Performance pay and the introduction of minimum wage : an empirical study

PERFORMANCE PAY AND THE

INTRODUCTION OF MINIMUM

WAGE: AN EMPIRICAL STUDY

Leo Huberts

Leo Huberts (10118462)

Bachelor Thesis

Faculty of Economics and Business

Supervisor Maurice Bun

23/12/2014

1

Performance pay and the introduction of minimum wage: an empirical study

Leo Huberts 1 INTRODUCTION

There has been a lot of discussion on bonuses and performance related pay in recent years. Public outrage with bonus structures in high level jobs providing the wrong incentives causes governments to consider bonus inhibiting regulations. At the same time plans are made to introduce performance pay for federal employees in both the US and Europe1_.

Including performance pay in compensation systems is one of the instruments used to provide workers with incentives to maximize their productivity. As the output of companies depends largely on the productivity of its workers, it is important to motivate employees to achieve maximum performance. Performance pay is the opposite of fixed pay, as it depends on the variable performance of the worker. In the Netherlands an increasing amount of firms implement performance related compensation and even the public sector is considering to incentivize teachers’ and police officers’ productivity by implementing performance related pay (Van Herpen, Van Praag & Cools, 2005).

In recent years there has been a lot of research on the incentives associated with

different pay structures. There is a large body of theoretical literature on how firms design pay structures to induce employees to maximize performance.2 Empirical research using company data tests the validity of the developed theories.3 As suitable data is hard to acquire, the majority of the empirical research depends on household, administrative or aggregated firm data, which allow for only indirect statistical inference (Pfeifer, 2012).4 This paper aims to contribute to the literature on pay structures by analyzing suitable personnel data of a single firm. This leads to the following central research question: how do worker behavior and productivity depend on performance pay systems?

1 _{See The New York Times (2014, July 31) for a report on British regulators setting strict rules for bonuses in}

banks, The Washington Post (2009, June 23) on US federal employees’ pay and the NRC Handelsblad (2012, October 23) and The London Times (2012, May 1) on performance pay plans in the Netherlands and the UK.

2 _{See Prendergast (1999) for a review of theoretical literature on incentive pay.}

3_{Examples of empirical research are given by Lazear (2000) on piece rates within Safelite Corp., Van Herpen,}

Van Praag & Cools (2005) on the effects of performance measurement and piece rates on motivation and Kishore, Singh Rao, Naramasimhan & John (2013) on bonuses and commissions.

4 _{An example of research using aggregated firm data is given by Gielen, Kerhofs & Van Ours (2006). They find}

2 A dataset provided by Pepperminds Nederland allows this question to be analyzed. Pepperminds Nederland B.V. is a large field marketing company based in the Netherlands. The company was founded in 2002 and offers retailers, publishers and charities a sales and fundraising channel. Although the company operates in four different markets in Europe, the data used for analysis in this paper was collected during their activities in the Netherlands only. The data covers all of 2013 and January through June of 2014.

As of the first of January 2014, the company changed its pay structure by introducing fixed minimum wages. Before then, the fixed pay was less than minimum wage. The variable part of the wage was divided into both piece rate pay and bonuses. Incentive pay in the form of piece rates is defined by payment on the basis of output (Lazear, 1986). Bonuses are

granted when a specific output number is reached, this can be both individually and as a team. Due to legislation Pepperminds was forced to change the fixed pay to minimum wage,

changing the character of the pay structure. As the fixed pay went up, the piece rates and bonuses went down. The data of roughly 3000 unique employees in 2013 and 2200 unique employees in 2014 are available, with worker specific shift performance as the appropriate performance measure. This shift information will be analyzed using a range of variables and estimation techniques, to identify the effects of the payment system change on workers’ performance.

The rest of this paper is organized as follows: In section 2 relevant literature will be reviewed to provide a sound theoretical background and a mathematical economic model will be presented to formalize the incentives and choices within the different payment systems. In section 3 the research design will be discussed. Section 4 will thoroughly analyze the three compensation systems covered in this paper. Section 5 will present the empirical analysis and its results and section 6 will conclude this paper with a review and discussion of these results and their implications.

2 BACKGROUND LITERATURE

In this section I discuss relevant theories to provide a theoretical background for interpreting the empirical analysis. The agency theory is important in modelling the choices and

relationships in the performance pay structures. Agency theory is supplemented with further theory on motivation. Theories on minimum wage are examined, they can be important if a shift in the composition of the workforce at Pepperminds occurs, which in turn may affect

3 total productivity. Furthermore, they may provide further understanding of observed effects and will help to develop implications for policy makers.

Agency theory

Agency theory offers an economic approach to human motivation and performance. It is the leading perspective used by scholars to support financial incentives to motivate or increase performance (Young, Beckman & Baker, 2012). An important assumption of the agency theory is that individuals make rational decisions based on their utility functions, which depend on some combination of income and leisure (Chambers & Quiggin, 2005). The principal-agent framework of the agency theory acknowledges that the principal, the risk-neutral employer, who rewards the agent, the risk-averse employee, might have different interests than the agent. The agent’s interests are not known to the principal, causing a situation of asymmetric information. As the agent pursues its own maximum utility, the principal can expect the agent to take actions to maximize his utility, possibly contrasting the principal’s objectives. The agent is assumed to have priorities that are not equal to the

principal’s, so the agent will likely pursue personal objectives ignoring their responsibilities to the principal if there is no financial consequence. Therefore an important issue for the

principal is how to structure the relationship and compensation to motivate the agent to work towards the principal’s objectives.

There are two general approaches the principal can take. The first approach entails specifying and monitoring the agent’s work activities. As the principal needs to have a clear view and understanding of the agent’s activities beforehand and monitor the activities during, this can be very expensive (Young, Beckman & Baker, 2012). The second approach involves forcing the agent to bear at least some of the production risk. An optimal incentive contract therefore would involve a performance pay structure which connects the agent’s pay to production and thereby indirectly to the agent’s effort (Kunz & Pfaff, 2002). This imposes measurement costs on the principal, as it is crucial to the performance pay contract that performance is measured accurately. Pepperminds B.V. combines these two approaches, using the first approach to filter its workers by dismissing underachieving workers and the second approach to motivate workers to achieve maximum sales.

Although intuitively it seems logical that including performance related pay in the

compensation structure will increase workers’ effort, this is not always the case. In an hourly wage system, workers are usually required to achieve some minimum level of output. If they

4 underachieve, they are let go. When a firm switches to a performance pay system, the chosen output by workers in a performance pay contract might be less than the minimum amount set by the firm under hourly wage (Lazear, 2000). This would cause a decrease in the workers’ output. Furthermore, the selection during hourly wage might cause only the most able to make the cut. This selection disappears when piece rates are introduced, possibly increasing the variance in performance resulting in a lower average workers’ output (Lazear, 2000). This causes inappropriately designed compensation systems to be possibly counterproductive.

Performance Pay and motivation

Psychological theories have been inserted into economic theory on motivation, in an attempt to explain results found by sociological experimental research; contradicting agency theory. As a result, motivation is split up into intrinsic and extrinsic motivation. Extrinsic motivation is influenced by external interventions, like monetary compensation, and widely used by agency theorists to assess the amount of effort an agent is expected to carry out (Van Herpen, Van Praag & Cools, 2005). Intrinsic motivation, largely ignored by agency theorists, causes employees to undertake tasks without monetary compensation; for immediate need satisfaction or for its own sake (Calder & Staw, 1975). Within sociology there are some theories that predict a possible reduction in workers’ output by adopting performance pay. Experimental research by Deci (1971) first showed that, at least in certain situations, intrinsic motivation to perform decreases when monetary rewards are introduced. Crowding theory extends this idea, by linking intrinsic motivation to agency theory; it states that under certain conditions the effects of monetizing incentives on individual performance might be negative (Van Herpen, Van Praag & Cools, 2005). As the agent perceives the performance related pay as a form of external control, thereby increasing the external pressure on the agent and causing the intrinsic motivation to be ‘crowded out’; causing a decline in productivity. The opposite effect occurs when the agent perceives the external intervention in the form of performance pay as supporting or informative, this causes ‘crowding in’ which may result in an increase in intrinsic motivation (Frey & Oberholzer-Gee, 1997). The total resulting effect of monetary compensation on motivation is therefore undetermined.

Both Van Herpen, Van Praag & Cools (2005) and Lazear (2000) find no evidence to support the crowding out theory. Van Herpen, Van Praag & Cools (2005) find no correlation between monetary compensation and intrinsic motivation. Lazear (2000) analyzed the effects on worker’s productivity of an auto glass company’s transition from a full fixed pay structure

5 to a full piece rate one. He found a dramatic increase in performance, with average output per worker increasing by 44 percent, contrasting theories in sociology predicting opposite

reactions. The effect of increasing the hourly wage part of the pay structure of Pepperminds is opposite in its direction to the investigated structure change by Lazear (2000). So a decrease in performance is expected according to Lazear’s (2000) results. In the conclusion of this paper the results will be compared to those found by Lazear (2000) and others.

Minimum wage

Economic regulation of firms and industries affect compensation structures, with minimum wage regulation in the Netherlands directly affecting the performance pay structure of Pepperminds. There exists a large body of research on the effects of minimum wages on a macro-economic scale.5 There is no real consensus on the effects of minimum wage on employment, although most of the literature seems to point towards a negative effect of minimum wages on employment (Neumark & Wascher, 2006). Empirical research on the subject is limited and often on a macro-scale. 6 The results of this research often contradict the standard textbook competitive model of the labor market, which predicts a decrease in

employment associated with minimum wages (Machin & Manning, 1997).

Some research has been done into the effects of minimum wages on different sectors, including sectors in which at least some of the pay is performance related, like the restaurant industry covered by Card (1991). He finds that an increase in the minimum wages in

California caused no major effects on employment within affected eating and drinking establishments, even noting a slight increase in employment.

The minimum wage system in the Netherlands differentiates across ages from € 2,59 for 15 year olds to € 8,63 for ages 23 and older (see table 1). This differentiation might cause the implementation of minimum wages to shift the composition of Pepperminds’ workers. This can have an effect on average worker output, as productivity might vary across different age groups. Economists believe teenage unemployment will see the greatest impact of

minimum wage regulation (Mankiw, 2013).

5 _{Examples of the effects of minimum wages are given by Machin & Manning (1997) on employment and}

income distribution in Europe and Neumark & Wascher (2006), who review existing literature on the effects of minimum wages in and outside of the United States.

6 _{Some empirical case studies have been done, notably the fast food case study by Card & Kruger (1993) and a}

state-wide case study by Card (1991) into the effects of a minimum wage hike in California. Gilman, Edwards, Ram & Arrowsmith (2012) reviewed the effect of a new minimum wage on small companies in the UK.

6 As Pepperminds employs a large group of teenagers, with

employees’ ages ranging from 17 to 24, it will be interesting to see if the effects are indeed large in this case. The effects on different age groups can be examined using the Pepperminds data, with the amount of shifts in different age groups being an indication of employment effects.

As recognized by Gilman, Edwards, Ram & Arrowsmith (2002) minimum wage can influence different pay structures in different ways. The minimum wage might simply raise hourly wage levels in fixed pay structures, but might change the entire pay structure when performance pay is involved, as in the case of

Pepperminds. The direct influence on hourly wage systems might be clearer than the influence on performance pay systems, as the

implementation of minimum wage might reverse the possible effect

on output the performance pay structure caused when implemented. The available data and

Age Hourly minimum wage 23 years and above € 8,63 22 years € 7,33 21 years € 6,25 20 years € 5,31 19 years € 4,53 18 years € 3,93 17 years € 3,41 16 years € 2,98 15 years € 2,59 Table 1: 2014 minimum

wage levels in the Netherlands

Change in performance pay system/ Minimum wage

implementation

Extrinsic motivation Intrinsic motivation

Prospects

Company

Composition of workforce

Workers’ effort _{Total output}

Incentives Perception + (Crowding in) - (Crowding out) / Utilities +/- Wages Costs Applying Hiring

7 subsequent analysis in this paper can provide some insight into the effects of Dutch minimum wage regulation within the performance pay structure used by Pepperminds.

To clarify how the different concepts discussed in this paper relate, the essence of the described model and theories is summarized in figure 1. The change in compensation system is represented in the top left corner of figure 1, where the three main theories link the change to the effects on total output in the bottom left. Agency theory provides a platform for the interpretation of the effect of the change through incentives on extrinsic motivations and subsequently on utilities and workers’ effort. Crowding theory links the change through workers’ perception and intrinsic motivation to workers’ effort. These two theories provide the link between the change and effects on workers’ effort, which affects workers’

productivity. The change in compensation system also affects the composition of the workforce through both the changed cost for the employer and the different wages for the prospect employees. Through changes in hiring by Pepperminds and applying by future employees the changes might cause sorting and change the composition of the workforce, which can in turn affect the productivity of the workers at Pepperminds.

Mathematical economic model

To further clarify the choices the employees at Pepperminds are faced with, the

situation is modelled below. This modelling is loosely based on the model proposed by Lazear (2000). The workers utility function is given by U, where we assume that the utility a worker obtains depends on income and effort.

Worker utility=U(Y,X) Y=income, X=effort

The output a worker achieves is assumed to depend on effort and ability, giving the following output function:

Output=e=f(X,A) X=effort, A=ability Where the derivative of f(X,A) to both X and A is positive: 𝜕𝑓(𝑋,𝐴)

𝜕𝑋 > 0, 𝜕𝑓(𝑋,𝐴)

𝜕𝐴 > 0.

The minimum amount an employee has to sell to work at Pepperminds is given by e0. To

achieve e0, a worker has to provide some minimum amount of effort which depends on the

ability of the specific worker, which gives us an X0(A); amount of effort that satisfies the

8 X0(A) satisfies e0=f(X0(A),A) e0=minimum output to stay on

The derivative of X0 is less than zero, as less effort is needed at higher abilities to achieve the

same output: 𝜕𝑋0

𝜕𝐴 < 0. For a certain A0, the utility will equal the utility from leisure:

U(W,X0(A0)) = U(0,0) U(0,0)=amount of utility from leisure

Anyone with ability greater than A0 will receive rents from working, as their utility for this

work will be higher than for leisure. To model which worker chooses to work at Pepperminds and which worker chooses to work at an alternative company, we assume that there is some upper cutoff Ah at which a worker chooses to work at the best alternative.

U(W,X0(Ah))=U(W’(Ah),X’(Ah))

Ah is cutoff at which worker is indifferent between working at best alternative and working at subject firm, with associated W’=wage and X’=effort at other firm. The total wage a worker earns at Pepperminds can be modelled by:

W=F+cf(X(A),A)+diTi+djIj F=fixed amount, c=commission, di=dummy team bonus i,

dj=dummy team bonus j, Ti=team bonus amount,

Ij=individual bonus j

Ability of workers at Pepperminds will be larger than A0 and smaller than Ah.

Utility of employee at Pepperminds will be:

U(F+cf(X*(A),A)+diTi+djIj,X*(A))

Where X*(A) is the chosen level of effort from the individual with ability level A. Now indifference curves for different ability levels can be drawn in the compensation graph, giving a representation of the dynamics within workers preferences and compensation structures. Indifference curves for two individuals are shown in figure 2. If the individual with purple indifference curves were to be a 17 year old promoter, the optimal solution in the new system would result in a lower utility and lower output (indicated by the purple dotted lines). If the individual with the blue indifference curves were a 23 year old captain, he or she would be able to achieve a much higher utility by performing less in the new compensation system (indicated by the blue dotted lines). The details of the compensation system will be discussed in the system analysis section.

9

Proposition 1

Average effort will not increase in the January 2014 system, as the average return on added performance is less in the new than in the old compensation system.

Proposition 2

Average ability will not increase, as higher aged workers are able to earn the same with less effort they need less ability to achieve the same compensation. Ah will not increase,

but A0 will decrease, as wages for lower scores will rise for most ages, lower ability workers

can now earn sufficient wages with the lower scores. Lower aged workers receive less extra compensation for the higher performances in the new system, making it less attractive for higher ability workers, possibly causing a decrease in Ah. Although they earn roughly the

same compensation in the lower score levels, at the higher scores (assumed to be achieved only by high ability workers) the reward in the new system is less than in the old system, making it less attractive for higher ability individuals to work at Pepperminds. This

proposition becomes especially clear if the team bonuses are excluded, as then the new system allows much less reward for higher performance.

€0,00 €50,00 €100,00 €150,00 €200,00 €250,00 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

17 All ages old 23+c All ages oldc

10

Proposition 3

Average age will not decrease. As average pay at any score for higher aged individuals rises with the new system, a decrease in average age would be counter intuitive and could point towards unobserved influences.

Proposition 4

The variance in scores will decrease. If we assume that a worker will choose to invest less effort if it can produce the same utility; older workers in the new system will likely be able to achieve higher levels of utilities with less effort, and younger workers will be able to achieve the same utility with the same effort at lower score levels, and only a lower level of utility at the higher levels.

These propositions will be discussed in the conclusion using the results of the empirical analysis.

3 RESEARCH DESIGN

This paper studies the effects of a change in compensation system on the performance at Pepperminds. In section 4 the compensation systems will be analyzed to clarify the system design and provide insight into the changes in monetary incentives for employees. This will be followed by an empirical analysis of the available dataset.

The dataset was provided by Pepperminds Nederland B.V., a large field marketing company based in the Netherlands. The data used for analysis in this paper was collected during their activities in the Netherlands and covers all of 2013 and January through October of 2014.

As of the first of January 2014, the company changed its pay structure by introducing fixed minimum wages. Before then, the fixed pay was less than minimum wage for all ages, except for ages 17 and 18 which minimum wages are slightly lower than the original fixed pay. The variable part of the wage was divided into both piece rate pay and bonuses. The data of roughly 3000 unique employees in 2013 and 2200 unique employees in 2014 are available. Descriptive statistics will be presented to elaborate on the basic properties of the data. These descriptive statistics will also provide a first insight into any changes in workforce

11 Almost 60,000 for 2013 and more than 44,000 unique shift per employee data points are available. Each observation consists of the performance of the employee and related payments and revenue, general information on the project and location and can be linked to employee information data in a different data set. These employee data will supply important variables like age, tenure, gender and schooling which will be included in the regression as control variables. Seasonal control variables need to be included, as different seasons produce very different results all-round. The main dependent variable will be the individual worker performance within shifts.

Multiple regressions will be performed, with different explanatory variables to measure the impact of the change in performance pay structure. By comparing average performance within shifts before and after the change date, the relationship between the different payment systems and performance can be examined. Since the pay structure changes differently for different age groups, as minimum wages rise with age, an interesting analysis into the changes due to the minimum wage implementation within different age groups is performed. Pepperminds employs three types of positions; talents, promoters and captains. The talents are employees that have completed less than four shifts, promoters are employees that have completed at least four shifts and have performed well enough to earn a promoter position. The compensation system for promoters and talents is the same. Finally the captains are the team leaders, they earn slightly more than the other two positions mainly due to a higher fixed wage to compensate them for organizational activities. The difference between these systems will be clarified in section 4. As the compensation systems differ, a variable for position will be included in every model and the results for different positions will be

discussed.

The main resulting regression models have the following form: yi = β’D1i + β’D2i + γ’Wi + εi

Where yi is the workers’ shift performance expressed in the achieved number of sales, Dji is a

dummy variable indicating the compensation system 1/1/2008 (D1i= D2i=0), 1/1/2014 (D1i=1,

D2i=0) or 1/7/2014 (D1i=0, D2i=1) and Wi represents a set of control variables. The latter

contains worker attributes (position, age, tenure, schooling, gender), project fixed effects, organization dummies and weather variables. This model will be estimated by ordinary least squares.

12 The coefficient of Dji will answer how, in the case of Pepperminds, worker behavior

and productivity depend on the performance pay system. The results will be translated into implications for performance pay policies within companies and governments and will be of value in the discussion on minimum wage legislation.

4 SYSTEM ANALYSIS

Compensation system

The compensation structure change needs to be analyzed in order to find the relative changes in compensation across ages and functions. Pepperminds’ employees fall into three categories; talent, promoter and captain. A ‘talent’ is an employee that has worked less than 5 shifts, a ‘promoter’ is an employee that has worked 4 shifts and performed well enough on these 4 shifts to be offered a contract. A captain is the leader of the shift team, he or she gets a higher base salary as they are paid extra for organizational activities as team leader. Talent and promoter functions have identical pay, a captain earns a larger base salary and in some cases higher bonuses.

Assumption 1

A necessary assumption when analyzing the compensation systems is made on the commission Pepperminds pays its employees on every sale. This commission varies with different types of products (higher commissions for higher donations for example), but does not vary between the old and new compensation system. Therefore a fixed commission of 3 euros per sale is included in the model. This fits the purpose of this analysis, as only the differences between the old and new system are important.

An overview of the compensation structures is shown in figure 3. The horizontal axis represents the employees’ performance in the number of sales in one shift, the vertical axis gives the resulting compensation in euros. The payment structure from before the change in January 2014 is given by the two dotted lines (red for promoters/talents and green for captains). The new payment structure is represented by the other 14 lines, with each line representing an age with either a promoter/talent function or captain function (red colors for promoters/talents and blue colored lines represent captain functions). The kinks in the graph represent the bonuses. Clear in the graph is the age differentiation in minimum wages and the resulting differences across ages for all results.

13

Promoters & Talents

If we look at promoters and talents aged 17 and 18 in figure 4a, we see that

compensation is quite close in the new and old systems up to a performance of 8. Above 8, the old system awarded higher compensations for all scores. The monetary incentive for ages 17 and 18 to perform above a score of 8 is therefore less in the new system.

If we look at promoters and talents with ages 22 and 23 in figure 4b we see a very different image. Compensation in the new system is higher for all scores for 23 year olds and higher for all but 2 scores for 22 year olds. The difference in compensation is largest in the lower scores. The older compensation system almost catches up in compensation for 23 year olds at a score of 9 and keeps quite close after. This suggests a decreased incentive to perform up to a score of 9 in the new system compared to the old system. Comparing graphs 3a and 3b shows the effects of the change will most likely differ across age groups because the absolute changes in wages are very different, which justifies analyzing the effects of the change in compensation system for different age groups.

Figure 3: Compensation structure old-new Promoters/Talents (red-red) and Captains (green-blue) €0,00 €50,00 €100,00 €150,00 €200,00 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 T o tal co m p en satio n p er s h if t

Number of sales per shift

17 18 19 20

21 22 23+ All ages old

17c 18c 19c 20c

14 If we look at the trend lines of the new and old compensation systems for promoters and talents, which are the same for all age groups, we find the slope of the old compensation system trend line is steeper than the trend line for the new compensation system (red for the old system with a coefficient of 10.6, blue/grey for the new system with a coefficient of 9.2). The fact that the slope is steeper shows that the increase in compensation of extra output for promoters and talents was larger in the original compensation structure than in the new

structure introduced in 2014. This would imply a smaller monetary incentive to perform in the newer system. y = 9,1765x + 31,58 SE trend: 0,333 R² = 0,9818 y = 10,585x - 0,475 SE trend: 0,448 R² = 0,9756 €0,00 €50,00 €100,00 €150,00 €200,00 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

22 23+ All ages old

b. New and old compensation structures for promoters/talents of ages 22&23 €0,00 €50,00 €100,00 €150,00 €200,00 €250,00 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

17c 18c All ages oldc

c. New and old compensation structures for captains of ages 17&18 y = 11,235x + 46,42 SE trend: 0,445 R² = 0,9785 y = 12,393x + 10,375 SE trend: 0,621 R² = 0,9661 €0,00 €50,00 €100,00 €150,00 €200,00 €250,00 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

22c 23+c All ages oldc

d. New and old compensation structures for captains of ages 22&23+ €0,00 €50,00 €100,00 €150,00 €200,00 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

17 18 All ages old

a. New and old compensation structures for promoters/talents of ages 17&18

15

Captains

The representation of the two compensation systems for captains is shown in figure 4c and 4d. The results are similar to the results for the promoter/talent employees. For 17 and 18 year olds the old system produces largely higher compensation than the new system. Ages 22 and 23 receive more compensation in the new system at almost any score, with the difference being especially large for scores under 9. The trend lines for captains are found to be less steep in the new system than in the system before 2014 (with a coefficient of 12.4 for the old system trend line (in red) and a coefficient of 11.2 for the new system trend line (in

gray/blue)), indicating a smaller monetary incentive to perform for captains in the new system analogous with the promoter/talent results. The coefficients of the trend lines also reveal a steeper performance compensation for captains compared to promoters and talents, which suggests higher monetary incentive for captains to perform compared to promoters/talents.

Note on team bonuses

Team bonuses are awarded only when the team performs well as a whole, with average scores of 6 and 9 in the old system and 4,6,8,10,12 in the new system. In the new system there is no more individual bonus and only team bonuses are awarded. This changes part of the dynamic of the incentives for the employees. In figure 5 the compensation structures for promoters/talents are reported in 5a and for captains in 5b, with the team bonuses removed from this data. This would present a situation in which an employee would be faced with an underperforming team incapable of achieving the score required for a team

y = 3x + 24,42 R² = 1 y = 7,8713x + 22,25 SE trend: 0,256 R² = 0,9854 €0,00 €20,00 €40,00 €60,00 €80,00 €100,00 €120,00 €140,00 €160,00 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

b. New and old compensation structures for captains with team bonuses excluded y = 3x + 16,58 R² = 1 y = 7,8713x + 6,65 SE Trend: 0,256 R² = 0,9854 €0,00 €20,00 €40,00 €60,00 €80,00 €100,00 €120,00 €140,00 €160,00 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

a. New and old compensation structures for promoters/talents with team bonuses excluded

16 bonus. Clear in both graphs is the linear form of the newer compensation structure, and the much steeper curve of the old compensation system (a slope of 7.87 for the old and 3 for the new system). It is reasonable to assume that in a situation like this, both a promoter/talent and captain will have less monetary incentive to achieve higher scores in the new than in the old system.

The third system

To complicate things, Pepperminds implemented a second change in its compensation structure in July of 2014. As they became aware of the drop in incentive that the system had caused, they decided to change the structure for captains. The captains once again received individual bonuses and their fixed wage was reduced to the level of promoters and talents. Figure 6 represents the new situation for captains (June 2014 system in blue, 2013 system in red) and shows that the trend line in the newest system is even higher than it was in the older system (with a coefficient of 16.1 in the newest and 12.4 in the older system), leading to the conclusion that for captains working in the newest wage system the monetary incentive to perform is much higher than in the January till June system and higher than the system from before 2014.

From the analysis of the compensation systems it is clear that in all situations and for all functions of Pepperminds’ employees, the monetary incentive to perform higher is lower in the compensation system implemented in January 2014 compared to the system in 2013.

y = 16,107x + 12,83 SE Trend: 0,547 R² = 0,9841 y = 12,393x + 10,375 SE Trend: 0,621 R² = 0,9661 €0,00 €50,00 €100,00 €150,00 €200,00 €250,00 €300,00 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

17 The newest system introduced in June 2014 increases the monetary incentive for captains only. This increase is substantial, especially compared to the January-June system for captains. The results of the data analysis in this paper need to be evaluated with the distribution of the different compensation systems in mind.

5 EMPIRICAL ANALYSIS

The data supplied by Pepperminds contains information on individual workers and shifts from the beginning of 2013 through October 2014. As discussed in the compensation system analysis, the consequences of the first and second wage system change are different among ages and positions. The final data used in the data analysis contains 57,967 shifts worked in 2013 (whole year) and 44,160 shifts in 2014 (January through October). This is panel data with individual employees performing multiple shifts at different moments in time. Descriptive statistics on important variables is shown in table 2. The productivity measure used in this analysis is the score an employee manages to achieve in one shift of five hours. This is the measure Pepperminds uses to evaluate their employees and is the basis for their performance based pay. The average score achieved in 2013 was 4.28 units sold with a standard deviation of 7.01. In the 1/1/2014 wage regime an average score of 2.94 with

Regime

Individual 1/1/2008 1/1/2014 1/7/2014

Total Captains (shifts) 14,950 7,460 4,288

Total Promoters (shifts) 34,401 15,863 10,259

Total Talents (shifts) 8,576 3,789 2,468

Avg. Score 4.28 2.94 3.34 Std. Dev. Score 7.01 3.23 3.45 Avg. Age 20.16 20.62 20.82 Std. Dev. Age 2.25 2.26 2.26 Female 21,948 10,534 6,512 Male 35,987 16,578 10,503 External Avg. Rainfall(mm) 20.16 17.44 25.78 Avg. Temperature (C°) 10.20 10.42 17.09 Unique projects 165 137 106

18 standard deviation 3.23 is found. The 7/1/2014 has a slightly higher average score of 3.34 with standard deviation 3.45.

As Pepperminds lets its employees work different projects over time, with different characteristics and results for each project, the project fixed effects are an important addition to the individual control variables. Of the 274 unique projects that Pepperminds has done in the data’s timespan, 29 span all three wage regimes, 76 have seen two and 169 only one of the three regimes. Furthermore, as external influences like rainfall and temperature on a given day can influence the amount and willingness to buy of people on the street they are included in the model. Data from the Royal Netherlands Meteorological Institute for temperature and rainfall in ‘de Bilt’ were used and added to the data. Descriptive statistics on the external data used is given in the bottom part of table 2.

The preliminary results of the descriptive statistics support the compensation system analysis and the propositions made. At first glance we see a large decrease in average score and an increase in age, as well as a decrease in the standard deviation of scores.

When regressing the score for the different shifts, the residuals show significant signs of heteroscedasticity and autocorrelation. The most important reason this occurs might be the fact that employees work in teams. The performances of the workers within these teams are given by different shift data points, but might be correlated. This is why clustered standard errors are used, to allow some correlation between the errors within clusters. These clusters are chosen to depend on the date of a shift. As not only team connections might cause correlation between shifts, other exogenous daily factors might cause correlation. Choosing these cluster will affect the results for the weather variables, as the weather effects were added on a daily basis. The clustered standard errors correct automatically for the heteroscedasticity in the error terms.

A linear regression of score on wage system, age, tenure, position and other control variables produces a large significant effect of the change in wage system (see table 3). All of the coefficients found are significant, except for temperature. The results of this regression are profound. The 1/1/2014 system reduced scores by 1.376 units compared to the 1/1/2008 system and the 1/7/2014 system produces scores of 0.954 units less than the 1/1/2008 system.

19 The explained variance of this regression is understandably low, as differences between projects and locations are ignored by this specification.

Including dummy variables for project category and Pepperminds location produces smaller, but still significant negative coefficients. Because the wage system and projects are correlated, since some of the projects ran through only one system and some through all, including variables related to the projects will most likely absorb some of the effect of the

# System 1/1/2014

System 1/7/2014

Promoter Talent Tenure Age R2 _Description

1 -1.376 (.042) -.959 (.055) -1.308 (.046) -1.953 (.066) -2.5e-3 (2.0e-4) 0.059 (8.7e-3)

0.023 Clustered SE’s, Rainfall,

Temperature, included 2 -.539 (.051) -.547 (.063) -.830 (.034) -1.286 (.075) 2.4e-3 (1.9e-4) .083 (7.1e-3)

0.386 Clustered SE’s, Rainfall,

Temperature, 35 project categories and 16 organization dummies included 3 -.427 (.045) -.436 (.066) -.859 (.028) -1.551 (.045) 2.6e-3 (1.4e-4) .076 (5.3e-3)

0.674 Clustered SE’s, Rainfall, Temperature and 273 project dummies included 4 -.445 (.067) -.457 (.093) -.096 (.037) -.0966 (.057) -.096 (.037 ) -.062 (.037)

0.730 Clustered SE’s, Rainfall, Temperature, 273 project dummies and 4,222 worker specific dummies included 5 -.179 (.0137) -.178 (.016) -.195 (8.5e-3) -.384 (.016) 8.5e-4 (4.8e-5) .022 (1.9e-3) AIC 4.886 Clustered SE’s, GLM Poisson regression including project categories, organization dummies and weather variables 6 -.131 (.013) -.141 (.016) -.198 (6.4e-3) -.454 (.010) 9.7e-4 (4.1e-5) .020 (1.4e-3) AIC 4.482 Clustered SE’s, GLM negative binomial regression including project categories, organization dummies and weather variables 7 -.395 (0.046) -.852 (0.032) -1.559 (0.051) 2.9e-3 (1.6e-4) 0.079 (6.0e-3) 0.693 Same as regression 3, except now only for the first 6 months of both years

8 0.068 (0.097) -0.910 (0.044) -1.526 (0.057) 2.5e-3 (2.4e-4) 0.084 (8.1e-3) 0.508 Same as regression 3, except now only the shifts between 1/7 and 15/10 9 -0.533 (0.052) -0.337 (0.067) -0.859 (0.028) -1.545 (0.044) 2.7e-3 (1.4e-4) .075 (5.3e-3)

0.675 Same as regression 3, now including 52 week dummies

Table 3: Coefficients and test statistics of multiple regressions, standard errors in brackets and insignificant

20 wage system change. This is confirmed by regressing score on the individual variables,

weather variables and 273 project dummies. Although the coefficients are smaller, they are still significant. Even including all the project fixed effects, giving a R2 _{of 0.674, the model}

indicates a drop of 0.4272 score per shift in the 1/1/2014 and 0.4356 in the 1/7/2014 system.

Regressions 7 and 8 in table 3 report the coefficients of regressions where the same periods are considered for both 2013 and 2014. Regression 7 considers only the first 6 months of both years, as this was the period the 1/1/2014 system was used in 2014. Regression 8 considers only the period from the first of July through the 15th of October. This is the period of available data for the 1/7/2014 system in 2014. The results of the 7th regression are very similar to the results in the third, which is as expected. The 8th regression shows no significant results for the pay system change, possibly due to a lower amount of observations. The last reported regression coefficients result from adding week dummies to the third regression specification. The coefficient for the 1/1/2014 system grows slightly compared to the specification without week dummies and the coefficient for the 1/7/2014 system is approximately one tenth lower than the coefficient in the third regression.

Individual fixed effects

The data can be viewed as panel data containing multiple observations over time for a set of individuals. To further control for unobserved heterogeneity between individual

workers, fixed effects regressions might be used including separate intercepts for every single individual. When adding 4,222 worker specific dummies to the model (regression 4 in table 3), the effects of the regime change remain large and significant. The coefficients given by this regression represent the effect of the change on the individual incentives for Pepperminds employees. This shows that the first change caused a decrease of approximately 0.445 unit sold per shift compared to the 1/1/2008 situation. The 1/7/2014 regime saw workers selling 0.457 units less per shift than the in the 1/1/2008 system. The average score is 4.28, which means the coefficient indicates a reduction in sales of more than 10%.

Count data

The dependent variable ‘score’ has so far been used as a continuous variable in a linear regression analysis. The results of such linear regressions produce significant results

21 and offer easier interpretation in the sense that the coefficients are marginal effects. The actual format of the scores, however, is not continuous. The score a worker achieves can be any of the integers in the set {0,1,2,3….}. This means the scores can be seen as count data. Different estimation techniques are available for count data dependent variables. The estimations results using the Poisson maximum likelihood count model are reported in row 5 in table 3. The estimations using a Negative Binomial distribution are given in row 6. A comparison of the two distributions and the score data is shown in figure 7, where it is clear that the negative binomial distribution has a better fit. A goodness of fit test supports this conclusion, with a p-value of less than 0.000 rejecting the null hypothesis of the Poisson distribution being a good fit. The detailed test results are reported in the appendix.

The results of these count data regressions are in line with the results found in the ordinary least squares regressions. The coefficients are not as easily interpreted as the OLS coefficients, but both regressions show a significant negative impact of the system changes. The incident rate ratios for the implementation of the 1/1/2014 regime of the Poisson regression equals 0.836 and 0.877 in the Negative Binomial model. They suggest that the implementation of the new compensation system caused a reduction in score by 16.4 percent

Figure 7: Comparison of score distribution and fitted Poisson and negative binomial distributions

0 .0 5 .1 .1 5 .2 Pro p o rt io n 0 2 4 6 8 10 k

observed proportion neg binom prob

poisson prob

22 using the Poisson and 13.3 percent using the Negative Binomial distribution. These are in line with the results from the linear models.

The large and significant effects of the compensation system changes are clear in all regressions presented. There are two possible interpretations of this effect. First the pure incentive effect of the regime change on extrinsic motivation, causing a decrease in effort and subsequently score. Second the effect might be due to sorting, as the workforce might have changed in composition due to the different compensation system.

There is no evidence in the results of presence of the Hawthorne effect7. This can be seen by examining the results of the regressions on promoters and talents in table 4. The reduction in scores for the 1/7/2014 system is larger than the reduction in the 1/1/2014 system, although nothing changed in compensation system for promoters and talents. This suggests the reduction in score was not a short-term response due to the change.

The results of the coefficients for the different position support the compensation system analysis. Captains are performing better in the 1/7/2014 system compared to the 1/1/2014 system. As the change was only positive for the captains, the absence of a positive change in performance for promoters and talents is as expected.

Sorting

To see the effects of the system changes on workforce composition, a quick look at average ages shows a steep increase in average age of almost 0.7 years (see table 5). This increase in older workers can be easily explained by the differentiation in minimum wages across ages, as higher ages earn more fixed pay than lower ages in the newer system. The increased average age can have both a negative and positive effect on output. Since age is significant across all regressions and higher ages indicate higher scores, one might expect

7 _{Hawthorne effect suggests that any change is likely to bring about some short-term gains in productivity.}

Position Coefficient system 1/1/2014 Coefficient system 1/7/2014

Captains -.5985837 (.0704787) -.5564017 (.1143034)

Promoters -.4507596 (.0499383) -.514737 (.0719901)

Talents -.1930379 (.0608271) -.2648825 (.1036064) Table 4: Regression coefficients for the three positions at Pepperminds

23 average scores to increase with an increase in the average age of the employees. In this case the opposite might be true. As higher aged employees now earn the same wages for lower score levels, lower ability individuals might now be able to earn satisfactory wages at

Pepperminds and apply. This might cause the ability of the average worker to decline and thus reducing score levels. It is not clear which of these effects is stronger in this case.

The results for different age groups are in line with the expectations formed in the compensation system analysis. As higher age groups earn more fixed pay and have the option to achieve the same utility with much less effort, the effect on performance is larger for the higher ages as can be seen in table 6. The effects become smaller down to age 18, then rise a little bit for the lowest ages. This might be explained by the fact that from ages 18 and younger, the employees in the new compensation system earn both less fixed pay and less performance based salary, resulting in a much lower monetary incentive for any score.

5 CONCLUSION

The results show that the effects of changes in compensation system on the employees’ performance at Pepperminds have been quite large. All of estimated

specifications indicate a significant reduction in performance as result of a change in pay system with lower monetary incentive to perform. Although the details of the compensation system and the changes for different positions and ages are quite intricate, the results for all

System Mean age

1/1/2008 20.156

1/1/2014 20.617

1/7/2014 20.824

Total 20.390

Table 5: Mean ages for the different compensation systems

Age group Coefficient system 1/1/2014 Coefficient system 1/7/2014

>23 -1.141 (.137) -1.346 (.170)

22-23 -.457 (.070) -.555 (.102)

20-21 -.308 (.062) -.302 (.084)

18-19 -.298 (.055) -.298 (.055)

<18 -.404 (.175) -.853 (.462)

24 combinations of worker positions and ages have been significant. The employees at

Pepperminds have reduced their sales output following the reduced monetary incentive given to them by the newer compensation systems at Pepperminds.

The results support the first proposition that average effort has at least not increased in the January 2014 system. The sharp drop in performance found, even when including worker specific dummies, shows the employees have most likely decreased their effort which resulted in the drop in sales performance. The second proposition that average ability did not increase is supported by the results. The significant reduction in performance by talents indicates that the workers applying at Pepperminds have not increased in ability or effort. The third

proposition is supported by a steep increase in mean ages in the new systems, most likely due to the higher compensation for the higher ages at all score levels. The final proposition states that the variance will decrease, which is confirmed by the standard deviation reported in the descriptive statistics. This decrease in variance may be caused by the effects of the decrease in effort and ability.

No evidence was found to support the crowding theory. Minimum wage

implementation will be likely seen as ‘fair’ and more supportive than a system with lower fixed pay. Crowding theory suggests this positive perception of the compensation system change might cause a possible increase in intrinsic motivation followed by an increase in performance (or at least not a decrease). The opposite was found, suggesting the effects of extrinsic motivation and monetary incentive are much larger in the case of Pepperminds.

The composition of the workforce at Pepperminds has changed slightly with the changes in wage regimes. Average age has risen, making at least the fixed wages higher on average in the new minimum wage systems. As fixed pay has risen, average age has gone up and minimum wage is higher for higher ages, it is safe to assume the changes in wage regimes have increased employment costs at Pepperminds. This was accompanied by a steep drop in performance across all ages and positions, suggesting a decrease of total operating results of Pepperminds.

The results found in the Pepperminds data do highlight a ‘hidden’ effect of minimum wage implementation. The effects of minimum wages on labor supply and demand are quite straightforward textbook material. Empirical research does not always support the classical theory (Machin & Manning, 1997). The indirect effect of minimum wages on worker productivity through its influence on compensation systems is less straightforward and less

25 discussed. This effect can influence the macro-economic results of minimum wage

implementation and might cause companies with performance pay systems to suffer a double rise in employment costs. Firstly, the minimum wage will most likely increase their

employment costs through higher fixed pay, second the minimum wages can cause a change in monetary incentives and, as seen in the case of Pepperminds, reduce productivity of its employees. This will cause both an increase in employment costs, as well as a decrease in production by the company. This ‘double dip’ effect can cause certain companies to suffer more from minimum wages than others. Although the implications of this double dip on a macroeconomic scale needs further research, with performance pay systems increasing in popularity, this paper makes a strong case for further analysis by both researchers and decision makers.

This study can be viewed as a case study and its results should not be too easily

generalized. The composition of the workforce at Pepperminds is quite specific, with a narrow age range and mostly part-timers who are still studying. Examples of comparable sectors for which these results might be especially relevant include the hospitality industry, marketing companies, supermarkets, the retail industry and sales companies. Less comparable sectors like education and banking might still find these results relevant for at least a part of their workforce. Further research is needed on what the effects of performance pay systems and minimum wages are in different branches and at different ages.

This paper clearly demonstrates a direct link between monetary incentives and performance. The performance based compensation system at Pepperminds has a great influence on the sales results of its workforce. A reduction of around ten percent in average added return on performance has caused similar reductions in actual sales. These results present hidden effects of minimum wage implementation. The data made available by Pepperminds shows that within companies using performance based compensation systems, minimum wages can negatively affect the productivity of the workers. This suggests that not only labor demand and supply on a macro-economic scale need to be considered when discussing minimum wage implementation, but the possible negative effects on the

productivity of employees at performance pay companies need to be taken into account. Both the discussion on the reduction of performance pay in the form of bonuses within banks, as the discussion on implementation of performance pay in education can benefit from the results this empirical study presents.

26

References

Anderson, J. (2014, July 31). British Regulators Set Strict Rules for Banker Bonuses. The

New York Times, pp. B7.

Calder, B.J., & Staw, B. M. (1975). Self-perception of intrinsic and extrinsic motivation.

Journal of Personality and Social Psychology, 31(4), 599.

Card, D., & Krueger, A. B. (1993). Minimum wages and employment: A case study of the fast food industry in New Jersey and Pennsylvania. The American Economic Review, 84(4), 772-793.

Card, D. (1991). Do minimum wages reduce employment? A case study of California,

1987-89(No. w3710). National Bureau of Economic Research.

Chambers, R. G., & Quiggin, J. (2005). Incentives and standards in agency contracts. Journal

of Public Economic Theory, 7(2), 201-228.

Davidson, J. (2009, June 23). Performance Pay for Federal Employees Still a Matter of Debate. The Washington Post, pp. A17.

Deci, E. L. (1971). Effects of externally mediated rewards on intrinsic motivation. Journal of

Personality and Social Psychology, 18(1), 105.

Frey, B. S., & Oberholzer-Gee, F. (1997). The cost of price incentives: An empirical analysis of motivation crowding-out. The American Economic Review, 87(4), 746-755.

Funnekotter, B. (2012, October 23). De leraar moet weer beter worden; Formeren is kiezen. Vandaag: het onderwijs. NRC Handelsblad, pp. 8.

Gielen, A. C., Kerkhofs, M. J., & Van Ours, J. C. (2010). How performance related pay affects productivity and employment. Journal of Population Economics, 23(1), 291-301. Gilman, M., Edwards, P., Ram, M., & Arrowsmith, J. (2002). Pay determination in small firms in the UK: the case of the response to the National Minimum Wage. Industrial relations

journal, 33(1), 52-67.

Hurst, G. (2012, May 1). Bad teachers should be paid less, say MPs; Flexible salary structure 'would reward success'. The Times (London), pp. 9.

27 Kishore, S., Rao, R. S., Narasimhan, O., & John, G. (2013). Bonuses versus commissions: A field study. Journal of Marketing Research, 50(3), 317-333.

Kunz, A. H., & Pfaff, D. (2002). Agency theory, performance evaluation, and the hypothetical construct of intrinsic motivation. Accounting, Organizations and Society, 27(3), 275-295. Lazear, E.P. (2000). Performance pay and productivity. The American Economic Review, 90(5), 1346-1361.

Lazear, E.P. (1986). Salaries and Piece Rates. The Journal of Business, 59(3), 405-431. Machin, S., & Manning, A. (1997). Minimum wages and economic outcomes in Europe.

European Economic Review, 41(3), 733-742.

Mankiw, N. G. (2013). Macroeconomics (8th edition). Houndmills, England: Palgrave Macmillan.

Neumark, D., & Wascher, W. (2006). Minimum wages and employment: A review of evidence

from the new minimum wage research (No. w12663). National Bureau of Economic Research.

Pfeifer, C. (2012). Base salaries, bonus payments, and work absence among managers in a

German company, University of Lüneburg Working Paper Series in Economics, No. 259.

Prendergast, C. (1999). The Provision of Incentives in Firms. Journal of Economic Literature, 37(1), 7-63.

Van Herpen, M., Van Praag, M. & Cools, K. (2005). The effects of performance measurement and compensation on motivation: an empirical study. The Economist, 153(3), 303-329.

Young, G. J., Beckman, H., & Baker, E. (2012). Financial incentives, professional values and performance: A study of pay‐for‐performance in a professional organization. Journal of

28

Appendix

When considering the score data as count data, a quick summary of the score shows that its variance is almost ten times as large as the mean. This indicates that the distribution of the score is most likely different from a Poisson distribution, as a Poisson distribution has equal variance and mean. Stata goodness of fit test Poisgof confirms this.

Using the Poisson regression on the score, a goodness of fit test can be performed. The results of a Stata goodness of fit test for the Poisson distribution, with H0: Poisson is a good fit:

Deviance goodness-of-fit = 239396.4 Prob > chi2(101945) = 0.0000 Pearson goodness-of-fit = 284397.4 Prob > chi2(101945) = 0.0000

Both are significant (p<0.0000) which indicates Poisson is not a good fit for the score function. Summary of Score Observations 102062 Mean 3.77 Std.Dev. 5.75 Variance 33.05 Skewness 10.04 Kurtosis 191.08