You are on page 1of 7

Justification for the Carzo-Yanouzas Experiment on Flat and Tall Structures Author(s): Rocco Carzo, Jr.

and John N. Yanouzas Source: Administrative Science Quarterly, Vol. 15, No. 2 (Jun., 1970), pp. 235-240 Published by: Sage Publications, Inc. on behalf of the Johnson Graduate School of Management, Cornell University Stable URL: http://www.jstor.org/stable/2391495 . Accessed: 05/08/2013 07:56
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 support@jstor.org.

Sage Publications, Inc. and Johnson Graduate School of Management, Cornell University are collaborating with JSTOR to digitize, preserve and extend access to Administrative Science Quarterly.

http://www.jstor.org

This content downloaded from 122.170.126.130 on Mon, 5 Aug 2013 07:56:05 AM All use subject to JSTOR Terms and Conditions

Rocco Carzo, Jr., and John N. Yanouzas

Justification for the Carzo-Yanouzas Experiment on Flat and Tall Structures


This paper replies to Hummon's criticism of the Carzo-Yanouzas experiment. Justification is presented for the use of a split-plot repeated measures design, and tests for homogeneity support the use of this design. Variables which Hummon considered to be not under control are explained as attributes of structure or as variables having inconsequential effects. The laboratory experiment involving flat and tall structures (Carzo and Yanouzas, 1969) was one of a series of studies (Carzo, 1963; Yanouzas, 1963; Cain, 1965; Jones, 1969; Carzo and Millman, 1970) which have simulated structural attributes of real organizations.' One attribute which has received much attention is tallness or flatness, previous research on its effects having been done in the field and data having been gathered by observation, interview and questionnaire. A laboratory experiment on tallness and flatness was undertaken on the premise that cause and effect could be determined better under controlled conditions. Although some real world conditions were simulated, the study was constrained by the requirements of laboratory research. The laboratory setting forces researchers to abstract much from the real world situation. College students are often used as subjects and are asked to perform functions in which they have had little or no experience. The setting is unreal because the environment is controlled. The subjects know that they are being observed and that their performance is being measured and recorded. Weick (1969) has placed these problems in their proper perspective. They need emphasis, however, because Hummon claims
1 The authors are grateful to Professor Harsha B. Desai for his assistance in the preparation of this report.

that our study violated the requirement of isomorphism. For example, he criticizes the experimental requirement that the president make all final decisions, stating that this restriction prevented the benefits that accrue to flat organizations through forced delegation. Hummon is referring to Worthy's comment that top managers in a flat organization become overburdened and are forced to delegate decision-making authority to subordinates. When subordinates are forced to manage, they learn to manage. The organization benefits, subordinate managers perform better and develop in ways that are not possible under the tall structure. In fact, the experiment was designed so that forced delegation could develop. The requirement that the president make all final decisions is not inconsistent with forced delegation and it is not inconsistent with the Worthy model. All organizations have limited resources. Executives at lower levels may have decentralized authority, but there has to be some overriding authority to resolve conflicts between their decisions and overall goals of the organization. Also, there is need for an overriding authority to determine how available resources shall be allocated among operating subunits whose demands conflict. Even routine, repetitive decisions are made within constraints imposed by a higher authority. Thus, real organizations usually have some final decision maker such as a president, and this requirement was doubtless assumed in the arguments of Worthy (1950).

235

This content downloaded from 122.170.126.130 on Mon, 5 Aug 2013 07:56:05 AM All use subject to JSTOR Terms and Conditions

236

ADMINISTRATIVE SCIENCE QUARTERLY

If flat organization forces more delega- are repeated measures on experience and tion of authority than the tall structure nonrepeated measures on structure. The dedoes, more delegation was possible in the sign is illustrated in Table 1. The letters s1 experiment. Subjects could conduct affairs as they pleased, as long as the requirements of the task and structure were ful- TABLE 1. CARZO AND YANOUZAS TWO FACTOR REPEATED MEASURES DESIGN filled. Subjects did delegate work and decision-making authority. It cannot be Factor B claimed, however, that delegation in the experience Factor A experiment was comparable to real delega... structure b1 b3 b60 b2 tion in amount or in substance, because it Flat was delegation peculiar to the constrained structure a1 S SI SI SI SI and artificial conditions of a laboratory. Given these conditions, our paper did Tall structure a2 s2 s2 s2 s2 not transcend the bounds of the laboratory S2 by extrapolating to the real world. The discussion and conclusions were limited to the results of the experiment. or S2 refer to sets of groups; there were two groups in each set. Groups in set si are Experimental Design observed under all levels of experience and The most important of Hummon's criti- in only one level of structure. Similarly, cisms is the one concerned with statistical groups in set S2 are observed under all analysis of dependent observations. The levels of experience and in only one level problem of dependency arose from our of structure. desire to simulate the repetitive nature of In the analysis, the effects of structure organizational tasks and to discern the and experience are split or separated so learning curve behavior found in industrial that they become analogous to randomized firms. The decision-making task offered designs. This split is accomplished by first three measures of performance: time to partitioning the total variation to betweencomplete the task, profit, and rate of return cell and within-cell variation as a 2 x 60 on revenue. On each trial, subjects had to nonrepeated measures factorial experiment. estimate demand which was a random vari- Then, the within-cell variation is divided able, and order an amount that satisfied into two error terms: one error term is the demand and eliminated inventories or variation due to experimental error and to backorders. differences between groups having the same Repeated 60 times, this task tended to structure-error (a) in Table 2 (Carzo and produce correlation between observations, Yanouzas, 1969:186). The other error term or statistically dependent measurements, and is variation due to experimental error and this apparent dependency led Hummon to to interaction of experience with differences claim that the assumptions of analysis of between groups having the same structure variance had been violated. The authors -error (b) in Table 2 (Carzo and Yanouwere aware of this possibility and, therefore, zas, 1969:186). When F ratios are formed used a split-plot repeated measure design Msstructure/MSgroupg with same structure is the which accommodates dependent observa- same as MSbetween groups/MSwithin groups in a tions. The design is justified if the correlation completely randomized and design, between all pairs of observations on the same MSexperience/MSinteraction between and experience subjects is constant. Tests to support the groups with same structure is the same as use of this design are shown below. MSexperience/MSresidual in a randomized block design (Kirk, 1968:252). Rationale of the Design The split-plot repeated measures design In this experiment, there are 2 levels of requires more restrictive assumptions about homogeneity of population variances than the structure variable-flat and tall-and 60 levels of the experience variable. There does the F test of uncorrelated data. Ac-

This content downloaded from 122.170.126.130 on Mon, 5 Aug 2013 07:56:05 AM All use subject to JSTOR Terms and Conditions

Carzo and Yanouzas: FLAT AND TALL ORGANIZATIONS cording to Box (1953), violation of the homogeneity assumption does not seriously affect the significance test of uncorrelated data but does bias the F test of correlated data. Thus, it was necessary to question the homogeneity of population variances. The basic assumption of this model is that the 60 population variances of error on each trial are equal, and that the population covariances of errors between trials are equal. In the variance-covariance matrix, all diagonal elements must equal y2 and all off-diagonal elements must equal p02, where p is the constant correlation coefficient. The matrix has the form shown in Table 2.
TABLE 2. VARIANCE-COVARIANCE FOR SPLIT-PLOT MEASURES DESIGN
... ... .p2

237

MATRIX

REQUIRED

REPEATED

bi

b2
p12
g2...

b59
p12

b60
p12

bi b2
*.

12

pr2
. .

pg2
. . . . . ..

b59 b6o

p12 pG2

p12 pG2

...

G2 .p12 o

p2

...

Greenhouse and Geisser (1959) have developed an approximation procedure to the variance-covariance matrix requirement. The procedure still assesses the significance of F ratios with the ordinary F table, but the critical values are determined with reduced degrees of freedom. When the F ratios for experience and the structure-experience interactions were evaluated in our previous report, maximum independence was assumed: 59 degrees of
TABLE

freedom in the numerator and 118 degrees of freedom in the denominator. However, if the degrees of freedom are reduced to 30 and 60, to 15 and 30, or even to 7 and 14, the observed F ratios for all three variables are still significant at the 0.01 level. At the extremes suggested by Greenhouse and Geisser, where almost total dependence is assumed, the observed F ratios are significant at 0.25 level. The tests are shown in Table 3. Thus, even under conservative, negatively biased tests, the hypotheses are not contradicted. Error (a) and Error (b) used to form F ratios in the study represent pooled variations from different sources. "One of the assumptions required in order that the F ratio actually follow an F distribution is that these sources of variation be homogeneous" (Winer, 1962:305). A test for the assumption of homogeneity of error variances is provided by the Fmaxstatistic which is the ratio of the largest of these sources of variation to the smallest. The test applied to each of the error terms is shown in Table 4 (Pearson and Hartley, 1958). Results on rate of return should be similar to those obtained on profits. Since the observed Fmax is less than the critical value in both cases, the hypotheses that the population variances are homogeneous are not contradicted. Therefore, the sources of variation can be pooled to form the error terms. Experimental Variables Some of the hypotheses of our study were based on field research (Worthy, 1950; Meltzer and Salter, 1960; Porter and Lawler, 1964); some were based on theory derived from industrial practice; others were derived from laboratory research and our own knowONE AND TWO DEGREES OF FREEDOM

3.

HYPOTHESIS

TESTS ASSUMING

Hypotheses
c22=0 c92=0

Observed Critical value F.75 [1,2] =2.57 F.75[1,21 =2.57 Time F=5.32* F=1.10 Profit F=7.89* F-4.39* ROR F=7.92* F=4.45*

aSi 4 Significant.

This content downloaded from 122.170.126.130 on Mon, 5 Aug 2013 07:56:05 AM All use subject to JSTOR Terms and Conditions

238
TABLE 4.

ADMINISTRATIVE SCIENCE QUARTERLY


TESTS FOR HOMOGENEITY ON SOURCES OF VARIATION THAT MAKE UP ERROR TERMS

Error terms
Error (a)

Critical value
Fmax .95 [2, 1] > 39.00 since Fmax 95 [2, 2] = 39.00 Fmax .95 [2, 59] = 1.67

Observed ___________________________ Time 12.04


Fmax -= 4.46 F max

Profit 108.11
= 11.23

2.70

9.63

224.00 Error (b)


Fmax

839.13
=150 Fmax = 1.47

150.30

570.35

ledge of organizational behavior. Hummon criticizes the study on what he believes is a lack of isomorphism between the hypotheses and the experiment. He states: "They did not meet the requirement of isomorphism very well because they did not control variables not covered by the hypotheses and because one of their independent variables, organization experience, was not a true experimental variable." There appears to be an inconsistency with the indication in the above statement that organization experience is an independent variable but not a true experimental variable and a later statement by Hummon, "Independent variables in experiments are synonymous with experimental treatments." If independent variables are the same as experimental treatments and organizational experience was an independent variable, then it must have also been an experimental treatment. Assuming that experimental treatment is the same as experimental variable, then, organizational experience was an experimental variable. In the next sentence, he seems to question organizational experience as a treatment because there was no between-group variance on the variable. The absence of a between-group variation on one factor is the very essence of a repeated measure design. Winer (1962:299) wrote: between In this kind of experiment,comparisons treatment combinations at different levels of factor A involve differences between groups as well as differencesassociated with factor A. On the other hand, comparisonsbetween different levels of factor B at the same level of A do not involve differences between groups. Since measurementsincluded in the latter comparisonsare

based upon the same elements, main effects associated with such elements tend to cancel. Hummon says the experiment did not control: (a) physical arrangement of subjects, (b) information supplied to subjects, (c) number of sales areas, and (d) communication nets. In laboratory as well as other experiments, the ideal of holding all variables constant, except the ones under study, is not always achieved because of physical or human considerations. Preliminary studies of this experiment indicated, however, that these conditions which introduced uncontrolled variables produced an inconsequential effect, or that their effect could reasonably be considered as a structure effect. For example, subjects in the flat organization were seated equally distant from the president. The seating arrangement of the tall organization did not place all subjects equally distant from the president. These arrangements were chosen because they fortified the subjects' impression that they were part of a tall or flat structure. Another example was the problem of creating positions and task assignments for each structure. Performance and attitudes are affected by group size (Thomas and Fink, 1963; Hare, 1952; and Slater, 1958), but since the effects of size were not of interest, it was held constant. Each structure had 15 members. Also the experiment was intended to compare the extremes of flat and tall structure. This meant that the tall structure would incorporate a span of control of two and four levels, and the flat structure would incorporate a span of control of 14 and two levels. Given these structural attributes, the next problem was

This content downloaded from 122.170.126.130 on Mon, 5 Aug 2013 07:56:05 AM All use subject to JSTOR Terms and Conditions

Carzo and Yanouzas: FLAT AND TALL ORGANIZATIONS to determine the number of market areas and the number of specialists assigned to each area. No way was found to make each structure have the same number of market areas and the same number of specialists assigned to each area; the final arrangement was shown in Figure 1 (Carzo and Yanouzas, 1969:181). The tall structure also required the creation of coordinating and decision-making tasks for managers and vice presidents. They had to combine, summarize, and analyze data submitted by subordinates and recommend how much to order from suppliers. As a basis for these judgments, they could use probability distributions of combined market data, supplied by the researchers, as well as the recommendations made by subordinates. These distributions were not supplied to the members of the flat organization. Hummon claims that the two structures imposed different communication patterns because the "need to communicate," each to every other group member was "obviated" by the "structure and nature of the task." The communication net design of the experiment was similar to the all-channel net of Guetzkow and Simon (1955), that is, all channels were open and any subject could communicate with any other subject. It was in this sense that the authors claimed equivalence between the communication nets of flat and tall structures. However, the actual usage of channels and communication patterns were different for each structure, as Hummon claimed. The authors assumed that these differences and their possible effect on performance could reasonably be considered as a function of structure. Each of these differences between flat and tall groups were carefully considered before the experiment. Pilot studies were conducted to determine the impact of these differences. In each case, the effect was considered minor or attributable to structure. Seating arrangements, for example, were considered as a structural attribute. The probability distributions given to managers and vice presidents had little or no effect on performance, either in terms of quality or time taken to complete decisions.

239

The Model
Analysis of variance with a split-plot repeated measures design was used. In addition to the structure effect, it was hypothesized that subjects would improve performance as they gained experience with the task. The ANOVA model was used to discern the form or shape of these patterns and to determine whether there were significant differences between the performance profiles for each structure. Hummon claims that the model chosen for the experiment was inappropriate when compared with the graphs of the data. As he recognized, the relationships between the independent and dependent variables were nonlinear. These findings and the charts support the idea of a learning curve. In the manner of learning curve practitioners, the data were presented in cumulative average charts of Figure 4 (Carzo and Yanouzas, 1969:188). It was not intended to imply that the ANOVA model and the model of cumulative averages were comparable. The cumulative average charts were an added feature-another way of arranging the data. The primary analytic procedure was the ANOVA with repeated measures. A related question raised by Hummon was whether the ANOVA model was consistent with learning theories which assume that performance on one trial is dependent upon performance on previous trials. As noted earlier, the ANOVA repeated measures design does involve dependent observations and, therefore, is appropriate for learning experiments. Winer (1962:132-135), for example, uses a learning experiment to illustrate the repeated measures design and trend analysis. As a matter of interest, other references which present or show an application of the repeated measures design to learning experiments include: Kogan (1948), Grant (1948), and Arima (1965). Our concern with trend analysis was not to study learning per se, but rather to discern patterns of behavior. The purpose was to study these patterns as they occurred under different organization structures. The analysis of variance repeated measures design was appropriate to determine the nature

This content downloaded from 122.170.126.130 on Mon, 5 Aug 2013 07:56:05 AM All use subject to JSTOR Terms and Conditions

240

ADMINISTRATIVE SCIENCE QUARTERLY


Hare, A. P. 1952 "Interaction and consensus satisfaction in different sized groups." American Sociological Review, 17:261-267. Jones, Halsey R., Jr. 1969 "A study of organization performance for experimental structures of two, three, and four levels." Academy of Management Journal, 12:351-365. Kirk, Roger E. 1968 Experimental Design. Belmont, Cal.: Brooke/Cole. Kogan, Leonard S. 1948 "Analysis of variance-repeated measurements." Journal of Social Psychology, 45:131-143. Meltzer, Leo, and James Salter 1960 "Organizational structure and the performance and job satisfactions of physiologists." American Sociological Review, 27:351-362. Pearson, E. S., and H. 0. Hartley 1958 Biometrika Tables for Statisticians. Vol. 1, 2nd ed. New York: Cambridge. Porter, L. W., and E. E. Lawler, III 1964 "The effects of 'tall' versus 'flat' organization structures on managerial satisfaction." Personnel Psychology, 17: 135-148. Slater, P. E. 1958 "Contrasting correlates of group size." Sociometry, 21:129-139. Thomas, Edwin, and Clinton F. Fink 1963 "Effects of group size." Psychological Bulletin, 60:371-384. Weick, Karl E. 1969 "Laboratory organizations and unnoticed causes." Administrative Science Quarterly, 14:294-303.

of these patterns and to determine whether the magnitude of the differences was significant. Rocco Carzo, Jr., is the head of the department of management, and John N. Yanouzas is a professor of management, both at The Pennsylvania State University.
REFERENCES

Arima, James K. 1965 "Human probability learning with forced training trials and certain and uncertain outcome choice trials." Journal of Experimental Psychology, 70: 43-50. Box, G. E. P. 1953 "Nonnormality and tests on variances." Biometrika, 40:318-335. Cain, Geraldine Smith 1965 "Some effects of organization structure on problem solving." National Business Education Quarterly, 33:2731. Carzo, Rocco, Jr. 1963 "Some effects of organization structure on group effectiveness." Administrative Science Quarterly, 7:393-424. Carzo, Rocco, Jr., and R. William Millman 1970 The Effects of Differential Rewards in Tall and Flat Organization Structures. Working paper, The Pennsylvania State University. Carzo, Rocco, Jr., and John N. Yanouzas 1969 "Effects of flat and tall organization structure." Administrative Science Quarterly, 14:178-191. Grant, David A. 1948 "The latin square principle in the de- Winer, B. J. 1962 Statistical Principles in Experimental sign and analysis of psychological exDesign. New York: McGraw-Hill. periments." Journal of Social Psychology, 45:427-442. Worthy, James C. 1950 "Organizational structure and emGreenhouse, S. W., and S. Geisser 1959 "On methods in the analysis of profile ployee morale." American Sociological data." Psychometrica, 24:95-112. Review, 15:169-179. Guetzkow, Harold, and Herbert A. Simon Yanouzas, John N. 1955 "The impact of certain communica1963 The Relationship of Some Organization nets upon organization and pertion Variables to the Performance of formance in task-oriented groups." Decision Groups. Doctoral dissertation, Management Science, 1:233-250. The Pennsylvania State University.

This content downloaded from 122.170.126.130 on Mon, 5 Aug 2013 07:56:05 AM All use subject to JSTOR Terms and Conditions

You might also like