How Changes in Compensation Plans Affect Employee Performance, Recruitment and RetentionAn Empirical Study of A Car Dealership Joanna

L.Y. Ho* University of California, Irvine Anne Wu National Chengchi University, R.O.C. Ling-Chu Lee National Pingtung Institute of Commerce, R.O.C

ABSTRACT: This study reports that changes in compensation from performance-sensitive (commission-based) to less performance-sensitive (base salary plus commission) schemes hurt employee performance. Also, we extend prior studies by examining how employees with different levels of ability are affected by the compensation plan change and we explore selection effects concerning employee recruitment and retention. This study involves performance data for 4,392 employees of a Taiwanese car dealership over 56 months. Our results show that high-performance employees were affected by the compensation plan change more than low-performance employees. In particular, salespersons with the second-highest performance ratings (not the top salespersons) were affected the most by the plan switch. Consistent with the predictions of selection effects, our results indicate that the less performance-sensitive plan retained fewer high-performance salespersons and recruited more low-performance sales staff. We also find the more loss for an employee, the more likely he/she would leave the dealership. Key Words: compensation scheme, incentive effect, selection effects, and performance Data Availability: The confidentiality agreement with the company that provided data for this study precludes the dissemination of detailed data without the companys consent. * Corresponding author: Joanna L. Ho, Paul Merage School of Business, University of California, Irvine, CA 92697-3125, (O) 949-824-4041, (F) 949-725-2833, jlho@uci.edu The authors thank Rajiv Banker, Chee Chow, Surya Janakiraman, Christo Karuna, Ken Cavalluzzo, Sanjay Kallapur, Bill Lanen, Ping Lin, Thomas Lin, James Wallace, and participants at the research colloquium of the Graduate School of Management of University of California, Irvine, the 9th IAAER World Congress of Accounting Educators, the 2003 Management Accounting Conference, and the 2004 American Accounting Association Annual Meetings for their helpful comments and suggestions.

Electronic copy available at: http://ssrn.com/abstract=1290646

. INTRODUCTION
Economic theory argues that performance-based compensation contracts increase employees incentives to exert effort, resulting in improved performance (Milgrom and Roberts 1992; Prendergast 1999). Previous empirical and laboratory studies on this topic have compared across compensation schemes or examined how changes to a more performance-sensitive incentive scheme influence employees compensation and performance (Waller and Chow 1985; Lazear 2000; Banker et al. 2001). Yet, no research has addressed the impact of changes to less performance-sensitive plans on employee performance. In the real world, many companies use or switch to less performance-sensitive incentive schemes. Examples include Sears (Driscoll 1994), the shoe manufacturing industry (Freeman and Kleiner 1998), and Fujitsu (Tanikawa 2001).1 Our study contributes evidence on how a switch to a less performance-sensitive incentive scheme affects an individual employees productivity and compensation. Furthermore, we examine whether employee ability affects their productivity in light of the plan change and which employee group is affected most by such a change. In addition to influencing employees, the compensation plan can affect company

Prior to 1992 Sears adopted a commission-based compensation system for auto repair salespersons. This

compensation scheme enticed salespersons to falsely diagnose brake and alignment problems, which cost Sears $15 million in refunds and other settlement costs. Consequently, Sears stopped the commission system. Also, during the 1990s, most shoe manufacturers adopted a survival strategy by shifting from a piece-rate to hourly-rate compensation system. In addition, in 2001 Fujitsu changed its compensation system from performance-based to non performance-based. This is because under the performance-based system workers tended to set goals as low as possible to avoid falling short of the goals in order to receive raises and promotions.

Electronic copy available at: http://ssrn.com/abstract=1290646

performance by impacting recruitment and retention (Stiglitz 1975; Salop and Salop 1976; Demski and Feltham 1978; Milgrom and Roberts 1992). For example, performance-based compensation contracts attract and retain high performers and differentiate high from low performers (e.g., Baron and Kreps 1999; Banker et al. 2001). A company benefits when low-performance employees leave, but suffers a setback when high-performance employees depart. Thus, it is important to consider who will join/leave the company when the performance sensitivity of the compensation contract is changed. This study examines when the compensation plan changes to a less performance sensitive one, what types of employees being attracted to the firm and also leaving the company. In addition, we investigate causes (e.g., ability, experience, compensation loss) that may account for employee separation. Our study focuses on a car dealership in Taiwan that changed its compensation scheme from being totally commission-based to a mix of fixed salary and lower commission rates. This change was in response to the requirement of the 1998 Taiwanese Labor Law amendment. Our database includes 4,392 pieces of detailed, individual-level data (e.g., salespersons compensation, sales quantities, performance ratings, and demographic information) and firm-level data (e.g., turnover rates, new hires) for a period of 56 months. We find that the switch in compensation plan lowered individual productivity and compensation, especially for high-performance salespersons. Moreover, the plan change had a lesser impact on top performance-rating salespersons but a higher impact on salespersons with the second-highest rating. As predicted by theory, we also observe that the less performance-sensitive plan attracted more low-performance salespersons and retained fewer high performers. This study adds to the extant literature since this is the first empirical study

examining the impact of change to a less performance-sensitive compensation scheme on low-level employee productivity. We directly measure the impact of a plan change (i.e., from being more to less performance-based) on individual productivity and the selection effects by using data prior to and after the plan change. Banker et al. (2001) examined how implementing a more performance-sensitive incentive scheme affects employee performance. However, they had only after-the-plan data of divisions implementing or not implementing the new plan. Banker et al. (2001, 322) were explicit about this limitation: This lack of objective, individual-level performance data has limited our knowledge of how an incentive plan affects employees selection of employment and effort level. Also, Banker et al.s (2001) inference regarding employees retention and separation was based on comparing sales productivity of those who stayed versus those who left only after the plan implementation. In this study, the availability of employee-level data pre- and post-the plan also allows more robust conclusions about effects on employee recruiting and separation. Furthermore, this study goes beyond related prior studies by investigating how employee ability affects their performance when a company changes its compensation plan to a less performance sensitive one. We capture employee ability by using both continuous variables (i.e., number of cars sold and reciprocal of time to the first promotion since he/she joined the dealership) and categorical variable (i.e., employee annual performance rating). The employee ability measure not only helps us identify specific performance groups that are most affected by the compensation plan change but also provides additional insights into causes of employee separation after the plan change. These findings have managerial implications since they can help top management anticipate possible impacts on incentives and efforts of different employee groups as well as on employee recruitment and separation

due to the compensation plan change to a less performance sensitive one. The remainder of this paper is organized as follows: Section 2 describes the research site and the nature of the changes made in the employee compensation plan. Section 3 reviews related prior studies as the basis for developing our hypotheses. Data and empirical models follow in Section 4. Section 5 presents the empirical results, and Section 6 summarizes our results with concluding remarks.

II. RESEARCH SITE

The research site is the largest car dealership in Taiwan, where the domestic automobile market has been nearly 400,000 cars per year.2 In 2001, the car dealership had 1,248 sales representatives and 87 sales outlets located throughout Taiwan and in some parts of Mainland China. In early 1998, the Taiwanese Labor Law Amendment required that, by the end of 1998, companies give a minimum wage (around US$466/month) for all forms of employee-employer relationships. At that time, the CEO heard that some competitors would superficially comply with the Laws requirement by providing a base salary and maintaining the same commission rates. However, they would demand a penaltythe base salary would be taken away from those not meeting the minimum sales requirement. It has been known that the Taiwanese government has not strictly enforced all the regulations, and that the penalty associated with failure to comply with the regulations is not severe. However, the situation is different for this research company since it is the largest dealership in Taiwan, which makes it the most likely target for governmental investigation

Quantities of sale are 361,615; 374,955; 394,160; 357,634; 350,461 from 1996 to 2000, respectively (sources:

Taiwanese Directorate General of Highways)

to see whether the industry is complying with the Law. After considering all factors (e.g., potential government investigation and penalty, high political cost), on July 1, 1998 the CEO decided not to follow its competitors tricks but to comply with the Laws requirement by adding base salary but with lower commission rates for all levels of salespersons. The reason of lowering its commission rates is to control total compensation expenses to the pre-plan change level. 3 This plan was implemented throughout the company rather than on a regional or step-by-step basis. As such, this provides a unique setting to study employee recruitment and separation since the research site and its competitors used different compensation plans in light of the government regulation. Furthermore, the dealership introduced an Improvement of Low Performance Plan. That is, if a salesperson cannot attain the minimum requirement (nine cars every three months, and then, given the economic downturn in 2001, changing to seven cars every three months), he/she receives an initial warning. If the salesperson cannot improve performance in the following two months, he/she will be terminated in the absence of extenuating circumstances.4 We chose the dealership for three reasons: (1) Changes in its compensation scheme

At that time 45.2% of the salespersons sold less than three cars per month (i.e., 20.1% not selling any car,

11.3% selling one car, and 13.8% selling two cars). Since almost half of the salespersons sold less than three cars, the total compensation costs will increase dramatically if the company would pay a base salary on top of the old (higher) commission rates. Therefore, to control total compensation costs and make it equal to the pre-plan switch level, CEO decided to lower the commission rates.
4

The research company refused to share with us documents revealing who had not met the minimum sales

requirement and been forced to leave the company. We were told that branch managers normally make an effort to convince a poor performer to quit voluntarily so there are no bad records on file.

from totally commission-based to a less performance-sensitive one (base salary plus lower commission rates) was not voluntary but in response to the 1998 Laws requirements. (2) It is the largest dealership in the automobile market, with nearly 18 percent of market share over the past five years, and attained an ISO 9002 certification in 1999. (3) It agreed to provide us with an archival database containing 4,392 sales representatives and 87 branches over a 56-month period, and top management also allowed us to review most internal documents and to interview their managers and sales forces, if needed.

. HYPOTHESES DEVELOPMENT
The premise of agency theory is that a principal designs a contract to align an agents actions with the principals best interests (e.g., Holmstrom 1979; Feltham and Xie 1994; Prendergast 1999; Lambert 2001). Classic agency models have primarily focused on the design of performance-based contracts to motivate employee effort (incentive effect) (e.g., Milgrom and Roberts 1992; Banker et al. 1996; Lazear 2000; Banker et al. 2001). Also, prior studies have suggested selection effects, i.e., performance-based compensation contracts attract and retain high performers (e.g., Milgrom and Roberts 1992; Baron and Kreps 1999; Lazear 2000; Banker et al. 2001). Below, we discuss both incentive and selection effects and the hypotheses to be tested. Incentive Effect Agency theory maintains that compensation contracts make prescriptions regarding the circumstances under which fixed (base salary) and variable (commissions and other forms such as bonuses or options) components can better align employer and employee interests. For example, suppose, in a commission-based compensation plan, linear individual compensation contracts of the form C = bx, where b is commission rate and x is productivity or output. Under this contract, C is very sensitive to both x and b. Therefore,
6

employees have incentives to work hard to earn more compensation. Now, assume the company changes compensation plan C to a new one with base salary and lower commission rates. Linear individual new compensation contracts of the form C = a + bx, where a is base salary, b is the new commission rate, and b < b. When x 3 (due to the requirement of the Improvement of Low Performance Plan), C is equal to a and b=0 (no commission). That is, employees who sell less than three cars per month receive only base salary. However, when x > 3, employees receive both base salary and commission, depending on their sales productivity. A lower b may induce less effort from employees to work hard, resulting in C < C. The graphical presentations of the two compensation contracts C and C are summarized in Figure 1. _________________________ Insert Figure 1 about here __________________________ Empirical studies have supported the effects of performance-based incentives on employee behavior (e.g., Banker et al. 1996; Bailey et al. 1998; Lazear 2000). Banker et al. (1996) reported that implementing a performance-based compensation plan increased sales and that the effect persisted and grew over time. Bailey et al. (1998) found that improvement rates in individuals initial and overall performances of an assembly task are higher when an incentive plan is in place. Similarly, Lazear (2000) examined the impact of piece rates on the performance of workers who install auto windshields. He found that worker output increased after the company switched its compensation scheme from hourly wages to piece rates. These findings suggest that if a company switches from contract C (performance-sensitive plan) to contract C (less performance-sensitive plan), employee

productivity will be reduced. However, experimental studies have reported mixed findings on how incentives affect performance. Ashton (1990) concluded that auditors bond-rating performance in the incentive plus feedback condition was not significantly better than their counterparts in the feedback only condition. Conversely, Sprinkle (2000) reported that participants receiving an incentive-based contract performed better on a production decision task than those who received a flat-wage contract. Also, the rate of performance improvement increased when participants either spent more time working on the task or analyzed the provided information more thoroughly. Sprinkle also pointed out that participants initial performance on a cognitive task is not very sensitive to the increased effort induced by incentives, but their subsequent performance improvement is. In reviewing laboratory studies of incentives, Bonner et al. (2000) also reported that less than half of their reviewed studies show incentives lead to significant performance improvement. Taken together, these findings suggest that removing incentives may not necessarily cause negative impact. In light of the arguments and empirical findings that performance-based incentives affect employee behavior, our first hypothesis is as follows: H1: Employees average sales productivity (cars sold) will decrease after the company changes its compensation plan to a less performance-sensitive scheme.

Figure 1 shows the relationship between compensation and cars sold under both the C and C compensation schemes. Clearly, when sales productivity increases, the differences in compensation under these two schemes increase. This suggests that switching from compensation plan C to C will cause high-performance employees to lose more benefits (due to lower commission rates) than low-performance employees, and the poorest

performers may actually gain the most (with a guaranteed base salary). In light of greater compensation loss, high-performance employees will have less incentive to work harder than lower-performance employees. This results in comparatively lower sales productivity. These arguments lead to the following hypothesis: H2: The decrease in sales productivity (cars sold) of high-performance employees after a company changes to a less performance-sensitive compensation scheme will be greater than that of low-performance employees.

Selection Effects Companies design compensation schemes not only to induce more employee effort but also to attract potential employees. Recent studies suggest that performance-based incentive plans effectively sort employees by ability (Lazear 2000; Banker et al. 2001). Selection effects include recruiting and separation effects. The former relates to the type of employees who join the company and the latter to the type of employees who leave. Lazear (2000) showed that contracts with higher piece rates attract high-ability employees. In a review paper, Prendergast (1999) argued that compensation contracts are important means for a company to recruit more capable workers, since they will benefit more from a performance-sensitive compensation plan than the less capable will. In turn, the more capable will be more likely to be attracted to the company than the less capable; therefore, the company will have a higher percentage of high performers. Conversely, when a company switches compensation schemes from C to C, it is likely that the guarantee of a base salary will attract more low-performance employees. Therefore, the recruiting effect suggests that the sales productivity of employees hired after the switch to contract C will be lower than that of those hired before the plan switch. These arguments lead to the following hypothesis:
9

H3:

The average sales productivity (cars sold) of employees hired under the less performance-sensitive compensation scheme will be less than that of employees hired under the performance-sensitive compensation scheme.

Hollenbeck and Williams (1986) pointed out an important distinction between the frequency and functionality of turnover. While the former refers to the number of turnover separations, the latter refers to the implication of those separations for an organization. Using a meta-analysis of 55 studies, Williams and Livingstone (1994) reported that poorer performers tended to leave when pay was based on performance and to stay when it was not. The implication of their findings is that performance-based compensation plans can lead to functional turnover. Lazear (2000) reported that about one-third of improved performance can be attributed to selection effects, i.e., less-productive workers leave the company and are replaced by more-productive workers. In Banker et al. (2001), results showed that a performance-based scheme in a retail firm attracted and retained more productive salespersons, while the performance of the less productive sales staff declined before they left. Feedback on compensation could help employees learn how they are affected by a less performance-sensitive compensation plan. High-performance employees who continually compare their compensation with past levels and outside opportunities are likely to be dissatisfied with their compensation. As such, the separation effect predicts that high-performance employees are more likely to leave because they are disappointed with their compensation under the less performance-sensitive plan. Therefore, we expect that sales productivity of employees who left after the change in compensation scheme from C to C to be higher than that of employees who left before the switch. We propose the following hypothesis concerning separation effects:

10

H4:

The average sales productivity (cars sold) of employees who leave under the less performance-sensitive compensation scheme will be higher than that of employees who leave under performance-sensitive compensation.

. MODEL DEVELOPMENT
Data The sampling period covered 97,541 person-months and 5,121 branch-months of data from January 1996 to April 2001 (56 months).5 We partitioned the sample into two sub-samples: (1) before the switch to the new plan (a total of 28 monthsfrom January 1996 to April 1998),6 and (2) after the switch to the new plan (a total of 28 monthsfrom January 1999 to April 2001). To identify the employee groups most and least affected by the compensation plan change, we use three different measures to capture employee ability (ABILITY): annual performance rating (RATING), reciprocal of time to the first promotion since they joined the dealership (PROMOTE), and number of cars sold prior to the month of interest (#CARSOLD). Gibbs (1995) argued that employee performance rating is a logical measure for expected ability and that expected ability declines monotonically with the number of promotions employees received over time. In the dealership, each employee is evaluated by his/her branch manager annually on a four-point scale with 4 being the best performance. Regarding the average cars sold, it reflects changes in a salespersons productivity from the first month in which the employee joined the dealership to the preceding month of interest. We classify each employee in our sample as follows:
5

The company did not have a complete data set available until January 1996. To make the monthly data

covered in the two subsamples equal, we only included data up to April 2001.
6

We excluded eight months (May to December) of 1998 from our sample in order to control both the

transition-period and seasonal-month-sales effects. 11

(1) OldPlan (DOLD =1, 0 otherwise): Salespersons who had joined the dealership before the switch to the new plan (July 1, 1998). (2) OldPlanQuitBefore (DOQB =1, 0 otherwise): Salespersons who had joined the dealership before the switch to the new plan and left before July 1, 1998. (3) OldPlanQuitAfter (DOQA =1, 0 otherwise): Salespersons who had joined the dealership before the switch to the new plan and left after July 1, 1998 but before the end of our sample period. (4) NewPlan (DNEW =1, 0 otherwise): Salespersons who had joined the dealership after the switch to the new plan. (5) NewPlanQuitAfter (DNQA =1, 0 otherwise): Salespersons who had joined the dealership after the switch to the new plan and left after July 1, 1998, but before the end of our sample period. Estimation Model We estimated the models parameters using pooled time-series data over a 56-month period and cross-sectional data for the 4,392 individuals and 87 branches. PLAN is equal to one if the salesperson was on the new plan during the given month. We measure employee sales productivity (IndivProd) as number of cars sold per person-month. In all the four models, we include employees experience and two types of competition control variables, i.e., competition types and competition intensity, to capture the general economic condition. There are two types of competitors to our research company: family competitors (FamComp) that sell the same brand of cars and non-family competitors (NonFamComp) that sell different car brands but compete for the same customer groups. Regarding competition intensity, we use CitySale (i.e., total monthly sales of competitors in the same city) as a proxy. The log forms of monthly car sales for FamComp, NonFamComp, and CitySale are used. We also include month dummies, Mmt, in the models to capture both new car model effect (i.e., when a car dealership launches a new car model, it attracts consumers attention, resulting in increased sales productivity that normally lasts for several months) and potential

12

seasonal monthly sales effects. Finally, to control for the impact of hiring additional salespeople on productivity, we include log#EMPLOY, the log form of number of employees in each branch. H1 (hypothesized incentive effect on sales productivity) is tested by examining whether the coefficient of PLAN (1) is negative in OLS regression model, which is specified as follows:
Indiv Pr od it = 0 + 1 PLAN it + 1EXPER it + 1LogFamComp t + 2 LogNonFamComp t + 3 LogCitySaleit + 4 Log # EMPLOYit + M mt + it
m =1 11

(1)

i = {1,.,4,392}, a subscript to denote employee i, t = {1,,56}, a subscript to denote the time-period in which the employee works, and it is a random error term. Because of repeated observations on the same set of cross-section units (Johnston and Dinardo 1997, 388), H1 can also be tested by examining if 1 is negative in panel data with salesperson fixed effects model. Recall that Model 1 explores how the compensation plan switch affected all employees and Model 2 examines whether high-ability employees were more affected by the plan change. Therefore, we included employees ability in Model 2, not in Model 1. IndivProdit = 0 + 1 PLAN it + 1 EXPERit + 2 ABILITYit + 3 ABILITYit PLAN it
+1LogFamCompt +2LogNonFamCompt +3LogCitySaleit +4Log#EMPLOYit + Mmt +it
m=1 11

(2)

As discussed earlier, we used three variables as proxies for a salespersons ability (ABILITY): annual performance rating (RATING), reciprocal of time to the first promotion since they joined the dealership (PROMOTE), and number of cars sold prior to the month of

13

interest (#CARSOLD). The interactions between these three variables and PLAN change are also included in the regression model. Our hypothesis predicts a negative sign of the coefficient of ABILITY PLAN (3). In sum, while Model 1 examines effects of the plan change on average individual sales productivity, Model 2 further investigates the magnitude of negative impacts on salespersons with different levels of ability by including ABILITY in the model. The fixed-effects model using panel data is not feasible for testing selection effects, because there was a high employee turnover rate, with some employees leaving the dealership and new entrants joining it at the same time. Consequently, we specify the following regression Models 3 and 4 to test hypothesized selection effects:
Indiv Pr od it =
11 0

+ 1 EXPER it + rNEW D NEW + 1 LogFamComp t

+ 2 LogNonFamComp t + 3 LogCitySale it + 4 Log # EMPLOYit + M mt + it

m =1

(3)

IndivProdit = 0 + 1EXPERit + rNQA DNQA + rOQA DOQA + 1LogFamCompt + 2 LogNonFamCompt + 3LogCitySaleit + 4 Log# EMPLOYit + Mmt + it
m=1 11

(4)

Model 3 tests the recruiting effect, i.e., comparing the sales productivity of OldPlan and NewPlan groups. Model 4 tests separation effects, therefore only employees who had left the dealership (i.e., DOQB, DOQA, and DNQA) are included. In both Models 3 and 4, we do not include ABILITY or PLAN, nor their interaction variable, because in these models we test only the differences between the groups NewPlan and OldPlan. However, we include EXPER (number of months employed) as a control variable since these groups had different times of employment and separation. We use variance inflation factor (VIF) to test potential multicollinearity problems.
14

Since all the VIFs are less than 10 and t-ratios (Maddala 1992; Hair et al. 1998) are significant, multicollinearity problems in our models are not serious.

. RESULTS
Descriptive Statistics
Table 1 summarizes the descriptive statistics of the key variables. Means (standard deviations) for the whole period and the two sub-periods are presented in Panel A of Table 1. The average number of cars sold per month per employee is 3.30, which is slightly higher than 3: the minimum requirement under the Improvement of Low Performance Plan. The turnover rate per month is 2.31 percent (or 27.45 percent per year). This high turnover rate suggests that we are able to examine selection effects. Also, the average age of salespersons hired under the NewPlan (32.88) is older than that under the OldPlan (31.46). __________________________ Insert Table 1 about here __________________________ As seen from Panel A of Table 1, the average of IndivProd under the new plan (3.15) is significantly (p < .0001) lower than that under the old plan (3.49), suggesting that the change to a less performance-sensitive compensation plan decreased individual salesperson productivity. Furthermore, additional data shows that in general the average sales productivity of NewPlanQuitAfter is below 3 cars over the 34-month period. All the results suggest that the Improvement of Low Performance Plan was effective and that poor performers had to leave the dealership. This can also be evidenced from Panel A of Table 1 that the variance of IndivProd significantly decreased after the change in compensation plans (from 4.09 to 3.45; p < .0001). Furthermore, the monthly employee turnover rates rose
15

from 2.19 to 2.41 percent (a 10 percent increase), but the difference is not statistically significant (t= 1.01, p=0.3152). Note that there is a decrease in the companys market share after the plan switch (from 18.2 percent to 17.0 percent); however, it is not statistically significant. Figure 2 displays the monthly average sales productivity of different employee types from January 1996 to April 2001. As seen in Panel A of Figure 2, the average sales productivity of Old Plan is better than that of New Plan, suggesting the less performance-sensitive plan attracts more low performers. Furthermore, Panel B of Figure 2 shows that the sales productivity of OldPlanQuitAfter (OLDQA) is higher than that of OldPlanQuitBefore (OLDQB), implying more high performers left the firm after the plan switch. __________________________ Insert Figure 2 about here __________________________

Panel B of Table 1 presents distributions of individual sales productivity for both Old Plan and New Plan periods. As shown in Panel B of Table 1, the change to a less performance-sensitive plan caused a decrease in sales productivity. For example, more salespersons sold no car each month under the new plan (24.8 percent) than under the old plan (20.1 percent). Similarly, a slightly higher percentage of salespersons sold less than three cars under the new plan than under the old plan (46.1 percent vs. 45.2 percent). Regarding the high performers, after the plan switch there was a decrease in the percentage of those who sold six or more cars per month (from 20.6 percent to 17.3 percent) and also a slight decrease in the percentage of those who sold ten or more cars per month (from 4.5
16

percent to 3.4 percent).

Tests of Incentive Effect (H1 and H2)

H1 predicts that employees average sales productivity decreased after the plan switch. Panels A and B of Table 2 show the panel data and regression results7 with individual sales productivity (IndivProd) as the dependent variable. As discussed above, we expect a negative coefficient of PLAN (1) in Panel A of Table 2 (columns 1 and 2). As predicted, 1 are negative (-0.375 using Fixed Effects and -0.354 using OLS) and statistically significant (at p < 0.0001 and p < 0.05). Recall that the dealership also introduced an Improvement of Low Performance Plan. In order to rule out the possibility that declined individual sales productivity is due to the threat of dismissal rather than the incentive scheme change, we separated the employees into two groupsthose with dismissal threat versus those without dismissal threat. Our classification is based on whether individuals average cars sold in the last month working in the dealership was above three cars (i.e., a requirement by the Improvement of Low Performance Plan). As shown in columns 3, the coefficient of PLAN for the sub-sample without dismissal threat is -0.464 (p < 0.0001). Therefore, our results support H1. ____________________________ Insert Table 2 about here ____________________________ Recall that H2 predicts the magnitude of the negative impacts on sales productivity is greater for high-performance salespersons than it is for low-performance salespersons.
7

To test the hypotheses, we conducted three different models: fixed-effects, OLS regression, and negative

binomial regression and we found qualitatively similar results. For discussion purposes, we only present results using the fixed-effects models and OLS regression.

17

Therefore, we expect the coefficient of ABILITY PLAN, 3, to be negative. As seen in Panel B of Table 2, the coefficient of RATING PLAN in column (1) has a negative sign (-0.195), consistent with the prediction of H2. Also, as shown in column (2) and (3) we observed qualitatively similar results as the use of the continuous variable PROMOTE PLAN (-1.024), #CARSOLD PLAN (-0.043) as a measure. Further, the positive coefficient of RATING also suggests that sales productivity is significantly influenced by an employees ability. To test the robustness of Model 2, we also explore the impact of employee ability on productivity by using RATING as a dummy variable. As seen in column (4) of Panel B, the plan change had a significantly higher negative impact on employees who received RATING3 (the coefficient of RATING3*PLAN is -0.323; p=0.0007) than that on RATING1 (bottom performers). However, the negative impact of the plan change on top salespersons (those who received RATING4) was similar to that on RATING1 (the coefficient of RATING4*PLAN is -0.236; p=0.715). Our interviews revealed that quite a number of top salespersons belong to the one thousand cars club (a rewards classification based on a minimum requirement of one-thousand cumulative car sales). This honor status not only wins them respect from their peers, but also provides them with benefits such as overseas travel and driving the best cars. As a result, top salespersons may possess stronger intrinsic motivation (Kunz and Pfaff 2002) and be less focused on their compensation losses. There is also no significant difference between the effect of the plan change on the two bottom groups (RATING2 vs. RATING1) since the coefficient of RATING2*PLAN is -0.010; p=0.894. It is likely salespersons with RATING2, knowing their productivity approached the minimum requirement, would put in more effort so that their productivity

18

would not fall any further. Taken as a whole, our results support H2 and reveal a different impact on the high-performance group. Specifically, while top salespersons were less affected by the plan change, salespersons who received second-highest performance ratings were affected the most among all the groups.

Tests of Selection Effect (H3 and H4)

To test the recruiting effect (H3), we examine whether sales productivity of the less performance-sensitive new plan is lower than that of the old plan. In Model 3, we include only DNEW to avoid potential multicollinearity problems. The average sales productivity of DOLD is 0, and the average sales productivity of DNEW is 0+NEW. Thus, H3 is tested by examining the following inequality: (0+NEW) -0 = NEW < 0. As seen in Panel C of Table 2, NEW is -0.439, which is significantly less than zero (at 0.0001 level). The result suggests that sales productivity was higher for employees hired under the old plan than for those hired under the new plan. Therefore, our results support H3 or a recruiting effect. To test separation effects (H4), we compare differences in the sales productivity of those who left before the change in compensation plans (OldPlanQuitBefore) with those of employees who had left after the switch (OldPlanQuitAfter). We expect the coefficient of average sales productivity of OldPlanQuitBefore (0) is less than that of OldPlanQuitAfter (0+ OQA). In other words, OQA is greater than zero. As shown in Model 4 of Table 2, Panel C, all the coefficients are statistically significant (with p values less than 0.01) and OQA is 0.969 (significantly > 0 at .0001 level). Recall the higher the OQA , the higher the chance that the dealership will lose high-performance employees by implementing the new plan. We then compare differences between the sales productivity of OldPlanQuitBefore
19

and that of those who were hired under the new plan (NewPlanQuitAfter). Recall that the dealership implemented an Improvement of Low Performance Plan. Therefore, NewPlanQuitAfter includes both poorer performers being forced out and high performers who quit because they were not satisfied with the compensation plan. As discussed earlier, we expect the coefficient of average sales productivities of NewPlanQuitAfter (0+ NQf) to be higher than that of OldPlanQuitBefore (0) or that NQA is greater than zero. As shown in Model 4 of Panel C, NQA is 0.120 (significantly > 0 at 0.0001 level). This finding suggests that employees who were recruited and later left under the new plan performed better than those who had been recruited and had left under the old plan. It also implies that more high-performance salespersons leave under the less performance-sensitive plan. Taken together, these results support H4 or separation effects. To delve further into the causes to employee separation, we ran two logit regression models (A and B). Model (A) examined how ability and experience affect employees leaving the dealership under old and new plans. As discussed earlier, we observed qualitatively similar results when using RATING, PROMOTE or #CARSOLD as proxies for salespersons ability. Therefore, in Model (A) and (B) we only used RATING as a measure for salespersons ability, and the logit regression results are summarized in Table 3. In Model (B), we tested how employees separation from the dealership was influenced by the compensation loss (CompLoss) under the new plan relative to the old plan as well as their ability and experience. CompLoss is the ratio of the change in the average compensation caused by the implementation of the new plan and the average compensation before the new plan. To explore the effect of CompLoss, we included only employees who joined the dealership before the plan switch. Since employees hired after the plan switch

20

were excluded, we did not include PLAN in Model (B). As seen from Table 3, in Model (A) RATING and EXPER are the only variables that significantly affect employees separation from the dealership (p = 0.0001). The negative signs of RATING (-1.313) and EXPER (-0.109) suggest that poorer and less experienced performers were more likely to leave. Interestingly, PLAN does not significantly affect employees departure decisions (p = 0.90), implying that the implementation of the new plan did not result in higher turnover. This is consistent with the descriptive statistics on monthly turnover (2.19 vs. 2.41) reported in Panel A of Table 1. Regarding the effect of compensation loss on employees departure decisions (Model B), Table 3 reveals a significant CompLoss effect (coefficient is -0.289; p = 0.0016). This suggests that the more loss for an employee, the more likely he/she would leave the dealership. In sum, our additional analyses show that poorer, less experienced, and higher compensation-loss performers tended to leave the dealership under both old and new plans. __________________________ Insert Table 3 about here ___________________________

. CONCLUDING REMARKS
Due to lack of access to objective, individual-level performance data, prior empirical studies of compensation schemes have only produced limited knowledge of how an incentive plan affects employees selection of employment and effort level (Banker et al. 2001). In this study, we have used data at the individual and company levels to empirically test how a companys change to a less performance-sensitive compensation plan affected employee performance. Our results support the inference from economic theory that a
21

switch to a less performance-sensitive compensation scheme hurts individual sales productivity. Also, we found more low-performance salespersons were attracted to the company and fewer high performers were retained. Also, this study investigates employee groups most affected by the switch to a less performance-sensitive plan. As predicted, we find that high-performance employees were affected more than low-performance employees. Importantly, our results show that impacts are different among high-performance employees, i.e., salespersons who received second-highest performance ratings were more affected by the plan switch than those with top ratings. Perhaps salespersons with highest ratings had stronger intrinsic incentive (e.g., better social status) and received status-related benefits (e.g., overseas travel, equipped with best cars) and thus placed less weight on sales-related compensation. These findings suggest that managers should anticipate impacts of alternate plans on the incentives and efforts of different employee groups (e.g., ones with different levels of ability) and refine the compensation contracts or take various actions to mitigate potential dysfunctional impacts. Furthermore, our results support the recruiting effect, i.e., the average sales productivity was higher for those hired under the old plan than for those hired under the new plan. Also, our results maintain the separation effect. We find that employees who were hired under the old plan and quit after the plan switch had lower average sales productivity than those who left before the plan switch. In addition, sales productivity was higher for those who were hired and left under the less performance-sensitive plan than for those who were hired before the switch. Results from additional analyses show that separation occurs mainly when employees have poor performance, less experience, and more suffering from compensation loss. Hence, these results can help management anticipate that employees

22

with certain characteristics will be more likely to leave if the compensation plan becomes less performance-sensitive. Taken as a whole, although our results support the predictions of economic theories (i.e., incentive and selection effects), some employee behaviors are better explained by organizational and behavioral theories (e.g., intrinsic incentive causes top salespersons to stay). While this study has advanced our understanding about the effects of performance-based incentive schemes and their related management practices, it is important to recognize its limitations. In particular, like prior studies on changing incentive schemes (e.g., Banker et al. 1996; Lazear 2000; Banker et al. 2001; Brickley and Zimmerman 2001), our analysis was restricted to a large data set from one organization. Therefore, our results may not be generalizable to other organizations and contexts. Furthermore, although we have included both competitor performance and local competition intensity in our models, we do not have information about competitors strategy, which may have also affected the performance of our research company.

23

REFERENCES
Ashton, R. H. 1990. Pressure and performance in accounting decision settings: Paradoxical effects of incentives, feedback, and justification. Journal of Accounting Research 28 : 148-180. Banker, R. D., S. Lee and G. Potter. 1996. A field study of the impact of a performance-based incentive plan. Journal of Accounting and Economics 21: 195-226. -----------, -----------, -----------, and D. Srinivasan. 2001. An empirical analysis of continuing improvements following the implementation of a performance-based compensation plan. Journal of Accounting and Economics 30: 315-350. Bailey C., L. Brown, and A. Cocco. 1998. The effects of monetary incentives on worker learning and performance in an assembly task. Journal of Management Accounting Research 10: 119-131. Baron, J. N. and D. M. Kreps. 1999. Strategic Human ResourcesFrameworks for General Mangers. John Wiley & Sons, Inc. Bonner, S.E., R. Hastie, G.B. Sprinkle, and S.M. Young. 2000. A review of the effects of financial incentives on performance in laboratory tasks: Implications for management accounting. Journal of Management Accounting Research 12: 19-64. Brickley, J.A. and J.L. Zimmerman. 2001. Changing incentives in a multitask environment: Evidence from a top-tier business school. Journal of Corporate Finance 7: 367-396. Demski, J. and G. Feltham. 1978. Economic incentives in budgetary control systems. The Accounting Review 53: 336-359. Driscoll, P. A. 1994. Sears to link incentives for auto service sales to customer satisfaction. Marketing News 28: 8. Feltham, G. and J. Xie. 1994. Performance measure congruity and diversity in multi-task principal/agent relations. The Accounting Review 69: 429-453. Freeman, R.B., and M.M. Kleiner. 1998. The last American shoe manufacturers: Changing the method of pay to survive foreign competition. Working paper, NBER. Gibbs, M. 1995.Incentive compensation in a corporate hierarchy. Journal of Accounting and Economics 19: 247-277. Greene, W.H. 2000. Econometric Analysis. Upper Saddle River, NJ: Prentice Hall.

24

Hair, J., R., Anderson, R. Tatham, and W. Black. 1998. Multivariate Data Analysis. 6th Ed. Prentice-Hall Inc. Hollenbeck, J.R. and C.R. Williams. 1986. Turnover functionality versus turnover frequency: A note on work attitudes and organizational effectiveness. Journal of Applied Psychology 71: 606-611. Holmstrom, B. 1979. Moral hazard and observability. Bell Journal of Economics 10: 74-91. Johnston, J. and J. Dinardo. 1997.Econometric Methods. Singapore: McGraw-Hill. Kunz, A. H. and D. Pfaff. 2002. Agency theory, performance evaluation, and the hypothetical construct of intrinsic motivation. Accounting, Organization and Society 27: 275-295. Lambert, R.A. 2001. Contracting theory and accounting. Journal of Accounting and Economics 32: 3-87. Lazear, E.P. 2000. Performance and productivity. The American Economic Review 90: 1346-1361. Maddala, G. S. 1992. Introductions to Econometrics. Prentice Hall International. Milgrom, P. and J. Roberts. 1992. Economics, Organization and Management. Englewood Cliffs, NJ: Prentice-Hall. Prendergast, C. 1999. The provision of incentives in firms. Journal of Economic Literature 37: 7-63. Salop. J. and S. Salop. 1976. Self-selection and turnover in the labor market. Quarterly Journal of Economics 90: 619-627. Sprinkle, G.B. 2000. The effect of incentive contracts on learning and performance. The Accounting Review 75: 299-326. Stiglitz, J. 1975. Incentives, risk, and information: Notes toward a theory of hierarchy. Bell Journal of Economics 6: 552-579. Tanikawa, M. 2001. Fujitsu decides to backtrack on performance-based pay. New York Times (Late Edition (East Coast)). New York, N.Y.: Mar 22, W.1. Waller, W. S. and C. W. Chow. 1985. The self-selection and effort effects of standard-based employment contracts: A framework and some empirical evidence. The Accounting Review 60: 458-476.

25

Williams, C.R. and L.P. Livingstone. 1994. Another look at the relationship between performance and voluntary turnover. Academy of Management Journal 37: 269-298.

26

Figure 1: The compensation and sales productivity under the commission-based and base-salary-plus-lower commission plans

Compensation

Commission-based plan (C)

b b

Base-salary-plus-commission plan (C)

a
3 Minimum requirement

Sales productivity (Cars sold)

27

Figure 2: Sales productivity by timing of employee employment and separation

Panel A Employees by different time of employment
OLD NEW

Panel B Employees by different time of separation

OLDQB OLDQA NEWQA

7
8

6
7 6 Average Sales 5 4 3

Average Sales
3 -6 -2 7 11 15 19 23 27 -30 -26 -22 -18 -14 -10 31

2
2 1 0 Month before( -30 to -1) /After PLAN ( 1 to 34)

0 3 -6 -2 7 11 15 19 23 27 -30 -26 -22 -18 -14 Month before( -30 to -1) /After PLAN ( 1 to 34) -10 31

Data cover the 64-month period from January 1996 to April 2001, Month -30 to -1 before PLAN and Month 1 to 34 after PLAN. As indicated in footnote 3, we excluded eight months (May to December) of 1998 from our sample in order to control both the transition-period and seasonal-month-sales effects. 1) OldPlan (NewPlan) is a salesperson who had joined the dealership before (after) the switch to the new plan. 2) OldQuitBefore is a salesperson who had joined the dealership before the switch to the new plan and left before July 1, 1998. 3) OldQuitAfter is a salesperson who had joined the dealership before the switch to the new plan and left after July 1, 1998 but before the end of our sample period. 4) NewQuitAfter is a salesperson who had joined the dealership after the switch to the new plan and left after July 1, 1998, but before the end of our sample period. 28

29

Table 1: Data description and descriptive statisticsa Panel A: Descriptive statistics of key variables
Variableb IndivProd Age EXPER Compensation #EMPLOY MthTurnover Market Share Categories All employees All employees All employees All employees Branches Months Branches Branch Months Months # of Observations 97,541 97,541 97,541 78,972 5,121 64 5,121 5,121 64 64 Entire Period Old Plan c New Plan c Difference d,e

3.30 (3.75) 3.49 (4.09) 3.15 (3.45) -0.3382*** (-13.71) 32.37 (5.72) 31.46 (5.62) 32.88 (5.71) 1.42*** (55.34) 1.08*** (79.74) -$360*** (-44.83) 1.29***(7.79) 0.22% (1.01) -1.2% (-1.87) -0.11***(-4.75) -0.05(-2.15) 3.20***(4.77)

3.98 (3.12) 3.30 (2.83) 4.37 (3.20) $1,413 ($1,136) 17.23 (5.93) 2.31% (0.88%) 17.55% (2.45%) $1,612 ($1,263) $1,255 (992)

16.52(6.04) 17.81(5.77) 2.19% (0.74%) 18.2% (1.90%) 2.41% (1.00%) 17.0% (2.8%)

LogCitySale LogNonFamComp LogFamComp

7.44(0.82) 7.50(0.81) 7.39(0.83) 10.01(0.22) 10.01(0.22) 9.96(0.21) 6.38(3.10) 4.68(3.90) 7.88(0.34)

29

30
Panel B: Distributions of individual sales productivity under the old plan (January 1996

April 1998) and under the new plan (January 1999 April 2001)

Quantities of Sales
0 1 2 3 4 5 6 7 8 9 10 > 10

Old Plan New Plan # of Cars Sold (%)Cumulative % # of Cars Sold (%) Cumulative %
7,977 4,504 5,475 5,436 4,730 3,433 2,582 1,748 1,257 793 584 1,247 ( 20.1) ( 11.3) ( 13.8) ( 13.7) ( 11.9) ( 8.6) ( 6.5) ( 4.4) ( 3.2) ( 2.0) ( 1.5) ( 3.0) 20.1 31.4 45.2 58.9 70.8 79.4 85.9 90.3 93.5 95.5 97.0 100.0 10933 3,361 6,058 6,906 5,242 3,990 2,714 1,699 1,011 681 469 1,061 ( 24.8) ( 7.6) ( 15.7) ( 15.7) ( 11.9) ( 9.0) ( 6.2) ( 3.9) ( 2.3) ( 1.5) ( 1.1) ( 2.3) 24.8 32.4 46.1 61.8 73.7 82.7 88.9 92.8 95.1 96.6 97.7 100.0

a. There are 97,541 person-months observations including 4,392 salespersons and 5,121 branch-months observations for 87 branches over a 56-month period. b. Definitions and measurements IndivProd: cars sold per person-month Age: Average age of salesperson EXPER: number of years given salesperson had been employed Compensation: compensation per person-month MthTurnover: Turnover rate per month #EMPLOY: a head count of salespersons in a branch and this variable excludes the administrative staff Market Share: market share in Taiwan car market LogFamComp: log monthly market sales of family competitors that sell the same brand of cars LogNonFamComp: log monthly market sales of non-family competitors that sell different car brands but compete for the same customer groups LogCitySale: log monthly market sales of local competitors. The branch and its local competitors are located in the same city. c. Standard deviations are in parentheses. d. t value in parentheses. e ***Statistically significant at or less than the 0.0001 level (two-tailed); ** Statistically significant at the 0.01 level (two-tailed).

30

31 Table 2 Statistical results of hypotheses testinga (Dependent variable: IndivProdb) Panel A: Test of incentive effects (H1)the plan change decreases average sales productivity Model 1 Full Sample Ordinary Least Squares (2) -18.944*** -0.354*** 0.216*** 0.068*** 1.848*** 0.208*** -0.032 Included 0.240 -0.344*** 1.02~4.57 2.16

Full Sample Individual Fixed Effects (1) -19.186*** -0.375* -0.062 0.139*** 0.406 2.691*** -1.310*** Included 0.276

INTERCEPT PLAN EXPER LogFamComp LogNonFamComp LogCitySale Log#EMPLOY Month dummy Adjusted R2 AR1 VIF Durbin-Watson

Sub-sample Without Dismissal Threat Ordinary Least Squares (3) -24.212*** -0.464*** 0.084*** 0.097*** 2.455*** 0.290*** -0.138 Included 0.164 -0.274*** 1.02~4.63 2.10

31

32 Panel B: Test of incentive effects (H2)the plan change on employee groups with different levels of ability Model 2
Ability proxy by RATING (continuous variable) (1) -23.780*** 0.209 0.111*** 1.216*** 0.602*** 1.935*** 3.521*** 0.856*** 0.771*** -0.195*** -0.010 -0.323** -0.236 -1.024*** -0.043** 0.099*** 2.067*** 0.255*** -0.063 Included 0.250 -0.298*** 1.02~9.43 2.12 0.070*** 1.855*** 0.219*** -0.044 Included 0.243 -0.342*** 1.10~4.52 2.16 0.118*** 1.615*** 0.154*** -0.082* Included 0.317 -0.174*** 1.02~4.52 2.04 0.083*** 1.866*** 0.202*** -0.068 Included 0.266 -0.300*** 1.02~4.49 2.12 PROMOTE (continuous variable) (2) -19.112*** -0.257*** 0.207*** #CARSOLD (continuous variable) (3) -17.830*** -0.098 -0.043*** RATING (dummy variables) (4) -19.534*** -0.158* 0.125***

INTERCEPT PLAN EXPER RATING RATING2 RATING3 RATING4 PROMOTE #CARSOLD RATINGPLAN RATING2PLAN RATING3PLAN RATING4PLAN PROMOTEPLAN #CARSOLDPLAN LogFamComp LogNonFamComp LogCitySale Log#EMPLOY Month dummy Adjusted R AR1 VIF Durbin-Watson
2

32

33 Panel C: Tests of H3 (recruiting effect) and H4 (separation effects) Model 3 (H3:Recruiting Effect) -20.231*** 0.193*** -0.439*** 0.120* 0.969*** 0.053*** 1.988*** 0.209*** -0.042 Included 0.242 -0.345*** 1.02~3.94 2.158 -0.030*** 1.936*** 0.073** -0.134** Included 0.330 -0.416*** 1.01~4.06 2.198 Model 4 (H4: Separation Effect) -18.678*** 0.106***

INTERCEPT EXPER D DNQA D

OQA NEW

LogFamComp LogNonFamComp LogCitySale Log#EMPLOY Month dummy Adjusted R2 AR1 VIF Durbin-Watson

a. Number of observations =95,490 b. Definitions and measurements IndivProd: cars sold per person-month PLAN: a dummy set equal to one if the salesperson is on the new plan during given month EXPER: number of years given salesperson had been employed RATING: a four-point scale annual performance rating with 4 being the best performance RATING1 to RATING4: dummy variables set equal to one if the salespersons annual rating is 1, 2, 3 or 4. PROMOTE: reciprocal of time to the first promotion since they joined the dealership. #CARSOLD:number of cars sold prior to the month of interest ABILITY (RATING, PROMOTE, #CARSOLD)*PLAN: interaction variable DNEW: An employee who joined the dealership after the new plan DOQA: An employee who joined the dealership before the new plan and who left the dealership after July 1, 1998 and before the end of our sample period. Comparison of two different groups--one pre-and one post-scheme change. DNQA: An employee who joined the dealership after the new plan and who left the dealership after July 1, 1998 and before the end of our sample period. LogFamComp: log monthly market sales of family competitors that sell the same brand of cars LogNonFamComp: log monthly market sales of non-family competitors that sell different car brands but compete for the same customer groups LogCitySale: log monthly market sales of local competitors. The branch and its local competitors are located in the same city. #Employee: a head count of salespersons in a branch. c. Panel data are repeated observations on the same set of cross-section units (Johnston and DiNardo, 1997). Therefore, salespersons who joined the dealership before the new plan and also remained until the end of our sample period are included in the panel data analysis. d. ***Statistically significant at or less than the 0.0001 level (two-tailed); **Statistically significant at the 0.01 level (two-tailed); *Statistically significant at the 0.05 level (two-tailed).

33

34

Table 3 Results of Logit regression of employee separation (dependent variable: LEFT)

N INTERCEPT PLAN CompLoss EXPER ABILITY (RATING) Log#EMPLOY Max-rescaled R-Square

Model A (not including compensation loss as a variable) 1713(LEFT=1) 1151(LEFT=0) 20.215 -17.849
-0.109*** -1.313*** 0.111 0.554

Model B (including compensation loss as a variable) 429(LEFT=1) 550(LEFT=0) 3.669***

-0.289** -0.322*** -1.180*** 0.148 0.371

a. Definitions and measurements LEFT=1 if the employee left the dealership, otherwise 0. PLAN: a dummy set equal to one if the salesperson is on the new plan during given month CompLoss= (the average compensation after the new plan the average compensation before the new plan)/the average compensation before the new plan EXPER=number of month of employment by an employee ABILITY (RATING): a four-point scale annual employee performance rating with 4 being the best performance #Employee: a head count of salespersons in a branch. *** b. Statistically significant at or less than the 0.0001 level (two-tailed);** Statistically

significant at the 0.01 level (two-tailed)

34