Rajiv D. Banker*, Seok-Young Leeb, Gordon Potter0, Dhinu Srinivasand *
aSchool of Management, University of Texas at Dallas, Richardson, TX 75083-0666, USA 
’’Department of Accounting, Sungshm Women’s University, Seoul 136-742, South Korea 
QSchool o f Hotel Administration, Cornell University, Ithaca, NY 14853-6902, USA 
dKatz Graduate School of Business, University o f Pittsburgh, Pittsburgh, PA 15260, USA
Received 14 February 2000; received in revised form 2 April 2001
Abstract
Performance improvements subsequent to the implementation of a pay-for-performance 
plan can result because more productive employees self-select into the firm (selection effect) 
and/or because employees allocate effort to become more effective (effort effect). We analyze 
individual performance data for 3,776 sales employees of a retail firm to evaluate these 
alternative sources of continuing performance improvement. The incentive plan helps the firm 
attract and retain more productive sales employees, and motivates these employees to further 
improve their productivity. In contrast, the less productive sales employees’ performance 
declines before they leave the firm.
JEL classification: D8, J33, M12, M30
Keywords: Salesforce compensation, Pay-for-performance, Self-selection, Incentive plans, 
Moral hazard, Productivity improvement.
W e gratefully acknowledge helpful comments and suggestions from S.P. Kothari (the editor), 
Eric Noreen (the referee), Jake Bimberg, John Core, Nandu Nagarajan, and seminar participants at the 
University o f  Arizona, Cornell University, University o f  Memphis, Michigan State University, Syracuse 
University, University o f  Pennsylvania, University o f  Pittsburgh, University o f  Texas at Dallas, the 
1996 Management Accountmg Research Conference at Vancouver, the 1997 Annual Meeting o f  the 
American Accounting Association at Dallas, the 1998 International Conference on Contemporary 
Accounting Issues at Taipei, and EAA 2000 - the 23rd Annual Congress o f  the European Accounting 
Association at Munich.
♦Corresponding author. Tel.: +1-412-648-1513; fax: +1-412-648-1693.
E-mail address: dhinus@katz.pitt.edu (D. Srinivasan)
1. Introduction
Many firms link compensation to performance by implementing performance-based 
incentive programs at every level of the organization [Schlesinger and Heskett, 1991; Coopers 
and Lybrand, 1992; Buchholz, 1996; McClain, 1998; Pfeffer, 1998; Karr, 1999; Hamilton 
1999].1 However, limited evidence exists on the performance impacts of such programs 
designed for frontline employees [Brown, 1992; Indjejikian, 1999; Murphy, 1999], While 
several research studies document that performance-based incentive plans result in 
performance improvements [Wagner, Rubin and Callahan, 1988; Banker, Lee and Potter, 
1996; Lazear, 1999], evidence on factors leading to those improvements is lacking. Also, 
formal empirical evidence on such plans’ multi-period effects on employee selection and 
retention is lacking [Gerhart and Milkovich, 1992], Using individual productivity data for 
3,776 sales employees for ten quarters following the implementation of a performance-based 
incentive plan, this study explores alternative sources of continuing performance 
improvements.
From agency theory perspective, performance-based incentives increase an 
organization’s overall productivity by attracting and retaining more productive employees 
(selection effect) and/or by inducing employees to increase or to better allocate their effort 
(effort effect). The selection effect occurs because a performance-based compensation 
contract can act as a screening device that encourages less productive employees to leave and 
that motivates more productive employees to join or remain with the firm. However, the 
impact of this sorting effect on workforce composition may not be instantaneous because 
employees may not know their own ability with certainty, and, may learn about it only as they 
receive feedback on their performance. Therefore, progressively higher levels of 
organizational productivity are expected over time.
1 Buck Employees estimate that in 1997, 37% of Fortune 1,000 companies offered bonus plans to hourly employees, up 
lfom 26% in 1993. Moreover, the bonuses amounted to a larger percentage of salary, 7.8% in 1997 versus 4.5% in 1993 
[Buchholz, 1996],
The effort effect occurs because a performance-based incentive plan motivates 
employees to learn more productive ways to perform their tasks. For example, in a selling 
context, employees can increase their “short-term” routine selling effort or their “long-term” 
effort in developing customer relations and loyalty that will eventually result in higher future 
sales. Economic theory suggests that employees with a long multi-period decision horizon, in 
the presence of performance-based incentives over multiple periods, are likely to devote effort 
to learn more productive ways of performing their tasks. To the extent employees invest effort 
in improving their productivity, performance should continue to improve over time (rather than 
instantaneously) following the introduction of a performance-based incentive plan.
Banker, Lee and Potter (1996) examine a performance-based incentive plan’s impact 
on storewide sales for 15 retail outlets. In addition to finding that plan implementation was 
associated with an immediate increase in storewide sales, their study documents a continuing 
improvement (see figure 1). However, the authors could not identify specific causes of 
continuing storewide performance improvement with store-level data. This paper examines 
individual level productivity (quarterly sales per hour) data from ten retail stores at Banker, 
Lee and Potter’s (1996) research site that had implemented the incentive plan. The study 
evaluates whether the continuing increase in sales performance is due to the attraction and 
retention of more productive employees (selection effect) and/or whether it is due to individual 
productivity gains driven by the improvements in employee effort (effort effect). Consistent 
with the predictions of agency theory, the results suggest that both these effects contribute to 
continuing performance improvements.
Insert figure 1 here
The remainder of this paper is structured as follows. Section 2 describes the research 
site and the nature of the performance-based incentive plan. Section 3 develops the research 
hypotheses. Section 4 describes the econometric model and estimation procedures. Section 5 
presents the empirical results, and section 6 offers concluding remarks.
2. Research site
The research site is a division of a Fortune 500 company that focuses exclusively on 
retailing. The division operates a number of retail outlets over a large geographic area. Each 
retail outlet sells similar merchandise to the general public. Most of the stores are located in or 
near shopping malls or urban areas. As one of America’s largest retailers, its stated selling 
strategy is to provide customers with exceptional value. Building on its history of excellence in 
customer service, the division initiated a performance-based incentive plan for sales employees 
in 1987. The main objectives of this plan are to provide superior customer service beyond the 
customer’s expectations and to attract, retain, and motivate productive employees through 
increased earnings opportunity. Prior to the implementation of the performance-based 
incentive plan, sales force compensation was based on fixed hourly wages.
Because division managers were unsure about the precise extent of the incentive plan’s 
impact on sales performance, they phased in the conversion of stores to the plan over a four- 
year period. Initially, the division implemented the plan in one store in 1987. By the end of 
1990, 15 stores out of a total of 34 stores in this geographic region had implemented the plan. 
Senior managers believed that by holding back stores they would be able to assess the impact 
of the incentive plan parameters on performance and hence, fine-tune the plan prior to 
implementing it for the entire salesforce.
The performance-based incentive plan is an add-on bonus program. The incentive 
plan’s key features are as follows: Each week employees get base salaries equal to their hourly 
rate times hours worked. This base salary is at the same rate as that prior to plan 
implementation. In addition, an employee earns a cash bonus each quarter if her quarterly sales 
exceed her pre-specified sales goal. Sales goals for each employee depend on her hourly wage 
rate, hours worked, her department or merchandise group, and other proprietary factors. An 
employee’s bonus is a percentage of the excess of her actual sales over her pre-specified sales 
goal in a quarter.
Failure to meet the quarterly goal does not result in a base salary reduction. However, 
the plan provides for possible termination of an employee if she does not meet her sales goal in 
two consecutive quarters. If an employee fails to meet her quarterly sales goals, she is subject 
to the "Below Requirement" process. The intent of this process is to communicate to
employees whose sales performance is below par that it must be improved within a specific 
period of time or termination will result. Managers, using a performance status monitor, meet 
with poor performing employees to inform them of their progress, and to instruct and advise 
employees on specific actions needed to improve sales performance. Employees do not receive 
annual merit increases, and promotions are rare. Thus, the bonus program is the only 
significant reward for high performance. On average, employees’ bonuses equal 20% of base 
salary. Because these employees have little opportunity for career advancement or other 
performance-based awards within the firm, this setting is particularly suitable for examining the 
effects of monetary incentives on individual performance.2
Senior managers regarded the incentive plan as a major change for the firm and its 
salesforce. Recall that prior to plan implementation employees were on straight salary. The 
firm did not formally track employee productivity and based employees’ pay raises on the 
length of their service experience at the firm. Managers believed that the new incentive plan 
would motivate many changes in employee behavior, leading to improved customer service and 
ultimately resulting in higher sales. For instance, managers expected that following the 
incentive plan employees would begin to build a client base in order to generate repeat sales. 
Selling efforts consistent with this approach include developing and updating customer address 
lists, following up on major purchases by customers, writing thank you notes, and contacting 
customers about upcoming sales and new merchandise. Managers also expected the 
employees to learn how to sell more effectively. This would occur as employees become more 
capable o f identifying a customer’s wants and preferences. Training sessions were held by 
central administration prior to plan implementation to facilitate employees’ transition to this 
new customer-focused strategy and to introduce the plan to the employees. Moreover, to
2 Interviews with managers at our research site indicate that they designed the new incentive plan to have two 
integral ‘carrot and stick’ components, a bonus component that rewards superior performance and a penalty 
component that imposes sanctions for persistent low performance in the form of ‘probation’ with a threat of 
possible termination if  an employee’s performance continues to be below expectation. Our interviews with the 
human resources manager at our research site reveal that very few employees were terminated because of poor 
performance. We acknowledge that both these positive and negative incentives can affect an employee’s sales 
productivity. Therefore, we cannot and do not attempt to disentangle these two possible effects on selection, 
retention and productivity.
encourage employees to be more responsive, plan implementation coincided with empowering 
them in merchandising and customer discount decisions.
3. Hypothesis development
Economic theory suggests that compensation contracts can affect organizational 
performance by sorting employees and by motivating employees to exert effort. Many firms 
face a selection problem of having to hire employees without knowing their skill levels. The 
adverse selection literature examines contracts that take into account employees’ different 
abilities [Spence, 1973; Salop and Salop, 1976], A performance-based compensation contract 
results in less productive employees leaving the firm because their expected future wages under 
the performance-based plan are lower than their prior wages and more productive employees 
joining the firm because they expect to earn more wages through incentive pay [Milgrom and 
Roberts, 1992], Thus, the introduction of the performance-based incentive plan at our 
research site coupled with the provision for possible termination of employment for poor 
performance is expected to have two sorting effects. Several experimental studies support 
these analytical results by documenting that high skilled individuals select performance-based 
incentive schemes when given a choice between piece rate and fixed pay [Chow, 1983; Waller 
and Chow, 1985; Dillard and Fisher, 1990; Shields and Waller, 1988], However, other than a 
few studies examining chief executive officer (CEO) turnover [Gibbons and Murphy, 1990; 
Warner, Watts and Wruck, 1988; Weisbach, 1988], little empirical research examines the role 
of pay-for-performance in attracting and retaining high-performing employees [Milkovich and 
Wigdor, 1991; Gerhart, Minkoff and Olsen, 1995], Therefore, we test the following 
hypotheses:
HI: The sales productivity of existing (pre-plan) employees who remain with the 
store after the implementation of the performance-based incentive plan is greater 
than that of both existing employees who leave and new employees who are hired 
but leave after plan implementation.
H2: The sales productivity of employees hired after the implementation of the 
performance-based incentive plan is greater than that of existing (pre-plan) 
employees who leave and new employees who are hired but leave after plan 
implementation.
Agency models typically assume that employees have perfect knowledge of their ability.
In reality, however, employees might learn about their ability levels only over a period of time
based on feedback on their performance. As a result, employee turnover is likely to continue
for several periods after plan implementation as employees update their beliefs about their
abilities. Therefore, we specify the following hypothesis:
H3: The change in the workforce composition with the more productive 
employees replacing the less productive employees occurs gradually over time 
rather than immediately after incentive plan implementation.
As a result, store productivity will continue to increase as the salesforce composition improves 
over several periods after plan implementation.
When it is not possible to monitor how hard employees are working, performance- 
based incentive plans help to alleviate potential moral hazard problems by providing incentives 
to exert more effort or to better allocate effort. The salesforce compensation literature 
suggests that compared to straight salary a performance-based contract improves a 
salesperson’s performance [Basu, Lai, Srinivasan and Staelin, 1985; Rao, 1990], In the 
context where the agent has to choose among multiple effort dimensions, performance 
improves over time if the new compensation contract provides incentives that favor better 
ways of performing tasks [Feltham and Xie, 1994; Banker and Thevaranjan, 2000], Ultimately 
performance will improve as employees adopt productive activities and eliminate unproductive 
procedures [Wagner, Rubin and Callahan, 1988],
Prior to plan implementation, at our research site, the fixed wage contract provided 
employees with little incentive to learn or to exert more than minimal amounts of effort. The 
new compensation plan provides incentives for employees to exert “short term effort” to 
effectively close sales and to quickly restock merchandise. In addition, the plan provides 
incentives to expend more “long term effort” to learn more productive ways to perform selling 
tasks, to build up a list of loyal well-serviced customers, and to encourage repeat purchases, as 
employees with a long decision horizon can trade-off the disutility of higher effort against the 
expected utility of higher future compensation. In contrast, employees with a short decision 
horizon have less incentive to devote effort to improve their selling skills. Therefore, we test 
the following hypotheses:
H4: The sales productivity of continuing employees improves over time.
HS: The sales productivity of continuing employees improves at a rate greater 
than that of employees who leave the firm.
Finally, because the incentive plan induces employees to go beyond their traditional
roles, in contrast to the pre-plan fixed wage contract that provided employees with little
incentive to learn or to exert more than minimal amounts of effort, the productivity
improvement rate for existing (pre-plan) employees resulting from their experience under the
incentive plan is likely to be greater than their productivity improvement rate prior to plan
implementation.3 Therefore, we specify the following hypothesis:
H6: The sales productivity of existing (pre-plan) employees who continue with
the store improves at a rate greater than the rate experienced prior to plan 
implementation.
At our research site the intent of managers was to develop, test and fine-tune a 
performance-based incentive plan for its salesforce. While we expect that the design of the 
plan considered the factors discussed above, we do not claim that the incentive plan is optimal 
nor do we test for its optimality [Jensen and Murphy, 1990], Moreover, because the firm did 
not track individual performance data prior to plan implementation, our study is limited to 
individual productivity changes subsequent to plan implementation.
4. Model development
4.1 Data
The data consist of 14,651 observations of actual quarterly sales for 3,776 employees 
from ten retail outlets between August 1989 and January 1992.4 Information on employee 
productivity prior to plan implementation is not available. Although the firm implemented the 
incentive plan in 15 stores, we limit our analysis to the ten stores for which individual 
performance information is available at the time of plan implementation. These ten stores are
3 Because productivity data for pre-plan period are not available, we estimate productivity improvement rates 
prior to plan implementation based on a regression model discussed in section 4.
4 Although the firm first implemented the incentive plan in one store in 1987, division headquarters started 
tracking and collecting individual performance information for plan stores in August 1989. The firm did not 
collect individual performance information for employees at non-plan stores during the sample period.
representative of the stores that implemented the incentive plan. Their average sales per hour 
and size as measured by square footage are not statistically different from those of the five 
stores that we exclude from our study. In addition to quarterly sales, individual employee 
information includes hours worked in a quarter, hourly wage rate, store name, merchandise 
group, and the number of years of service at the store by an employee prior to plan 
implementation.
To examine the selection and effort effects, we classify each employee in our sample 
into one of five employee types, and define corresponding indicator variables, as follows:
1) TEMPW ORK ( D ^ l ,  0 otherwise). Employees who remain with the store for no 
more than two quarters during the sample period.
2) NEWSTAY (DNS=1, 0 otherwise). Employees who join the store after incentive 
plan implementation and who remain until the end of our sample period.
3) NEWQUIT (Dnq=1, 0 otherwise). Employees who join the store after incentive 
plan implementation and who leave the store prior to the end of our sample period.
4) OLDQUIT (Doq=1, 0 otherwise). Employees who are with the store before plan 
implementation and who leave prior to the end of our sample period.
5) OLDSTAY (Dos= l, 0 otherwise). Employees who are with the store before plan 
implementation and who remain until the end of our sample period.
To evaluate the impact of the plan over time, we construct a variable PLANQTR that 
measures the number of quarters an employee works under the plan. The variable PLANQTR 
takes on the value one in the employee’s first quarter under the plan and increases by one for 
each quarter completed thereafter. We also include a variable PRIORQTR that measures the 
number of quarters of an existing (pre-plan, i.e. OLDQUIT and OLDSTAY) employee’s 
experience at the store prior to plan implementation.
We measure an employee’s productivity as her average sales per hour (expressed in 
constant dollars after deflating by CPI) in a quarter (HSALES). The demographic and 
competitive environment and, to a limited degree, management style are different for each 
store. Because hourly sales productivity may vary by store due to these factors, we include an 
indicator variable (denoted by the subscript s) for each store. Full-line department stores have
merchandise groups for children’s, men’s and women’s wear, as well as for accessories. These 
merchandise categories within a store may require varying degrees of employee effort and 
store support in the form of advertising and promotional sales. We, therefore, include 
indicator variables for the 14 merchandise groups (denoted by the subscript w) that are present 
in a store. Finally, because the selling season has a large impact on sales productivity, we also 
include indicator variables for the fiscal quarter (denoted by q).
4.2 Estimation model
We model hourly sales of an individual employee i as:
10 14 4 w) Nq
HSALES.t = a  + v i + w^ 2 ? w D w. + q^ q D qt + 8 PLANQTRit
+5NSD]NSPLANQTRlt + S^D^PLANQTR^ + 8 ° SDpS PLANQTR it +elt
( 1 )
where,
v- = ,   T PD_  TP ^ . NQ X j +A, D-N Q + ,A . N SD_  jN S +^  A.  OQ DjO Q + A. O SD-O| S  
+ q ° Q D ° Q PRIORQTRi +  q ^ D ^ P R IO R Q T R j +\i{ (2)
i = 1.. .3776, a subscript to denote employee i,
t = 1... 10, a subscript to denote the time-period in which the employee works,
HSALESit = employee i’s sales per hour (in constant dollars) in time-period t,
Ds; = a dummy variable to represent the store s in which the employee works,
=1 if employee i works in store s (= 1... 10), = 0 otherwise,
Dwj = a dummy variable to represent the merchandise group w in which the employee works,
=1 if employee i works in merchandise group w (= 1... 14), = 0 otherwise,
Dqt = a dummy variable to represent the fiscal quarter q (=1.. .4),
= 1 if the time period t is the same as fiscal quarter q, = 0 otherwise,
D™ = a dummy variable that represents a temporary employee,
= 1 if employee i is temporary, = 0 otherwise,
Similarly, DjNQ, D,NS, D,fJQ, and Di°s denote NEWQUIT, NEWSTAY, OLDQUIT and 
OLD STAY employees, respectively,
PLANQTR = Number of quarters of service by an employee under the plan,
PRIORQTR = Number of quarters of service by an employee prior to the plan,
Sit and pi are error terms.
In equation (1), an employee’s average hourly sales for a quarter is a function of the 
store (s), merchandise group (w), fiscal quarter (q), plan experience measured in number of 
quarters (PLANQTR), an individual specific term v, and a random error component S;t. We
model V\ as a function of employee type, past experience (PRIORQTR) and a random
individual component p.i that represents an individual specificd isturbance term, assumed
randomly distributed across individuals with mean zero. Substituting equation (2) into (1), we 
obtain the following model:
10 14 4
HSALES lt = « +  I  Ps D si + 2  y w D w  + H q D qt 
s—j* w—z q_
+ .T P „DTj P  + , NQ D- NQ +X.  NSD j NS +X. O Q ^OX X ^Q  + , OS ..OSX
+ 5NQ D jNQ PLANQTR it +8NS D PLANQTR it.......................................... ....................  (3)
+ 8()Q D P L A N Q T R  it + 8 OS D ° S PLANQTR it
+ r,C)Q PRIORQTR l + riOS d P S PRIORQTR ; + P; + e it 
We assume:
E[£.t ] = E[p. ] = 0, E[e 2xi ] = a l , E \ v ?  ] = o J ,
E[e . u 1 = 0 for all i, j and t,E [e. e . 1 = 0 if  i and t^ ,
lt^ j J it js J
E [n^^] = 0 if  i *j.
We use the random effects model for estimation purposes because the error component 
Pi is randomly distributed across individuals [Greene, 1993, pp. 469-475], The appendix 
outlines the estimation procedure.5 An alternative specification is the fixed effects model. The 
fixed effects model is not feasible given the specification in (3) because many of the variables in 
the model are linear combinations o f the individual dummy variables required in a fixed effect 
model to represent each individual employee. More importantly, the individual effects for 
different employees are best modeled as varying around the average due to unidentifiable 
stochastic factors. This view is consistent with Mundlak (1978) who argues that individual 
effects should always be treated as random rather than fixed.
5 We also examine all of the coefficients and hypotheses tests presented in this paper using OLS estimates of 
equation (3), and find the results to be robust. However, the OLS estimates are inefficient as the Breusch-
Pagan (1980) test rejects the null hypothesis that the individual effects are zero (Ho: a /  = 0). Empirically, o  2 
is approximately 85% of h \  .
4.3 Tests of hypotheses
The first two hypotheses state that incentive plan implementation results in the 
attraction and retention of more productive employees. To test these hypotheses, we take into 
account prior experience (PRIORQTR) of existing (pre-plan) employees and experience under 
the incentive plan (PLANQTR) for all employees. For instance, the average productivity of 
NEWSTAY employees at the start of the fifth quarter is + 48NS. The median PRIORQTR
experience for OLDQUIT employees is four quarters. Consequently, the average productivity 
of a typical OLDQUIT employee after four quarters is Xt)Q + 4 5C)q + 4r]OQ to reflect the 
productivity improvement of 4r|OQ over the four quarters of prior experience. Therefore, 
ascertaining whether the productivity of NEWSTAY employees at the start of the fifth quarter 
under the plan is greater than that of OLQUIT employees at the same point in time requires 
not just the comparison of XNS and A.(>Q, but the test of the following inequality: X.NS + 45NS > 
+ 4 5°° + 4qOQ. Based on this analysis, we test hypotheses HI and H2 by comparing the 
productivity o f OLDSTAY and NEWSTAY employees to that of OLDQUIT and NEWQUIT 
employees at various points in time (after 0, 4 and 8 quarters under the plan) after taking into 
account the impact of prior experience of existing (pre-plan) employees and the impact of plan 
experience for all employees.
To evaluate Hypothesis H3 about the changing composition of the workforce, we 
specify the following model:
10 4 2 (4 \
PROPORTION (i)st = a + ZpsDs + Z  £qD t + Yj PLANQTR st + y2PLANQTR ^  + est w
s=2 q=2
where,
PROPORTION(i)st = Proportion of an employee type i in a store’s workforce in a quarter t, 
i = TEMPWORK, NEWQUIT, NEWSTAY, OLDQUIT and OLDSTAY, 
est is an error term and other variables are as before.
In this model, the proportions of different employee types in a store’s salesforce in 
quarter t are regressed against PLANQTR, the time elapsed after plan implementation and 
PLANQTR2, controlling for store-specific factors (Ds) and seasonality (Dqt). A positive 
(negative) yi coefficient of the time variable PLANQTR for an employee type indicates that the 
proportion of that employee type is increasing (decreasing) over time. Similarly, a positive
(negative) y2 coefficient of the quadratic term PLANQTR2 indicates that the proportion of an 
employee type is changing at an (a) increasing (decreasing) rate.
To evaluate hypothesis H4 about changes in individual productivity over time we test 
whether the 8 coefficients in equation (3) are positive. Observe that because PLANQTR is 
measured as the number of quarters of experience under the plan, 5 represents average 
productivity improvement per quarter of plan experience. A positive 8 coefficient indicates a 
productivity improvement over time. To test hypothesis H5 we compare the 8 coefficients of 
OLDSTAY and NEWSTAY employees with those of OLDQUIT and NEWQUIT employees. 
Hypothesis H6 states that existing (pre-plan) employees who continue with the store have 
higher productivity improvement rates under the plan than prior to plan implementation. 
Because we do not have employee productivity data prior to plan implementation, we infer the 
average productivity attributable to OLDSTAY and OLDQUIT employees’ prior experience 
based on equation (3). Observe that rj, the average productivity improvement per quarter of 
prior experience is estimated in the regression based on the cross-sectional variation in hourly 
sales among existing (pre-plan) employees, whereas the post-plan productivity improvement 
parameter, 8, for these employees is estimated based on changes in their hourly sales over time 
after plan implementation. We evaluate Hypothesis H6 by comparing r| and 8 coefficients for 
the OLDSTAY and OLQUIT employees.
5. Results
5.1 Preliminary store-level results
Transition from fixed wages based on tenure and hours worked to a performance-based 
incentive plan causes changes to the salesforce. These changes evolve slowly as the salesforce 
responds to the incentives in the new plan. If the performance-based plan is effective in sorting 
employees by attracting more productive employees and encouraging low performers to leave, 
then turnover after plan implementation should be higher than turnover before plan 
implementation. Turnover should gradually subside as the composition of the changed 
workforce begins to reflect the equilibrium induced by the new performance-based plan. 
Turnover data are available only after plan implementation. The first two columns of table 1 
report empirical analysis of the trend in employee turnover at the ten stores. The parameter
estimate in the first column, -0.76, indicates that the proportion of employees leaving the store
declined by 0.76% per quarter from the level experienced immediately on plan implementation.
The results for the quadratic model in the second column also suggest that employee turnover
at a store declined gradually after plan implementation. This evidence is consistent with the
plan inducing a sorting effect that reduces over time as a better matching of the new incentive
system and employee type occurs. Company documents also reveal that some managers at the
company believed that the new incentive plan was impacting turnover. The following quote
from a human resources manager’s internal memorandum is consistent with our findings:
“Initially, turnover in plan stores is higher than the non-plan stores as individuals 
work through changes... Although turnover is high during the first year of the 
plan, there seems to be a dramatic decrease after the store has been on the plan for 
one year.”
Insert table 1 here
The last two columns display the change in storewide hourly sales productivity 
(measured as quarterly store sales divided by total employee hours in that quarter) after 
controlling for economy-wide factors and firm-specific trends in hourly retail sales using the 
average sales productivity at the 19 control stores that did not implement the incentive plan. 
The estimated coefficient in the third column relating the average storewide hourly 
productivity to the number of quarters on the plan suggests that storewide sales productivity 
increases about $0.67 an hour (in constant dollars) on average per quarter. The estimated 
coefficients for the quadratic model shown in the last column indicate that storewide sales 
productivity increases at a slightly higher rate on average per quarter. In the analysis that 
follows, we explore the sources of this continuing improvement.
5.2 Descriptive statistics
Table 2 presents descriptive information on the mix of employees for the ten retail 
outlets used in the study. Comparing the number of observations (14,651) with the number of 
individuals (3,776) indicates that there are on average about four observations for each 
employee. However, the number of observations per employee varies by employee type. While 
we classify only about 17% of the employees as OLDSTAY, this group represents almost 40%
of the observations. Many of the 3,776 employees are temporary (27%). About 55% of the 
permanent employees turn over in these stores in a span of ten quarters.
Insert table 2 here
Table 3 presents summary statistics for the variables used in the study. The mean 
constant dollar sales per hour (HSALES)6 is $105.14 and is close to the median ($99.44). 
There appears to be a large variation in sales per hour as the standard deviation is $38.17, and 
the range between the first and ninth deciles is $86.49. Much of this variation can be explained 
by the seasonal nature of retail sales. The average sales per hour in the fourth quarter exceeds 
that of the first quarter by about $18 (not reported in the tables). This large difference in 
quarterly sales has motivated at least one retailer, Macy’s, to pare down its commission rate 
during the fourth quarter (New York Times, 1995). The median employee works about 24.6 
hours a week (= 320.80 hours per quarter / 13 weeks per quarter). This is typical in the retail 
industry where full time employees can select from a number of weekly work schedules 
ranging from 16 to 40 hours. PLANQTR variable indicates that about 50% of the 
observations are for employees with at least three quarters of experience under the plan.
Insert table 3 here
Table 4 reports sales per hour in constant dollars (HSALES), number of quarters under 
the plan (PLANQTR), number of quarters at the store before plan implementation 
(PRIORQTR), and average hours worked per quarter (HOURS) for each employee type. The 
average sales per hour for NEWSTAY and OLDSTAY employees are greater than those of 
TEMPWORK, NEWQUIT and OLDQUIT employees. However, these averages cannot be 
used to directly compare groups of employees because they do not control for selling season, 
merchandise category, pre-plan or plan experience, and economy-wide factors such as growth 
in retail sales. The average quarterly hours worked by TEMPWORK are less than those for 
any other group. OLDSTAY employees work the most hours. Because all employee types 
except OLDSTAY have at least one quarter (their first quarter or last quarter) in which they
6 Raw sales per hour are on average $ 6  - $10 more than the deflated sales per hour reported in the tables.
may have worked for a fraction of the period, their average quarterly hours are understated. 
Interestingly, of the employees who are with the firm at the start of the plan, those who remain 
(OLDSTAY) have 12 (=16-4) more quarters of pre-plan experience (PRIORQTR) than those 
who quit (OLDQUIT).
Insert table 4 here
5.3 Regression results
Table 5 presents the regression results relating employee type, plan experience and 
quarters of service to sales per hour in constant dollars. Our basic model (model 1) controls 
for inflation during the sample period because the dependent variable is in constant dollars after 
deflating by CPI. However, time-series data may be sensitive to temporal changes in retail 
sales productivity due to other economy-wide, industry- and firm-specific events. Therefore, 
we include three alternative sales growth indexes (GRWTHINDX) as control variables to 
reflect changes in hourly sales attributable to changing economic conditions. In model 2, we 
examine hourly sales productivity by including a retail sales index (a measure of retail-industry 
specific inflation) published by the Department of Labor Statistics as a control variable. While 
this measure controls for economy-wide factors in the retail industry that may affect sales, it 
does not capture other economy-wide, regional and firm-specific factors that may affect sales 
over time. Therefore, in model 3 we use as a control variable a sales growth index computed 
using the total quarterly sales for the 19 control stores that did not implement the plan. In 
model 4, we employ as a control variable a sales productivity growth index computed as the 
average quarterly sales per hour for the 19 control stores. Observe that in our research setting, 
the 19 stores that did not implement the incentive plan provide a natural control because they 
experience the same impact o f inflation as well as other economy-wide, regional and firm- 
specific factors as the ten stores that we analyze. Thus, the growth indexes used in models 3 
and 4 are superior control variables that account for not only inflation but also other factors
that affect sales.7 Because the results o f the four models are similar, we limit our discussion to 
model 4.
Insert table 5 here
5.3.1 Selection and productivity
The X coefficients estimate the amount by which a particular employee type’s 
productivity differs from that of OLDSTAY employees before taking into account pre-plan 
experience. The estimated X coefficients in panel A of table 5 are all significantly negative 
indicating that the OLDSTAY employees initially have higher productivity than other types of 
employees. We present the tests of hypotheses based on model 4 in panel B of table 5. Labels
1.1 through 1.6 in panel B denote tests of hypothesis HI. Test 1.1 compares the productivity 
of OLDSTAY and OLDQUIT employees at the beginning of the first quarter under the plan 
holding their prior experience constant. Tests 1.2 and 1.3 compare the productivity of 
OLDSTAY and OLDQUIT employees after 4 and 8 quarters of plan experience, respectively. 
In these comparisons, we consider median prior experience of 16 quarters (productivity 
increases at the rate of ri°s per quarter) for OLDSTAY employees and four quarters (at the 
rate of per quarter) for OLQUIT employees. The tests indicate that at all these points in 
time, the productivity of OLDSTAY employees exceeds that of OLDQUIT employees. 
Similarly tests 1.4 through 1.6 reveal that the productivity o f OLDSTAY employees exceeds 
that of NEWQUIT employees after 0, 4 and 8 quarters of plan experience.
The first three tests of hypothesis H2 (labeled as 2.1 through 2.3) reveal that 
NEWSTAY employees’ productivity exceeds that of the OLDQUIT employees even after 
taking into account the prior experience of the OLDQUIT employees. Tests 2.4 through 2.6
7 We also estimate the regression models using raw (undeflated) sales per hour. In this case, model 1 does not 
control for any trends over time that may affect sales. Model 2 controls for inflation by using the retail-industry 
specific inflation index as a right hand side variable. Models 3 and 4 implicitly control for inflation because 
the growth indexes used in these models are based on the 19 control stores that experience the same impact of 
inflation and other factors as the experimental stores. However, Glejser’s (1969) tests based on all the four 
models using raw sales before deflation by CPI indicate that the residual errors are proportional to CPI (and 
square of CPI), implying heteroskedasticity in CPI. Therefore, we estimate all the models after deflating by 
CPI and report results that are based on the deflated models. Glejser’s (1969) tests based on the deflated 
models used in the paper indicate that the residual errors are homoskedastic in CPI. Results based on the 
undeflated models (not shown in the paper) are qualitatively similar.
indicate that NEWSTAY employees perform significantly better than the NEWQUIT 
employees at various time periods.
5.3.2 Selection and turnover
Figure 2 displays the changes in the proportions of different types of employees over 
time and provides informal evidence supporting hypothesis H3 that selection effect occurs 
gradually over time rather than instantaneously. Observe that the proportion of NEWQUIT 
employees increases for the first six quarters after plan implementation at about the same ratio 
as the proportion of NEWSTAY employees.8
Insert figure 2 here
Table 6 presents tests of hypothesis H3. Consistent with figure 2, the coefficients of 
PLANQTR in both the linear and quadratic models for NEWSTAY employees are positive and 
that of OLDQUIT employees are negative, and support hypothesis H3 that the transition from 
a less productive (OLDQUIT employees) to a more productive (NEWSTAY employees) 
workforce is gradual. The coefficient o f PLANQTR for NEWQUIT employees is positive and 
the coefficient of PLANQTR2 in the quadratic model is negative and consistent with figure 2, 
suggesting that the proportion of NEWQUIT employees initially increases but eventually 
decreases over time. These findings support hypothesis H3 that some of the new employees 
(NEWQUIT) who are less productive recognize the limits of their ability only after working 
for a few quarters and eventually self-select out of the firm. Results for TEMPWORK and 
OLDSTAY employees suggest that the proportion of temporary employees and existing (pre­
plan) employees who stayed on with a store remains nearly constant. In summary, the results 
shown in table 6 along with the results discussed earlier document that the performance-based 
incentive plan induces a gradual selection effect on the composition of the workforce.
Insert table 6 here
8 As discussed in footnote 3, our sample period is between August 1989 and January 1992 consisting of a 
maximum of ten quarters of data for a store. In our figures, we display the results only for eight quarters to 
document the trend over time. Clearly, the proportions of NEWQUIT and OLDQUIT employees are zero by 
the end of ten quarters, the length of our sample period.
5.3.3 Impact of effort effect on productivity
We evaluate the effect of performance-based incentives in motivating the employees to 
learn how to sell more effectively by examining the 5 coefficients in panel A of table 5. The 
coefficient estimates shown in panel A reveal that only the NEWSTAY and OLDSTAY 
employees appear to improve their sales productivity each period under the plan, and the 
OLDQUIT employees experience a productivity decline before quitting. This finding is 
consistent with hypothesis H4 that the effort effect is positive for the continuing employees. 
Hypothesis H5 asserts that the productivity gains of the continuing employees exceed those of 
the employees who leave. Panel B shows tests of pair-wise comparisons of the 8 coefficients 
(labeled as 5.1 through 5.4). Consistent with the point estimates of the 5 coefficients shown in 
panel A, these tests indicate that the OLDSTAY and NEWSTAY employees realize higher 
productivity gains over time than either the OLDQUIT or NEWQUIT employees.
5.3.4 Prior experience versus plan experience
Evaluation of hypothesis H6 requires a comparison between pre- and post-plan 
experience effects. The estimated q  for OLDSTAY and OLDQUIT are positive and 
statistically significant, indicating that an employee’s pre-plan experience has a positive impact 
on sales productivity. This finding is consistent with the positive link between experience and 
productivity documented for Navy recruiters by Kostiuk and Follmann (1989) and Asch 
(1990). A comparison of q and 5 coefficients for the OLDSTAY employees indicates that 
their productivity increases at a significantly more rapid rate after the implementation of the 
plan (0.15 per quarter in the pre-plan period versus 0.31 after implementation). This suggests 
that the incentive plan motivates the experienced OLDSTAY employees to exert effort to sell 
more effectively. However, there is no evidence of productivity improvement for the 
OLDQUIT employees. All four models indicate deteriorating productivity for the OLDQUIT 
employees after plan implementation.
In addition to testing the hypotheses, we compare the productivity growth rates and the 
productivity of OLDSTAY and NEWSTAY employees at various points in time, as shown at 
the bottom of panel B of table 5. A comparison of 5NS and 5os indicates that the productivity 
improvement rate for NEWSTAY employees is significantly higher than that of OLDSTAY
employees. A comparison between the OLDSTAY and the NEWSTAY employees suggests 
that initially the productivity of NEWSTAY employees is lower than that of the OLDSTAY 
employees. However, the difference in productivity between these two types diminishes with 
greater experience under the incentive plan because the rate of productivity improvement (8NS) 
for the NEWSTAY employees is higher than that for OLDSTAY employees.
Figure 3 portrays the trends in overall productivity for the different types of employees 
after plan implementation, based on each type’s coefficient estimates for model 4 shown in 
table 5. We use the median values for OLDSTAY and OLDQUIT pre-plan experience, 16 and 
4 respectively, to compute the pre-plan experience effect on productivity. The graphs in figure 
3 display evidence of both a selection effect and an effort effect. The employees who stay with 
the store have greater hourly productivity than the employees who quit, indicating a positive 
selection effect. Interestingly, the figure suggests that productivity o f the two employee types 
who remain with the store seems to converge over time. The two groups of continuing 
employees also appear to become more effective, as evidenced by their increasing productivity 
over time. The same is not true for the employees who quit.
Insert figure 3 here
5.4 Non-linear model
While figure 3 is based on the assumption that the impact of the effort effect on the 
sales productivity of different types of employees is linear in plan experience, it is possible that 
productivity increases at a decreasing rate [Carroll, Lee and Rao, 1986]. To examine a non­
linear relation between effort effects and plan experience, we also include a quadratic term 
(PLANQTR2) in our regression models. Panel A of table 7 presents the results based on this 
alternative specification. The tests of hypotheses, presented in panel B of table 7, are similar to 
the linear model, but in calculating productivity we also take into account the quadratic term. 
For example, the productivity of OLDQUIT employees after four quarters of plan experience 
under the quadratic model is given by X(XJ + 4rjOQ + 48OQ + 160C)Q in contrast to .̂OQ + 4r|OQ + 
45oq of the linear model. 160OQ is the square of plan experience multiplied by its coefficient 
9°° Consistent with our earlier findings, the results of tests 1.1 through 1.6, and 2.1 through
2.6 shown in panel B of table 7 also support hypotheses HI and H2 that the productivity of 
continuing employees is greater than the productivity of employees who quit. The change in 
our model specification does not affect tests of hypothesis H3. The estimated 5 coefficients for 
the four quadratic models, shown in panel A of table 7, indicate that productivity growth rates 
under the incentive plan are positive for OLDSTAY and NEWSTAY employees, but the 
productivity gains occur at a decreasing rate as captured by the negative 0 coefficients. Unlike
Insert table 7 here
the linear model where the 5 coefficients capture the constant productivity growth rate under 
the plan, the 5 coefficients of the quadratic model do not completely capture the productivity 
growth rate. Because of the quadratic term, the productivity growth rate is equal to the slope 
of the quadratic curve (8 + 2*PLANQTR*0) and hence will vary with plan experience. In 
panel B of table 7, we use a median value of 3 for PLANQTR to test hypotheses H4-H6.9 For 
example, test 4.1 examines if the productivity growth rate of OLDSTAY employees evaluated 
at the median value of 3 for PLANQTR (5os + 6*0OS) is greater than 0. Tests of hypotheses 
H4-H6, shown in panel B of table 7, are consistent with the results of the linear model. Figure 
4, based on model 4 with quadratic terms, shows the overall productivity for each type of 
employee over time and, consistent with prior results, supports both selection and effort effect 
hypotheses.
Insert figure 4 here
5.5 Examination of an alternative explanation
The intent of the incentive plan at our research site is to attract and retain productive 
employees while encouraging less productive employees to leave (selection effect), and to 
motivate the remaining employees to sell more effectively over time (effort effect). Our 
empirical results suggest that the incentive plan achieves these objectives. However, because 
our tests are based on an ex post partitioning of employee types, a survivor bias, that results if
9 Clearly, lower (higher) values of PLANQTR will result in a stronger (weaker) support for these hypotheses 
because 0 coefficients are negative.
sales are random occurrences and sorting occurs based on random sales as opposed to ability 
and effort, may explain our results. For example, some of the least skilled employees are lucky 
(obtain high sales) and thus stay with the firm until the end of our sample period. Similarly, 
some of the most skilled employees are unlucky (obtain low sales) and as a result decide to 
leave the firm. Thus, instead of the incentive plan effectively sorting employees of different 
ability and effort levels, the documented effects can be explained by the luck of the draw.10
The first approach we use to examine this issue is to test whether realized sales are 
randomly distributed. Our null hypothesis is that the sales productivity of employees is a 
random variable assuming two levels, lucky sales and unlucky sales, without any employee 
specific factors driving performance. If realized sales are random occurrences, then an 
individual’s quarterly sales should be independent of her sales in other quarters. Under this 
assumption sales of each employee will follow a binomial distribution (Mood, Graybill and 
Boes, 1990, pp. 88-90). Based on the binomial distribution, we can evaluate the null 
hypothesis by analyzing the time series of each employee’s quarterly sales.
To remove shocks in an employee’s quarterly sales due to non-employee-specific 
factors such as store, merchandise, seasonality, and economic factors, we estimate model 4 
after omitting employee type, plan experience and pre-plan experience variables to obtain the 
residual sales. A positive residual denotes lucky sales and a negative residual denotes unlucky 
sales. Because our hypothesis assumes binary values for the dependent variable, we use the 
least absolute value (LAV) estimation model. The LAV regression model results in residuals 
with both an expected mean value of zero and an expected median of zero (Koenker and 
Bassett, 1978). The null hypothesis is that the probability of a positive residual (lucky sales) in 
a quarter for any employee regardless of type is equal to 0.5. If, for example, the sales 
residuals for an employee in nine out of possible ten quarters are positive, then based on the 
binomial distribution, we can reject the null hypothesis of random sales for this employee 
(probability (lucky sales) = 0.5) at a p-value of 0.011. However, the binomial tests of 
randomness for each employee (unconditional on the employee type) have very low power
10 We thank the referee for pointing out this possibility.
because of the short length of stay for most employees.11 For example, NEWSTAY employees 
have a median length of stay of six quarters and OLDQUIT employees have a median length of 
stay of four quarters. Therefore, we analyze the residuals using aggregated data for the 
employees who stay till the end of our sample period (STAY) and those that quit before the 
end of our sample period (QUIT), and use the normal distribution because of the large sample 
size. The results are as follows:
Employee type Sample probability of a p-value of
positive sales residual null hypothesis
in a quarter
STAY 0.56 0.01
QUIT 0.43 0.01
These results suggest that there are systematic differences in sales between these two 
groups. For both STAY and QUIT, our tests reject the probability o f positive residuals 
equaling 0.5. The null hypothesis also implies that the probabilities of positive residuals for 
STAY and QUIT are equal. Tests indicate that the probability of positive residuals for STAY 
(NEWSTAY and OLDSTAY employees) is significantly higher than for QUIT (OLDQUIT 
and NEWQUIT employees). In summary, the results of these tests of quarterly time series of 
sales residuals indicate that the employees who stay until the end of our sample period are 
more likely to have systematically higher sales in more quarters than those who quit.
The second approach we take to rule out the alternative explanation of our results is to 
determine if employees’ ability and effort rather than random factors drive sales performance 
and survival. For the incentive plan to be effective it should attract or retain employees with 
high selling ability, encourage these employees to exert effort resulting in continuing sales 
increases and encourage employees without the necessary selling ability and effort to leave. 
Thus, sales performance and hence survival of employees should be a function of employee 
ability and effort. We construct a measure of an employee’s ability as her first quarterly 
residual sales based on the modified regression model described earlier. We construct a
11 Runs or serial correlation tests based on each employee’s residual sales (unconditional on the employee type) 
are other possible approaches to test randomness [Mendenhall, Wackerly, and Scheaffer, 1990, 708-714]. 
However, these tests also have similar (or worse) limitations than the binomial tests because of the short time- 
series of observations for most employees.
measure of an employee’s effort as the change in sales residuals from the first quarter to the 
second quarter normalized by the sales residual in the first quarter. Observe that these ex ante 
measures do not depend on ex post classification of employees into STAY and QUIT.
We use observations from an employee’s third quarter until the end of sample period or 
departure to estimate the relation between sales performance and, ability and effort. The 
estimated regression coefficients based on both OLS and LAV regressions (not shown here) 
suggest that employee sales are significantly and positively associated with both employee 
ability and effort. Thus, our measures of ability and effort are positively related to future sales 
performance. In addition, we analyze the relation between the failure rates (inverse of 
survival) of employees, and our ability and effort measures. Survivor bias may be present in 
our results if the tenure of an employee is not related to her ability or effort. Based on the 
accelerated failure time model (Greene 1993, pp. 721-22), we estimate the failure rate (the 
probability of an employee who stays until time t quitting in time t+1) as a function of ability 
and effort. Based on the actual length of stay for various employees and for a specified 
underlying distribution such as Weibull or lognormal distribution for the survival function, this 
model estimates the failure or hazard rate as a function of ‘covariates’ or other independent 
variables based on maximum likelihood method.12 The estimated coefficients of the ability and 
effort measures based on this model indicate that the failure rate of employees is negatively and 
significantly (p=0.01) associated with both ability and effort of employees.13 Thus, employee 
ability and effort help explain the sorting we observe as a result of the incentive plan at our 
research site. As reported below, STAY had significantly higher levels of ability and effort 
when compared to QUIT. This finding suggests that continuing productivity gains from 
incentives are likely to be a function of employee ability and effort.
12 The employees who remain until the end of our sample period have right-censored length of stay unlike the 
people who quit before the sample period. However, the maximum likelihood estimation procedure in SAS 
handles this problem easily by allowing specification of the censored observations.
13 We also measure effort as the change in residuals from the 1st quarter till the last quarter of stay. The results 
are robust to this change of measurement. We estimate the models using Weibull and lognormal distributions, 
but the results are robust to these specification changes.
Employee type Mean ability Mean effort
STAY 2.91 1.52
QUIT -6.16 -0.60
We acknowledge that short time-series of observations for many employees, our
measures o f employee ability and effort, and the availability of data only during post-plan
period constrain our tests. However, because our alternative approaches yield consistent
results, we believe that survivor bias arising from random sales is unlikely to explain all of our
results. Overall, the results indicate that there are differences in the productivity of the four
types of employees and these differences are consistent with predictions from economic theory.
Specifically, the incentive plan appears to be successful in attracting and retaining employees
who are more productive than those that quit. Moreover, only the employees who remain with
the firm show continuing productivity gains. This is consistent with the continuing employees
responding to incentives to sell more effectively.
5.6. Managerial significance
Regression test results alone are not informative about the relative magnitudes of the
selection and effort effects on a store’s average hourly sales productivity. Determining their
magnitudes requires the consideration of the relative mix of employee types for each quarter
under the incentive plan as well as each type’s initial productivity and subsequent productivity
gains. Therefore, we first compute the average store’s employee mix for each plan quarter.
Using this empirical distribution of employee types, we compute the average hourly sales
productivity for the first eight quarters under the plan by weighting the employee-type
productivity estimates (derived from the coefficient estimates for model 4 in table 5) by the
proportion of different employee types. Specifically, we denote the overall productivity in
period t as p^ = Zw .^p.^, where 
i
wfi = proportion of workforce of employee type i in period t, 
i = OLDSTAY, NEWSTAY, OLDQUIT, and NEWQUIT, 
pit = sales per hour of employee type i in period t.
The increase in store productivity from period 0 to period t is:
pt - po = ?1 witpit 1“ ? WiO1PiO = s ( w it - WiO)piO + ?1 w it(pit - pi0 ) 
In this expression, the first term is the selection effect and the second term is the effort effect.
In figure 5, we further separate the effort effect of the existing (pre-plan) employees (i =
OLDSTAY, OLDQUIT) from that of the new employees (i -  NEWSTAY, NEWQUIT).
Figure 5 displays estimated average sales per hour for a representative store over an 
eight-quarter period beginning with the first period on the plan. It is apparent from the figure 
that both selection and effort effects contribute to the continuing increase in store sales 
productivity. Little is known about how selection impacts organizational performance. Figure 
5 indicates that in our study a large portion of the continuing productivity gains is related to 
selection. Specifically, in addition to the pure selection effect, the new employee effort effect 
can be viewed as indirectly related to the selection of new employees.
Insert figure 5 here
6. Concluding remarks
Attracting, retaining and motivating effective frontline employees are some of the most 
critical challenges facing many service firms today. Economic theories predict that 
performance-based incentives increase an organization’s overall productivity by attracting and 
retaining the more productive employees (selection effect) and/or by inducing employees to 
increase or to better allocate their effort (effort effect). However, while prior studies 
document that performance-based pay results in improved performance, empirical evidence on 
the selection and effort impacts of incentive compensation is lacking, possibly because of lack 
of objective individual-level performance data [Kostiuk and Follmann, 1989; Asch, 1990], In 
this paper, we conduct an analysis of detailed performance data for individual employees to 
ascertain if these effects explain continuing performance improvements after the 
implementation of a performance-based incentive plan.
We find evidence that the continuing improvements following the implementation of a 
performance-based incentive plan are related to these economic theories of behavior. 
Implementation of the plan leads to the attraction and retention of more productive employees, 
supporting the hypothesis that a pay-for-performance plan acts as an effective screening device
by sorting employees by ability. Moreover, the plan motivates the employees remaining with 
the firm to continually improve their productivity, which suggests that pay-for-performance 
provides incentives to invest in effort that has long-term performance effects. Thus, the results 
of this field study support economic theories that analytically demonstrate the importance of 
both selection and effort effects as drivers of improvements in an organization’s performance 
following the implementation of a pay-for-performance plan.
References
Asch, B.J., 1990. “Do incentives matter? The case of navy recruiters,” Industrial and Labor 
Relations Review, Vol. 43, February (Special Issue), 89S-106S.
Banker, R.D., S. Lee, and G. Potter, 1996. “A field study of the impact of a performance-based 
incentive plan,” Journal o f Accounting and Economics, Vol. 21, 195-226.
Banker, R.D. and A. Thevaranjan, 2000. “Goal congruence and evaluation of performance measures,” 
Working paper, University of Texas at Dallas, Richardson, TX.
Basu, A.K., R. Lai, B. Srinivasan and R. Staelin, 1985. “Salesforce compensation plans: An agency 
theoretical perspective,” Marketing Science, Vol. 4, Fall, 267-291.
Breusch, T. and A. Pagan, 1980, “The LM Test and Its Applications to Model Specification in 
Econometrics,” Review of Economic Studies, Vol. 47, 239-254.
Brown, C., 1992. "Wage levels and method of pay," RAND Journal o f Economics, Vol. 23, No. 3, 
Autumn, 366-375.
Buchholz, B., 1996. “The bonus isn’t reserved for big shots anymore,” New York Times, October 
27, F10.
Carroll, V.P., Lee, H.L. and A. G. Rao, 1986. “Implications of salesforce productivity heterogeneity 
and demotivation: a navy recruiter case study,” Management Science, Vol. 32, No. 11 November, 
1371-1388.
Chow, C., 1983. "The effects of job standard tightness and compensation scheme on performance: 
An exploration of linkages," The Accounting Review, October, 667-685.
Coopers and Lybrand, 1992. “Editor’s page: compensation,” Stores, August, 6.
Dillard, J. F. and Fisher, J. G., 1990. "Compensation schemes, skill level and task performance: An 
experimental examination," Decision Sciences, 121-137.
Feltham, G. and J. Xie, 1994, “Performance measure congruity and diversity in multi-task 
principal/agent relations,” The Accounting Review, July, 429-453.
Gerhart, B. and G. Milkovich, 1992. “Employee compensation: Research and practices,” 
Handbook o f Industrial and Organizational Psychological, 2nd Edition, Consulting 
Psychologists Press, Inc., Palo Alto, CA.
Gerhart,B., H.B. Minkoff and R.N. Olsen, 1995. “Employee compensation: theory, practice and 
evidence,” Handbook o f Human Resource Management, Blackwell Business.
Gibbons, R. and K..J. Murphy, 1990. "Relative performance evaluation for chief executive 
officers," Industrial and Labor Relations Review, Vol. 43, February (Special issue), 30S-51 S.
Glejser, H., 1969. “A New Test for Heteroskedasticily,” Journal o f the American Statistical 
Association, Vol. 64, 316-323.
Greene, W., 1993. Econometric analysis, Macmillan Publishing Company.
Hamilton, D.P., 1999. “H-P chief shuffles top roles, says server sales may pinch net,” Wall Street 
Journal, October 4, A4.
Indjejikian, R.J., 1999. Performance evaluation and compensation research: an agency perspective,”
Accounting Horizons, Vol. 13 No. 2, June, 147-157.
Jensen, M.C. and K.J. Murphy, 1990. “Performance pay and top-management incentives,” Journal 
o f Political Economy, Vol.98, No. 2, 225-264.
Karr, A.R., 1999. “Work week: A special news report about life on the job — and trends taking
shape there,” Wall Street Journal, April 6, A1.
Koenkar, R. and Bassett, Jr. G., 1978, “Regression Quantiles,” Econometrica, Vol. 46, No. 1, 33- 
50.
Kostiuk, P.F. and D.A. Follmann, 1989. “Learning curves, personal characteristics, and job 
performance,” Journal o f Labor Economics, Vol. 7, No. 2, 129-146.
Lazear, E.P., 1999. "Performance pay and productivity," Working paper, National Bureau o f  
Economic Research.
McClain, D.L., 1998. “Tricks for varying the pay to motivate the ranks,” New York Times, 
November 15, BU5.
Mendenhall, W., Wackerly, D.D., and Scheaffer, R.L., 1990. Mathematical Statistics with 
Applications, Duxbury Press, Belmont, CA.
Milgrom, P. and J. Roberts, 1992. Economics, Organization and Management, Prentice Hall, 
Englewood Cliffs, NJ.
Milkovich, G.T. and A.K. Wigdor, 1991. Pay for performance: Evaluating performance appraisal 
and merit pay, National Research Council, National Academy Press, Washington, D.C.
Mood, A.M., Graybill, F.A., and Boes, D C., 1990, Introduction to the Theory o f Statistics, 
McGraw-Hill, New York.
Mundlak, Y, 1978. “On the pooling of time-series and cross sectional data,” Econometrics, 46, 69- 
86 .
Murphy, K.J., 1999. “Executive compensation,” Handbook o f Labor Economics (3B), North 
Holland, Amsterdam.
New York Times, 1995. Macy’s marks down some employees’ holiday sales commissions, 
December 15, C1-C4.
Pfeffer, J., 1998. "Six dangerous myths about pay," Harvard Business Review, May/June, 109-119.
Rao, R.C., 1990. "Compensating heterogeneous salesforces: Some explicit solutions," Marketing 
Science, 9, 319-341.
Salop, J. and S. Salop, 1976. "Self-selection and turnover in the labor market," Quarterly Journal o f 
Economics, November, 617-627.
Schlesinger, L.A. and J.L. Heskett, 1991. “The service-driven service company,” Harvard Business 
Review, September-October b, 71-81.
Shields, M.D. and Waller, W., 1988. "A behavioral study of accounting variables in performance- 
incentive contracts," Accounting, Organizations and Society, 581-594
Spence, M., 1973. "Informational aspects of market structure: An introduction," Quarterly Journal 
o f Economics, November, 591-597.
Wagner, J.A., P.A. Rubin, and T.J. Callahan, 1988. “Incentive payment and non-managerial 
productivity: An interrupted time series analysis of magnitude and trend," Organizational 
Behavior and Human Decision Processes, Vol. 42, 47-74.
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, 
July, 458-476.
Warner, J.B., R.L. Watts and K.H. Wruck, 1988. "Stock prices, event prediction, and event studies: 
An examination of top management changes," Journal o f Financial Economics, Vol. 20, No. 1/2, 
January/March, 461-492.
Weisbach, M S., 1988. "Outside directors and CEO turnover," Journal o f Financial Economics, 
Vol. 20, No. 1/2, January/March, 431-460.
Appendix
Estimation of Random Effects Model
We estimate parameters and their standard errors as follows. We first estimate equation (3)
using OLS. Using the errors, ei, from (3) we obtain the sum of the variances of the random
error and individual component based on the following asymptotic relation:
PLIMeie, 9 j
n T l = 4 + 4  <A 1 >
Next, we compute the means by individual for the variables in (3). An OLS regression of these 
means provides the error terms, e2, which have the following property:
Ĝ L  = ? L  + o2 (A2)
n - k  T H
Since T is not a constant across groups, we replace it with
i f "  1 ^
r = -  Y - (A3)
Tt.
Using equations (Al), (A2) and (A3), we compute estimates for cte2 and Op2.
The next step in the estimation requires the transformation of all of the variables in equation 
(3) as follows:
r ; = r „ - g r „  and r ^ x ^ - e x ,
where 0  is computed as:
0 = 1----------—------r  (A4)
To2 + o 2 ^
P e .
We re-estimate equation (3) using OLS with these transformed variables to obtain consistent 
estimates for the parameters in (3) and their standard errors.
Regression Results Relating Plan Experience to Storewide Performance Measures
(p-values in parentheses, n = 83 store-quarters)
10 4 j
TURNOVERs( = a +  £  p s Ds + £  ^qD  t +y ,  PLANQTR s t + y 2P LA N Q TR ^+est 
s= 2 q=2 H
10 4 2
HSALESs t = a + £ p sDs + £ £qD t + yyPLANQTRst + y2PLANQTR^t + ̂ GRWTHINDXt +est 
s= 2 q - 2
Dependent
Variable TURNOVER HSALES
Predictor
PLANQTR -0.76** -1.96** 0.67* 1.53*
(0.01) (0.01) (0.03) (0.05)
PLANQTR2 - 0.11 - -0.08
(0.09) (0.16)
GRWTHINDX _ _ 30.05* 29.96*
(0.05) (0.05)
p(model) 0.01 0.01 0.01 0.01
Adjusted R2 0.29 0.31 0.89 0.89
TURNOVERst= Percent of a store's permanent employees who quit in a quarter,
HSALESst = Storewide quarterly sales divided by storewide selling hours in a quarter,
GRWTHINDXt = Storewide quarterly sales divided by storewide selling hours in a quarter for
the 19 control stores,
PLANQTRst = Number of quarters of a store under the incentive plan,
Ds = a dummy variable to represent the store s (= 1... 10),
Dqt = a dummy variable to represent the fiscal quarter q (=1... 4),
= 1 if the time period t is the same as fiscal quarter q, = 0 otherwise,
* indicates significant at 1% level, 
indicates significant at 5% level.
Composition of Salesforce
Emp oyees Employee -Quarters
Type Number Proportion Number Proportion
TEMPWORK 1,042 27.60% 1,286 8.78%
NEWQUIT 460 12.18% 1,648 11.25%
NEWSTAY 581 15.39% 2,378 16.23%
OLDQUIT 1,055 27.94% 3,692 25.20%
OLDSTAY 638 16.89% 5,647 38.54%
Total 3,776 100.00% 14,651 100.00%
TEMPWORK (D17) = An employee who remains with the store for no more than two 
quarters during the sample period,
NEWQUIT (Dnq) = An employee who joins the store after incentive plan implementation 
and who leaves the store prior to the end of our sample period,
NEWSTAY (Dns) = An employee who joins the store after incentive plan implementation 
and who remains until the end of our sample period,
OLDQUIT (D°°) = An employee who is with the store before plan implementation and 
who leaves prior to the end of our sample period,
OLDSTAY (Dos) = An employee who is with the store before plan implementation and 
who remains until the end of our sample period,
Dj17 = a dummy variable that represents a temporary employee,
= 1 if employee i is temporary, -  0 otherwise,
Similarly, D;NQ, DiNS, D,oq, and D;os denote NEWQUIT, NEWSTAY, OLDQUIT, and 
OLDSTAY employees, respectively.
Descriptive Statistics 
(n=  14,651 employee-quarters)
Standard 10th 90th
Variables Mean Deviation Percentile Median Percentile
HSALES $105.14 $38.17 $64.34 $99.44 $150.83
HOURS 310.01 123.68 134.30 320.80 464.10
PLANQTR 3 3 1 3 8
PRIORQTR 14.74 25.59 0 4 48
HSALES = An employee’s sales per hour (in constant dollars),
HOURS = Hours worked by an employee in a fiscal quarter,
PLANQTR = Number of quarters of service by an employee under the incentive plan, 
PRIORQTR = Number o f quarters o f service by an employee prior to the incentive plan.
Information by Employee Type 
(Means with medians in parentheses, n = 3,776 employees)
^ " \ T y p e TEMPWORK NEWQU1T NEWSTAY OLDQUIT OLDSTAY
V a r ia b le ^ \
HSALES $98.75 $91.27 $103.93 $99.26 $114.99
(94.25) (87.61) (99.44) (95.90) (107.66)
HOURS 192.60 280.90 329.30 303.00 341.70
(169.80) (272.10) (350.70) (302.80) (357.60)
PLANQTR 1.41 3.56 5.60 3.88 9.14
0 ) (3) (6) (4) (10)
PRIORQTR 12.98 29.76
(4) (16)
See tables 2 and 3 for variable definitions.
HSALES i t = a +  I  f i s D  t + I  r wD wj + X $q D t + £GR W1HINDX ,  + ZTP d J P + XNQ D ^ Q + XNS D ^ S 
s=2 w = 2 q = 2
+ x ° Q D ° Q  + s N Q D ? Q  PLANQTR if + S NS D ? S  PLANQTR jf + s °QDOQPLANQTR it
+  S OS D f S  PLANQTR u  + q ° 9  PRIORQTR  +  r;OS D ° S PRIORQTR i + P i + £ it
Panel A: Parameter Estimates
Coefficient Estimates
Variable Coefficient Model 1 Model 2 Model 3 Model 4
D n‘ ATP -7.84** -8.39** -7.66** -8.77**
(0 .0 1 ) (0 .0 1 ) (0 .0 1 ) (0 .0 1 )
D nq r e -9.98** -10.31** -9.88** -10.44**
(0 .0 1 ) (0 .0 1 ) (0 .0 1 ) (0 .0 1 )
D m A m -5.10** -5.56** -4.95** -5.82**
(0 .0 1 ) (0 .0 1 ) (0 .0 1 ) (0 .0 1 )
D OQ r e _9 -9.12** _9 1 9 ** -9.13**
(0 .0 1 ) (0 .0 1 ) (0 .0 1 ) (0 .0 1 )
D nqPLANQTR S NQ -0.24 -0.30 -0 . 2 0 -0.33
(0.27) (0.23) (0.31) (0 .2 0 )
D nsPLANQTR S NS 1.28** 1 .2 1 ** 1.32** 1.14**
(0 .0 1 ) (0 .0 1 ) (0 .0 1 ) (0 .0 1 )
DoqPLANQTR S ° Q -0.49* -0.45* -0.67**
(0 .0 2 ) (0 .0 2 ) (0.03) (0 .0 1 )
D os PLANQTR S°s 0.47** 0.37* 0.51** 0.31**
(0 .0 1 ) (0 .0 2 ) (0 .0 1 ) (0 .0 1 )
D oqPRIORQTR J J OQ 0 . 1 1 * * o n * * o n * * O i l * *
(0 .0 1 ) (0 .0 1 ) (0 .0 1 ) (0 .0 1 )
D os PRIORQTR rf s 0.15** 0.15** 0.15** 0.15**
(0 .0 1 ) (0 .0 1 ) (0 .0 1 ) (0 .0 1 )
p(model) 0 . 0 1 0 . 0 1 0 . 0 1 0 . 0 1
Adjusted R2 0.30 0.30 0.30 0.30
* *  indicates significant at 1 % level one-tail 
* indicates significant at 5% level one-tail
See tables 2 and 3 for variable definitions.
*
oI
Model 1 Model 2 Model 3 Model 4 Model 4 
(p-values) (p-values) (p-values) (p-values) (point 
estimates®)
Hypothesis HI (Productivity of OLDSTAY versus 
OLDQUIT or NEWQUIT after 0,4 or 8 quarters)
1.1 p(16r|os = Loq + 4t|oq) 0.01 0.01 0.01 0.01 11.09
1.2 P(16tios+48os = ).oq + 4tioq + 45oq) 0.01 0.01 0.01 0.01 15.01
1.3 p(16r|os+ 88os = >„0Q + 4t1oq + 88°°) 0.01 0.01 0.01 0.01 18.93
1.4 p(16rtos = XMQ) 0.01 0.01 0.01 0.01 12.84
1.5 p(16r|os + 48os = >.NQ + 48nq) 0.01 0.01 0.01 0.01 15.40
1.6 P(16t|0S+ 88os = X.NQ + 88nq) 0.01 0.01 0.01 0.01 17.96
Hypothesis H2 (Productivity of NEWSTAY versus 
OLDQUIT or NEWQUIT after 0,4 or 8 quarters)
2.1 p(XNS = X0Q + 4r|0Q) 0.02 0.04 0.01 0.01 2.87
2.2 p(XNS + 48ns = X0Q + 4tioq + 48oq) 0.01 0.01 0.01 0.01 10.11
2.3 p(XNS+ 88NS = XOQ + 4riOQ + 88OQ) 0.01 0.01 0.01 0.01 17.35
2.4 p(HNS = ?iNQ) 0.01 0.01 0.01 0.01 4.62
2.5 p(XNS+ 48NS = XNQ + 48NQ) 0.01 0.01 0.01 0.01 10.50
2.6 p(XNS+ 88ns = >.NQ + 88nq) 0.01 0.01 0.01 0.01 16.38
Hypothesis H5 (Productivity growth rate of OLDSTAY 
or NEWSTAY versus OLDQUIT or NEWQUIT)
5.1 p(8os = 8OQ) 0.01 0.01 0.01 0.01 0.98
5.2 p(8os = 8nq) 0.04 0.05 0.04 0.05 0.64
5.3 p(8NS = S0*3) 0.01 0.01 0.01 0.01 1.81
5.4 p(8NS = 6nq) 0.01 0.01 0.01 0.01 1.47
Hypothesis H6 (Productivity growth rates before and 
after plan implementation)
6.1 p(qos = Sos) 0.01 0.10 0.01 0.10 -0.16
6.2 p(iiOQ = 80Q) 0.02 0.02 0.01 0.01 0.78
OLDSTAY versus NEWSTAY (Productivity growth rate 
and productivity after 0,4 or 8 quarters)
p(8os = SNS) 0.01 0.01 0.01 0.01 -0.83
p(16rios = >-NS) 0.01 0.01 0.01 0.01 8.22
P(16tios + 48os = >.NS + 48ns) 0.01 0.01 0.01 0.01 4.80
P(16t,os + 8Sos = >.NS + 88ns) 0.56 0.48 1 0.61 0.39 1 1.58
@ point estimates of the difference between the left and right hand sides of the hypothesis test, 
Median of PRIORQTR for OLDQUIT = 4,
Median of PRIORQTR for OLDSTAY = 16,
GRWTHINDXt = Growth index, a control for changing economic conditions during the sample period. Model 1 does not 
have such a control. In model 2, GRWTHINDX = a retail sales index published by Department of Labor Statistics. In 
model 3, GRWTHINDX = a sales growth index computed as the total quarterly sales for the 19 control stores that did not 
implement the plan. . In model 4, GRWTHINDX = a sales productivity index computed as the average quarterly sales per 
hour for the 19 control stores that did not implement the plan.
10 4
PROPORTION(i)s t = a +  £  P sDs + E  tqD , t  + y,PLANQTR s t + est 
s= 2 q=2 *
10 4 j
PROPORTION(i)s t = a +  £  /?SZ)V + £  %q D  t +y,PLANQTRs t + y-P L A N Q T R ^ + est 
s= 2 q=2 ^
Variable TEMPWORK NEWQUIT NEWSTAY OLDQUIT OLDSTAY
(parameter)
PLANQTR -0.07 0.20 0.60 9.83** 3.86** 3.27** -6.21** -13.83** 0.06 3.48**
(Yi) (0.76) (0.83) (0.10) (0.01) (0.01) (0.01) (0.01) (0.01) (0.75) (0.01)
PLANQTR2 -0.03 -0.80** 0.05 0.73** -0.33**
(Y2> (0.77) (0.01) (0.65) (0.01) (0.01)
Adj. R2 0.24 0.23 0.40 0.65 0.85 0.85 0.90 0.95 0.75 0.82
PROPORTION(i)st = Proportion of an employee type i in a store’s workforce in a quarter, 
i = TEMPWORK, NEWQUIT, NEWSTAY, OLDQUIT, and OLDSTAY,
PLANQTR = Number of quarters of a store under the incentive plan,
Ds = a dummy variable to represent the store s (= 1... 10),
Dqt = a dummy variable to represent the fiscal quarter q (=1.. .4),
= 1 if the time period t is the same as fiscal quarter q, = 0 otherwise,
** indicates significant at 1% level.
HSALES jf  =  « +  I  P s D si  +  X  y w D wj  +  Z  £ q D qt +  gGRWTHINDX ,  +  XIP  d J P  +  XNQ  +  X NS o f S
S—2  IT—2  q=2
+ x ° Q d ° Q  + 6 n Q d ? Q  PLANQTR j( + S NS D f S  PLANQTR u  +  o f ®  PLANQTR u
+ S OS D 9 s  PLANQTR j( +  0 N @ D ? ®  PLANQTR * + 0 N S D ? S  PLANQTR ^  + 0 °@  PLANQTR *
+ 0 ° S o f S  PLANQTR ft + f ] ° Q o f Q PRIORQTR  ■ + tjOS D ? S PRIORQTR , + M, + s it
Panel A: Parameter Estimates
Variable Parameter Model 1 Model 2 Model 3 Model 4
d tp X™ -7.27** -7.78** _7 1 4 ** -8.23**
(0 .0 1 ) (0 .0 1 ) (0 .0 1 ) (0 .0 1 )
d nq ^NQ -10.77** -11.09** -10.71** -11.29**
(0 .0 1 ) (0 .0 1 ) (0 .0 1 ) (0 .0 1 )
d ns N̂S -5.15** -5.60** -5.04** -5.97**
(0 .0 1 ) (0 .0 1 ) (0 .0 1 ) (0 .0 1 )
Doq X0* -9.27** _9 1 7 ** -9.26** -9.03**
(0 .0 1 ) (0 .0 1 ) (0 .0 1 ) (0 .0 1 )
DnqPLANQTR gNQ 1.91* 1 . 8 6 1.96* 1.90*
(0.03) (0.06) (0 .0 2 ) (0.03)
DnqPLANQTR2 0NQ -0.41* -0.42* -0.42* -0.43**
(0 .0 2 ) (0 .0 2 ) (0 .0 2 ) (0 .0 1 )
DnsPLANQTR gNS 2.24** 2 .2 0 ** 2.26** 2 .2 1 **
(0 .0 1 ) (0 .0 1 ) (0 .0 1 ) (0 .0 1 )
DnsPLANQTR2 0NS -0.16 -0.16 -0.16 -0.18
(0 .1 2 ) (0 . 1 1 ) (0 . 1 2 ) (0.08)
DoqPLANQTR 5oq 0.40 0.19 0.40 -0.03
(0.52) (0.78) (0.52) (0.96)
DoqPLANQTR2 0°Q -0.13 -0 . 1 2 -0.13 -0.09
(0.14) (0 .2 1 ) (0.15) (0.30)
DosPLANQTR g°s 1.00** 0 .8 8 * 1 .0 1 ** 0.82*
(0 .0 1 ) (0 .0 2 ) (0 .0 1 ) (0 .0 2 )
DosPLANQTR2 0°S -0.06 -0.06 -0.06 -0.06
(0.13) (0.16) (0.14) (0.14)
d ^ p r i o r q t r 0 . 1 1 * 0 .1 1 * 0 . 1 1 * 0 . 1 1 *
(0 .0 2 ) (0 .0 2 ) (0 .0 2 ) (0 .0 2 )
DosPRIORQTR T1°S 0.15** 0.15** 0.15** 0.15**
(0 .0 1 ) (0 .0 1 ) (0 .0 1 ) (0 .0 1 )
p(model) 0 . 0 1 0 . 0 1 0 . 0 1 0 . 0 1
Adjusted R2 0.31 0.31 0.31 0.31
** indicates significant at 1 % level,
* indicates significant at 5% level,
See tables 2 and 3 for variable definitions.
M odel 1 Model 2 M odel 3 Model 4 Model 4 
(p-values) (p-values) (p-values) (p-values) (point 
estimates)®
Hypothesis HI (Productivity of OLDSTAY versus OLDQUTT 
or NEWQUTT after 0,4 or 8 quarters)
1.1 p(16r|os = X0Q + 4t|oq) 0.01 0.01 0.01 0.01 10.99
1.2 p(16tios + 450s + 160os = ),0Q + 4q0Q + 4S0Q + 160°°) 0.01 0.01 0.01 0.01 14.87
1.3 p(16r|os + 85os + 640OS = >,OQ + 4ti°q + 85OQ + 640OQ) 0.01 0.01 0.01 0.01 19.71
1.4 p(16r)os = X.NQ) 0.01 0.01 0.01 0.01 13.69
1.5 p(16r|os + 480s + 160os = >,NQ + 45NQ + 160NQ) 0.01 0.01 0.01 0.01 15.29
1.6 p(16r|os + 880s + 640OS = >,NQ + 88NQ + 640NQ) 0.01 0.01 0.01 0.01 28.73
Hypothesis H2 (Productivity of NEWSTAY versus OLDQUIT 
or NEWQUTT after 0,4 or 8 quarters)
2.1 p(>.NS = + 4q0Q) 0.01 0.04 0.01 0.05 2.62
2.2 p(XNS + 48ns + 160ns = X0Q + 4t]0Q + 48OQ + 160OQ) 0.01 0.01 0.01 0.01 10.14
2.3 p(X.NS + 88ns + 640ns = >,0Q + 4qOQ + 88OQ + 640OQ) 0.01 0.01 0.01 0.01 14.78
2.4 p(XNS = XNQ) 0.01 0.01 0.01 0.01 5.32
2.5 p(XNS + 48ns + 160ns = >.NQ + 4Snq + 160NQ) 0.01 0.01 0.01 0.01 10.58
2.6 p(X.NS + 88ns + 640ns = >,NQ + 88nq+ 640nq) 0.01 0.01 0.01 0.01 23.80
Hypothesis H4 (Productivity growth rates for OLDSTAY and 
NEWSTAY)
4.1 p(80S + 6 0OS = 0) 0.01 0.01 0.01 0.01 0.46
4.2 p(8NS + 6 0NS = 0) 0.01 0.01 0.01 0.01 1.13
Hypothesis H5 (Productivity growth rate of OLDSTAY or 
NEWSTAY versus OLDQUIT or NEWQUTT)
5.1 p(8os+ 6  0os = 8OQ+ 6  0OQ) 0.01 0.01 0.01 0.01 1.03
5.2 p(80S + 6 0OS = 8NQ + 6 0NQ) 0.01 0.02 0.02 0.01 1.14
5.3 p(8NS + 6 0NS = 8OQ+ 6  0OQ) 0.01 0.01 0.01 0.01 1.70
5.4 p(SNS + 6 0NS = 8nq + 6 0NQ) 0.01 0.01 0.01 0.01 1.81
Hypothesis H6 (Productivity growth rates before and after plan 
implementation)
6.1 p(tios = Sos+ 6  0os) 0.01 0.04 0.01 0.06 -0.31
6.2 p ^ " 3 = 6 0OQ) 0.05 0.05 0.06 0.01 0.68
OLDSTAY versus NEWSTAY (Productivity growth rate and 
productivity after 0,4 or 8 quarters)
p(8os+ 6 0OS= 8ns+ 6 0NS) 0.01 0.02 0.03 0.02 -0.67
p(16qos = X™) 0.01 0.01 0.01 0.01 8.37
p(16r|OS + 48os + 160os = ),NS + 48ns + 160NS) 0.01 0.01 0.01 0.01 4.73
pQ6qos + 88os + 640OS = >.NS + 88NS + 640NS) 0.18 0.16 0.19 0.08 4.93
@ point estimates of the difference between the left and right hand sides o f the hypothesis test See table 5 -  panel B for variable definitions. 
Median ofPRIORQTR for OLDQUIT = 4. Median of PRIORQTR for OLDSTAY = 16. Median of PLANQTR=3.
Productivity growth rate: For the quadrat ic model, the productivity growth rate is given by 5 + 2 PLANQTR*0. Hypothesis tests reported above 
are evaluated at the median value of PLANQTR = 3 so that the productivity growth rate is given by Si+ 6 0i for each employee type i.
Figure 1
Plot of Average Incremental Sales Around Date of Implementation of
Plan
Model: SALESst = a s + ps AVGCOSAL, + est for each s = l , ... 15,
where
SALESst = Sales of store s in month t,
AVGCOSALt = Average monthly sales of the 19 company stores in the same geographic region that did 
not implement the incentive plan, and
a  s and (3 s are estimated using monthly data for 24 months prior to the plan implementation, 
Incremental Salessl (standardized) = X  (S A L E S ^  — — (3^A V G C O SA L ^)/15^V ar(8
Fort = -24, ...-1,0, ...24, where t = 0 is the month of plan implementation.
Source: Banker, Lee and Potter. Journal of Accounting and Economics 21 (1996, p.212).
Figure 2
Proportion of Employee Types Over Time
PLANQTR
TEMPWORK = An employee who remains with the store for no more than two quarters during the sample period, 
NEWQUIT = An employee who joins the store after incentive plan implementation and who leaves the store prior 
to the end of our sample period,
NEWSTAY = An employee who joins the store after incentive plan implementation and who remains until the end of our 
sample period,
OLDQUIT = An employee who is with the store before plan implementation and who leaves prior to the end of our sample 
period,
OLDSTAY = An employee who is with the store before plan implementation and who remains until the end of our sample 
period,
PLANQTR = Number of quarters of a store under the plan,
PROPORTION = Proportion of an employee type in a store’s workforce in a quarter.
PLANQTR
TEMPWORK = An employee who remains with the store for no more than two quarters during the sample period, 
NEWQUIT = An employee who joins the store after incentive plan implementation and who leaves the store prior to the 
end of our sample period,
NEWSTAY = An employee who joins the store after incentive plan implementation and who remains until the end of our 
sample period,
OLDQUIT = An employee who is with the store before plan implementation and who leaves prior to the end of our sample 
period,
OLDSTAY = An employee who is with the store before plan implementation and who remains until the end of our sample 
period,
PLANQTR = Number of quarters of service by an employee under the plan,
Sales Per Hour = Average sales per hour (in constant dollars) for employee type i in a quarter.
OLDSTAY
NEWSTAY
100 --
aoS C
US
i. •
oa 95 -- NEWQUIT
AVh
at
A OC OLDQUITC/5 w
85
80
4 5 6
PLANQTR
TEMPWORK = An employee who remains with the store for no more than two quarters during the sample period, 
NEWQUIT = An employee who joins the store after incentive plan implementation and who leaves the store prior to the 
end of our sample period,
NEWSTAY = An employee who joins the store after incentive plan implementation and who remains until the end of our 
sample period,
OLDQUIT = An employee who is with the store before plan implementation and who leaves prior to the end of our sample 
period,
OLDSTAY = An employee who is with the store before plan implementation and who remains until the end of our sample 
period,
PLANQTR = Number of quarters of service by an employee under the plan,
Sales Per Hour = Average sales per hour (in constant dollars) for employee type i in a quarter.
% 3
W ©
u
Pm c
on
cc S§
PLANQTR
PLANQTR = Number of quarters of a store under the plan,
Sales Per Hour = Average storewide sales per hour (in constant dollars) in a quarter.