MATH GURU ONLY

Where the Grass Is Greener: Voluntary Turnover and Wage Premiums

Voluntary Turnover and Wage Premiums

MARLENE KIM*

Standard economic and compensation theories suggest that voluntary turnover should decline when a firm pays wages that are higher than those of its competi- tors. Turnover behavior in the State of California’s Civil Service, however, does not support this prediction. Using a fixed-effects estimator to control for job-specific characteristics, I find that the wages California pays relative to those of its competitors has little or no effect on turnover. In addition, estimates of the elasticity of turnover with respect to alternative wages indicate that higher wage rates do not pay for themselves through lower turnover costs. Instead, the absolute wage level and wage growth have large effects. In other words, it appears that workers are less likely to quit jobs that pay high wages and have larger wage increases no matter how their wages compare with those paid by other employers.

Introduction

A number of theories suggest that voluntary turnover should decline when a firm raises its wages relative to what others are paying. Standard competitive economic theory predicts that if employers fail to match the market wage rate, they will lose their workers to higher-paying competi- tors. Because they believe that the wages paid for similar jobs outside their firm influence employees’ turnover decisions, human resource managers rely on wage surveys to keep their wages competitive (Mobley, 1982; Milkovich and Newman, 1996).

584

* Department of Labor Studies and Employment Relations, School of Management and Labor Rela- tions, Rutgers University. David Card, Bill Dickens, Doug Kruse, Mark Killingsworth, and Cathy Weinberger provided exceptionally helpful comments.

INDUSTRIAL RELATIONS, Vol. 38, No. 4 (October 1999). © 1999 Regents of the University of California Published by Blackwell Publishers, 350 Main Street, Malden, MA 02148, USA, and 108 Cowley Road,

Oxford, OX4 1JF, UK.

Job search and efficiency wage theories formalize the relationship between turnover and comparative wage rates. Models of employed job search assume that workers will switch jobs if the expected discounted lifetime earnings in an alternative job are greater than the expected discounted lifetime earnings in the present job plus the disutility and costs of changing jobs. At the margin, these models imply that the num- ber of quits will increase if the alternative wage rate increases relative to workers’ current expected wage streams (see Burdett, 1978; Weiss, 1984).

In addition, the turnover version of efficiency wage theory suggests that firms with high turnover costs can pay wage premiums in order to lower voluntary turnover (Salop, 1979; Summers and Bulow, 1986). If such firms bear part of the turnover costs, firms can maximize profits by increasing wages, since higher costs in wages will be offset by lower turn- over costs.1 Firms can therefore set wages to economize on turnover costs, with firms having higher turnover costs paying wage premiums (see Salop, 1979).

These theories imply that the wage a firm pays relative to its competi- tors has a direct effect on turnover and that for some firms this relation- ship may be cost-effective. Empirically, therefore, we should observe a strong negative relationship between relative wages (the wage a firm pays relative to its competitors’ wages for the same job) and voluntary turnover. However, the empirical findings are mixed. Although there is some indication that relative pay among industries and individuals is a factor in determining variations in turnover (Mobley, 1982; Bartel, 1982; Antel, 1988), evidence at the firm level suggests that this is not cost-effective.

For example, Leonard (1987) finds a negative correlation between aver- age wages and turnover among a cross section of 200 high-technology firms; however, the wage premiums are too costly to justify savings from lower turnover. Powell et al. (1994) reach similar conclusions after exam- ining the wage structure of 205 child care centers. Campbell’s (1993) study of more than 5000 firms’ most recent hires also confirms that turn- over costs are not sufficient by themselves to account for wages that exceed market-clearing levels.2

This article builds on this firm-level research by examining turnover in the State of California’s Civil Service (California or California Service).

Voluntary Turnover and Wage Premiums/ 585

1 Turnover costs include direct hiring and training costs and lost output during this time period. 2 Campbell attributes the cause of his weak findings to long-term turnover strategies that may not show

up in his cross-sectional study.

The California Service is used as a case study because its data allow me to examine a number of different issues that previous studies were unable to address. First, because the data are a panel by occupation, I am able to control for constant job-specific characteristics that affect turnover. It is important to control for job characteristics, since much of turnover results from the nature of the job itself (Slichter, 1921; Telly et al., 1971). Previ- ous analyses at the firm level (Campbell, 1993; Leonard, 1987) included controls for job characteristics (such as blue- and white-collar composi- tion of firms and shares of occupation and industry categories); however, these may not have adequately controlled for much of the variation in turnover among jobs within firms. Their failure to find turnover cost-effective may have resulted from using average turnover rates within firms, which would have masked much of the intrafirm variation in turn- over at the job level.

Second, because the data include California’s own wage surveys, which estimate what its labor competitors pay for the same occupations, I am able to control for exact measures of market wages rather than esti- mate them by using wage regressions, as does Campbell (1993).3 These data allow me to examine how the wages paid by California relative to what its labor market competitors pay (calledrelative wages) affect turn- over rates within an occupation.

Finally, I am able to examine separately the effects of relative wages and absolute wages on turnover. In other words, I am able to examine whether turnover responds to wage levels that increase relative to what other firms are paying, or whether turnover responds merely to higher wages—regardless of how they compare with other firms’ pay. The next section describes a simple turnover model and the hypothesized effects of wages and labor market conditions. The following section describes the data set used to test the model, with the results contained in the final section.

A Turnover Model of Efficiency Wages

Many models of turnover assume that higher wage rates reduce turn- over (see, for example, Flacco and Zeager, 1989), especially when com- pared with the alternative wages available (Campbell, 1993; Salop, 1979;

586 / MARLENE KIM

3 Because predicted wages estimated by wage regressions may not adequately measure one’s true alter- native wage rate, measurement error could bias the coefficients on the alternative wage rates downward, resulting in Campbell’s (1993) estimates being insignificantly different from zero. In addition, using the same regressors to estimate quit rates and alternative wage rates may have biased the coefficients on the alternative wages downward as well.

Stiglitz, 1974).4 The model I describe is a simplified version of Stiglitz (1974).5 Following Stiglitz (1974), assume that firms produce outputQ using a production process that employs capitalK and laborL. The production process is described by a production function

Q = F(K, L)

whereFL > 0 andFLL < 0.

For simplicity, assume that labor is homogeneous and that the capital stock is fixed. Labor costs include training and hiring costs as well as wages. Training and hiring costsT include the direct costs of training and hiring per worker, such as recruitment, screening, testing, and orientation, plus indirect costs, such as lost output when current workers take time off from their responsibilities to help newer workers, the costs of equipment that is broken, and lost output for new workers while productivity is lower. For simplicity, assume thatT is constant. Total training costs will depend on the turnover rate of workers, which in turn will depend on the wage inside the firmw vis-à-vis the wage rate paid outside the firmwa. Turnover also will depend on the unemployment rateu, which serves as a proxy for labor market slack. Workers are less likely to quit if jobs are scarce, since their chances of obtaining alternative employment are smaller (Salop, 1979). The quit rateq is therefore defined as

q = q(w, wa, u) (1)

whereqwa > 0, qw < 0, andqu < 0.

As the wage rate paid by the firm increases, turnover should decline. Turnover also should decline when the unemployment rate increases, but it should increase when the wages paid by other employers increase.

Voluntary Turnover and Wage Premiums/ 587

4 There are slight differences among these models; however, most of the general arguments are the same. Campbell includes the firm’s wage, alternative wage, and unemployment rate; however, the wages are separate arguments instead of a ratio. Salop specifies quits as a function of the ratio of the firm’s wage to a general measure of labor market tightness, such as the average wage and nonpecuniary benefits adjusted for the probability of receiving a job.

5 Stiglitz (1974) has a more complex model, since he estimates rural and urban wage rates and mobility between these two sectors. In this article, I reduced this model to only one sector.

Total labor costs are the following:

w* L = wL + qTL (2)

wherew* is the total labor cost per employee. Setting the price of output to unity, the firm’s profits are

F(K, L) − wL − TqL (3)

If the firm chooses its employment level to maximize profits,

FL = w + Tq (4)

The marginal product of labor will equal the wage plus training costs. If this wagew exceeds the wage that occurs when labor supply equals labor demand, unemployment will result.

For a givenL, the firm will minimize the cost per employee:

min(w + qT) (5)

The firm will minimize wage and training costs, taking the unemployment rate and other firms’ wages as given. The first-order condition yields

1 + Tqw = 0 (6a)

or

Tqw = −1 (6b)

The extra wage costs will equal the marginal savings in turnover costs.6 In other words, a profit-maximizing firm will continue to increase wages

588 / MARLENE KIM

6 Allowing for intertemporal profit maximization yields similar results (see Stiglitz, 1974).

until the last dollar spent is equal to the savings in turnover costs. This result is standard for all firm turnover models (see, for example, Campbell, 1993). In addition, the greater the training and hiring costsT, the more likely it is that the efficiency wagew will exceed the market- clearing wage rate, resulting in involuntary unemployment.

The Data

To examine the relationship between turnover and relative wages, I use unique data from the State of California’s Civil Service. The California Service collects market wage data through annual salary surveys. These data can be matched to California’s own wage scales and turnover data to obtain estimates of wages, market wages, and turnover rates by occupa- tion. California’s salary surveys are one of the most sophisticated surveys conducted. The Bureau of Labor Statistics established the survey method- ology and until the 1980s oversaw the data collection. Currently, the sur- vey includes a stratified (by firm size) random sample of 1000 firms in the State of California. The Bureau of Labor Statistics continues to collect data on trades and clerical jobs for the California Service during its area wage surveys, while California collects data on professional and technical occupations. Since the 1980s, a separate nonprofit agency works with businesses and government representatives in the state to continue these surveys, using the same methodology as before. An entire year is devoted to collecting the data, with analysts personally visiting firms and firms using job descriptions to match their jobs to those surveyed. The survey data estimate market wages for approximately 40 of California’s 4000 detailed occupations.7,8 Although proportionately few occupations are surveyed, the California Service surveys those which it believes are most affected by external market forces. The surveyed occupations are common to many other firms and require less specialized (i.e., govern- ment-only) or firm-specific knowledge. Therefore, these occupations are the ones most likely to be influenced by labor market competition.9 The

Voluntary Turnover and Wage Premiums/ 589

7 The number of occupations surveyed varies by year. In addition, in some years the sample sizes for various occupations are too small to compute market averages; thus the actual number of occupations with valid market data varies from year to year.

8 Occupationanddetailed occupationare used interchangeably. They refer to the California Service’s own detailed occupations, or what are commonly termedjob titles.

9 Like many government agencies, the California Service suffers from job title proliferation. Approxi- mately one-fourth of its job titles contain only one person in them; many are specific to government or unique to the California Service. Thus many of the 4000 job titles have no counterpart in other firms that can be surveyed.

results obtained in this study therefore willoverestimatethe effect of rela- tive wages on turnover over all occupations.

For each occupation surveyed, the market data are summarized into four measures—the median, average, first, and third quartiles. The Cali- fornia Service conducts these wage surveys because it believes that recruitment and retention will suffer if it fails to keep up with the market. Thus, examining the effects of relative wages on turnover will confirm the extent to which retention does suffer.

There are two measures of wages that California pays. The public wage scales contain the monthly entry (minimum) rate and the top (maximum) rate paid to each occupation (unfortunately, the average wage paid is not available). For this analysis, I also calculated the midpoint rate, the simple average of the minimum and maximum rates for the occupation. The wages used exclude benefit costs; however, because the same bene- fits are offered to all employees and remain constant over the time period examined, omitting benefit costs should not affect the results.10

Data for voluntary turnover in California are from public records of voluntary turnover per 100 workers by occupation from July through December 1969, 1970, 1974, and 1975 and from January through June 1970, 1971, 1972, 1974, and 1982–1987.11 Voluntary turnoversare defined as voluntary resignations from temporary or permanent positions or absences without leaves; retirements are excluded.

The data used in this analysis were constructed as follows: For each occupation surveyed, the market wage rates and California’s wage rates were collected, as well as the appropriate turnover data. If the occupation was surveyed in October, the July through December turnover rate was collected. If the occupation was surveyed in July, the January through July turnover rate was collected. The total data set includes 494 observa- tions from approximately 40 occupations over 14 time periods.12

Although these data allow for excellent measures of market wages and turnover for a sample of occupations, they omit other factors. Work experience, recent job history (such as recent promotions and quits), and demographic variables, all of which are associated with turnover (Antel, 1988; Solnick, 1988; Weiss, 1984; Bartel, 1982; Van den Berg, 1992), are not available by occupation in California and are therefore excluded from

590 / MARLENE KIM

10 Benefits for the private sector may fluctuate, however. The results should be interpreted with this caution.

11 Data more recent than 1987 were not used because they did not match the time frame of the salary sur- veys, and I did not want to contaminate the estimates with measurement error.

12 Because of some missing data, and because the occupations surveyed changed during these years, the total number of observations is 494.

the analysis. In addition, it is possible that the voluntary turnover rates may include some “voluntary resignations” that may have been one step away from “involuntary separations.”13

Finally, the State of California’s Civil Service may not be a typical enterprise. Because it is public, it may not face the same cost-minimization constraints as private firms.14 Yet the California Service believes that turn- over will result if it fails to keep up with the market, and it collects the surveyed wage data and sets its wages with this in mind. It surveys those occupations which it believes to be the most susceptible to competitive market conditions, and it uses surveyed wage measures that it estimates to be market wage rates. Thus the results can address the effectiveness of California’s own wage policy and whether the California Service suffers higher turnover should it fail to pay market wages. In addition, although the California Service does not consciously follow an efficiency wage strategy, it is still instructive to examine whether paying wages that are higher than the market leads to cost savings that justify wage premiums.15

Although only one enterprise is examined, it is certainly a major one. The California State Civil Service is one of the largest public employers, with 150,000 employees, of which 120,000 work full time. Between January and June 1987, there were more than 8000 voluntary turnovers, of which 6000 were full-time workers. Table 1 displays voluntary turn- over rates for the sample of occupations in 1987. As the table indicates, voluntary turnover rates vary considerably by occupation. In addition, the wage rates paid to each occupation relative to what California claims the market pays vary over time for each occupation and among occupations as well.

A closer examination of the data indicates that wage rates in the California Service often deviate from market wage rates, and these departures are often permanent. In other words, wage rates do not

Voluntary Turnover and Wage Premiums/ 591

13 On the demand side, these “hidden terminations” may increase with slack labor markets, if employers are more willing to reduce their workforce during hard times when labor is relatively more plentiful. If this is true, turnover may increase with the unemployment rate, offsetting the tendency for “true” voluntary turnover to decline during slack markets. This could explain the insignificant result I obtain on unemploy- ment in the first regressions. On the other hand, theoretically, behavior on the labor supply side may have the opposite effect: “Hidden terminations” may decline as unemployment increases, since workers may decrease shirking or otherwise increase productivity in order to keep their jobs. This behavior would be consistent with a negative coefficient on the unemployment rate.

14 Cost minimization is not essential in order to examine the turnover behavior of individuals and esti- mate turnover elasticities with respect to wages, however.

15 This is standard. Previous empirical analyses of the turnover version of efficiency wage theory also examine firms and industries that do not necessarily follow an efficiency wage strategy (Leonard, 1987; Campbell, 1993; Powell et al., 1994).

592 / MARLENE KIM

TABLE 1

VOLUNTARY TURNOVER RATES PER100 WORKERS FOR THE FIRST 6 MONTHS OF1987

California Service Job Title Voluntary Turnover (Per 100 Workers) Relative Wagea

Physical therapist I 11.56 89.97 Security guard 11.29 98.00 Key data operator 9.75 98.50 Food service worker, level B 9.55 86.89 Assistant engineer (various) 9.09 89.54 Electrician, level A 8.93 76.85 Truck driver, level A, B 8.33 75.24 Office assistant II (general), level A 8.06 116.03 Psychologist, clinical 7.79 82.97 Janitor, level A, B 7.69 85.72 Office assistant II (typing), level A, B 7.55 98.32 Delineator 7.08 108.46 Accounting clerk II, level A, B 7.04 99.65 Word processing technician 7.04 87.31 Buyer I, II 6.25 88.82 Laborer 6.25 77.17 Painter, level A 5.97 81.56 Carpenter I, level A 5.75 75.51 Secretary 5.65 105.08 Associate engineer (various) 5.26 90.73 Plumber I, level A 4.23 77.00 Programmer II 4.77 86.10 Laundry worker, level B 4.50 108.41 Clinical laboratory technician 3.70 92.59 Stock clerk, level A 2.92 106.45 Associate management analyst 2.83 105.43 X-ray technician 2.78 87.91 Computer operator, level B 2.58 90.93 Pharmacist I 2.54 96.16 Associate budget analyst 2.48 98.95 Stationary engineer, level A, B 2.00 107.09 Accountant trainee 2.00 91.56 Licensed vocational nurse 2.00 87.52 Associate personnel analyst 1.94 101.96 Auto mechanic 1.47 76.65 Mechanical helper 0.00 78.60

NOTE: Voluntary turnover includes voluntary resignations or absences without leaves. Retirements are excluded. aMidpoint of California Service’s salary range [(entry level + top level)/2] divided by median of the surveyed wage rate times 100.

SOURCE: California State Personnel Board.

necessarily converge with the market wages.16 Given the numerous re- sources the California Service expended to estimate market wage rates, it appears odd that it would not follow them more faithfully. But the Cal- ifornia Service, like many employers, followed multiple compensation goals. Its explicit compensation policy states that besides paying market wage rates, it also aims to maintain internal equity in its pay system, i.e., paying jobs it evaluates to require greater skill higher pay (California State Personnel Board, 1975). Because the California Service believed that changing the relationships dictated by internal equity would invite grievances and cause morale problems (Crain, 1986; Atwood, 1986), it chose not to follow the market “in lock step” if doing so would disrupt the internal salary structure (California State Personnel Board, 1975). The result is that its wage rates often diverge from the market wages. It is because of the existence of these deviations that we can examine whether turnover responds according to the predictions of economic theory.

The Empirical Model

I used the following regression to examine whether turnover declines if the wage the California Service pays increases relative to the wages paid by other firms for the same occupation. Because the data are a panel, I was able to control for job-specific characteristics that determine turnover by using a fixed effects (dummy variable) estimator:

Turnoveri,t = αi + α2 ln(survey)i,t + α3 ln(wage)i,t + α4 (pchwage)i,t + α5 ln(unt) + ei,t (7)

where turnoveri,t is the voluntary turnover rate per 100 workers for theith occupation in California in timet, surveyi,t is the surveyed wage rate for the ith occupation in timet, wagei,t is California’s wage rate for theith occupation in timet, pchwagei,t is the percentage change in California’s

Voluntary Turnover and Wage Premiums/ 593

16 The following regression,∆ ln wagei,t = β1 + β2 ∆ ln marketi,t + β3 ∆ ln marketi,t−1 + β4 ln(wagei,t − marketi,t) + ui,t, showsβ2 = 0.5,β3 = 0.1, andβ4 = −0.0001, where wagei,t is the wage in the California Service for occupationi at timet, and marketi,t is the market wage rate (estimated by the salary survey) for occupationi at timet. Thus it appears that wage rates change according to changes in market wage rates, but they do so incompletely. Moreover, larger gaps between the market and existing wage rates are not remedied with larger wage increases, so wages that deviate from the market do not necessarily catch up with the market wage rates.

wage rate (in constant dollars) for theith occupation in timet17, unt is the unemployment rate in California in timet, αi is the fixed effect for each occupation surveyed in California (estimated as a dummy variable), and ei,t is the error term.

The unemployment rate was included to control for the cyclic nature of voluntary turnover due to relatively slack or tight labor markets. The coefficient of this variable is expected to be negative. The dummy vari- ables for each occupationαi will control for constant occupation-specific attributes such as bad working conditions that simply make some jobs more desirable than others (Viscusi, 1979; Bartel, 1982). I am assuming that these characteristics do not change over the 18-year time period.18

The coefficient on wagesα3 is expected to be less than zero; the higher the wage rate in California, holding the surveyed wage rate constant, the lower is the turnover rate in California. Likewise, the higher the surveyed wage rate, the higher is the turnover rate, holding California’s wage rate constant. Thus I expectα2 > 0.

The change in wage variable is included because workers may be less likely to quit their jobs if they recently were awarded real wage gains, especially if they form expectations about future wage growth based on recent increases (Solnick, 1988). They also may be more likely to quit their jobs if they receive lower wage increases than they are accustomed to receiving. In addition, if information about alternative wage rates is imperfect, greater nominal wage increases may fool workers into thinking that they are receiving wage rates that are higher than the market, and it may take time for workers to learn how their wages compare with those of similar employees.

I used four different measures of wages to see if the results were sensi- tive to the measurements used. Following the California Service’s own convention, I compared the first quartile of the surveyed rates with the entry-level rate in California, the third quartile of the surveyed rates with the maximum rate in California, and the median and mean surveyed rates with the midpoint of California’s rates.19 Table 2 displays the means of all the variables used.

594 / MARLENE KIM

17 [(Wagei,t − wagei,t-1/wagei,t−1)] × 100 18 This assumption is plausible given how I constructed the data set. Many occupations indeed changed

during the 18-year span. However, because of job descriptions and analysts’ detailed notes regarding the occupations’ evolution, I was able to determine if the changes significantly affected the job. If an occupa- tion changed significantly, I simply started a “new” occupation and discontinued the old one. Because of such occupation changes and changes in which occupations were surveyed, few occupations were present during the entire 18-year period.

19 California compared the first quartile with the entry rate, the third quartile with the maximum rate, and the median and mean surveyed rates with its weighted average.

To verify that the fixed-effects model was appropriate rather than sepa- rate regressions by occupation, the regression was initially run separately for each occupation to examine if the coefficients varied by occupation (not shown). Generally, the findings were consistent across occupations, indicating that turnover by occupation varies as a shift factor and not through interactions with the independent variables. This confirms that the fixed-effects model is appropriate.20 Because initial estimates indicated the

Voluntary Turnover and Wage Premiums/ 595

TABLE 2

VARIABLES AND DEFINITIONS

Variable Definition Mean Standard Deviation

Quitrates Six-month quitrate per 100 employees 2.56 7.15 Unemployment rate Unemployment rate in California 7.62 1.49 Relative wages

Midpoint/ median

Midpoint of Calif.’s wages/median of surveyed wages) * 100

91.37 9.51

Midpoint/ average

Midpoint of Calif.’s wages/average of surveyed wages) * 100

90.65 8.79

Minimum/ 1st quartile

Minimum of Calif.’s wages/1st quartile of surveyed wages) * 100

89.80 9.84

Maximum/ 3d quartile

Maximum of Calif.’s wages/3d quartile of surveyed wages) * 100

92.11 9.82

Wage levels minimum Minimum of Calif.’s monthly wagesa 1719.6 464.81 maximum Maximum of Calif.’s monthly wagesa 2056.1 584.11 midpoint Midpoint of Calif.’s monthly wagesa 1887.8 520.30 ln (minimum) Natural log of the minimum of Calif.’s monthly wagesa 7.40 0.27 ln (maximum) Natural log of the maximum of Calif.’s monthly wagesa 7.58 0.27 ln (midpoint) Natural log of the midpoint of Calif.’s monthly wagesa 7.49 0.27

Surveyed wages Median Median of surveyed wagesa 2084.56 551.60 Average Average of surveyed wagesa 2149.98 605.53 1st quartile First quartile of surveyed wagesa 1944.34 524.40 3d quartile Third quartile of surveyed wagesa 2241.12 589.25 ln (median) Natural log of median of surveyed wagesa 7.61 0.28 ln (average) Natural log of average of surveyed wagesa 7.63 0.29 ln (1st quartile) Natural log of 1st quartile of surveyed wagesa 7.53 0.28 ln (3d quartile) Natural log of 3d quartile of surveyed wagesa 7.68 0.27 Pchmid Percent change in midpoint of Calif.’s salary

from t − 1 to ta −0.36 6.59

Pchmin Percent change in minimum of Calif.’s salary from t − 1 to ta

−0.46 6.64

Pchmax Percent change in maximum of Calif.’s salary from t − 1 to ta

−0.45 6.65

a All wages are in constant (1982–1984) dollars.

20 In addition, the fixed-effects estimator increased the explained variance by a factor of 10.F tests indicate that the fixed effects are statistically significant.

presence of first-order serial correlation, the model was estimated using a Prais-Winsten (1954) GLS estimator.

Results

Table 3 displays the GLS regression results. As expected, higher unem- ployment rates lower turnover. A 10 percent increase in unemployment (about a 0.8 point increase in the unemployment rate) reduces turnover by 0.7 percentage points. Given an average quit rate of 2.6 percent per six months, this yields a reduction in the quit rate of 27 percent. In addition, the coefficients on the wage variables in California are all significant. Increasing wages 10 percent decreases turnover by 1.6 points, or by 62 percent (−1.6/2.6). Finally, the wage growth variables are also significant,

596 / MARLENE KIM

TABLE 3

TURNOVER, WAGE LEVELS, AND SURVEY WAGESa

(1) (2) (3) (4)

Wage levelb

ln(midpoint) −16.629** (5.614)

−18.490** (5.503)

ln(minimum) –15.580** (5.529)

ln(maximum) –16.736** (5.567)

Surveyed wage ln(median) 0.85548

(5.516) ln(average) 8.5411*

(4.329) ln(1st quartile) 2.7641

(5.418) ln(3d quartile) 4.4657

(5.477) Changes in Calif. Service’s wages

pchmid −0.20698** (0.06555)

−0.23042** (0.06613)

pchmin −0.19415** (.06488)

pchmax −0.17075** (0.06077)

ln(unemployment rate) −7.0419** (2.600)

−6.6566** (2.591)

−5.4848* (2.692)

−7.0049** (2.682)

Adj R2 0.413391 0.420011 0.407745 0.408086 N 418 418 419 416

* p < 0.05. ** p < 0.01. a Dependent variable = 6-month voluntary turnover per 100 workers. Standard errors are in parentheses. Fixed-effects occupa-

tional dummy variables are included. See Table 2 for the definitions of other variables. b All wages are in constant (1982–1984) dollars.

with a 1 percent increase reducing the quit rate by 0.2 percentage points, or by 8 percent (−0.2/2.6).

What is surprising is that the coefficients on the surveyed wage ratesremain insignificant in three of the four regressions. Moreover, despite examining a number of other specifications, I could not find stronger estimates of these coefficients. When changes in the surveyed rates are included in the regres- sion, for example, the coefficients are all insignificant (see Table 4). In

Voluntary Turnover and Wage Premiums/ 597

TABLE 4

TURNOVER WITH CHANGES IN SURVEYED WAGESa

(1) (2) (3) (4)

Wage levelb

ln(midpoint) −16.376** (5.737)

−17.813** (5.713)

ln(minimum) −16.615** (5.722)

ln(maximum) −16.473** (5.680)

Surveyed wage ln(median) −0.41665

(6.155) ln(average) 6.6046

(5.353) ln(1st quartile) 4.9484

(6.233) ln(3d quartile) 2.7399

(6.144) Changes in Calif. Service’s wages

pchmid −0.2246** (0.07453)

−0.24533*** (0.07086)

pchmin −0.17082* (0.07276)

pchmax −0.19152 (0.06812)

Changes in surveyed wages ∆median −0.035395

(0.06878) ∆average 0.02875

(0.04649) ∆1st quartile −0.046202

(0.06504) ∆3d quartile 0.043082

(0.06717) ln(unemployment rate) −7.3976**

(2.689) −6.5473** (2.605)

−4.9393** (2.801)

−7.2317** (2.717)

Adj R2 0.411910 0.41874 0.4068807 0.4068779 N 417 417 419 414

aDependent variable = 6-month voluntary turnover per 100 workers. Standard errors are in parentheses. Fixed-effects occupa- tional dummy variables are included. See Table 2 for the definitions of other variables.

bAll wages are in constant (1982–1984) dollars. * p < 0.05 ** p < 0.01. *** p < 0.001.

addition, a number of other specifications were run in order to examine whetherα2 was underestimated because of using short-term annual wage rates instead of long-term wage measures. If turnover responds to long-term trends in wage rates, including only annual wages could underestimate the impact of relative wages on quit rates. Thus moving averages of wages rather than the wage levels were included; however, the coefficients on mar- ket wage rates were seldom significant, and when significant, none came close to those shown in the second regression in Table 3.21 In addition, when long-term changes in turnover were regressed on long-term changes in wage rates, the coefficients were insignificant.22

These results suggest that turnover is not affected by what other employers pay for comparable jobs. Instead, what is important for turn- over behavior are the wages one receives and recent wage changes. In other words, workers are less likely to quit jobs that pay higher wages and that have received larger wage increases, no matter how these wage rates compare with those of other employers.

As Leonard (1987) suggests, it is useful to examine the estimates ofα3 to see if its magnitude supports the efficiency wage theory’s contention that it may be cost-effective for firms to increase wages in order to reduce turnover. Equation (6b) allows me to do so. Usingqw = [α3/w]/100, α3 = −16.5, and the average wage of $1888 per month multiplied by 6 for a 6-month estimate ofw, the estimatedqw is equal to [−16.5/11328]/100, or −0.000015. I find that paying wage premiums is profitable only if the marginal cost of turnover exceeds $66,667 over 6 months,23 which is much larger than estimates of turnover costs.24 Thus these results indicate that quit rates do respond to wage levels, but the magnitude is insufficient

598 / MARLENE KIM

21 However, if turnover responds to long-term relative wage rates and these remain constant, turnover behavior would not change at all, resulting in the low coefficients on the relative wage variables. This would occur if relative wage rates are stable in the long run but vary temporarily in the short run. But rela- tive wage rates do not follow such a pattern of stable wage rates in the long run with transitory changes year to year. In the long run (the time period in the sample), the wage ratio changed by an average of 9.5 percent in absolute value, and one-third of the occupations in the sample had changes in the double digits, with half of these above 20 percent. Thus it does not appear to be the case that the year-to-year changes in relative wage rates are only temporary changes that mask a constant wage ratio.

22 Ten-year changes in turnover were regressed on 10-year changes in wage rates. In addition, changes in turnover were regressed on changes in wage rates when using 3-year averages of the final 3 years of data minus 3-year averages of the first 3 years of data.

23 Given that wage increases affect the wage-change variables as well as the wage-level variables, however, the total estimate ofqw may be examined by using bothα3 andα4. Using−16.5 forα3 and−0.2 for α4, qw is −0.000032. This means that increasing wages to reduce turnover is cost-effective if turnover costs exceed $31,036 over 6 months. This is still not within reason.

24 Campbell (1993) suggests that turnover costs for salaried employees average $10,000 in 1980, as estimated by the M and M Association (1980).

to justify wage increases based on the efficiency wage-turnover theory when only cost savings from turnover are examined.

Despite examining a variety of different specifications, I could not find values ofqw in which higher wages would pay for themselves through reduced turnover costs. For example, using wage levels without a log- linear form (not shown) produces an estimate ofqw = −0.00006; this indicates that wage increases are cost-effective if training costs exceed $16,667 over 6 months. Using wage levels rather than the natural loga- rithm of the wage produces similar results (qw = −0.00006) if the wage- change variables are excluded. Using moving averages of wage rates also produced smaller estimates ofqw, and regressing long-term changes of turnover on long-term changes on wage rates produced insignificant results.

One possibility for the low coefficients on the wage variables may be due to California Service employees having preferences for working in state government or from receiving higher nonpecuniary benefits, such as job security, that is unmatched by private-sector employers. These would make employees less responsive to changes in relative wage rates. Yet, because other studies find low (and even lower) estimates of quit elastici- ties among private employers, the unresponsive turnover behavior does not seem particular to the California Service.

In fact, overall, these findings confirm those in previous studies that although higher wages do reduce quit rates, the elasticity of quit rates with respect to wages is too small to justify paying efficiency wages. My estimates ofqw are larger than Campbell’s (1993) but smaller than those of Leonard (1987) and Powell et al. (1994). Most likely, using panel data, omitting demographic characteristics of workers, adding controls for job- specific characteristics, and using different data and specifications account for these differences.

Of course, higher wages may be profitable if firms obtain a greater number of job applicants, better labor quality, and a greater amount of effort expended per worker. These, plus reduced turnover, could in com- bination be sufficient to justify paying higher (efficiency) wages. Past research indicates that the number of job applications rather than tenure duration may adjust to relative wages (Krueger, 1988). Unfortunately, I was unable to examine worker effort and the number of job applications in my data set. However, I was able to examine whether labor quality adjusted to wage premiums.

I examined this in two ways. First, using CPS data, I examined the correlations between the average age and education for state government workers in California and a relative wage variable (wages in the California

Voluntary Turnover and Wage Premiums/ 599

Service divided by the market wage rates). If labor quality adjusts to wage premiums, one would expect these correlations to be positive. The correla- tions between the average education of California state government work- ers and relative wages were negative, however, while the correlations between the average age of these workers and relative wages were positive but insignificantly different from zero. Second, I examined the correlation between a labor-quality variable and relative wages. Labor quality was estimated from wages predicted from a wage regression on age and educa- tion25; these predicted that wages serve as a proxy for labor quality. If labor quality increases as a result of higher wage premiums in the California Service, one would expect the correlations to be positive. Correlations between the predicted wages and the relative wage rates were negative and insignificant from zero, however.26 Thus these estimates, although not conclusive, suggest that labor quality in the State of California does not seem to be the major response to changes in relative wages.

My results differ from previous studies in finding that instead of relative wages affecting turnover, wage levels and wage growthwithin a firm are important predictors. In other words, workers are less likely to quit if they are paid higher wages and receive wage increases, no matter how these compare with the wages paid by other employers for similar work. This implies that wage levels may be important relative to many alternative uses of one’s time, only one of which includes working for another employer. In addition, the changes in wages have more significant effects on quit behav- ior than do the wage levels. Most previous work on turnover behavior treats quit rates as a function of the wage level rather than its rate of change.

600 / MARLENE KIM

25 I did this as follows: First, I ran a simple wage regression using March CPS data:

wi = B0 + B1agei + B2edi + ui wherewi is real annual earnings for each full-time year-round workeri in state government other than California, agei is the age of each full-time year-round workeri in state government other than California, edi is the education level for each full-time year-round workeri in state government other than California, andui is the error term. Second, using the estimated coefficients, I then estimated predicted wages for each year for California state government workers using CPS data:

where is predicted wages for full-time year-round California state government workers in yeart, aget is average age for full-time year-round California state government workers in yeart, and edt is average edu- cation level for full-time year-round California state government workers in yeart. Third, I examined the correlation between and the average relative wage variable. I also ran the first regression using local government and private-sector workers in California, but the correlations between and the average rela- tive wage variable that I obtained were unchanged.

26 To the extent that the average quality of workers would change for new hires rather than for the total workforce, the measurement of predicted wages underestimates the predicted wages of new hires and biases the correlations downward.

$w

$w $w

$ $ $ $w B B Bt t t= + +0 1 2age ed

These results are consistent with Reynolds (1951), who claims that workers simply do not change jobs due to better wages paid elsewhere. There are many reasons for why voluntary turnover may be insensitive to relative wages: Information about wages paid by other employers is imperfect, the cost of changing jobs may be prohibitive, fixed employ- ment and geographic limitations may inhibit turnover, and turnover may simply take time. It seems plausible, as these results suggest, that workers have greater knowledge about wages in their own firm—especially recent wage changes—than in other firms, causing them to base their decisions about quitting mainly on recent wage changes in their own firm and on how wage levels and wage growth compare historically within their occu- pation. Lucas (1972) has proposed such an asymmetrical information model for the economy, with agents knowing prices and price changes in their own market rather than in the aggregate economy, since information is costly. Certainly, further research is warranted to further investigate these explanations and their full implications.

Conclusion

Theoretically, voluntary turnover should decline as a firm’s wages increase relative to what others are paying. However, evidence from the State of California’s Civil Service does not support this. When control- ling for occupation-specific job characteristics and using the California Service’s own measurements of alternative wages available, I find that market wages have little orno effect on voluntary turnover. However, the absolutewage level paid for the occupation andwage changesin that occupation, rather than therelative wage level, have an effect. In other words, for a firm, it does not seem to matter how much one pays relative to what other firms pay; rather, what matters are recent wage increases and whether the job is simply low paid.

Other than the wage level and wage changes, the unemployment rate also had a strong effect on turnover. In addition, job-specific characteris- tics explained the greatest amount of variation in turnover. Finally, similar to previous findings, the results indicate that by itself, the elasticity of the quit rate with respect to the wage paid is too small to justify paying higher wages in order to reduce turnover costs. These results persist despite varia- tions in the model’s specifications.

These results cannot explain why turnover is not more responsive to wage rates. It is possible that increasing wages above the market-clearing level can still be profit-maximizing if there are productivity effects because of increased morale, reduced shirking, or feelings of gratitude.

Voluntary Turnover and Wage Premiums/ 601

The sum of all these effects could justify paying higher wages. Further research is needed to examine these issues.

These results also cannot explain why turnover is not more responsive to market wage rates. Workers may lack information about alternative wages available, face high fixed costs from changing employers (such as loss of seniority or pension), or otherwise face significant mobility costs. Further research is needed to verify whether these results hold in the pri- vate sector. If firms do not suffer the consequences of costly turnover should they fail to match their competitors’ salaries, paying market wage rates may not bring the benefits of lower turnover that firms have long assumed (although, of course, it may be beneficial for recruitment or labor quality). This has implications for standard compensation practices and the extent to which firms need to invest in salary surveys and adjust their salaries according to what their competitors pay. Instead, what may be more important for turnover is the growth and level of paywithin a firm.

REFERENCES

Antel, John. 1988. “Interrelated Quits: An Empirical Analysis of the Utility Maximizing Mobility Hypothesis.”Review of Economics and Statistics70(1):17–22.

Atwood, Jay. 1986. Deposition of Jay Atwood, July 7, 1986.CSEAv. State of California et al., C-84-7275 MHP. Federal District Court, San Francisco.

Bartel, Ann P. 1982. “Wages, Nonwage Job Characteristics, and Labor Mobility.”Industrial and Labor Relations Review35(4):578–89.

Burdett, Kenneth. 1978. “A Theory of Employee Job Search and Quit Rates.”American Economic Review68(1):212–20.

California State Personnel Board. 1975. “Overview of the Salary-Setting Process.” Undated docu- ment attributed to the mid-1970s.

Campbell Carl M., III. 1993. “Do Firms Pay Efficiency Wages? Evidence with Data at the Firm Level.” Journal of Labor Economics11(3):442–70.

Crain, Bruce. 1986. Deposition of Bruce Crain, May 14, 1986.CSEAv. State of California et al., C-84-7275 MHP. Federal District Court, San Francisco.

Flacco, Paul R., and Lester A. Zeager. 1989. “The Competitive Firm with Uncertainty in the Rate of Labor Turnover.”Southern Economic Journal56(October):457–66.

Krueger, Alan B. 1988. “The Determinants of Queues for Federal Jobs.”Industrial and Labor Relations Review41:4.

Leonard, Jonathan S. 1987. “Carrots and Sticks: Pay, Supervision, and Turnover.”Journal of Labor Economics5(4, part 2):S136–52.

Lucas, Robert E., Jr. 1972. “Expectations and the Neutrality of Money.”Journal of Economic Theory4(2):103–24.

Milkovich, George T., and Jerry M. Newman. 1996.Compensation. Boston: Irwin. Mobley, William H. 1982.Employee Turnover: Causes, Consequences, and Control. Menlo Park,

CA: Addision-Wesley. Powell, Irene, Mark Montgomery, and James Cosgrove. 1994. “Compensation Structure and

Establishment Quit and Fire Rates.”Industrial Relations33(2):229–48. Prais, S., and C. Winsten. 1954. “Trend Estimation and Serial Correlation.” Discussion paper 383,

Cowles Commission, Chicago.

602 / MARLENE KIM

Reynolds, Lloyd G. 1951.The Structure of Labor Markets. New York: Harper. Salop, Steven C. 1979. “A Model of the Natural Rate of Unemployment.”American Economic

Review69(1):117–25. Solnick, Loren M. 1988. “Promotions, Pay, Performance Ratings and Quits.”Eastern Economic

Journal 14(1):51–62. Summers, Lawrence H., and Jeremy I. Bulow. 1986. “A Theory of Dual Labor Markets with Appli-

cation to Industrial Policy, Discrimination, and Keynesian Unemployment.”Journal of Labor Economics4(3, part 1):376–414.

Slichter, Sumner. 1921.The Turnover of Factory Labor. New York: Appleton and Co. Stiglitz, Joseph E. 1974. “Alternative Theories of Wage Determination and Unemployment in

LDC’s: The Labor Turnover Model.”Quarterly Journal of Economics88(2):194–226. Telly, Charles S., Wendell L. French, and William G. Scott. 1971. “The Relationship of Inequity to

Turnover Among Hourly Workers.”Administrative Science Quarterly(June):164–72. Van den Berg, Gerard J. 1992. “A Structural Dynamic Analysis of Job Turnover and the Costs

Associated with Moving to Another Job.”The Economic Journal102(414):1116–33. Viscusi, W. Kip. 1979.Employment Hazards: An Investigation of Market Performance. Cambridge,

MA: Harvard University Press. Weiss, Andrew. 1984. “Determinants of Quit Behavior.”Journal of Labor Economics2(3):

371–87.

Voluntary Turnover and Wage Premiums/ 603

Copyright of Industrial Relations is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.