Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Journal of Applied Psychology Copyright 1985 by the American Psychological Association, Inc.

1985, Vol. 70, No. 4, 777-781 0021-9010/85/$00.75

Role of Leadership in the Employee Withdrawal Process:


A Constructive Replication
Gerald R. Ferris
Department of Management,
Texas A&M University

The contributions of average and dyadic leadership style to explaining variance


in employee turnover were examined in an investigation designed to constructively
replicate the study by Graen, Liden, and Hoel (1982). The results showed leader-
member exchange (dyadic leadership style) to be a stronger predictor of turnover
than average leadership style, although the magnitude of these results was not as
strong as in the Graen et al. study. Also, dyadic leadership style predicted turnover
better than did employee attitudes, even though employee attitudes seemed to
mask the dyadic leadership style-turnover relation.

Efforts to more clearly articulate the determi- (averaged) leadership style components. Katerberg
nants of turnover continue to highlight the impor- and Horn (1981) employed such an approach and
tance of employee attitudes (e.g., Mobley, Griffeth, found that the within-unit component more
Hand, & Meglino, 1979). Mobley (1982) discussed strongly predicted reactions than did the between-
the important role of the immediate supervisor in unit component, although both components' pre-
turnover management and called for research that dictive power were significant. They called for
moves beyond the traditional focus on leader- more research that examined leader behavior com-
satisfaction ratings and begins to address the dyadic ponents on nonattitudinal dependent variables.
relationship between leader and subordinate. Additionally, Mobley (1982) recently cautioned
The investigation of the leadership-subordinate that much turnover research has confused and
turnover relation is not a recent phenomenon (e.g., crossed levels of analysis by using grouped or
Fleishman & Harris, 1962). However, the way that aggregated data (e.g., job satisfaction mean scores)
leadership is measured and subsequently related to predict individual employee turnover. In order
to outcomes has been the source of considerable to understand, in a more informed sense, which
discussion. One approach suggests a between- individuals will leave or stay, individual-level data
groups focus assuming the same behavior toward needs to be used in the prediction.
all subordinates in a given group or unit, so that In a study that addressed both of these concerns,
an average leadership style score would be assigned Graen, Liden, and Hoel (1982) investigated the
to each group member. Another approach contends role of leader behavior in the employee turnover
that the appropriate level of analysis is the dyadic process with both within-unit (dyadic) and be-
relationship between leader and subordinate, tween-unit (averaged) leadership components. Their
adopting the assumption that leaders vary their results supported the superiority of dyadic lead-
behavior across subordinates (e.g., Dansereau, ership style over averaged leadership style in pre-
Graen, & Haga, 1975; Graen, 1976). Although dicting turnover of employees.
research exists to support each position, few com- The present study was designed to constructively
petitive tests have been conducted on the same replicate that investigation, examining the effec-
data set to permit stronger conclusions to be tiveness of different leadership components in the
drawn concerning leader behavior. prediction of employee turnover. Although there
Dansereau and Dumas (1977) suggested a vari- are some differences in the measurement methods
ance-partitioning procedure that would permit the used in this study compared to Graen et al., the
examination of both within (dyadic) and between main difference is with respect to the sample
employed.

The author would like to thank Fred Dansereau, Method


George Graen, Frank Landy, and two anonymous review-
ers for their helpful comments on an earlier draft of this Sample
note.
Requests for reprints should be sent to Gerald R. Eighty-one registered nurses and their supervisors
Ferris, Department of Management, Texas A&M Univer- drawn from all major nursing services subunits of
sity, College Station, Texas 77843. a midwestern hospital comprised the sample for
777
778 SHORT NOTES

this study. All were female, the average age was Employee attitudes. The Job Descriptive Index
about 34 years, and the median tenure in the (JDI; Smith, Kendall, & Hulin, 1969) was used to
organization was 2.6 years, with a range of 1 measure employee attitudes toward select facets
month to 32.25 years. A restriction that a unit or aspects of the job. Although the only JDI
consist of at least three subordinates in order to component that has conceptual relevance to the
obtain acceptable estimates of average leadership Vertical Dyad Linkage (VDL) theory is satisfaction
style (see Fleishman & Harris, 1962) was imposed with supervision, four satisfaction components from
and reduced the sample to 68 nurses from 18 of the JDI were examined consistent with the Graen
the 20 different units in the hospital. et al. (1982) investigation. Graen et al. used the
four JDI subscales of Work, Supervision, Pay, and
Promotion as revised for professional-level positions
Procedure (see Dansereau et al., 1975), whereas the present
Questionnaires were administered by a re- study simply used the original, unmodified JDI
searcher to all nurses at the end of their work scales. Coefficient alphas for the four JDI scales
shifts. Data were collected from nurses on the day were as follows: Work = .59, Supervision = .79;
shift (7:30 a.m.-3:30 p.m.) and evening shift (3: Pay = .79; Promotion = .85.
30 p.m.-11:30 p.m.). Subordinate reaction to the overall job situation
A 1-year follow-up found that 9 of the 68 staff was assessed with a new JDI scale, Satisfaction
nurses had left the organization. All departures with the Job in General. This scale was developed
were voluntary on the part of the employee. by Smith and her associates as somewhat of a
Because performance data on nurses was not made composite measure of affective reactions toward
available, it was not possible to determine the the entire job situation. Furthermore, it was de-
extent to which turnover was functional or dys- signed to circumvent potential psychometric dif-
functional. Furthermore, it should be noted that ficulties encountered when researchers sum the
the turnover rate in this study (about 13%) differs five JDI scales to devise an overall measure of
substantially from the turnover rate in the Graen satisfaction with the job (a description of this
et al. study (about 42%). scale's development and psychometric properties
is in progress). Coefficient alpha for this scale was
.89. Graen et al. (1982) used the Hoppock Job
Measures Satisfaction Blank (Hoppock, 1935) as an overall
Leader-member exchange (LMX). The five- measure of satisfaction.
item leader-member exchange scale used by Graen
et al. (1982) was employed to assess the quality of Analysis
supervisor-subordinate interaction. A 1 to 5 re-
Leader-member exchange variance was sepa-
sponse format was used instead of the 1 to 4 scale
rated into an average component and a deviation
used in the Graen et al. study. Thus, in summing
from average component in order to examine the
the five items for each subordinate, the possible
relative contribution of between-unit (averaged)
range of scores extended from 5 to 25. Several
variance and within-unit (individualized) variance
other measures were collected using 5-point scales
to explaining turnover. The deviation score was
(collected as part of a larger survey and thus not
calculated as LMX score minus the ALS (average
relevant to the present study), and, thus, a 1 to 5
leadership) score. Thus, as indicated by Graen et
response format was used instead of the 1 to 4
al. (1982), an inverse relation would indicate that
scale employed by Graen et al. primarily out of
subordinates with LMX scores above the unit
concern for maintaining symmetry across scales.
average tended to leave less often than those with
The first, third, and fifth scale points were anchored
scores below the unit average. Within and Between
with descriptions similar to those reported by
Analysis (WABA) was conducted according to the
Graen et al. The remaining two scale points were
procedures discussed by Dansereau, Alutto, and
not anchored. The change in format, however,
Yammarino (1984), in order to more specifically
reflected respectable internal consistency reliability
investigate within- and between-unit variance rel-
with a coefficient alpha of .83, which was not
ative to turnover. Also, employee attitudes were
dissimilar from the coefficient alpha of .80 reported
correlated with turnover while holding constant
by Graen et al.
LMX, and the LMX-turnover relation was ex-
Average leadership style (ALS). Average lead-
amined, partialing out employee attitudes.
ership style was assessed in the same manner as
in the Graen et al. (1982) study. The LMX ratings Results
of all subordinates reporting to a particular super-
visor were averaged, and that average was assigned Table 1 shows the relations of LMX, ALS, and
to each group member as their ALS score. the deviation of LMX from ALS with turnover.
SHORT NOTES 779

Table 1
Correlations of Leadership Components With Turnover f'N = 68)
LMX scores ALS Deviation LMX

Component Score Var. Score Var. Score Van


Average Leadership Style (ALS)
Present study .68*** .46
Graenetal. (1982) .48* .23
Deviation LMX
Present study .73*** .53 .00 .00
Graenetal. (1982) .89*** .79 .02 .00
Turnover -.19 .04 -.11 .01 -.25* .06
-.44** .19 -.22 .05 -.38** .14
Note. N = 48 for results reported by Graen et al.; Var. = proportion of common variance values (coefficient of deter-
mination); LMX = leader-member exchange.
* p<, .05. **/><; .01.*** p&.OOl.

As can be seen, there was mixed support for the the extent to which a single set of scores is most
findings reported by Graen et al. (1982). Although influenced or best represented by within- or be-
the pattern of results relative to the directionality tween-unit scores. The between-unit eta ()? = .68)
of relation is consistent, the magnitude of the and within-unit eta (n = .73) for LMX suggest
obtained relation only partially supports the pre- that variability is reasonably high both within and
vious investigation. between units. In fact, Dansereau et al. suggest
The present results demonstrated that LMX calculation of an E ratio, which is simply the
and ALS explained 4% and 0% of the variance in between-unit eta correlation divided by the within-
turnover, respectively, compared to 19% and 5% unit eta. They claim that an E ratio is equal to
accounted for in the Graen et al. study. In both 1.00 when between- and within-unit variation
studies, the deviation leader-member exchange ex- contributes equally to the total deviations. The E
hibited significant relations with subordinate turn- ratio in the present case was .92. Adjusting this
over, both suggesting that subordinates with LMX ratio for degrees of freedom, an F test provides a
scores higher than the group average tended to more statistical perspective on the ratio of within-
leave less frequently than those with LMX scores and between-unit variance. Using N — J (i.e., total
below the group average. sample size minus the number of units or groups)
The predictive power of the LMX and ALS and J - \ degrees of freedom, a significant F ratio
measures with respect to turnover was further was obtained, F(50, 17) = 2.49, p < .05, suggesting
examined through the development of a hit-rate that even though both within- and between-unit
table, creating high and low groups in the same
manner as did Graen et al. (1982) by using a
median-split procedure. The hit-rate analysis is Table 2
presented in Table 2. Hit-Rate Analysis for Leader-Member Exchange
The results of the hit-rate analysis for LMX and and Average Leadership Style in the
ALS in the prediction of turnover do not strongly Prediction of Turnover
support LMX. Although the hit rate for LMX
approaches statistical significance, x20)> N = 68 Prediction
2.87, p < . 10, the results do not appear substantially
different from those for ALS. Leave Stay
WABA was conducted according to the rationale Leadership Hit
and procedures discussed by Dansereau et al. component Hit Miss Hit Miss rate" x20)
(1984) to elaborate on the analyses reported by
Leader-member
Graen et al. (1982). Specifically, one issue that exchange 7 28 31 2 56% 2.87*
emerges is whether there is sufficient within- and Average-leader-
between-unit variance on leadership to permit an ship style 6 28 31 3 54% 1.15
adequate test of the two models. The within- and
1
between-unit eta correlations, in \\ABA, indicate Base rate = 87%. * p < .10.
780 SHORT NOTES

Table 3
Zero-Order and First-Order Partial Correlations With Turnover
Zero-order First-order partial
correlation correlation

Present Oraen et al. Present Graen et al.


Predictor study (1982) Controlling study (1982)

LMX -.19 .44**


LMX JDI-Supervision -.27* -.33**
LMX JDI-Work -.20* -.41**
LMX JDI-Promotion -.27* -.41**
LMX JDI-Pay -.19 -.35**
LMX JDI-Job in General -.15 -.39**
JDI-Supervision .00 LMX .19 -.07
JDI-Work -.05 LMX .06 -.11
JDl-Promotion .06 LMX .20* -.13
JDI-Pay -.04 LMX .04 -.29*
JDI-Job in General -.14 LMX -.06 -.18

Note. N = 48 for results reported by Graen et al.; LMX = Leader-member exchange, Graen et al. (1982) did not use
the Job Descriptive Index-Job in General scale as a overall measure of satisfaction with job situation. They used the
Hoppock Job Satisfaction Blank (Hoppock, 1935).
* p<, .05. **p<; .01.

variance were reasonably high, there seemd to be each of the employee attitude scales, the LMX-
evidence of greater between-unit variability. turnover relations remained statistically significant
Another issue, which was not addressed by and only slightly diminished in all five cases, as
Graen et al. (1982), was the extent to which can be seen in Table 3 (the values in parentheses).
between-unit and within-unit correlations account In the present study, however, the nonsignificant
for a particular total correlation (i.e., between zero-order correlation between LMX and turnover
LMX and turnover) and the difference between was elevated to statistical significance (p < .05) in
these two correlations. Graen et al. calculated three of the five cases when first-order partial
within- and between-group scores but they neither correlations were computed, holding constant em-
conducted statistical tests of the difference between ployee attitudes. Not only did partialing out JDI-
such correlations nor did they specify the number Supervision not substantially reduce the LMX-
of units or groups on which their study was based turnover correlation, as might be expected (but
so that such tests could be conducted. Within- not found to be the case by Graen et al. 1982), it
and between-unit correlations reflect the relation actually inflated the relation to r = —.27, p < .05
of LMX and turnover based on either within-unit (from a zero-order correlation of r = —.19, ns).
deviation scores or between-unit deviations. In this Similar results were obtained when LMX was
case, the within-unit correlation (r = —.25) is larger partialed out of the employee attitudes-turnover
than the between-unit correlation (r = —.11), relations.
indicating that the within-unit deviations serve as In the present case, several of the employee
a better predictor of turnover than deviations attitudes (i.e., JDI scales) apparently served to
between units. The difference between these cor- mask the relation between LMX and turnover,
relations, however, is not statistically significant which actually brought the obtained values of the
(Z = .49, ns). first-order partial correlations closer in magnitude
The zero-order and first-order partial correlations to the findings reported by Graen et al. (1982).
of LMX, employee attitude scales, and turnover Leader-member exchange and each of the em-
appear in Table 3. The first set of partial correla- ployee attitudes were highly correlated (r - .51),
tions examined the relation between LMX and with the highest relation found for LMX and JDI-
turnover while holding constant each of the satis- Supervision (r = .68, p < .001).
faction scales. The second set of partial correlations
held constant LMX in each of the attitude-turnover Discussion
relations. The results of these analyses appear
somewhat different from those reported by Graen The results of this study tend to support Graen
et al. (1982). In their study, after controlling for et al.'s (1982) finding that leader-member exchange
SHORT NOTES 781

is a stronger predictor of turnover than average Dansereau, F., & Dumas, M. (1977). Pitfalls and pratfalls
leadership style, even though the magnitude of the in drawing inferences about leadership behavior in
relations obtained in the present study are not as organizations. In J. G. Hunt & L. Larson (Eds.),
compelling, particularly as indicated by the hit- Leadership: The cutting edge (pp. 68-83). Carbondale:
Southern Illinois University Press.
rate analysis. Further support was provided for the
Dansereau, E, Graen, G., & Haga, B. (1975). A vertical
validity of the Graen et al. findings concerning dyad linkage approach to leadership within formal
LMX and employee attitude predictors of turnover. organizations. A longitudinal investigation of the role
Both studies found LMX to be superior in the making process. Organizational Behavior and Human
prediction of turnover than employee attitudes. Performance, 13, 46-78.
On balance, the present constructive replication Fleishman, E. A., & Harris, E. E (1962). Patterns of
seemed to provide support for and extend the leadership behavior related to employee grievances and
generality of Graen et al.'s findings concerning turnover. Personnel Psychology, 15, 43-56.
leader-member exchange and turnover. In light of Graen, G., (1976). Role-making processes within complex
the variance restriction on the criterion variable organizations. In M. D. Dunnette (Ed.), Handbook of
industrial and organizational psychology. Chicago: Rand
and thus the conservative nature of the present McNally.
tests, these results may be viewed as even stronger Graen, G., Liden, R., & Hoel, W. (1982). Role of
support for the Graen et al. findings. A point leadership in the employee withdrawal process. Journal
needs to be made, however, concerning the inter- of Applied Psychology, 67, 868-872.
pretation of the present results. The WABA sug- Hoppock, R. (1935). Job satisfaction. New York: Harper.
gested that both within- and between-unit variation Katerberg, R., & Horn, P. (1981). Effects of within-group
is reasonably high, but within-unit deviations served and between-groups variation in leadership. Journal of
as a stronger predictor of turnover than did be- Applied Psychology, 66, 218-223.
tween-unit deviations (although the difference be- Lykken, D. T. (1984). Statistical significance in psycho-
logical research. In T. S. Bateman & G. R. Ferris
tween the two correlations was not statistically (Eds.), Method and analysis in organizational research
significant). This suggests that perhaps the LMX- (pp. 54-64). Reston, VA: Reston Publishing Co.
turnover relationship is due more to individual Mobley, W. H. (1982). Employee turnover: Causes, con-
differences than to work group differences. Another sequences, and control. Reading, MA: Addison-Wesley.
potential explanation concerns a different perspec- Mobley, W. H., Griffith, R. W., Hand, H. H., & Meglino,
tive on dyads. It might be the case that supervisor- B. M. (1979). Review and conceptual analysis of the
subordinate dyads, in the present context, exhibited employee turnover process. Psychological Bulletin, 86,
variance both within and between work groups 493-522.
(Dansereau, 1984). Further research might consider Smith, P. C, Kendall, L. M., & Hulin, C. L. (1969). The
measurement of satisfaction in work and retirement.
more completely examining these notions. Chicago: Rand McNally.
References
Dansereau, E, Alutto, J. A., & Yammarino, F. J. (1984).
Theory testing in organizational behavior: The varient Received November 27, 1984
approach. Englewood Cliffs, NJ: Prentice-Hall. Revision received January 7, 1985

You might also like