8
ORGANIZATIONAL BEHAVIOR AND HUNIAN PERFOR1V~AINCE 9, 369-376 (1973) Further Discussion of Fiedler's Contingency Model of Leadership Effectiveness AHMED SAKR ASHOUR 1 Cali/ornia State University, Fresno Fiedler's claim that his model is empirically valid was refuted on methodological and empirical grounds. The analysis of cumulative results, using Fisher's method of combining correlations, indicated that the model failed the validity test in six of the eight situational octa~ts. Further analysis indicated, contrary to F~edler's claim, that the model and its re- lated research have serious theoretical and methodological flaws. The principal thrusts of Professor Fiedler's (1973) comments on my paper (Ashour, 1973) consist of the following: (1) a claim that the contingency model is empirically valid, (2) a defense of the methodology prescribed by the model, and (3) an assertion that the model is theo- retically adequate. I shall discuss each of these points. Empirical Validity and Criteria o] Testing the Model Fiedler argues that the binomial test provided a reasonable test of the empirical validity of the model based on the number of correlations in the predicted direction. This argument is in sharp contrast to the rules of empirical testing of deductive inference (Meehl, 1967; Popper, 1965). These rules suggest that it is not so much the number of cor- roborating instances which determines the degree of corroboration of a theory as the severity of the various tests to which the theory in question can be, and has been, subjected (Popper, 1965, p. 267). Furthermore, the use of the binomial test requires independence of observations. This requirement was not fulfilled, since Fiedler counted results of different octants as independent instances. We cannot consider results of different oetants generated from the same study as independent. Fiedler, however, recognized that the binomial test was not the method of choice, and used: (1) a method of combining probabilities, (2) a method of averaging the z values, and (3) a method of correlating 1Requests for reprints should be sent to the author who is now at the Depart- ment of Business Administration, Faculty of Commerce, Alexandria University, Alexandria, Egypt. 369 Copyright O 1973 by Academic Press, Inc. All rights of reproduction in any form reserved.

1973 Further Discussion of Fiedler's Contingency Model of Leadership Effectiveness

Embed Size (px)

Citation preview

Page 1: 1973 Further Discussion of Fiedler's Contingency Model of Leadership Effectiveness

ORGANIZATIONAL BEHAVIOR AND HUNIAN PERFOR1V~AINCE 9, 369-376 (1973)

Further Discussion of Fiedler's Contingency Model of Leadership Effectiveness

AHMED SAKR ASHOUR 1

Cali/ornia State University, Fresno

Fiedler's claim that his model is empirically valid was refuted on methodological and empirical grounds. The analysis of cumulative results, using Fisher's method of combining correlations, indicated that the model failed the validity test in six of the eight situational octa~ts. Further analysis indicated, contrary to F~edler's claim, that the model and its re- lated research have serious theoretical and methodological flaws.

The principal thrusts of Professor Fiedler 's (1973) comments on my paper (Ashour, 1973) consist of the following: (1) a claim tha t the contingency model is empirically valid, (2) a defense of the methodology prescribed by the model, and (3) an assertion tha t the model is theo- retically adequate. I shall discuss each of these points.

Empirical Validity and Criteria o] Testing the Model

Fiedler argues tha t the binomial test provided a reasonable test of the empirical val idi ty of the model based on the number of correlations in the predicted direction. This argument is in sharp contrast to the rules of empirical testing of deductive inference (Meehl, 1967; Popper, 1965). These rules suggest tha t it is not so much the number of cor- roborat ing instances which determines the degree of corroboration of a theory as the severity of the various tests to which the theory in question can be, and has been, subjected (Popper, 1965, p. 267). Furthermore, the use of the binomial test requires independence of observations. This requirement was not fulfilled, since Fiedler counted results of different octants as independent instances. We cannot consider results of different oetants generated from the same study as independent.

Fiedler, however, recognized tha t the binomial test was not the method of choice, and used: (1) a method of combining probabilities, (2) a method of averaging the z values, and (3) a method of correlating

1 Requests for reprints should be sent to the author who is now at the Depart- ment of Business Administration, Faculty of Commerce, Alexandria University, Alexandria, Egypt.

369 Copyright O 1973 by Academic Press, Inc. All rights of reproduction in any form reserved.

Page 2: 1973 Further Discussion of Fiedler's Contingency Model of Leadership Effectiveness

3 7 0 AHMED SAKR ASHOUR

median correlations of original studies with median correlations of validation studies to demonstrate that the model is empirically valid.

A closer examination of Fiedler's use of the method of combining probabilities, and the method of combining z values reveals serious methodological pitfalls. First, combining results belonging to different oetants tends to obscure the differences in the predictive power of the model among the eight oetants. Second, the procedure allows oetants that have more coefficients to carry a greater weight in the final com- posite than those with fewer coefficients. Third, the results of the dif- ferent oetants when combined in the manner used by Fiedler cannot form any meaningful composite; the model postulated these oetants to be different and made different predictions for each one of them. Fourth, the procedure is misleading, since it is possible to have significant results in only one or two oetants that have more coefficients and high enough probability values to bias the overall evaluation and lead to the incorrect conclusion that the model as a whole is empirically valid.

Fiedler states: "the requirement that the combined probability of the correlations within every one of the oetants be significant i s . . . untenable. The median correlation for oetant VII was, for example, .0.5 in the original studies. Why should we not, for example, expect similarly non- significant correlations in validation studies?" (Fiedler, 1973, p. 359). This argument is rather interesting. First, it implies that the model is not capable of passing the empirical test of the predictions it offered for each oetant. Second, to have a nonsignificant and trivial prediction (proportion of variance explained = .0025) in oetant VII based on original results is an indication that the prediction of a positive cor- relation for that oetant was not empirically justified in the first place.

Fiedler's procedure of correlating median correlations of original studies with median correlations of validation studies was as inap- propriate as his procedure of combining results of different oetants. The procedure of correlating medians ignores ~he variability of the correla- tions about the medians and makes no provision for the statistical re- liabilities of the medians correlated. Furthermore, the procedure obscures the differences in the predictive power of the model among the oetants.

A Validity Test Based on Cumulative Results

The most rigorous and direct method of combining correlations ob- tained from small samples has been suggested by Fisher (1946, p. 204). The method provides an overall estimate of the correlation based on correlations obtained from different samples. The estimated value could then be subjected to a test of significance, and this would be equivalent to testing the significance of cumulative results of the samples included.

Page 3: 1973 Further Discussion of Fiedler's Contingency Model of Leadership Effectiveness

FURTHER DISCUSSION ON FIEDLER~S MODEL 371

Fisher's method and the operational procedure suggested by DuBois (1965, pp. 337-340) were applied to: (1) the results of validation studies that adhered to the methodology prescribed by the model (Table 2, Ashour, 1973), and (2) the results of validation studies that did not adhere to the prescribed methodology but were considered by Fiedler to provide further support to his model (Table 7, Fiedler, 1971). Table 1 shows the composite results derived by the use of this method for each situational condition separately.

The results in Section A of Table 1 show that only for oetants I and IV of the eight octants proposed by the model were the composite correlations of all studies significant. The cumulative results of these studies failed to provide significant support to the model in octants II, III, V, VI, VII, and VIII. The results of laboratory studies were similar to those of all studies. The results of field studies provided even less support.; the cumulative results of these studies were significant only in oetant I. The proportions of variance explained were provided to indicate the practical significance of the composite correlations obtained. In octant I, for example, where the eomposite correlation based on all studies was significant at the .01 level, the explained variance was 35% leaving 65% of the variance unaccounted for. The proportions of variance explained in most of the octants under whatever classification of the studies were of negligible praetical significance (over 90% of the variance was unaccounted for). These are the data upon which Fiedler claimed emi~irical validity of his model!

Fiedler also argues that. studies not. adhering to the prescribed method- ology of the model provide further support for the model and indicate that. the model is not. as specific as I asserted (Fiedler, 1973, p. 358). An examination of the cumulative results of these studies in Section B of Table 1 reveals that they are not any better than those in Section A. Out of the nine conditions of situational favorableness, only conditions I and IV had statistically significant results. The cumulative results failed to provide significant support to the model in conditions II, III, V, VI, VII, VIII, and IX.

The results indicate that. the hypothesis that situational conditions moderate the relationship between LPC and group effectiveness holds only for two of the eight conditions prescribed by the model. Cumulative evidence such as this suggests, contrary to Fiedler's claim, that the model for the most part is not empirically valid. An identification of the factors behind this poor showing requires a critical examination of the research procedure the model prescribes and the theoretical foundation of its hypothesis. This is a necessary step toward improving research and theory in this area.

Page 4: 1973 Further Discussion of Fiedler's Contingency Model of Leadership Effectiveness

TA

BL

E

1 C

UM

UL

AT

IVE

I~

ESU

LT

S O

F ST

UD

IES

TE

ST

ING

T

HE

C

ON

TIN

GE

NC

Y

MO

DE

L

A.

STU

DIE

S A

DH

ER

ING

TO

TH

E

I°R

ESC

RIB

ED

M

ET

HO

DO

LO

GY

--4

bo

Octa

nts

Stu

die

s I

II

III

IV

V

VI

VII

V

III

All

stu

die

s C

orre

lati

ons

-.

59 b

.02

-.

16

.40

~ .1

9 .0

8 .1

9 --

. 26

P

rop

ort

ion

s of

.3

5 .0

0 .0

3 .1

6 .0

4

.01

.04

.07

var

ian

ce e

xpla

ined

F

ield

stu

die

s Co

rrel

atio

ns

--.

56 ~

--.

23

--.

15

.16

.21

.06

.38

.41

Pro

po

rtio

ns

of

.31

.05

.02

.03

.04

.00

.14

.17

var

ian

ce e

xpla

ined

L

abo

rato

ry s

~ud

ies

Cor

rela

tion

s --

. 62

c .1

0 --

. 16

.4

7 b

.18

.11

.03

.07

Pro

po

rtio

ns

of

.38

.01

.03

.22

.03

.01

.00

.0

0 v

aria

nce

exp

lain

ed

B,

STU

DIE

S D

EV

IAT

ING

FR

OM

TH

E

PRE

SCR

IBE

D

ME

TH

OD

OL

OG

Y

Sit

uat

ion

al

con

dit

ion

s (d

egre

es o

f fa

vo

rab

len

ess)

~

I II

II

I IV

V

V

I V

II

VII

I IX

All

stu

die

s C

orre

lati

ons

-.

36 b

-.

43

.16

.72

b

--.

16

-.

10

--.

49

-.

20

.02

Pro

po

rtio

ns

of

.13

.18

.03

.52

.03

.01

.24

.04

.00

var

ian

ce e

xpla

ined

Th

ese

are

the

degr

ees

of j

ud

ged

fav

ora

ble

nes

s of

th

e le

ader

ship

sit

uat

ion

. T

he

furt

her

to

the

left

, th

e m

ore

fav

ora

ble

th

e si

tuat

ion

(F

iedl

er,

1972

, p.

143

).

b p

< .0

1.

~p

<: .

05.

Page 5: 1973 Further Discussion of Fiedler's Contingency Model of Leadership Effectiveness

FURTHER DISCUSSION ON FIEDLER'S MODEL 373

Methodological Rigor

LPC. Fiedler (1973) reported results of test-retest reliability studies, most of which were above .70. Fiedler also capitalized on the high end of these results. What Fiedler failed to report., however, were the low test-retest reliabilities he included in his book (Ficdler, 1967, p. 48). These reliabilities were .57, .47, .31, and .41. Such results indicate that the LPC measure is not without reliability problems as Fiedler seemed to imply. Why capitalize only on the relatively high test-retest re- liabilities and the high correlation with external criterion? Why not also consider the test-retest reliability that was as low as .31 and the correlation with external criterion that was as low as .00?

As to the meaning of LPC, it has gone through four different, and competing, interpretations. The first of these interpretations suggested that LPC was a measure of emotional reaction to people with whom one cannot work (Fiedler, 1967, p. 47). The second was the interpretation of' relations vs task orientations (Fiedler, 1967, p. 60). The third was the cognitive complexity and differentiation interpretation (Fiedler, 1971, p. 129). The fourth and the latest interpretation (Fiedler, 1973) suggests that LPC is measuring goal hierarchy. This interpretation asserts my argument that LPC is not. interpretable independently of the situation. It is interesting to note that each of the four interpretations has been substantiated by meager and, in some instances, conflicting evidence. Thus, we have been provided with at. least four competing interpretations, each of which was at one time supported by Fiedler. We can then con- elude, given these four competing interpretations and the lack of solid empirical support for each, that the independent measure in the model is equivocall

Situational classification. Fiedler agrees with my criticism that the measures of the component variables of situational classification require substantial improvements. However, he ignores the question that I raised about the meaning of that classification. There is a Iack of empirical support to situational favorableness as an interpretation. Thus, the meaning of this composite variable is as equivocal as that of LPC.

Another problem of the situational classification is the question of what variables constitute that classification. The model prescribes that the component variables are leader-member relations, task structure, and position power. However, Fiedler and his associates have continued to use an inductive strategy in determining the situational classification in studies testing the model. Thus, in the Belgian Navy Study (Fiedlcr, 1966), task structure was deleted from the classification, and the median correlations were plotted separately for the struetnred and the un-

Page 6: 1973 Further Discussion of Fiedler's Contingency Model of Leadership Effectiveness

374 A H M E D SAKR ASHOUR

structured tasks. In effect, this procedure implies that task structure is not part of the situational classification. In another study (Fiedler, O'Brien & Ilgen, 1969'), group atmosphere was assessed, but position power and task structure were not systematically measured a priori.

In most of the studies, the measure of leader-member relations and of position power have been dichotomized on the basis of an average score derived from the sample itself. Such a procedure tends to create different cutting scores for different studies and to produce inconsistency in the "critical values" of the situational dimension. The fact that "critical values" of the situational dimension exist and have been recently utilized, as Fiedler pointed out, does not invalidate my assertion that the determination of these values a priori is important, and that most of the studies on the contingency model have determined these values on an ad hoc basis.

Samples and research settir~g. Fiedler is not correct in underrating the sampling problems of research on the contingency model. The limited representation of octants, characterizing most studies on the model, has increased the chance that the results obtained might, have been con- laminated by sample and setting characteristics other than those pre- scribed by the model. Full representation of octants in each study helps to control the extraneous variables. This is a critical methodological requirement that should guide the researcher in his choice of sample and research setting. Apparently, original studies as well as later studies have compromised greatly on this requirement for the sake of con- venience, but at the expense of research rigor.

Fiedter underrates the effect of sample size on the precision and clarity of research results. There is an inverse relationship between sample size and the difficulty of arriving at correct conclusions based on sample results. The smaller the sample, the larger the errors of type I and type II, and, consequently, the greater the diffficuRy in correctly interpreting the research results; small samples increase the ambiguity of research results. Thus, as long as small samples remain a "way of life" in leadership research, as Fiedler asserts, ambiguity and error also remain to characterize the results obtained in this area.

Theoretical Adequacy

Fiedler considers the contingency model to be theoretically adequate based on his claim that the model has the ability to make valid and nontrivial predictions. I t is obvious from the analysis of cumulative results (see Table 1) that his claim is not fully supported by empirical evidence.

The contingency model does not fulfill another criterion of theoretical

Page 7: 1973 Further Discussion of Fiedler's Contingency Model of Leadership Effectiveness

FURTtIER DISCUSSION ON FIEDLER'S MODEL 375

adequacy. Not only does a theory have to o#er valid and nontrivial predictions, but it also has to provide meaning]ul explanations of the relations and predictions it proposes. The contingency model merely suggests a set of relationships without offering meaningful explanations of them. This is evident in the failure of the model to provide the be- havioral linkages and conceptual explanations that tie group performance outcomes to the independent and moderator variables in the model. The model also fails to provide valid conceptual constructs to clarify the meaning of LPC and the situational classification. The model is an empiric generalization that. is not even fully supported by empirical results; it is not a theory. When it provides explanations of valid and nontrivial predictions and' also provides valid and meaningful conceptual constructs of the operational measures it prescribes, it would then qualify as a theory.

Fiedler also claims that his model is not a static one and provided evidence that it applies to a system undergoing change. Thus, according to Fiedler, the amount of training and experience the leader acquires alters the situational favorableness, and, hence, leads to a change in group performance criteria. Does the addition of a new variable, leader's training, make the model a dynamic model? A dynamic model is one that accounts for: (1) feedback effects of the dependent variables on the independent variables and the intervening (moderator) variables as they take place over time, (2) long-range interactions among the independent variables and also their effect on one another, and (3) change in the relationships between the independent and the dependent, variables and in the nature of the variables themselves. Where does the contingency model stand with respect to these features of a dynamic model? We do not know the effect of group performance feedback on leadership style, or on situational variables. We do not know the longitudinal effect of position power, for example, on leader-member relations. We do not know how lasting or stable is the performance of groups working under different leadership styles and different, situational conditions. The model has to provide valid answers to these and similar questions to be con- sidered as a dynamic model. To say that leader's training alters situa- tional favorableness and group performance would not add a dynamic feature to the model unless that training were derived from leader- situation interaction over time and were assessed through a longitudinal study. Apparently, longitudinal designs have not yet been incorporated in research on the contingency model.

Other theoretical limitations of the contingency model are: (1) the model and its related research deal only with small interacting groups, and (2) the medel omits the important dependent variable of group and

Page 8: 1973 Further Discussion of Fiedler's Contingency Model of Leadership Effectiveness

376 A H M E D S A K R A S H O U R

individual satisfaction as a criterion of leadership effectiveness; it is a classical model in this respect.

In conclusion, Fiedler 's contingency model and its related research have serious empirical, methodological, and theoretical problems. An a t t empt to improve the empirical val idRy of the model and its meaning- fulness would have to s tar t with solving the methodological and theo- retical problems pointed out in the previous paper (Ashour, 1973) and in this paper.

REFERENCES

ASHOUR, A. S. The contingency model of leadership effectiveness: An evaluation. Organizational Behavior and Human Per]ormance, 1973, 9, 339-355.

DuBols, P. H. An Introduction to Psychological Statistics. New York: Harper & Row, 1965.

FIEDLER, F. E. A Theory of Leadership Effectiveness. New York: McGraw-Hill, 1967.

FIEDLE~, F. E. The contingency model: a reply to Ashour. Organizational Behavior and Human Per]ormance, 1973, 9, 356-368.

FIED~ER, F. E. The effect of leadership and cultural heterogeneity on group per- formance: A test of the contingency model. Journal o] Experimental Social Psychology, 1966 2, 237-264.

FmDL~R, F. E. Validation and extension of the contingency model of leadership effectiveness: A review of empirical findings. Psychological Bulletin, 1971, 76, 128-148.

FIEDLSR, F. E., O'BRIE~, G. E., & IL~EN, D. R. The effects of leadership style upon the performance and adjustment of volunteer teams operating in stressful foreign environment. Human Relations, 1969, 22, 503-514.

FISHER, R, A. Statistical Methods ]or Research Workers. London: Oliver and Boyd, 1946. Tenth edition.

ME~L, P. E. Theory testing in psychology and physics: A methodological paradox, Philosophy o] Science, 1967, 34, 103-115.

PoweR, K. R. The Logic of Scientific Discovery. New York: Harper & Row, 1965.

RECEIVED: August 11, 1972