Effects of Human Resource Systems on Manufacturing Performance and Turnover

Author(s): Jeffrey B. Arthur

Source: The Academy of Management Journal, Vol. 37, No. 3 (Jun., 1994), pp. 670-687
Published by: Academy of Management
Stable URL: http://www.jstor.org/stable/256705 .
Accessed: 13/09/2013 21:30

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].

Academy of Management is collaborating with JSTOR to digitize, preserve and extend access to The Academy
of Management Journal.

http://www.jstor.org

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
? Academy of Management Journal
1994, Vol. 37, No. 3, 670-687.

EFFECTS OF HUMAN RESOURCE SYSTEMS ON

MANUFACTURING PERFORMANCE AND TURNOVER

JEFFREY B. ARTHUR
Purdue University

Using an empirical taxonomy identifying two types of human resource

systems, "control" and "commitment," this study tested the strategic
human resource proposition that specific combinations of policies and
practices are useful in predicting differences in performance and turn-
over across steel "minimills." The mills with commitment systems had
higher productivity, lower scrap rates, and lower employee turnover
than those with control systems. In addition, human resource system
moderated the relationship between turnover and manufacturing per-
formance.

Long a concern among organizational contingency theory researchers,

the concept of the congruence, or fit, between diverse sets of organizational
policies and practices has recently emerged as an important subject of study
for human resources management researchers. This new strategic, macro,
human resource management perspective differs markedly from the more
traditional approach focusing on the effects of separate human resource
practices on individual-level outcomes (Butler, Ferris, & Napier, 1991; Jack-
son, Schuler, & Rivero, 1989; Mahoney & Deckup, 1986; Snell, 1992). In
contrast, the strategic human resource management perspective integrates
macro-level theories and concepts to explore the impact of specific config-
urations, or systems, of human resource activities on organization-level per-
formance outcomes (Dyer & Holder, 1988; Fisher, 1989; Wright & McMahan,
1992).
Dobbins, Cardy, and Carson pointed out that although a macro approach
to studying human resource issues appears promising and conceptually very
rich, "the validity of its propositions is ultimately an empirical question"
(1991: 33). Empirical evidence demonstrating the predictive value of the
strategic human resource perspective, however, has not been forthcoming.
Conceptual typologies abound in this literature, but empirically based tax-
onomies of human resource strategies are rare. As a result, basic hypotheses
concerning the implications for firm performance that flow from the strate-
gic human resource perspective have generally not been tested. A recent

I would like to thank Steven G. Green, Margaret L. Williams, Michael A. Campion, Chris J.
Berger, Harry C. Katz, and three anonymous reviewers for helpful comments on previous drafts
of this article.

670

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
1994 Arthur 671

review of strategic human resource management, for example, concluded

that "there is little empirical evidence to suggest that strategic HR directly
influences organizational performance or competitive advantage" (Leng-
nick-Hall & Lengnick-Hall, 1988: 468).
In this study, I addressed this important gap in the existing literature by
empirically testing specific organizational performance hypotheses flowing
from a strategic human resource management perspective. To accomplish
this, I drew on the results of a previous study that used a cluster analysis
technique to empirically identify two types of human resource systems,
labeled "control" and "commitment" systems, in a sample of steel minimills
(Arthur, 1992).1 I developed and tested propositions regarding the utility of
this human resource system taxonomy for predicting both manufacturing
performance, measured as labor efficiency and scrap rate, and the level of
employee turnover in steel minimills. In addition, I tested the proposition
that the relationship between turnover and manufacturing performance dif-
fers significantly across the two systems.

THEORETICAL DEVELOPMENT AND HYPOTHESES

Testing the strategic human resource perspective first requires catego-

rizing organizations into a meaningful typology of human resource systems.
Using the strategic perspective, a number of authors have suggested typol-
ogies (e.g., Dyer & Holder, 1988; Miles & Snow, 1984; Osterman, 1987;
Schuler & Jackson, 1987; Walton, 1985). Underlying the use of these typol-
ogies is the proposition that organizations differ in their basic approaches or
objectives in managing human assets. These objectives, often stated in terms
of desired employee characteristics, attitudes, and behaviors, are (or should
be) derived from a firm's overall business goals and may be moderated by
factors internal and external to the organization (Schuler, 1992; Wright &
McMahan, 1992). However, because these typologies have not been consis-
tently measured, their validity and predictive power have not been assessed.
Control and Commitment Human Resource Systems

My earlier research (Arthur, 1992) is one of the first published attempts

to develop an empirical classification of firms based on human resource
system characteristics. Applying a cluster analysis technique to data from
human resource managers, I found that the variety of human resource pol-
icies and practices in 30 U.S. steel minimills could be meaningfully de-
scribed by six clusters, or systems. Further, I grouped those systems into two
broad categories based on their characteristics and the functions they served
and labeled them "cost reducers" and "commitment maximizers." To main-
tain consistency with previous research on human resource strategy (Lawler,

1 Minimills are relatively small steel-producing facilities in which metal scrap is melted in
electric furnaces and continuously cast into a variety of shapes and grades of steel. Detailed
discussion of minimills can be found in Barnett and Crandall (1986) and Hogan (1987).

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
672 Academy of ManagementJournal June

1986; Walton, 1985), I have labeled those systems "control" and "commit-
ment" in this study.
Control and commitment represent two distinct approaches to shaping
employee behaviors and attitudes at work. The goal of control human re-
source systems is to reduce direct labor costs, or improve efficiency, by
enforcing employee compliance with specified rules and procedures and
basing employee rewards on some measurable output criteria (Eisenhardt,
1985; Walton, 1985). In contrast, commitment human resource systems
shape desired employee behaviors and attitudes by forging psychological
links between organizational and employee goals. In other words, the focus
is on developing committed employees who can be trusted to use their
discretion to carry out job tasks in ways that are consistent with organiza-
tional goals (e.g., Organ, 1988).
The control and commitment approaches to human resource manage-
ment are expected to be represented by different sets of programs and prac-
tices. In my previous research (Arthur, 1992) I found that in general, com-
mitment human resource systems were characterized by higher levels of
employee involvement in managerial decisions, formal participation pro-
grams, training in group problem solving, and socializing activities and by
higher percentages of maintenance, or skilled, employees and average wage
rates. The present study's methods section presents further details of these
system patterns.
The existence of the control and commitment variations in organiza-
tions is generally thought to be associated with certain organizational con-
ditions. Most human resource strategy researchers have taken a behavioral
perspective (cf. Snell, 1992). Research using this perspective rests on the
often implicit assumption that the successful implementation of a business
strategy requires a unique set of employee behaviors and attitudes and that
a unique set of human resource policies and practices will elicit those be-
haviors and attitudes (Cappelli & Singh, 1992). Alternatively, control theory
researchers (e.g., Eisenhardt, 1985; Ouchi, 1979; Snell, 1992) have noted that
the use of a control system depends on managers having a relatively com-
plete knowledge of the transformation process (inputs to outputs) and a high
ability to effectively set performance standards and measure employee out-
puts. These conditions enable employers to directly monitor and reward
employee behavior or the specific outcomes of that behavior. In the absence
of these conditions, an input, or clan, system is predicted, in which selec-
tion, training, and socialization policies that try to align employee interests
with those of the firm are emphasized (Eisenhardt, 1985).
Hypotheses
Human resource systems and manufacturing performance. Although
the above theoretical approaches suggest a contingency view of human re-
source system effectiveness, there are a number of reasons to believe that a
smoothly functioning commitment human resource system will be associ-
ated with higher organizational performance than will a control system, the

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
1994 Arthur 673

more traditional human resource approach in manufacturing (Walton, 1985).

By decentralizing managerial decision making, setting up formal participa-
tion mechanisms, and providing the proper training and rewards, a commit-
ment system can lead to a highly motivated and empowered work force
whose goals are closely aligned with those of management (Thomas & Velt-
house, 1990). Thus, the resources required to monitor employee compliance,
such as those needed to maintain supervision and work rules, can be re-
duced (Locke & Schweiger, 1979). In addition, employees under these con-
ditions are thought to be more likely to engage in organizational citizenship
behaviors (Organ, 1988), nonrole, unrewarded behaviors that are believed to
be, nonetheless, critical to organizational success (e.g., Katz, 1964).
A parallel set of arguments on the superior performance of commitment
systems can be found in the recent industrial relations literature (e.g.,
Kochan, Katz, & McKersie, 1986). Management's attempt to implement a
classic control system for reducing labor costs by unilaterally increasing
performance standards and maintaining wages and benefits is likely to be
met by strong resistance from a unionized work force. Resistance in the form
of strikes, high grievance rates, and adversarial labor relations have been
found to be extremely costly to firms in terms of productivity and quality
(Cooke, 1992; Katz, Kochan, & Weber, 1985).
Finally, although there continues to be debate concerning the condi-
tions under which the participative management practices associated with
commitment systems are effective (e.g., Levine & Tyson, 1990; Locke &
Schweiger, 1979), recent evidence suggests that these practices are espe-
cially critical for the effective implementation and utilization of advanced
manufacturing technology (Majchrzak, 1988; Dean & Snell, 1991). The
present study focused specifically on manufacturing performance, defined
by two production-process-related measures of organizational effectiveness,
labor efficiency and scrap rate. These performance outcomes are predicted to
be most directly affected by differences in employee behaviors and charac-
teristics as shaped by systems of human resource activities.
Hypothesis 1: Plants with commitment human resource
systems will have better manufacturing performance than
plants with control human resource systems.
Human resource systems and turnover. A vast literature exists on the
determinants of employee turnover, long considered an important outcome
for both individuals and organizations. Most of this research has focused on
individual-level variables, such as employees' satisfaction with their jobs
and their organizational commitment (e.g., Cotton & Tuttle, 1986). In this
work, I tested the proposition that organization-level human resource char-
acteristics are also significantly related to the overall turnover in firms, ex-
pecting higher turnover in organizations with control systems than in those
with commitment systems.
The basis for this prediction is the different objectives of control and
commitment systems as manifested in different combinations of human re-

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
674 Academy of Management Journal June

source policies and practices. As noted above, the driving force behind a
control system is to reduce direct labor costs. This goal is expected to be
manifested in the use of relatively simple, well-defined job tasks. Because
employees with a minimum amount of training and experience can perform
such tasks, wages and the costs of employee search, selection, and training
can also be minimized. Under these conditions, the costs of employee turn-
over to a firm are expected to be relatively low, so employers have very little
incentive to try to minimize turnover through human resource policies and
policies designed to increase employee commitment or attachment. In fact,
employee commitment might be considered dysfunctional since compensa-
tion is generally higher for senior employees than for similarly qualified new
employees.
Hypothesis 2: Turnover will be higher in control human
resource systems than in commitment human resource
systems.
Turnover and manufacturing performance. Since the late 1970s, re-
search on the consequences of employee turnover has generally compared
the cost and performance of individuals who leave an organization with
those of (1) their replacements, (2) those who stay with the organization, or
(3) both (e.g., Boudreau & Berger, 1985; Hollenbeck & Williams, 1986). A
potential limitation of this approach is that the effect of organizational con-
text is largely ignored. In other words, the departure of an individual with a
given level of assessed performance is assumed to have the same effect on
organizational performance across organizations.
In contrast, the human resource strategy perspective suggests that the
effect of turnover level on organizational performance depends critically on
the nature of the context or system in which the turnover occurs (cf. Miller
& Friesen, 1984). System characteristics can be seen as affecting the perfor-
mance impact of a number of the predicted consequences of turnover, such
as disruption of social and communication structures, training and assimi-
lation costs, and decreased cohesion and commitment of members who stay
(Dalton & Todor, 1979; Mobley, 1982; Staw, 1980). For example, the fact that
the jobs in organizations with commitment systems often require high train-
ing and skill levels suggests a stronger relationship between organizational
tenure and performance than exists in control systems. Individuals in such
jobs will take longer to reach top performance than individuals in the sim-
pler jobs of control systems (Campion, 1989).
In addition, because production employees in commitment systems take
on more managerial-level decision-making tasks, their organizational cen-
trality, and hence the potential for their departure to disrupt organizational
functioning, is expected to be greater than the disruptive potential of the
typical employee in a control human resource system who does not have
these vertical task responsibilities.
Hypothesis 3: There will be a stronger negative relation-
ship between turnover level and manufacturing perfor-

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
1994 Arthur 675

mance in commitment human resource systems than in

control human resource systems.

METHODS
The data used for this study come primarily from questionnaire re-
sponses by human resource managers at 30 of the 54 existing U.S. steel
minimills. Data were gathered between November 1988 and March 1989.
The average age, size, geographic region represented, and union status of the
mills surveyed are not statistically different from those of the total minimill
population (Arthur, 1992).
The modal minimill firm is an independent, privately held U.S. firm
that owns one minimill. There are, however, a growing number of larger
domestic (and some foreign) companies that own multiple plants (Hogan,
1987). Because of the possibility that a manager's scope of responsibility
might span several mill locations, managers were specifically instructed to
direct their comments only to the mill location to which the questionnaire
was addressed. In addition, a different set of managers provided data for
each mill so that in no case were data from more than one mill collected from
a single manager.2
Measures
Minimill human resource systems. In Arthur (1992), I used the minimill
questionnaire data to construct ten variables measuring various aspects of
mills' workplace human resource systems. Table 1 defines these variables,
which I standardized and submitted to a cluster analysis using Ward's
method in order to empirically identify the minimill human resource sys-
tems. This procedure yielded a six-cluster solution based in part on an
analysis of the change in the fusion coefficient, defined as the error sums of
square for this procedure (Arthur, 1992: 504).
Theoretical considerations, such as a desire to test for differences be-
tween two dominant human resource systems, and sample size limitations
(degrees of freedom were not available to include all six clusters in the
analysis) required some additional data aggregation. I accomplished this by
categorizing the patterns of cluster scores into control and commitment sys-
tems. As reported in my previous study, this judgment was informed by
descriptions of alternate forms of control and commitment human resource
systems found in the relevant literature as well as by detailed primary and
secondary case descriptions of the mills.
Conceptually, this method of data aggregation is consistent with the

2
Some recent studies have found that different human resource practices may be applied
to different occupational and functional groups within firms (ackson et al., 1989; Snell & Dean,
1992). Because this study only focused on the maintenance and production workers in min-
imills, I could not assess the possibility that different sets of human resource activities were
applied to other functional and occupational groups in the organizations.

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
 _ #

TABLE 1
Description of Human Resource Variables and Comparison of Means for Control a
-_ . r ..u . . .. -. ...t
w ',,I -:,' -:- . ...I. qi ......-......-. . . ... .-.- .. I X,:,
A . . I..-..-
Variables Contr
Name Description Means
Decentralization Degree to which nonsupervisory employees monitor data on quality, costs, 2.42
productivity, and scrap; determine work flow or order of tasks; invest in
new equipment and technology; develop new products (1 = very little, 6
= very much)
Participation Percentage of mill employees who received training in group problem 36.57
solving, meet on a regular basis in small groups to discuss production or
quality problems, or are involved in joint union-management or
employee-management committees
General training Part of general training and development activities for production and 1.92
maintenance employees accounted for by seminars, classes, training not
directly related to employees' immediate work area; general skills training not
directly related to the employee's current job; or people skills training such
as communication or group problem solving (1 = no part, 6 = large part)
Skill Number of maintenance and craft workers as a percentage of all mill employees 0.14
Supervisor Number of production workers per supervisor 7.05
Social Number of times per year management organizes social gatherings for 3.94
employees, such as company picnics and bowling
Due process Percentage of total employee complaints or grievances that are handled 20.56
through formal grievance procedures involving several steps, up to and
including binding third-party arbitration
Wages Average total employment cost per production and maintenance worker, 18.07
including wage rate, benefits, bonus or incentive payments, and taxes
Benefits Percentage of total average employment cost accounted for by employee 27.31
benefits
Bonus Percentage of total average employment cost accounted for by bonus or 29.96
incentive payments
Percentage unionized 38
a N = 16, control; N = 14, commitment.
b One-tailed
tests; approximation t-test for unequal sample variances were used when indicated by a sig
* p < .05
**p < .01

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
1994 Arthur 677

systems concept of "equifinality," according to which the same end (control

or commitment) can be achieved by multiple means (configurations of hu-
man resource policies and practices) depending on specific organizational
constraints. The categorization procedure also allows for the possibility of
interpreting an individual variable or set of variables differently depending
on its overall pattern or context (cf. Miller & Friesen, 1984). This freedom is
especially useful for interpreting a variable such as due process (see Table 1),
which means different things in union and nonunion contexts (Arthur,
1992). By examining the complete pattern of scores as well as the union
status of the different clusters, I could identify human resource cluster pat-
terns that were consistent with the descriptions of alternate forms (union
and nonunion) for achieving cost control and employee commitment found
in the human resource and industrial relations strategy literature.
As Table 1 shows, the pattern of scores for the aggregated clusters shows
considerable face validity, appearing to match the descriptions of control
and commitment systems provided earlier. All the mean differences be-
tween the two cluster groupings are in the predicted directions. In particu-
lar, the average scores on measures of decentralizing decision making, gen-
eralized training, skill, and wage rate are all significantly higher in commit-
ment systems than in control systems. The value of the bonus variable is
significantly higher for control systems, which is consistent with the exis-
tence of an output control strategy (Eisenhardt, 1985).
Additional evidence for the validity of the cluster groupings was pro-
vided in Arthur (1992) by a significant biserial correlation between two of
Porter's (1980) business strategies, low cost and differentiation, and the ag-
gregated cluster groupings (r = .46, p < .05). As predicted, mills with low-
cost business strategies were more likely to have control human resource
systems, and mills with differentiation strategies were more likely to have
commitment human resource systems. Business strategy data for the mills
were gathered independently from the data on human resource practices.
This evidence of predictive validity is especially relevant for use with clus-
ter analysis results that, in the absence of statistical tests for the "right"
number of clusters, must be judged primarily on their usefulness in predict-
ing outcomes of variables not used in the clustering procedure (cf. Aldender-
fer & Blashfield, 1984).
To help determine the reliability of the categorization, I gave six raters
(advanced graduate students in human resource and organizational behav-
ior) a questionnaire that included the descriptions of control and commit-
ment systems and asked them to judge independently whether each of the
six patterns of standardized variable scores presented in Arthur (1992) rep-
resented control or commitment. The six raters unanimously selected the
category used in this study for five of the six cluster patterns, and five of the
six raters correctly categorized the sixth cluster. These results indicate a very
high degree of interrater reliability for the human resource system categori-
zation used in this study. In the analysis, control systems were coded 0 and
commitment systems, 1.

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
678 Academy of Management Journal June

Manufacturing performance and turnover. The respondents were also

asked to report on their mills' labor efficiency, scrap rate, and turnover over
the year before the survey. The variable labor hours was defined as the
average number of labor hours (both direct and indirect) required to produce
one ton of steel at a mill. Scrap rate was the number of tons of raw steel that
had to be melted to produce one ton of finished product. Industry journals,
interviews with mill managers, and pretest feedback indicated that these
were common performance metrics used in the industry with which top
human resource managers would be familiar. Because only human resource
managers were asked to report on these measures, however, interrater reli-
ability could not be assessed.3
The managers were also asked to indicate the number of production and
maintenance employees who had "left the mill (either voluntarily or invol-
untarily) over the past year." This figure was divided by the total number of
production and maintenance employees in a mill, derived from a separate
question asked at the beginning of the questionnaire, to measure turnover.
Control variables. The age, size, union status, and business strategy of
the mills were included as control variables. Age was the number of years a
mill had been in existence (1993 minus year of founding). Data for this
measure were gathered from archival sources (Hogan, 1987; International
Trade Administration, 1986). It is included in the model as a proxy for the
newness and sophistication of the plant design and production technology.
Size was the number of employees at each mill reported by its human re-
source manager.
Union status and business strategy were expected to vary with the hu-
man resource system variable and to have potential direct effects on manu-
facturing performance. Union status data were gathered from secondary
sources (e.g., Hogan, 1987); 47 percent of the minimills studied were union-
ized. In all but one of the unionized mills, production and maintenance
employees were represented by the United Steelworkers of America.
Business strategy data were gathered through the use of a separate ques-
tionnaire completed by the top line manager at each minimill. I then cluster-
analyzed these data and used them to classify the mills into low-cost and
differentiation strategy categories following the work of Porter (1980) and
others (e.g., Dess & Davis, 1984). Details of the business strategy variables

3 Two studies, however, have reported performance levels for minimills measured at the
industry level very similar to those found in this study and thus provide some support for the
validity and reliability of the performance measures. The average productivity rate of 2.35 labor
hours per ton in this study compares favorably with the average of approximately 2.00 labor
hours per ton found in Hogan's (1987) survey of U.S. minimill firms. A U.S. Department of
Commerce study also supports the validity of the average scrap rate of 18 percent found in this
study, reporting an 80 percent "yield of raw steel to finished product" (a 20 percent scrap rate)
in U.S. minimills (International Trade Administration, 1986: 12). However, given the lack of
independently reported plant-level performance data with which to assess the reliability and
validity of the performance measures over a specified period of time, readers may want to
exercise some caution in interpreting these results.

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
1994 Arthur 679

and classifications are presented in Table 2. The strategy and union variables
were coded as dichotomous variables, with 1 equaling the presence of a
union and a differentiation business strategy, respectively.
Table 3 presents the means, standard deviations, and correlations for all
the dependent and independent variables in the study.

RESULTS

Hypothesis 1 predicts that the presence of a commitment human re-

source system will be related to significantly higher manufacturing perfor-
mance than the presence of a control system. I tested this hypothesis using
regression analysis for labor hours and scrap rate in models that included
system type and the control variables.4 Table 4 presents results of the anal-
yses. The negative coefficients for the human resource system variable in
both models indicates that commitment is significantly related to both fewer
labor hours per ton and lower scrap rates. Because the results of the overall
regression model for scrap rate are not significant, however, the significance
of the human resource system variable in this model must be interpreted
with some caution.
The second hypothesis predicts that turnover will be higher in min-
imills with control systems than in those with commitment systems. Turn-
over was over twice as high in the former (x = .07, s.d. = .07, control, and
x = .03, s.d. = .03, commitment). A t-test showed that this difference was
statistically significant (t = 2.19, p < .05).
Finally, Hypothesis 3 states that the negative relationship between turn-
over and manufacturing performance will be higher in commitment human
resource systems than in control systems. Table 5 presents the results of a
moderated regression analysis that included the main effects of turnover and
system and their multiplicative interaction. Union status was used as a con-
trol variable in this analysis because previous research has found that union-
ization is related to both turnover level and manufacturing performance
(e.g., Freeman & Medoff, 1984). The interaction term is significant for both
labor hours and scrap rate, and its inclusion results in a significant change
in R2.
Table 6 shows results of a comparison of the sizes of the correlations
between turnover and performance in the two types of system, with union
status again controlled. This subgroup comparison was appropriate for test-
ing the strength of the moderating effect of system type (Arnold, 1982). The
table shows significant, positive correlations between turnover and both
labor hours and scrap rate for minimills with commitment systems-
manufacturing performance is worse the higher the turnover. For mills with
control systems, the correlations are negative and insignificant. Results of

Missing data reduced the number of usable observations by three in the labor hours
4

equationand by six in the scraprateequation.The mills with missing datawere not statistically
different from the rest of those studied in age, size, or union status.

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
 TABLE 2
Description of Business Strategy Variables and Comparison of
Low-Cost
Variables Producer
Name Description Means s.d.
Low-cost producer Provide customers with basic steel products at prices 4.67 1.11
below those of competitors.b
Differentiator Provide customers with specialized sizes, shapes, or 3.23 1.00
grades of steel; develop and produce new steel shapes,
sizes, or grades; switch quickly between the production
of different shapes, sizes, or grades to respond to
fluctuations in market demand; produce shapes, sizes,
and grades for the higher-priced segment of steel
market; develop new techniques and methods to
market products.b
Focus Serve needs of customers within a specific geographic 4.60 1.37
area; create recognition of company's name and
reputation in the industry.b
Product types Number of different product types produced 1.64 0.74
Product sizes Number of different sizes of steel products produced 50.54 69.30
Product grades Number of different grades of steel products produced 9.07 8.34
Contract customers Percentage of 1987-88 production delivered to contract 17.69 18.33
customers
Open market customers Percentage of 1987-88 production sold on the open, 25.38 32.30
spot, market
aData are from Arthur (1992). N = 15, low-cost producer; N = 25, differentiator.
b
Respondents were asked to indicate importance, with 1 = unimportant, 6 = very important.
* p < .05
**p < .01
*** p < .001

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
 TABLE 3
Means, Standard Deviations, and Correlations
Variables N Means s.d. 1 2 3 4
1. Human resource system 30 0.47 0.51
2. Business strategy 29 0.69 0.47 .46*
3. Age 29 20.44 20.01 .06 .14
4. Unionization 30 0.47 0.51 .27 .35 .38*
5. Size 30 463.40 318.30 .45* .24 .36 .39*
6. Labor hours 28 2.30 0.98 -.13 .07 -.65*** .43*
7. Scrap rate 25 0.22 0.22 .14 -.08 -.22 .20
8. Turnover 29 0.05 0.06 -.37 -.27 -.10 -.26
* p < .05
** p < .01
***
p < .001

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
682 Academy of Management Journal June

TABLE 4
Results of Regression Analysis, Systems on Performance
Labor Hours Scrap Rate
Variables b s.e. b s.e.
Constant 1.46*** 0.28 0.12* 0.05
Age 2.23** 0.67 -0.24* 0.11
Size 0.08 0.06 0.02 0.01
Unionization 0.27 0.32 0.08 0.05
Business strategy 0.17 0.33 0.03 0.05
Human resource system -0.79* 0.36 -0.14* 0.06
R2 0.65 0.42
df 5,21 5,18
F 6.25** 1.85

p < .05
**p < .01
***p< .001

tests of the difference between the correlations using Fisher's Z transforma-

tion were significant for both labor hours and scrap rate.

DISCUSSION
The regression results in this study indicate that the human resource
system taxonomy developed in my previous research (Arthur, 1992) was
significantly associated with variation in steel minimills' performance. More
specifically, these results support observations made by Walton (1985) and

TABLE 5
Results of Regression Analysis, Turnover and Human Resource Systems
on Performance
Labor Hours Scrap Rate
Restricted Model Full Model Restricted Model Full Model
Variables b s.e. b s.e. b s.e. b s.e.
Constant 2.34*** 0.36 2.56*** 0.31 1.13*** 0.05 1.16*** 0.05
Unionization 0.95* 0.36 0.74* 0.33 0.08 0.05 0.06 0.05
Turnover 1.74 3.06 -0.90 2.90 0.24 0.43 -0.08 0.41
Human resource
system -0.28 0.38 -1.18* 0.48 -0.01 0.06 -0.14t 0.07
Turnover x system 25.22* 9.43 2.99* 1.30
R2 0.25 0.43 0.12 0.30
df 3,23 4,22 3,21 4,20
F 2.51t 4.17* 0.92 2.16
AR2 0.18 0.18
F 6.92*** 5.14**

tp < .10
*p < .05
** p < .01
***p < .001

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
1994 Arthur 683

TABLE 6
Comparisons of Partial Correlations
Control Commitment
Variables r N r N za
Laborhours -.046 16 .830** 11 1.95*
Scrap rate -.057 15 .900*** 10 2.01*
a One-tailed tests.
p < .05
** p < .01
**p < .001

others concerning the effectiveness of commitment-type human resource

systems, at least in the context of manufacturing plants using technologi-
cally intensive and relatively integrated continuous production processes
(e.g., Dean & Snell, 1991). Interestingly, no tradeoff between production
quality and efficiency emerged. Commitment was associated with both
lower scrap rates and higher labor efficiency than control.
In addition, the significant differences found between the two types of
systems in both turnover and the relationship between turnover and manu-
facturing performance suggest the importance of including a measure of
human resource system as a moderator in future research on the conse-
quences of turnover. These results support the recent observations of a num-
ber of researchers who have noted the importance of identifying organiza-
tion-level consequences of employee turnover (e.g., Dalton & Todor, 1979;
Mobley, 1982; Schwab, 1991).
Finally, the lack of a significant correlation between turnover and man-
ufacturing performance in control systems may indicate a nonlinear rela-
tionship between these two variables. In other words, organizations with
control human resource systems may benefit from high employee turnover
up to some point, but after that point is reached there begins to be a detri-
mental effect on manufacturing performance. In contrast, the results show a
negative linear relationship between turnover and manufacturing perfor-
mance in the commitment systems. The finding that the right amount of
turnover varies with system type has important implications for practition-
ers seeking to manage this process.
Limitations and Future Research
The relatively small number of existing minimills limited the types of
statistical analyses possible, including the use of a full set of control vari-
ables in testing Hypotheses 2 and 3. In addition, the specific characteristics
of the organizations studied limit the generalizability of these results. To
what extent do the human resource systems found in this study also exist in
large public sector organizations such as schools and hospitals? What is the
relationship between human resource system and organizational perfor-
mance in less technologically intensive, service-oriented organizations? Fi-

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
684 Academy of Management Journal June

nal evaluation of the evidence for the human resource strategy perspective
will need to await the accumulation of results from studies conducted in
multiple industry contexts.
In addition, although the findings of this study are consistent with a
conceptual model in which the choice of human resource system leads to
changes in manufacturing performance, the cross-sectional data used here
did not permit any tests of the causal ordering between effects of system and
performance. It is possible that better performing mills also have additional
resources that facilitate management's choosing commitment systems.
Further, research progress on the human resource strategy perspective
depends critically on the development of conceptually and methodologi-
cally sound measures of the human resource system construct. Although the
taxonomy used in this study shows some conceptual and predictive prom-
ise, much more work needs to be done concerning definition and measure-
ment of the dimensions of human resource systems. A key related issue is
the performance implications of mixed systems. A premise of this work is
that control and commitment represent conceptually distinct ideal systems
and that any deviation from the ideal types will weaken performance. Al-
ternatively, control and commitment can be conceptualized as the opposite
ends of a continuum of possible human resource systems and the most
effective system seen as existing somewhere between the two extremes.5
Empirical tests are needed to determine which conceptualization more ac-
curately describes the construct.
Other parts of the human resource strategy model are also in need of
empirical investigation. For example, because of sample size limitations, I
was unable in this study to test for the performance effects of the fit between
business strategy and human resource strategy. Finally, there is a need to
demonstrate that certain combinations of human resource programs, poli-
cies, and practices lead to specific employee attitudes, such as trust in man-
agement or organizational commitment, that in turn lead to specific em-
ployee behaviors beneficial to effectively implementing a given business
strategy. Exploring these intermediate links explicitly will undoubtedly lead
to further refinements and insights into the process by which combinations
of human resource activities can lead to competitive advantages for firms
(e.g., Cappelli & Singh, 1992).
Conclusions
In spite of its limitations, this research shows that a number of insights
can be gained through the use of a human resource strategy perspective and
methodology. By empirically testing whether certain combinations of activ-
ities are associated with higher manufacturing performance, this study pro-
vides one of the first pieces of empirical evidence with which to evaluate the
prescriptions in the human resource strategy literature. Many authors have

5 I would like to thank an anonymous reviewer for drawing my attention to this point.

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
1994 Arthur 685

called for such evidence (Dobbins et al., 1991; Fisher, 1989; Jackson et al.,
1989; Lengnick-Hall & Lengnick-Hall, 1988; Snell & Dean, 1992; Wright &
McMahan, 1992). In addition, the study has shown that identification of
human resource systems promises to add significantly to understanding the
relationship between turnover and organizational performance. Although
these results should be seen as preliminary because of data limitations, this
study provides future researchers with some empirical evidence supporting
a promising new perspective with which to study important human resource
and organizational outcomes.

REFERENCES
Aldenderfer, M. W., & Blashfield, R. K. 1984. Cluster analysis. Beverly Hills, CA: Sage.
Arnold, H. J. 1982. Moderator variables: A clarification of conceptual, analytical, and psycho-
metric issues. Organizational Behavior and Human Performance, 29: 143-174.
Arthur, J. B. 1992. The link between business strategy and industrial relations systems in Amer-
ican steel minimills. Industrial and Labor Relations Review, 45: 488-506.
Barnett, D. F., & Crandall, R. W. 1986. Up from the ashes: The rise of the steel minimill in the
United States. Washington, DC: The Brookings Institution.
Boudreau, J. W., & Berger, C. J. 1985. Decision-theoretic utility analysis applied to employee
separations and acquisitions. Journal of Applied Psychology, 70: 581-612.
Butler, J. E., Ferris, G. R., & Napier, N. K. 1991. Strategy and human resources management.
Cincinnati: South-Western.
Campion, M. A. 1989. Ability requirement implications of job design: An interdisciplinary
perspective. Personnel Psychology, 42: 1-24.
Cappelli, P., & Singh, H. 1992. Integrating strategic human resources and strategic management.
In D. Lewin, O. S. Mitchell, & P. D. Sherer (Eds.), Research frontiers in industrial relations
and human resources: 165-192. Madison, WI: Industrial Relations Research Association.
Cooke, W. N. 1992. Product quality improvement through employee participation: The effects
of unionization and joint union-management administration. Industrial and Labor Rela-
tions Review, 46: 119-134.
Cotton, J. L., & Tuttle, J. M. 1986. Employee turnover: A meta-analysis and review with impli-
cations for research. Academy of Management Review, 11: 55-70.
Dalton, D. R., & Todor, W. D. 1979. Turnover turned over: An expanded and positive perspec-
tive. Academy of Management Review, 4: 225-235.
Dean, J. W., & Snell, S. A. 1991. Integrated manufacturing and job design: Moderating effects of
organizational inertia. Academy of Management Journal, 34: 776-804.
Dess, G. G., & Davis, P. S. 1984. Porter's (1980) generic strategies as determinants of strategic
group membership and organizational performance. Academy of Management Journal,
27: 467-488.
Dobbins, G. H., Cardy, R. L., & Carson, K. P. 1991. Examining fundamental assumptions: A
contrast of person and system approaches to human resource management. In K. Roland &
G. Ferris (Eds.), Research in personnel and human resources management, vol. 9: 1-38.
Greenwich, CT: JAI Press.
Dyer, L., & Holder, G. W. 1988. Toward a strategic perspective of human resource management.
In L. Dyer (Ed.), Human resource management: Evolving roles and responsibilities:
1-46. Washington, DC: Bureau of National Affairs.

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
686 Academy of Management Journal June

Eisenhardt, K. M. 1985. Control: Organizational and economic approaches. Management Sci-

ence, 31: 134-149.
Fisher, C. D. 1989. Current and recurrent challenges in HRM. Journal of Management, 15:
157-180.
Freeman, R. B., & Medoff, J. L. 1984. What do unions do? New York: Basic Books.
Hogan, W. T. 1987. Minimills and integrated mills: A comparison of steelmaking in the
United States. Lexington, MA: Heath.
Hollenbeck, J. R., & Williams, C. R. 1986. Turnover functionality versus turnover frequency: A
note on work attitudes and organizational effectiveness. Journal of Applied Psychology,
71: 606-611.
International Trade Administration. 1986. Growth patterns of the U.S. minimill steel industry.
Washington, DC: U.S. Government Printing Office.
Jackson, S. E., Schuler, R. S., & Rivero, J. C. 1989. Organizational characteristics as predictors of
personnel practices. Personnel Psychology, 42: 727-786.
Katz, D. 1964. Approaches to managing conflict. In R. L. Kahn & E. Boulding (Eds.), Power and
conflict in organizations: 105-114. New York: Basic Books.
Katz, H. C., Kochan, T. A., & Weber, M. 1985. Assessing the effects of industrial relations and
quality of working life on organizational effectiveness. Academy of Management Journal,
28: 509-526.
Kochan, T. A., Katz, H. C., & McKersie, R. B. 1986. The transformation of American industrial
relations. New York: Basic Books.
Lawler, E. E., III. 1986. High involvement management. San Francisco: Jossey-Bass.

Lengnick-Hall, C. A., & Lengnick-Hall, M. L. 1988. Strategic human resource management: A

review of the literature and a proposed typology. Academy of Management Review, 13:
454-470.
Levine, D. I., & Tyson, L. D. 1990. Participation, productivity, and the firm's environment. In
A. S. Blinder (Ed.), Paying for productivity: A look at the evidence: 183-243. Washing-
ton, DC: Brookings Institution.
Locke, E. A., & Schweiger, D. M. 1979. Participation in decision-making: One more look. In
B. M. Staw (Ed.), Research in organizational behavior, vol. 1: 169-339. Greenwich, CT:
JAI Press.
Mahoney, T. A., & Deckup, J. R. 1986. Evolution of concept and practice in PA/HRM. Journal of
Management, 12: 223-241.
Majchrzak, A. 1988. The human side of factory automation. San Francisco: Jossey-Bass.
Miles, R. E., & Snow, C. C. 1984. Designing strategic human resource systems. Organizational
Dynamics, 31(1): 36-52.
Miller, D., & Friesen, P. H. 1984. Organizations: A quantum view. Englewood Cliffs, NJ: Pren-
tice-Hall.
Mobley, W. H. 1982. Some unanswered questions in turnover and withdrawal research. Acad-
emy of Management Review, 7: 111-116.
Organ, D. W. 1988. Organizational citizenship behavior: The good soldier syndrome. Lex-
ington, MA: Lexington Books.
Osterman, P. 1987. Choice of employment systems in internal labor markets. Industrial Rela-
tions, 26: 46-67.
Ouchi, W. G. 1979. A conceptual framework for the design of organizational control mecha-
nisms. Management Science, 25: 833-848.

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions
1994 Arthur 687

Porter, M. E. 1980. Competitive strategy. New York: Free Press.

Schuler, R. S. 1992. Strategic human resource management: Linking people with the strategic
needs of the business. Organizational Dynamics, 21(1): 18-31.
Schuler, R. S., & Jackson, S. E. 1987. Linking competitive strategies with human resource man-
agement practices. Academy of Management Executive, 1(3): 207-219.
Schwab, D. P. 1991. Contextual variables in employee performance-turnover relationships.
Academy of Management Journal, 34: 966-975.
Snell, S. A. 1992. Control theory in strategic human resource management: The mediating role
of administrative information. Academy of Management Journal, 35: 292-327.
Snell, S. A., & Dean, J. W. 1992. Integrated manufacturing and human resource management: A
human capital perspective. Academy of Management Journal, 35: 467-504.
Staw, B. M. 1980. The consequences of turnover. Journal of Occupational Behaviour, 1: 253-
273.
Thomas, K. W., & Velthouse, B. A. 1990. Cognitive elements of empowerment: An "interpre-
tive" model of intrinsic task motivation. Academy of Management Review, 15: 666-681.
Walton, R. A. 1985. From control to commitment in the workplace. Harvard Business Review,
63(2): 77-84.
Wright, P. M., & McMahan, G. C. 1992. Theoretical perspectives for strategic human resource
management. Journal of Management, 18: 295-320.

Jeffrey B. Arthur is an assistant professor of management at the Krannert Graduate

School of Management, Purdue University. He received his Ph.D. degree in industrial
relations from Cornell University. His current research interests include strategic hu-
man resource management and industrial relations patterns and transformations.

This content downloaded from 152.15.236.17 on Fri, 13 Sep 2013 21:30:36 PM

All use subject to JSTOR Terms and Conditions