Professional Documents
Culture Documents
Managing The Impact of Employee Turnover On Performance - The Role of Process Conformance
Managing The Impact of Employee Turnover On Performance - The Role of Process Conformance
net/publication/211395193
CITATIONS READS
250 3,696
2 authors, including:
Robert S Huckman
Harvard University
48 PUBLICATIONS 1,981 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Robert S Huckman on 29 May 2014.
W e examine the impact of employee turnover on operating performance in settings that require high levels of knowledge
exploitation. Using 48 months of turnover data from U.S. stores of a major retail chain, we find that, on average,
employee turnover is associated with decreased performance, as measured by profit margin and customer service. The effect
of turnover on performance, however, is mitigated by the nature of management at the store level. The particular aspect of
management on which we focus is process conformance—the extent to which managers aim to reduce variation in store
operations in accordance with a set of prescribed standards for task performance. At high-process-conformance stores,
managers use discipline in implementing standardized policies and procedures, whereas at low-process-conformance stores,
managers tolerate deviations from these standards. We find that increasing turnover does not have a negative effect on
store performance at high-process-conformance stores; at low-process-conformance stores, the negative effect of turnover
is pronounced. Our results suggest that, in settings where performance depends on the repetition of known tasks, managers
can reduce turnover’s effect by imposing process discipline through standard operating procedures.
Key words: employee turnover; process management; knowledge exploitation; retail operations
History: Published online in Articles in Advance December 11, 2007.
reduce variation in store operations in accordance with other and the organization (Leana and Van Buren 1999,
a set of prescribed standards for task performance. Dess and Shaw 2001). A second source of indirect costs
Our study design takes advantage of the fact that we is the demoralization of employees who remain with
are able to examine longitudinal observations of store- a firm (Staw 1980, Steers and Mowday 1981, Mobley
level performance across sites that are owned and oper- 1982). This demoralization may be due to the loss of a
ated by the same parent company. We are thus able to respected colleague or the fact that turnover may require
control for firm-level characteristics (e.g., employment additional work to be absorbed by remaining employ-
policies, training procedures, etc.) that may otherwise ees whose capacity is already stretched (Mowday et al.
Additional information, including rights and permission policies, is available at http://journals.informs.org/.
INFORMS holds copyright to this article and distributed this copy as a courtesy to the author(s).
a source of new knowledge for firms. Furthermore, an Bowen and Lawler 1992, Adler and Cole 1993, Sitkin
experimental study by Argote et al. (1995) finds that, et al. 1994, Bowen and Lawler 1995, Adler et al. 1999).
although the average effect of turnover on work group In support of the role of process conformance in mit-
performance is negative, this effect is less pronounced igating turnover’s effect on performance, Argote and
for complex tasks than for simple tasks. They attribute Epple (1990) state that “turnover may matter more in
this difference in turnover’s effect to the fact that the per- organizations where jobs are not standardized and pro-
formance of complex tasks requires greater innovation cedures do not exist for transmitting knowledge to new
(i.e., exploration) than simpler tasks requiring repetition members” (p. 922). In motivating our second hypothesis,
we note that this quote uses the word “may,” suggest-
Additional information, including rights and permission policies, is available at http://journals.informs.org/.
INFORMS holds copyright to this article and distributed this copy as a courtesy to the author(s).
(i.e., exploitation).
We maintain that performance in retail chain settings, ing that the role of such standardization in mitigating
such as the setting considered in this study, is more turnover’s negative effects merits empirical examination.
accurately characterized as requiring exploitation than We also stress a point that is important to our analysis,
exploration. We refer to March’s choice of terms to though not addressed in the above quote—that standard-
ization is a necessary but not sufficient condition for
describe exploitation (e.g., refinement, production, effi-
process conformance. In our setting, the level of stan-
ciency, implementation, execution) versus exploration
dardization as designated by the corporate office is the
(e.g., search, variation, risk taking, experimentation,
same at all 268 stores, each of which is instructed to
discovery, innovation) (March 1991). As Winter and follow the policies and procedures described in the 334-
Szulanski (2001) also argue, companies that operate page corporate manual. Because the chain uses central-
multiple units often evolve from a phase of exploration ized planning, compliance with standards is important
during which they experiment with different approaches, from the perspective of the corporate office. The extent
to a phase of exploitation during which they “freeze the to which store managers enforce these standards, how-
design” and replicate the established approaches. As a ever, differs considerably across sites. While some store
result, the performance at mature retail chains is highly managers are quite disciplined in implementing these
dependent on the successful execution of known activi- prescribed standards (high-process-conformance stores)
ties such as processing of incoming inventory, shelving others are more tolerant of deviations from these stan-
merchandise, responding to customer queries, and trans- dards (low-process-conformance stores).
acting sales on the cash registers. In our retail setting, to the extent that stores operate
In this environment, we expect employee turnover to with a high degree of process conformance, we expect
have a negative effect on firm performance due to oper- that knowledge concerning task performance will be
ational disruption from employee departures, additional transferred more easily to new employees (i.e., it will
work that must be absorbed by remaining employees, be less likely to be lost when employees depart). This
and the loss of tacit knowledge and accumulated experi- expectation leads to our second hypothesis:
ence held by departing employees. While activities such Hypothesis 2. In settings requiring high levels of
as shelving merchandise may benefit from increased knowledge exploitation, the level of process confor-
effort of new employees, we expect this positive effect to mance will moderate the impact of turnover on operat-
be outweighed by the negative effects of turnover. Based ing performance. In particular, stores with lower levels
on this characterization and the prior literature, we arrive of process conformance will be more negatively affected
at our first hypothesis: by turnover than stores with higher levels of process
conformance.
Hypothesis 1. In settings requiring high levels of
knowledge exploitation, turnover will, on average, have
a negative effect on operating performance. 3. Research Design
3.1. Empirical Setting
To the extent that the hypothesized negative effect We test our hypotheses using data from Borders Group
of turnover on operating performance in exploitation- (Borders), a Fortune 500 retailer of entertainment prod-
based settings exists, we are interested in understand- ucts such as books, CDs, and DVDs. At the end of 2003,
ing the tools that managers might use to mitigate these Borders operated 445 superstores (under the name Bor-
effects. Below, we argue that one management lever in ders) and about 720 mall-based stores (under the name
this regard is process conformance—the degree to which Waldenbooks), and employed approximately 32,000 peo-
managers aim to reduce variation in operations in accor- ple. In this study, we focus solely on Borders super-
dance with a set of prescribed standards for task perfor- stores (hereafter called Borders stores). Like most retail-
mance. This concept is similar to notions of attention to ers, Borders stores experience high levels of employee
detail and adherence to rules and procedures (Naveh and turnover. Between 1999 and 2002, the average annual
Erez 2004). Several researchers claim that a process con- full-time employee turnover across Borders stores ranged
formance approach may be appropriate in settings that from 49% to 69%, and the average annual part-time
produce standardized products or services (Levitt 1972, employee turnover ranged from 94% to 114%.
Ton and Huckman: Managing the Impact of Employee Turnover on Performance
Organization Science 19(1), pp. 56–68, © 2008 INFORMS 59
Borders stores provide an ideal setting in which to 3.2. Data and Measures
examine whether the impact of turnover on performance Our sample includes data from all 268 Borders stores
can be moderated by process conformance. Because that opened before August 1999. We obtained monthly
Borders uses centralized merchandise planning, it has turnover and performance data for each store from 1999
developed standard operating procedures across all of to 2002. Below, we describe our measures of store per-
its stores. This standardization supports customer ser- formance, employee turnover, and process conformance.
vice efforts at the store level. For example, computer We then describe our empirical analysis and the asso-
terminals at Borders stores allow customers to check the ciated control variables. Summary statistics and simple
Additional information, including rights and permission policies, is available at http://journals.informs.org/.
INFORMS holds copyright to this article and distributed this copy as a courtesy to the author(s).
availability and location of specific books. The success correlations among all variables used in our analysis are
of this system depends on books being shelved in spe- provided in Tables 1 and 2, respectively. We note that
cific sections, as prescribed by the corporate standards. Table 2 should be interpreted with caution, as it uses
The shelving standards, as well as other standard oper- pooled observations from the 268 sites across 48 time
ating procedures, are described in a 334-page policy and periods.
process handbook. Although all stores are instructed to
conform to these standards, we observed during our store 3.2.1. Store Performance. We examine turnover’s
visits that the extent of conformance varies across sites. effect on two measures of store performance. The first,
While some store managers impose discipline in con- customer service score, is a subjective measure. Each
forming to prescribed standards, others tolerate devia- store is shopped once a month by a mystery shopper.
tions. For example, at one store, in violation of shelving These shoppers fill out a form that includes approxi-
standards salespeople organized art books by artist rather mately 50 questions about the store environment and
than by author. The manager of that store noted customer service interactions.2
we don’t shelve art books like other Borders stores do. A store’s overall customer service score is based on
Our salespeople are extremely brainy. Other stores don’t the mystery shopper’s answers to these questions. Our
shelve by artist, they shelve by author. Like some of titles second performance measure, profit margin (operating
up there, I don’t know which artists they are and look in income divided by sales), is an objective measure of
the computer, and it’s like a Degas book, and I wouldn’t financial performance. We prefer this measure to other
know that by looking at it, but the salespeople know. measures of financial performance, such as total sales or
So they put titles with Degas that don’t have the word total profit, as these latter measures are affected by the
“Degas” in the name. They do that with biographies as absolute level of activity at the store. In many instances,
well. If a biography doesn’t have a name of a person
they will know that and say, “Hey, this is a biography
the absolute level of activity depends on factors that
about this person. I’m going to shelve it with this per- are beyond the control of the store manager, such as
son’s work.” weather conditions, corporate promotions, author sign-
ings, or idiosyncratic special events. Because it is dif-
At another store, salespeople placed overstocked ficult to account for the impact of all of these external
books underneath display tables on the selling floor factors, we use profit margin as a performance measure
instead of placing them in appropriate storage areas. that controls for differences in the absolute level of activ-
Similar to the above example, the store manager was
ity across stores.
well aware of this violation and even encouraged it, as
she thought it saved her salespeople time in restocking
shelves. Table 1 Summary Statistics (Sample of 268 Stores Across
This approach, however, was not consistent across 48 Months)
stores. In fact, we asked another store manager why N Mean Std. dev. Min Max
he did not encourage his employees to deviate from
Customer 11325 8507 1121 2778 100
shelving standards when they disagreed with them. He service score
replied, “That’s exactly how books get lost at a store.” In Profit margin 12717 744% 991% −9365% 5566%
a related vein, a five-year employee at another store told RPL 10343 8770% 1270% 430% 10000%
us that, while she often did not agree with the shelving Store conditions∗ 714 7543 1133 3700 9920
Full-time turnover 12718 437% 537% 000% 5000%
standards, it was clear to her that she was expected to Part-time turnover 12709 749% 1050% 000% 25000%
follow them. While picking up the new merchandise that Total turnover 12717 548% 532% 000% 8519%
arrived at the store, she pointed at two books that were Turnover of 12722 0022 0149 0 2
related to the bestselling book The Da Vinci Code, and managers
Total payroll ($) 12711 63,759 21,437 10,416 273,653
observed Proportion full 12717 062 013 018 100
this one goes to “Christianity,” but this one is in “meta- Percent of 12717 429 173 110 1930
physical.” See, that’s another dumb thing. We don’t keep unemployment
Number of 12715 099 097 0 6
them together in sections. This is not our thinking, but
competitors
this is what we are asked to do. The label says Christian-
ity, so I’ll put it in Christianity. ∗
Store conditions is observed at a store-year level.
Ton and Huckman: Managing the Impact of Employee Turnover on Performance
60 Organization Science 19(1), pp. 56–68, © 2008 INFORMS
Customer service 1
score
Profit margin −0022 1
0022
Full-time turnover 0002 −0048 1
Additional information, including rights and permission policies, is available at http://journals.informs.org/.
INFORMS holds copyright to this article and distributed this copy as a courtesy to the author(s).
0871 <00001
Part-time turnover −0011 −0066 0161 1
0248 <00001 <00001
Total turnover 0007 −0060 0441 0389 1
0436 <00001 <00001 <00001
Turnover of −0033 −0004 0048 0028 00313 1
managers
0001 0650 <00001 0002 0001
Total payroll 0059 0287 −0018 −0020 −0022 0037 1
<0001 <00001 0038 0024 0013 <00001
Proportion full 0046 −0110 −0004 0169 00421 0004 00418 1
<00001 <00001 0684 <00001 <00001 0693 <00001
Unemployment −0223 −0072 −0089 −0031 −0090 −0005 −00408 00565 1
<00001 <00001 <00001 0000 <00001 0553 <00001 <00001
Competitors −0024 0006 −0036 −0043 −0051 0003 0008 −00923 −0072 1
0007 0369 <00001 <00001 <00001 0771 0361 <00001 <00001
3.2.2. Employee Turnover. Employees voluntarily 1977, Glebbeek and Bax 2004, Shaw et al. 2005). As
leave Borders stores for a variety of reasons including shown in Table 1, average monthly full-time turnover
career changes, geographical moves, pursuit of educa- in our sample is 4.4%, while average monthly part-time
tional opportunities, or retirement. As shown in Figure 1, turnover is 7.5%.
across all Borders stores, employee turnover tends to be 3.2.3. Process Conformance. To assess the level of
highest during the month of August due to employees’ process conformance at stores, we use two measures
leaving Borders to return to school. that are regularly tracked by Borders management. The
Given the nature of the performance measures we first measure, internally called store conditions, is taken
observe, we focus on turnover of employees who per- quarterly and measures conformance to a wide range of
form day-to-day operational activities at the stores and processes. The second measure, internally called returns
exclude employees who are engaged in office coordi- pull list, or RPL, is taken monthly and measures confor-
nation or community relations activities. In addition, mance to a specific process. Below we describe each of
we focus solely on permanent employees. Like most the measures and explain how we use them to create a
retailers, Borders hires temporary employees during composite measure of process conformance.
high-traffic periods (e.g., holiday shopping in Decem-
3.2.4. Store Conditions. Borders classifies its stores
ber). These temporary employees are excluded in our into different regions, each of which is led by a regional
measures.3 In addition, Borders reports employees’ manager. Regional managers regularly make store vis-
departures only when they leave the company perma- its to assess the performance of stores and to set goals
nently. Consequently, our turnover measures do not track for store managers. Once a quarter, each store’s regional
store transfers. Finally, we note that most turnover at manager also performs a thorough inspection of its con-
Borders stores is voluntary. Though we would have formance to a wide range of processes. The inspec-
liked to distinguish between voluntary and involuntary tion covers four categories: books, café, music and
turnover, we were not able to obtain the data required to video, and store operations. Each category is divided
make such distinctions. into multiple sections. For example, the book category
We calculate the turnover rate in a particular period is divided into five sections: alphabetization, shelf orga-
as the number of employees who left a store during nization, endcaps, overstock, and product flow. Each of
that period divided by the average number of employees these sections is further divided into different items. The
working at the store during that period. While there are alphabetization section, for example, includes overall
several other methods for calculating employee turnover alphabetization, category-critical alphabetization, biog-
(Mobley 1982, Price 1977), our measure is widely used raphy alphabetization, and alphabetization when an
and allows for comparability to earlier findings (Price author has multiple books.
Ton and Huckman: Managing the Impact of Employee Turnover on Performance
Organization Science 19(1), pp. 56–68, © 2008 INFORMS 61
3
99
00
00
01
01
02
02
2
-9
-9
-9
-0
-0
-0
-0
-0
-0
b-
b-
g-
b-
g-
b-
g-
ay
ov
ay
ov
ay
ov
ay
ov
Fe
Au
Fe
Au
Fe
Au
Fe
Au
M
N
Time
Stores receive a score for each of the items. The aver- in manufacturing plants. Preventative maintenance often
age of these scores determines the score for the specific is neither tracked nor used to evaluate plant perfor-
section. Specific section scores are first multiplied by a mance, nor used to compensate plant managers. It does,
weighting scale and then added up to create the total however, reflect managerial attention to process con-
store conditions score. formance; managers who place the greatest emphasis
3.2.5. Returns Pull List. The second component of on process conformance are likely to ensure that pre-
our composite measure of process conformance is a ventative maintenance is performed as scheduled, while
store’s RPL score. In this setting, retailers are allowed those who are less concerned with process conformance
to return unsold books to the publishers for a full refund may sacrifice preventative maintenance for other press-
minus the costs of shipping and handling. Processing ing tasks at the plant. In addition, like preventative
returns is a major task at Borders stores: a typical maintenance, although returning books does not have an
store returns approximately 110,000 units, or 40% of its immediate effect on performance, it may have a longer-
inventory, every year. At the beginning of each month, term and indirect effect on store performance. Removing
Borders’s corporate office sends a returns list to each books that do not sell well from the selling floor allows
store. The standard operating procedures require stores stores to use the limited shelf space for books that sell
to return all books on the list to the retailer’s distribu- better, presumably leading to higher sales.
tion centers by the end of the month. At the end of each
returns period, the stores receive a returns conformance 3.2.6. Composite Measure of Process Conformance.
score (internally called an RPL score), based on the To develop our composite measure of process con-
number of units returned divided by the total number of formance, we use the average store conditions score
units that were supposed to be returned. The returns pro- and the average RPL score for each store for each
cess, described in detail in the policy and process book, year.4 Using these scores, we calculate the mean and
involves finding the books, packing them, and shipping standard deviation of store conditions and RPL scores
them to the distribution centers. across all Borders stores for each year. For each store,
RPL scores do not affect stores’ short-term financial we then standardize the yearly store conditions and
performance because inventory management is central- RPL scores by subtracting the mean and dividing by
ized at Borders. Because stores do not pay for the books the standard deviation. We combine these standardized
they carry, they do not receive a payment when they scores to create the composite process conformance
return the books. In addition, RPL performance is not measure. Using this measure, we rank stores and divide
used by corporate management to evaluate store man- them into high-process-conformance and low-process-
ager performance. Consequently, maximizing the RPL conformance stores. We use these two categories—
score is not necessarily a high priority for store man- rather than using process conformance as a continuous
agers. Precisely because RPL is not a high priority, it variable—to account for the fact that process confor-
reflects the degree to which store managers place empha- mance is a proxy for high and low levels of process
sis on process conformance. In this sense, RPL is similar conformance rather than a precise measure of this under-
to activities such as preventative machine maintenance lying variable. As such, we are reluctant to suggest that
Ton and Huckman: Managing the Impact of Employee Turnover on Performance
62 Organization Science 19(1), pp. 56–68, © 2008 INFORMS
small variations in process conformance scores repre- control for changes over time in factors such as eco-
sent a meaningfully different level of process confor- nomic conditions and corporate policies, while the
mance. Furthermore, our simple categorization simpli- calendar-month effects control for seasonality.
fies the interpretation of our empirical results. To make sure that turnover is not endogenous to per-
formance, we asked store managers, store employees,
3.3. Empirical Model and corporate managers what drives turnover at Borders
We estimate the following equation using ordinary least stores. None of the managers we interviewed mentioned
squares (OLS) to test for the effect of employee turnover poor store performance as a driver of employee turnover.
Additional information, including rights and permission policies, is available at http://journals.informs.org/.
INFORMS holds copyright to this article and distributed this copy as a courtesy to the author(s).
on various dimensions of store performance: Furthermore, they noted that involuntary turnover due
to the poor performance of individual employees is
Performanceimy quite rare. Finally, several of our empirical models use
= i + y + m + 1 Employee Turnover imy lagged measures of turnover to reduce concerns of endo-
geneity. Nevertheless, to test whether employee turnover
+ Ximy 2 + imy
(1) remains endogenous, we examine the effect of past store
We estimate versions of (1) using several different mea- performance on employee turnover. We use one-month
sures of employee turnover. In our initial model, we use and two-month lags for both customer service and profit
separate variables for part-time and full-time turnover. margin. The results, not reported in our paper, show no
Given the similarity between the tasks performed by evidence of significant effect of either measure of store
these two types of employees in this setting, we also performance on employee turnover. The absence of sig-
estimate (1) using a single turnover measure for both nificant effects in the reverse regressions is reassuring,
types of workers, which we refer to as total turnover. and suggests that we need not be overly concerned about
Regardless of whether we split or combine full- and the exogeneity of our turnover measure.
part-time turnover, we measure each value in two ways. Several have hypothesized that the relationship
First, we measure turnover over the one-month period between employee turnover and performance is nonlin-
ranging from the 15th day of the prior month to the ear (Price 1977, Bluedorn 1982). To determine whether
15th day of the current month. We use this time period the effect of turnover on performance depends on the
due to the uncertainty surrounding the speed with which level of turnover at the stores, we estimate models in
one would expect turnover to affect performance. Using which we interact monthly turnover with two categories
turnover lagged by one month may be misleading if for the average level of turnover—high and low—at the
the effects of turnover occur immediately. Alternatively, stores. This model appears below:
using turnover in the current month would be problem- Performanceimy
atic if performance effects occur with a lag. We use our
measure as a compromise between these two extremes. = i + y + m + 1 Employee Turnover imy
To address concerns about the potential noise in monthly + 2 Employee Turnover imy
levels of turnover, we also estimate our models using a
measure of turnover during the prior three months. ∗ High Turnover imy Ximy 4 + imy
(2)
In addition to employee turnover, (1) includes a vec-
tor, Ximy , that contains several store-level variables that The high- and low-turnover categories were assigned by
vary over time. These include an indicator for turnover ranking and dividing the stores into halves based on their
by store managers during the current month (to con- turnover in each year. We also estimate versions of (2)
trol for management changes);5 full-time employees as where we divide stores into three categories based on
a percentage of total employees (to control for employee their yearly turnover performance—high, medium, and
mix); total store payroll (to control for the total amount low. The results (not reported in this paper) are very
of labor used by the store); the number of competitors similar to those using two categories.
in the local market;6 and the unemployment rate in the Finally, to determine whether process conformance
store’s metropolitan statistical area (to control for labor moderates the relationship between turnover and per-
supply).7 We would have liked to control for average formance, we estimate models in which the level of
employee tenure at the store. We were not able, however, employee turnover is interacted with the two categories
to obtain these data from the company. of process conformance—high and low. This model
This specification also includes fixed effects for each appears below:
store i , each year y from 1999 to 2002, and each
Performanceimy
month m of the calendar year. Store fixed effects con-
trol for time-invariant, unobserved heterogeneity across = i + y + m + 1 Employee Turnover imy
stores8 (Hausman and Taylor 1981), which might oth-
+ 2 Employee Turnover imy
erwise affect both employee turnover and store perfor-
mance, leading to biased estimates. The year effects ∗High Process Conformanceiy +Ximy 4 + imy
(3)
Ton and Huckman: Managing the Impact of Employee Turnover on Performance
Organization Science 19(1), pp. 56–68, © 2008 INFORMS 63
As a test of robustness, we also estimate a version of (3) are not large—an increase of one-standard deviation10
where we divide stores into three categories based on in full-time turnover and part-time turnover at an aver-
their yearly process conformance performance—high, age store lead to reductions of 0.33 and 0.20 points in
medium, and low. In addition, we also estimate versions customer service scores, respectively (0.4% and 0.2%
of (3) where we use yearly RPL scores, yearly store con- relative to the average score of 85.1 points). Using total
ditions scores, and variations in RPL as single measures turnover, the impact of an increase of one standard devi-
of process conformance.9 ation is a reduction of 0.39 points (0.5% relative to the
average score).
Additional information, including rights and permission policies, is available at http://journals.informs.org/.
INFORMS holds copyright to this article and distributed this copy as a courtesy to the author(s).
Notes. ∗ , ∗∗ , ∗∗∗ denote statistical significance at the 10%, 5%, Notes. ∗ , ∗∗ , ∗∗∗ denote statistical significance at the 10%, 5%,
and 1% levels, respectively. Standard errors are heteroskedasticity and 1% levels, respectively. Standard errors are heteroskedasticity
robust and clustered by store. Regressions include a constant term robust and clustered by store. Regressions include a constant term
not shown in the table. not shown in the table.
Ton and Huckman: Managing the Impact of Employee Turnover on Performance
64 Organization Science 19(1), pp. 56–68, © 2008 INFORMS
deviation is 2.41%, relative to the average profit margin, measures of turnover, we find that the impact of turnover
and amounts to $11,700 of lost profit per store per year. on performance is negative at low-turnover stores for
This seemingly modest negative effect on profit margin, both customer service and profit margin. Using customer
however, is economically significant for the company. service as the dependent variable, high-turnover stores
A loss of $11,700 profit per store amounts to a total show an effect of turnover that is 8.1 points greater than
loss of $3.1 million for 268 stores, or approximately the effect for low-turnover stores using monthly mea-
6% of the net income that came from Borders stores in sures of turnover. The combined effect of turnover of
1999.11 4.33 (=−3
72 + 8
05) for high-turnover stores is not
Additional information, including rights and permission policies, is available at http://journals.informs.org/.
INFORMS holds copyright to this article and distributed this copy as a courtesy to the author(s).
Although our findings are sensitive to the length of significantly different from zero. Similarly, using profit
time over which turnover is observed, the results from margin as the dependent variable, high-turnover stores
Tables 3 and 4 provide some support for Hypothesis 1. show effects of turnover that are 0.06 and 0.02 percent-
Using turnover rates for the prior three months, on aver- age points greater than the effect for low-turnover stores
age, turnover has a negative effect on both measures of using monthly and three-month measures of turnover,
store performance. We note that the results using sepa- respectively. Again, the combined effects of turnover
rate measures of full- and part-time turnover and a single for high-turnover stores are not significantly different
measure of total turnover are very similar. In subsequent from zero. Overall, our results show that the effect
sections, we discuss our results using the measure of of turnover on performance is worse for low-turnover
total turnover only. stores than for high-turnover stores. This is consistent
with Price’s (1977) prediction and a recent empirical
4.2. Nonlinear Effect of Employee Turnover study of turnover in the trucking and concrete pipe
Columns 1 and 2 in Table 5 present estimates of (2) industries (Shaw et al. 2005).
with customer service as the dependent variable, and
Columns 3 and 4 provide similar estimates of models 4.3. Moderating Effect of Process Conformance
with profit margin as the dependent variable. For all Table 6 presents estimates of (3) with customer service
and profit margin as the dependent variable.12 Regardless
Table 5 Regressions Testing the Nonlinear Effect of of whether we use customer service (Columns 1 and 2)
Employee Turnover on Both Measures of
Performance (Standard Errors in Parentheses)
or profit margin (Columns 3 and 4) as the performance
variable, our results are similar. The effect of turnover
Dependent variable: Dependent variable: for low-process conformance stores is negative and sig-
Customer service Profit margin
nificant. This negative effect is offset as we move to
Variable 1 2 3 4 stores with high levels of process conformance. Using
Percent of total turnover −3720∗ −0038∗∗ both monthly and three-month measures of turnover, the
2036 0018 effects for high-process conformance stores are signif-
Percent of total turnover 8052∗∗∗ 0062∗∗∗
∗ High turnover 3292 0021
icantly higher than those for low-process conformance
Percent of total turnover −4266∗∗∗ −0024∗∗∗
stores. The combined effects for the high- conformance
(3 months) 1224 001 group are not significantly different from zero in any
Percent of total turnover −0057 0022∗∗ of the four columns. For example, when using monthly
(3 months) 1774 0010 turnover, and examining the effect of turnover on cus-
∗ High turnover
Any manager turnover −0605 −0390 −0004 −0005∗
tomer service, the combined effect of turnover for
during period? 0729 0739 0003 0003 high-process conformance stores of 4.431 (=−7
777 +
Percent of full-time −1814 −1769 −0007 −0001 12
208) in Column 1 is not significantly different from
employees 1334 1401 0011 0010 zero. Note that we obtained very similar results when
Total payroll 0569∗∗∗ 0555∗∗∗ −0006∗∗∗ −0006∗∗∗ we divided stores into high-, medium-, and low-process
(in ten thousands) 0150 0160 0001 0001
Number of local 0589∗ 0425 −0028∗∗∗ −0027∗∗∗
conformance categories.
competitors 0311 0358 0003 0003 In Table 7, we report results in which we replace our
Percent of local −0037 0025 −0004∗∗∗ −0004∗∗∗ composite measure of process conformance with three
unemployment 0137 0146 0001 0001 separate measures of process conformance—high yearly
Store fixed effects? Yes Yes Yes Yes RPL scores, high yearly store conditions scores, and
Calendar month fixed Yes Yes Yes Yes high variation in RPL. The first two measures were cre-
effects?
Year fixed effects? Yes Yes Yes Yes
ated by ranking stores with respect to their yearly RPL
Observations 11,321 10,567 12,709 11,928
and yearly store condition scores and dividing them into
Adjusted R2 03922 03699 07893 07949 high- and low-RPL and high- and low-store conditions
categories, respectively. Regardless of whether we use
Notes. ∗ , ∗∗ , ∗∗∗ denote statistical significance at the 10%, 5%,
and 1% levels, respectively. Standard errors are heteroskedasticity
high-RPL or high-store conditions, we obtain similar
robust and clustered by store. Regressions include a constant term results. The effect of turnover on both customer service
not shown in the table. and profit margin for stores that are in low categories is
Ton and Huckman: Managing the Impact of Employee Turnover on Performance
Organization Science 19(1), pp. 56–68, © 2008 INFORMS 65
Table 6 Regressions Testing Process Conformance as a 1991). Our longitudinal analysis of Borders stores shows
Moderator of Turnover’s Effect on Customer Service that, on average, increased employee turnover is asso-
Score and Profit Margin (Standard Errors in ciated with decreased store performance, as measured
Parentheses)
by customer service score and profit margin. Consis-
Dependent variable: Dependent variable: tent with Price (1977), we also find that turnover has
Customer service Profit margin
a nonlinear effect on performance, with low-turnover
Variable 1 2 3 4 stores being more affected by turnover than their
Percent of total turnover −7777 ∗∗
−0031 ∗∗ high-turnover counterparts. At low-turnover stores, the
Additional information, including rights and permission policies, is available at http://journals.informs.org/.
INFORMS holds copyright to this article and distributed this copy as a courtesy to the author(s).
Table 7 Regressions Testing Various Measures of Process Conformance as Moderators of Turnover’s Effect on Customer
Service Score and Profit Margin (Standard Errors in Parentheses)
Notes. ∗ , ∗∗ , ∗∗∗ denote statistical significance at the 10%, 5%, and 1% levels, respectively. Standard errors are heteroskedasticity robust
and clustered by store. Regressions include a constant term not shown in the table.
single firm allows us to avoid having to control for het- and process conformance are correlated, the moderating
erogeneity across stores in ownership. We thus avoid effect of process conformance may be partly due to the
having to control for unobservable firm-level factors that involuntary nature of turnover at these stores. Although
are correlated with both performance and turnover. Fur- we do not have any reason to believe that there is more
thermore, our findings may be applied to other forms of involuntary turnover at high-process-conformance stores,
retailing as well as to other settings—such as call cen- we are not able to test this relationship with our data.
ters, hospitals, and food services—that operate with high Third, our data do not allow us to determine whether a
levels of employee turnover. According to the Bureau store’s level of process conformance is the result of man-
of Labor Statistics (2005a), employee turnover rates in agerial policy or employee will. For example, we are not
the retail industry and in the accommodation and food able to observe whether a store has low process confor-
services industries were 55.1% and 75.4%, respectively, mance because its manager decides not to enforce corpo-
in 2005.13 These turnover rates include turnover of all rate standards or because its frontline employees decide
employees (e.g., workers who perform daily operational not to act in accordance with mandates from the store
activities as well as managers and assistant managers). manager. Regardless of the underlying cause of low con-
Turnover rates for workers who perform operational formance, our study highlights the impact of its occur-
activities are likely to be even higher. At many fast-food rence on the effect that turnover has on performance.
chains, for example, employee turnover rates of 200% a Fourth, it is possible that there is an alternative expla-
year for hourly workers are common (White 2005). Fur- nation of our results. Specifically, employees at stores
thermore, workers performing daily operational activi- with low process conformance may be more skilled in
ties in these industries represent a significant portion of selling books and interacting with customers than they
the workforce in the United States.14 are in following standardized processes. Because these
Second, we cannot distinguish between voluntary and employees directly affect customer service, and thus
involuntary turnover. Borders managers maintain that sales and profit, their departure may disproportionately
most turnover at their stores is voluntary and, as a result, hurt operational performance at stores with low process
they do not systematically track the causes of turnover for conformance. This explanation is unlikely in our setting.
store employees. To the extent that involuntary turnover We find that stores with high process conformance have
Ton and Huckman: Managing the Impact of Employee Turnover on Performance
Organization Science 19(1), pp. 56–68, © 2008 INFORMS 67
higher average customer service scores than those with • Were facilities in the restroom in working order?
low process conformance (88.2 versus 86.6). This dif- • Were bookshelves fully stocked, without gaps or the back
ference of 1.6 points is statistically significant at the of shelves showing?
1% level.15 As a result, our setting does not appear to Examples of questions relating to customer service interac-
be characterized by a trade-off between customer ser- tions:
• Did any employee initiate verbal acknowledgment or
vice and process conformance. Furthermore, we have
greet you while you were in the store?
no reason to believe that, at stores with low process • Was at least one employee visible and approachable at
conformance, employees who leave are more customer the information desk?
Additional information, including rights and permission policies, is available at http://journals.informs.org/.
INFORMS holds copyright to this article and distributed this copy as a courtesy to the author(s).
service and sales oriented than are those employees • Did employees make additional item suggestions?
who stay. Although we are unable to observe individual 3
We note that the distinction between permanent and tem-
employees’ inclinations toward process conformance or porary employees is not the same as that between full-time
customer service, we expect some degree of similarity and part-time employees. Permanent employees are those who
across employees within a store, especially given that are employed for an unspecified duration; they are considered
hiring at a store is generally done by a single manager. either full time or part time based on the number of hours they
Finally, it is not obvious that our findings translate to work each week. Temporary employees are hired for a speci-
a more exploration-oriented setting in which there may fied period of time (e.g., one month). During that period, they
may work in either a full- or part-time basis.
exist a trade-off between process conformance and per- 4
We had access to monthly RPL scores. We were not able,
formance measures, such as customer service. In such however, to obtain quarterly store conditions scores. We were
settings, a push toward greater process conformance and able to collect data only on average store conditions scores for
standardization will likely limit process flexibility; the each year.
theoretical effects of this trade-off on overall store or 5
We also repeated our analysis using store manager turnover
firm performance are not clear. As such, our findings during the past three months. Although the magnitude of the
are most applicable to exploitation-oriented settings (i.e., store manager turnover coefficient as well as its significance
those settings in which the returns to having consistent changed slightly in different models, the direction remained
and replicable processes are greater than those to having the same. We do not report the results using this alternative
a high level of flexibility). measure of store manager turnover.
6
Managers at Borders consider Barnes & Noble and other Bor-
Despite these caveats, our study has important man-
ders stores in the area as the main competitors to a specific
agerial implications. To reduce the costs associated Borders store. Consequently, Borders tracks the opening and
with employee turnover, most recommend that managers closing of Barnes & Noble and other Borders stores near each
focus on reducing the level of turnover. Here we sug- existing Borders store.
gest that, in certain settings, managers may also be able 7
Unemployment data were obtained from the Bureau of Labor
to reduce the turnover’s effect (regardless of its absolute Statistics.
level) by focusing on process management and the cre- 8
In this setting, these unobserved, time-invariant aspects of a
ation of standard operating procedures. Our findings sug- store include factors such as physical size and layout.
9
gest that the current managerial and financial resources Variation in RPL scores was calculated using monthly RPL
dedicated to reducing the occurrence of turnover might scores for each store.
10
be redirected, at least in part, to improving the firm’s The standard deviation of three-month turnover using pooled
observations from all cross sections and time periods is 0.109
ability to manage turnover’s impact.
for full-time employee turnover, 0.242 for part-time turnover,
and 0.101 for total turnover. The standard deviation of three-
Acknowledgments month turnover within a store varies from 0.024 to 0.245 for
The authors thank Linda Argote and three anonymous review- full-time turnover, from 0.067 to 0.801 for part-time turnover,
ers for their thoughtful suggestions that helped improve this and from 0.033 to 0.27 for total employee turnover. In inter-
paper. The authors are grateful to Lee Fleming, Susan Kulp, preting our results, we use the standard deviation of turnover
Ananth Raman, Noel Watson, and Steven Wheelwright for within an average store.
providing feedback on earlier drafts. The authors also thank 11
In 1999, Borders Group’s net income was $90.3 million.
participants in the Technology and Operations Management Approximately 60% of the company’s sales came from Bor-
Seminar at Harvard Business School for helpful comments, ders stores. If we assume a similar share of the profits, Bor-
and they acknowledge support from the Division of Research ders stores contributed to about $54 million in profit. We note
and Faculty Development. that the 2.72% decrease in a store’s profit margin leads to a
larger decrease in corporate net income as the former does not
Endnotes include part of the selling, general, and administrative (SG&A)
1
For example, see Sasha Corporation (2007) for a list of stud- expenses (it includes store labor, which is reported under
ies that estimate the cost of turnover. SG&A), it also does not include other investment expenses or
2
Examples of questions relating to the customer service envi- taxes.
12
ronment: Note that the number of observations dropped significantly,
• Was area outside the door and around the display win- as we were able to obtain only three years of data for store
dows clean and neat? conditions.
Ton and Huckman: Managing the Impact of Employee Turnover on Performance
68 Organization Science 19(1), pp. 56–68, © 2008 INFORMS
13
The turnover rate is the number of total separations as a Hom, P., R. Griffeth. 1995. Employee Turnover. South-Western Pub-
percent of total employment in these industries. lishing, Cincinatti, OH.
14
According to the Bureau of Labor Statistics (2005b), retail Huselid, M. 1995. The impact of human resource management prac-
salespeople, cashiers, food preparation and serving workers tices on turnover, productivity, and corporate financial perfor-
(including waiters and waitresses), and cooks in fast-food mance. Acad. Management J. 38(3) 635–672.
chains accounted for 13.9 million employees, or about 10.6%
Jovanovic, B. 1979. Job matching and the theory of turnover. J. Polit-
of total employment in 2005. ical Econom. 87(5) 972–990.
15
We do not observe a difference in average profit margins
between high-conformance and low-conformance stores. Leana, C., H. Van Buren. 1999. Organizational social capital and
Additional information, including rights and permission policies, is available at http://journals.informs.org/.
INFORMS holds copyright to this article and distributed this copy as a courtesy to the author(s).
Dess, G., J. Shaw. 2001. Voluntary turnover, social capital, and orga- Steers, R., R. Mowday. 1981. Employee turnover and post-decision
nizational performance. Acad. Management Rev. 26(3) 446–456. accommodation processes. Res. Organ. Behav. 3 235–281.
Glebbeek, A., E. Bax. 2004. Is high employee turnover really harm- White, E. 2005. To keep employees, Domino’s decides it’s not all
ful? An empirical test using company records. Acad. Manage- about pay. Pizza chain attacks turnover by focusing on man-
ment J. 47(2) 277–286. agers. Wall Street Journal (February 17) A1.
Hausman, J., W. Taylor. 1981. Panel data and unobservable individual Winter, S., G. Szulanski. 2001. Replication as strategy. Organ. Sci.
effects. Econometrica 49(6) 1377–1398. 12(6) 730–743.