1、SEARCHING FOR THE OPTIMAL LEVEL OF EMPLOYEETURNOVER: A STUDY OF A LARGE U.K.RETAIL ORGANIZATIONW. STANLEY SIEBERTUniversity of BirminghamNIKOLAY ZUBANOVCPB Netherlands Bureau for Economic Policy AnalysisWe study the relationship between sales assistant turnover and labor productivity in325 stores of
2、 a large U.K. clothing retailer tracked over 199599. We find that theturnover-productivity relationship is contingent on type of work system. For a largegroup of part-timers, managed under a “secondary” work system, the relationshipclearly has an inverted U-shape, but for the smaller group of full-t
3、imers, managedunder a “commitment” system, the relationship is the conventional negative one.Implications for the contingency view of the link between turnover and productivityare discussed.Research on the link between employee turnoverand organizational performance has long been abattlefield of con
4、flicting views. Thus, Dess andShaw (2001) contrasted “human capital” theory,predicting losses in performance as turnover erodesfirm-specific human capital, with cost-benefit ap-proaches predicting an optimal level of turnovermaximizing the difference between its benefits andcosts. A particular benef
5、it of turnover is replace-ment of poorly performing workers with better jobmatches (Ableson by Kacmar, Andrews,Van Rooy, Steilberg, and Cerrone (2006), of restau-rants in the Burger King chain; by Morrow andMcElroy (2007), of bank branches; and by Shaw,Duffy et al. (2005), also of restaurants. Admit
6、tedly,there is the disadvantage that a single organizationis not representative. Nevertheless, lessons can belearned, so long as its relevant characteristics aremade clear. Basically, we believe that our resultswill apply to most situations requiring manage-ment of many unskilled workers over disper
7、sedsites.We conducted our analysis in the context of thework systems for sales assistants in a large U.K.retail organization, whose stores we tracked over199599. We leave aside the more skilled store andcentral management groups, as well as central ad-ministration and warehousing. For these groups,t
8、he inverted U-shaped turnover-performance asso-ciation is likely to be ruled out because substantialhuman capital is lost with turnover; a considerableliterature on such losses exists (Arthur, 1994; Batt,2002; Huselid, 1995; Shaw, Duffy, et al., 2005;Shaw, Gupta et al., 2005). But for sales assistan
9、tsthe position may be different. Their turnover ishigh, with an average headcount separation rate ofabout 25 percent a year in our organization. Thejob might be thought of as low skilled, with lowinvolvement, and so some level of employeeturnover, and perhaps quite a lot, may be appro-priate. Theref
10、ore, here we would expect to findan inverted U-shaped turnover-performancerelationship.But even within a seemingly homogeneous em-ployee group such as sales assistants, differentwork systems may be necessary. We needed to bealert to this possibility, following Lepak and Snells(1999: 42) questioning
11、of the “one-size-fits-all” ap-proach to analyzing work systems. In fact, as wewill show below, two work systems appear to co-exist for sales assistants within our organization: a“commitment” system for the core full-time group(20 percent of the workforce), and a “secondary”system for the many part-t
12、imers. For the core work-ers, we expected a straight negative turnover-performance link, and we expected the inverted Uto apply only to the secondary group. Some evi-dence for this contingency view exists, in thatboth Arthur (1994: 683) and Guthrie (2001: 186)found that turnover-performance relation
13、shipswere negative only for organizations using “high-involvement” work systems (the commitment sys-tem in our terms) and were positive for organiza-tions with low-involvement work systems (oursecondary work system) (but see Shaw, Gupta et al.2005: 64 for a no-contingency finding). The citedstudies
14、were conducted in multiple organizations;ours was within one organization. We thus carrythe contingency debate further (see Datta, Guthrie,Lepak, Taylor, Tekleab, Marrone, Youndt, Snell, Dean, see below for man-agement methods for inducing “voluntary” separa-tions), and on his/her contracted hours.
15、These dataallowed us to analyze turnover not simply in termsof headcount, as is the norm, but with adjustmentfor full-time equivalentsthat is, hours gained orlost from hirings or separations. In fact, as emergesbelow, most turnover occurred among the short-hours, part-time workers. There are many pa
16、rt-tim-ers in our studied organization, but not many hours2009 295Siebert and Zubanovare lost when one leaves. Thus, the adjusted turn-over rates are below the headcount measures. Weshow that the performance-turnover link is muchclearer when turnover rates are measured in theirfull-time equivalents,
17、 a reasonable conventiongiven that output primarily depends on workerhours, not headcounts.The second innovation our approach offers wasderived from taking advantage of the panel struc-ture of our data, which allowed every store to betracked over five years. Having five annual obser-vations for each
18、 store meant that we could includein the regression equation a term representingstore-specific unobservables, or “omitted vari-ables” (Becker 17 percentworked fewer than five hours a week. Their careeraspirations lie elsewhere, which is why their head-count turnover rate is much higher than that of
19、thesales force as a whole (21 percent per year). Natu-rally, this group receives less responsibility, lessspecialist training, and fewer promotion opportu-nities. Their pay is flat and is determined by salarysurveys of similar occupations in the county (thatis, by the external labor market). Althoug
20、h not for-mally on temporary contractsonly 5 percent ofthis organizations employees are on such con-tractsthey are temporary workers in all but name.Thus, following Bamberger and Meshoulams(2000) categorization, we identify this group asmanaged under a secondary work system. Giventhe part-time natur
21、e of employment in our organi-zation, we propose that sales assistants workingfewer than 30 hours per week belong to this group.These part-timers constitute 80 percent of all thefirms sales assistants.At the other end of the spectrum, our organiza-tion contains sales assistants working regular full-
22、time hours (30 or more per week). These workersform the core of every store. They are the mostresponsible and committed and are less likely toseparate than the part-timers. In addition to per-forming the basic functions outlined above, theyopen and close the stores in which they work,reconcile cash
23、with receipts (“cash up”), help in-duct and train newcomers, and generally act ascommunication nodes, for example sometimes con-ducting team briefings instead of the manager.They receive specialist training in administration,wage control, and health and safety. The processthrough which they are sele
24、cted is tougher: theyearly intake of full-timers is lower than that ofpart-timers and, of part-timers with a given tenure,2009 297Siebert and Zubanovsome, but not all, become full-timers. It is thisgroup that forms the pool from which future storemanagers are generally drawn, particularly for thesma
25、ller stores (65 percent of store managers derivefrom internal promotions, according to our inter-views). This possibility of promotion provides animplicit link of pay to performance. Therefore, thisgroup fits the description of being managed under acommitment work system.Our allocation of sales assi
26、stants into two differ-ent work systemsthe basis of our contingencyview of the turnover-performance linkis testable.It may be objected that sales assistants, even themost committed ones, are not sufficiently skilled tobe managed under a true commitment system (theyare far from Lepak and Snells 1999:
27、 36 Intel en-gineers). Yet differentiation makes sense if the full-timers job is more strategic, thereby requiringdisproportionately greater investment (Becker they are quite small (containingon average about seven full-time-equivalent work-ers); and they sell a similar range of goods distrib-uted f
28、rom the companys warehouses. In these cir-cumstances, the company has naturally adoptedcentralized management.The wage policy is quite simple. Central manage-ment sets out a standard hourly wage schedule forsales assistants, determined every year, for allstores. During our period of study, this sche
29、dule setwages that differed by region and age. Four regiongroups (rural, town, city, and central London) andtwo age groups (18 or younger, and older than 18)are specified. There are no wage increments forseniority and no immediate pecuniary rewards forperformance for sales assistants. Training varie
30、s bygroup (see above) but is basic for the majority ofsales assistants.The companys management rule, “right people,right time,” is a good summary of its personnelpolicies with respect to both part- and full-timers.Within the constraints of central policies, storemanagers select workers, negotiate ho
31、urs, and reg-ulate turnover. By hiring workers to make up forseparations, store managers match the workforce tothe seasonal pattern in demand and also to yearlyfluctuations, which are quite marked (see below).In addition, the managers negotiate contract hours,maintaining a pool of part-timers to be
32、employed atthe “right time,” and can if necessary pressure aworker to leave by reducing hoursa useful alter-native to terminating contracts, which is difficultunder U.K. employment law. Most separations are“voluntary,” therefore. In these circumstances,high employee turnover among part-timers is to
33、beexpected (Batt, Colvin, Shaw et al.,1998), particularly if wages are set relatively low, asis the case. This strategy is required by the turbu-lence of the retail environment, which we delineatein the next subsection. On the other hand, coreworkers whom managers particularly want to keepare given
34、longer hours and more training. A com-mitment approach is required for managing thisgroup of “right people” to ensure coherent commu-nications and a supply of management talent.The retail environment leads to both seasonaland year-to-year fluctuations in sales, employment,hirings, and separation tha
35、t are worth illustrating.Figure 1 illustrates the seasonal turbulence affect-ing the average store (though we emphasise that weused only annual data for our empirical work). Wesee a relatively steady headcount separation rate ofaround half a person per month. However, peaks inthe hiring rate occur a
36、t the beginning of summer,and then in September-October, which is the sea-son for childrens clothing sales. Store managershave to try to even out labor productivity over theyear, so as to avoid having idle staff on the onehand, and overlong customer queues on the other.Store managers also have to me
37、et year-to-yearfluctuations in demand. Figure 2 shows how diffi-cult this problem can be, graphing large changes inlabor input and in sales for many stores during199599. In fact, around one-fifth of the stores ex-panded or contracted hours worked by more than50 percent over our study period. Sales w
38、ere almostas changeable. Further, the link between changes inlabor inputs 199599 and changes in sales is weak,correlated at only .26. Therefore, labor productivitywas difficult to predict, and store managers had towork hard to provide the appropriate labor inputsto service their customers. In such a
39、 turbulent en-FIGURE 1Hires and Separations in an Average Store by Month and Headcount1.25 1.05 0.70 0.52 0.28 0.76 0.28 0.30 0.47 0.59 0.51 0.52 0.70 0.50 0.38 0.42 0.46 0.53 0.65 0.60 0.60 0.50 0.40 0.36 0 0. 2 0. 4 0. 6 0. 8 1 1. 2 1. 4 Jan. Feb. Mar.Apr. May JuneMonths Person s Hires Separations
40、 July Aug. Sept. Oct. Nov . Dec.2009 299Siebert and Zubanovvironment, it is as important to downsize as upsize,so a certain amount of employee turnover, at leastamong part-timers, has clear advantages. The ques-tion is, What is the optimum level of suchturnover?DataOur data set contains detailed inf
41、ormation onemployee turnover and personnel characteristics. Itconsists of three parts. The first part is based onpersonnel records of all the employees who workedat any time between 1995 and 1999. Altogether,there are 72,669 records of 32,546 individuals. In-dividual records include age, gender, dat
42、e hired,date left employment, and weekly contract hours,our measure of hours worked. We thus could ob-serve job matches as short as one day and accu-rately adjust our measures of turnover to their full-time equivalents. Since individual productivitydata were not available, we derived employee av-era
43、ge data for each store in order to compute storeannual average productivity.Then, to the employee information we matchedthe second part of our data, which contains storeinformation, including the output variable, real an-nual sales per store, store square footage (our cap-ital variable), and other p
44、otentially relevant char-acteristics (see below). The third part of the dataconsists of areawide wage and unemployment datafor the county in which each store was situated.These data were obtained from the UK LabourForce Survey, which is based on samples of indi-viduals 16 to 65 years of age in 63 U.
45、K. counties. Toavoid dealing with store entry and exit events,which might present special effects of downsizingon performance (see McElroy, Morrow, Shaw, Gupta et al., 2005).Turnover variables. Our measure of turnover isthe separation rate, as is common (see, for example,Glebbeek and Bax 2004 and Sh
46、aw, Duffy et al.FIGURE 2Changes in Store Employment and Sales, 199599a1.50.511 .5 0 .5 1Proportionate Change in Store Annual Hours, 199599Proportionate Changein Store Annual Sales,199599ar H11005 .26, p H11021 .001, 325 stores.300 AprilAcademy of Management Journal2005b; see Morrow and McElroy 2007
47、andShaw, Gupta et al. 2005 for studies using the quitrate alone). We calculated separations from em-ployee records for the sales assistants workingfewer than 30 hours per week (defined as part-timeseparations) and 30 or more hours per week (de-fined as full-time separations), as described in theTheo
48、ry section above. (We also conducted a sensi-tivity test, using 20 hours a week as the cut-off; seebelow.) Although the records do not reveal thereason for worker separation, our measure of sepa-rations included few terminations because it iscompany policy not to terminate.We calculated the full-tim
49、e equivalent separationrate as the number of hours that leavers would haveworked had they not left, relative to annual totalhours worked. (Here we used our daily data onwhen workers left and were hired; but we mustemphasize that all our variables are in annualterms, so seasonality is not an issue.) Hence, ifsomeone separated at the very end of a year, theimpact of his/her leaving on labor productivity inthat year would be zero, and the converse wouldapply if he or she left at the beginning of the year.Table 1 gives an example of our calculation. Thus,our
