale of 1-5: Higher values denote higher belief that control is warranted
Principal agent relationship 4.25 1 5 Scale of 1-5: Higher values denote closer relationship with manager
Preference for control 3.06 1.5 5 Scale of 1-5: Higher values denote higher preference for control
Social environment, team cohesion
4.66 3.5 5 Scale of 1-5: Higher values denote closer team cohesion
Enforcement of punishment 4.38 2 5 Scale of 1-5: Higher values denote greater belief punishment will be enforced Fairness 4.02 1.5 5 Scale of 1-5: Higher values denote greater
belief that management control is fair Playing hard to get 3.10 1 5 Scale of 1-5: Higher values indicate idea of
acting with the crowd
Outside option 3.32 1 5 Scale of 1-5: Higher values indicate ease of outside recruitment possibilities
Job satisfaction 4.50 3 5 Scale of 1-5: Higher values indicate greater enjoyment of position
Pre-existing norm 4.32 3.3 5 Scale of 1-5: Higher values denote higher motivation to work harder than necessary
To identify differences between the COI and non-COI group we compare survey response between both groups of individuals in Table 9. We see on average that crowded out individuals choose a slightly higher rating on most questions than do the group where no crowding out is evident. Beyond cursory examination only five variables were identified as
27
being significantly different. Again significance was tested using a two-sided t-test, with the assumption that each survey answer is independent of any other; there is no reason to believe that the answer on one question will affect the probability of the answer on any other question. The null hypothesis in all these t-tests is that the difference between means is zero. We reject the null hypothesis for variables 3, 9, 12 and 13 – t-stats can be seen in the far right column of Table 9. The answers to these questions do differ between groups. These are investigated in greater detail below.
Table 9: Summary of survey respondents Individuals
Original sample size
Survey
respondents Proportion
Exhibiting crowding out (COI) 11 8 0.73
No crowding out identified
(Non-COI) 15 14 0.93
26 22
Characteristic means between groups Characteristic
No Crowding
out Crowding Out T-stat
Risk aversion 4.93 4.50 -
Loss aversion 3.86 4.00 -
Intrinsic motivation 4.09 4.65 (-2.45)
Legitimacy of control 3.43 3.50 -
Principal agent relationship 3.96 4.75 (-1.80)
Preference for control 3.07 3.04 -
Social environment, team cohesion 4.57 4.81 -
Enforcement of punishment 4.25 4.63 -
Fairness 3.71 4.56 (-2.52)
Playing hard to get 3.07 3.13 -
Outside option 3.29 3.38 -
Job satisfaction 4.29 4.88 (-2.15)
Pre-existing norm 4.17 4.59 (-2.22)
Considering the statistically significant variables:
Intrinsic Motivation: t-stat (-2.45) p-value (0.024). Survey results show that individuals displaying crowding out rate themselves as having a higher level of intrinsic motivation than the no crowding out group. Existing theory does not indicate a direction for this effect, simply stating that intrinsic motivation is necessary for the crowding out effect to occur. Frey and Jegen (2001) and Boly (2010) show that without some level of intrinsic motivation any drop in performance is simply
28
attributable to the decreased effectiveness of the control, not evidence of the crowding out effect.
Principal agent relationship: t-stat (-1.80) p-value (0.085). Despite not being strictly significant, this variable was included in further analysis because it falls closest to significance than any other non-significant variable. Survey results show that individuals displaying crowding out rate themselves as having a better relationship with their manager than the no crowding out group. This aligns with previous literature by Frey (1993), Barkema (1995), Dickinson and Villeval (2008). Agents with a close relationship to the Principal are likely to view the introduction of a control as a sign of distrust. Indeed the main experiment conducted by Falk and Kosfeld (2002) rests on this notion; an agent being controlled will supply lower effort than one not being controlled. One reason presented is that distrust is a negative signal to the agent who responds in kind.
Fairness: t-stat (-2.52) p-value (0.020). Survey results show that individuals displaying crowding out rate the introduction of the MER by management to be fairer than the no crowding out group. Literature suggests this relationship should be reversed, with unfair control leading to more crowding out. However, this may be attributed to the fact that a control considered ―unfair‖ is often deemed this way if it expects unachievable results. In our sample where 55% of COI reached MER at least once, this control may not have been perceived to be unachievable.
Job satisfaction: t-stat (-2.15) p-value (0.044). Survey results show that individuals displaying crowding out enjoy their job more than the no crowding out group. As with the previous finding, this goes against the existing literature by Nagin et al (2002) who indicate that lower job satisfaction leads to less response to control and more crowding out. To understand these differing results, it is important to notice that our sample of COI contain a number of high performers, it is understandable why these high performers would indicate high job satisfaction. The link between crowding out and higher job satisfaction may be similar to that of the principal agent relationship. Individuals who enjoy their work may feel more marginalized by the implementation of a control (than those who enjoy their job less) and respond negatively.
Pre-existing norm: t-stat (-2.22) p-value (0.039). Survey results show that individuals displaying crowding out show a higher pre-existing norm to do well than the no
29
crowding out group. This contrasts with Kessler and Leider (2013), who propose that a pre-existing norm of high effort is negatively related to crowding out. This result is counter intuitive and difficult to rationalize from an economic point of view, going against the existing economic literature. This is most likely a result of the limitations of the survey design, see section on possible limitations.
These results do not show strong support for Hypothesis 4: Characteristics of the crowded out subset will fit those characteristics identified via previous literature in Table 1. With respect to all conditions we can reject the null-hypothesis that there is no difference between the mean COI group and non-COI group answers for only four characteristics: Intrinsic Motivation, Fairness, Job Satisfaction and Pre-existing Norm. For all these characteristics the COI group answers differ from the non-COI group answers and this difference is statistically significant from zero. (two-sided t-test, t-stat= (-2.45), (-2.52), (-2.15) and (-2.22) respectively and p = (0.022), (0.020), (0.044), (0.038) respectively.
One possible limitation comes from the structure of the survey, which cannot measure the variables of interest directly but only through the proxy of a personal ranking system. This is addressed in more detail in the section on possible limitations. Another explanation simply comes from the fact that a field study is qualitatively different to a laboratory experiment, from which the comparisons are made. Job satisfaction in the real world encompasses numerous aspects of an individual‘s experience, while in a laboratory it is far easier to give individuals an unpleasant task and then introduce a control to elicit crowding out.
Econometric Analysis
In order to support these statistically significant characteristics we turn to an econometric analysis using the following forms, where MERx3 is an interaction variable between MER and Intrinsic Motivation, MERx5 is an interaction variable between MER and principal agent relationship, and so forth.
( ) ( ) ( ) ( ) ( ) ( )
‗Made target‘ (MT) and ‗Daily calls‘ (DC) are as defined in the previous section. Due to the panel data nature of the variables, it is necessary to run a Hausman test. This test gives us an
30
insignificant p-value and as a result we use random effects for our regression form, the results of which are presented in Specifications 1 and 2. Since the Made Target variable is of a binary nature, a logistic regression was also run on specifications which included its use.
Table 10: Econometric Analysis of Significant Characteristics - Coefficients
Specification 1 2 3
Dependent Variable DC MT MT VIF MERx11
Form OLS OLS Logistic Correlation
MER 20.87 0.423 2.439 34.48 0.98
(-4.00) (-3.10) (-3.11)
MERx3 (Intrinsic Motivation) -1.045 -0.032 -0.145 65.48 0.99
(-0.61) (-0.75) (-0.64)
MERx5 (Principal agent relationship) -4.376 -0.085 -0.459 29.67 0.93
(-4.19) (-3.06) (-2.91)
MERx9 (Fairness) 1.978 -0.001 -0.049 61.55 0.95
-1.22 (-0.02) (-0.22)
MERx12 (Job satisfaction) 3.039 0.229 1.422 100.55 0.98
-1.54 -4.54 -4.83
MERx13 (Pre-existing norm) -3.772 -0.204 -1.313 93.17 1
(-1.94) (-3.96) (-4.44) Prob > Chi2 0 0 0 R2 Within 0.0195 0.0166 rho 0.502 0.265 T-statistics in parenthesis.
Significant coefficients are in bold, with t-statistics in parenthesis. We see support for interaction of MER with variables 5, 12 and 13 with regards to affecting the way individuals would otherwise respond to the MER introduction. Principal agent relationship and pre-existing norms respectively are in agreement with Table 9; a higher rating on these variables decreases an individual‘s response to the MER than would otherwise be the case. Job satisfaction is only significant when using the Made Target performance measure, and indicates a higher rating on Job satisfaction will increase the response to MER. This fits in line well with traditional theory, but strangely is the reverse found in Table 9. Of course
31
Table 9 looks at mean differences in rankings on the survey and does not take into consideration the weight of these effects on number of days made target.
A conscious decision was made to focus only on those characteristics identified as significant in Table 9, the main issue facing this regression analysis is the problem of multicollinearity, in fact VIF and correlation coefficients provide ample sign of multicollinearity and dissuade us from searching for more significant characteristics not grounded in theory, as we are likely to find many. Unfortunately these tests do impact the significance of the regression results; however we feel confident that Table 10 simply supports Table 9 and should be viewed as confirmation of the importance of principal agent relationships and pre-existing norms.
Possible Limitations
Before moving on to a final discussion it is worth reflecting on some of the possible limitations that may be confounding the results of this paper. It is important to reiterate that field data by nature is subject to a number of external unmeasurable factors.
The first external effect is sales team revenue. In a telephonic sales environment more calls are likely to equate to more revenue, but the effect is a causality dilemma, more revenue equates to more calls. To understand this it is necessary to realize consultants will typically follow up on all incoming calls with a number of outgoing calls (quotes, information etc.) before a sale is closed. Therefore in months where many customers are calling in and requesting products, revenue will rise; as will ADC. However the revenue increase is not being created by the increased ADC.
This may present an issue whereby the ADC increase or subsequent crowding out is being influenced by a particularly strong or weak revenue period. To investigate this possible effect we look at the revenue values of 2013 compared to the previous three years for the same sample period.
Table 11: Total revenue, period of sample
Year 2010 2011 2012 2013
Revenue 20.55 17.59 16.76 17.86
Difference from mean 12.97% -3.30% -7.86% -1.81%
32
Table 11 indicates that over the seven month sample period, 2013 displays no large deviations from the last three years in revenue terms; however, our months of interest are July and August. The below Figure 4 gives a summary of 2013 revenue as well as 3-year averages and the ADC rate for the full sample group from Table 4. There seems to be little cursory correlation between the ADC rate and the 2013 revenue.
Looking at August we do see that 2013 underperforms the previous 3-year average as well as the month of July 2013. This change is accompanied by a decrease in ADC as has been extensively covered. July 2013 improves markedly on June 2013 accompanied by an ADC increase. While it is tempting to view these two explainable changes in ADC as a correlation with revenue, this three month period (June-Aug 2013) is the only period in which this correlation holds. Looking instead at September to October and the period April to June, the correlation does not hold. This lack of any strong pattern between revenue and ADC across all periods should provide evidence against this potential problem. That is, it is unlikely that ADC is influenced by revenue.
The second limitation that may hold validity is that agents are substituting quantity for quality, by reducing time spent on each call. In such a case our main determinant for identifying crowding out, the ADC, may be underreporting or exaggerating the crowding out
25 27 29 31 33 35 37 39 41 43 45 10 15 20 25 30 35
April May June July Aug Sept Oct
ADC p er con su lta nt Revenu e Average 2013 ADC
Figure 4. Sample period revenue 2013 and 3 year average overlaid with ADC of full group. Note ADC axis is on right
33
effect. To check this, a randomly selected group of thirteen individuals was tracked each month and the time spent on an ADC illustrated as in Figure 5.
The September Effect can be seen slightly, forcing individuals to reduce ADC length. However there is little evidence that individuals sacrificed call quality for quantity. In fact, the difference between May (the longest pre-MER ADC month) and July equates to just six seconds per call.
Intuitively these results make sense; a telephonic sales consultant can only compress a call so much (a certain amount of information needs to be discussed). In addition short calls jeopardize the probability of a sale. It is unproductive to reduce call length too significantly. Consequently it appears that ADC length only deviates slightly around a natural average. We thus find no support for this possible issue.
Besides the intrinsic differences between a laboratory and field study already mentioned, poor survey results may play a large role in the discrepancies found in Table 9 and the disagreement with theory of a number of variables. Although extensive steps were taken to explain the survey and remove all potential issues of privacy, a number of issues likely still remain stemming mainly from the difficulty of measuring unquantifiable factors such as relationships or fairness. With a response of 22 individuals, sample size may not have been large enough to adjust for individual biases in answers and identify an underlying trend; this small sample size expresses itself as variation between individuals when answering. Two top performing individuals may answer the question ―I work harder than needed‖ as a 3 and a 5.
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
April May June July Aug Sept Oct
AD C le ng th in mi nut es
34
Ideally they should both have answered with a 5. This effect will reduce statistical significance when comparing groups and may even reverse direction of effect.
Lastly, subjectivity comes into play when dealing with answers based on personal opinions and intensity ratings. Questions like ―I get on well with the rest of my team‖ are unlikely to attract either a 5 or a 1 response, since most respondents will choose to go with middle-of-the-road responses such as 3 or 4. We see this issue clearly in Table 8, where only one average answer falls below a 3.5. Combining this bias with natural variation makes it difficult to establish true opinions.
SECTION V
Discussion
With regards to our original hypothesis, Support was found for Hypothesis 1 (p.19), Hypothesis 2 (p.24) and Hypothesis 3 (p.19). The crowding out effect was clearly identified and tested. In net effect, although there was some variation of response among individuals, an increase in ADC was almost unanimous; indicating the strong dominance of the traditional disciplining effect outweighing the crowding out. This increased effort level was not sustainable and within a month post-control introduction ADC had dropped back to pre-MER introduction levels.
This provided us with two main periods to test for the crowding out effect which was identified in 11 individuals; a surprisingly high number which was reduced to be more in line with theoretical predictions once statistical significance was checked. These results are confirmation of the recommendations of previous authors that the economic field of incentive theory needs to encompass the very real crowding out effect. The number of under and over performers was closely matched in the crowded out sample group and the non-crowded out group. These results contribute to the literature because they occur in a situation of control rather than incentive and appear in a field study rather than a laboratory experiment. As such the author hopes these findings will add weight to the small body of existing literature with similar designs.
35
Survey data did not provide as strong support for previously identified characteristics of crowding out as was hoped in Hypothesis 4 (p.29). Although intrinsic motivation and principal agent relationship were in line with this literature, the remainder of the characteristics found either no statistical support or a reversed relationship. This can be attributed to the nature of the survey data and small sample size, rather than through any fault of pre-existing studies.
In the introduction it was stated that the importance of accurate outcome predictions is paramount for organizations. Our finding provides real world evidence that there is a downside to placing control on agents within an organization; these consequences need to be taken into account when planning a policy of control.
Conclusion
This study set out to identify whether crowding out of intrinsic motivation would occur in a real world field study under implementation of a minimum effort level. This effect was found with strong statistical evidence. As predicted, the traditional disciplining response effect still outweighed any crowding out effect in both number of individual responses and net effect on the targeted variable. Overall the control mechanism put in place by management achieved its desired effect, nevertheless the results of this field study show that serious consideration needs to be placed on the possible crowding out effects that may occur.
36 APPENDIX
Appendix I: Falk and Kosfeld (2006) Table 1. Illustration of the crowding out effect under
various control treatments (C5, C10, C20) Notice apart from C20, agents always choose higher effort levels when the principal does not control.
Appendix II – discussion of the 35calls hard floor MER is calculated.
In mid June 2013, management announced its intention to introduce a minimum number of daily calls (MER). On July 1st the control was introduced, however concessions were made based on agent feedback in the last week of June. In theory each consultant was required to make 80% of a 50 outgoing call requirement a day = 40calls. However in reality leeway of another 10% was given before punishment was imposed. As such the hard ‗floor‘ minimum effort requirement I will use going forward is 35 calls a day. Consultants were made aware of the number of calls they had made each day and as the month progressed. Punishments were handed out on a monthly basis using the simple formula; essentially making the calculation (Days at work x 35).
Punishment: In theory a consultant who fell below the monthly requirement was handed a letter of warning. This is a very serious punishment as three letter of warning provided adequate grounds for dismissal5. In reality very few letters of warning were actually handed out. However management made it public who the under performers were and individual
5
