Seeking an Empirical Development Taxonomy for Manufacturing Smes

SEEKING AN EMPIRICAL DEVELOPMENT TAXONOMY FOR MANUFACTURING SMEs

USING DATA FROM AUSTRALIA’S BUSINESS LONGITUDINAL SURVEY

Professor Richard G.P. McMahon,

Head, School of Commerce,

The Flinders University of South Australia.

SCHOOL OF COMMERCE RESEARCH PAPER SERIES: 00-1

ISSN: 1441-3906

Summary

This paper reports on the pilot stage of a proposed research effort to derive, characterise and employ an empirically-based development taxonomy for small and medium-sized enterprises (SMEs) in the manufacturing sector using panel data recently made available from Australia’s Business Longitudinal Survey. Cluster analysis is used with key enterprise size, age and growth variables to discover if there appear to be any stable development pathways evident in the data. Each of three annual data collections is separately examined, and then comparisons are made of the resulting cluster analysis outcomes over time. Descriptive statistics for various enterprise characteristics facilitate initial interpretation of the cluster analysis solutions. The findings match those of a prior taxonomic study reasonably well; and they provide general support for a stage-wise SME development pattern. Certain development pathways – subsistence, capped growth and continued growth in particular – seem to be stable and persistent.

Introduction

Small and medium-sized enterprise (SME) growth and development have, for some time, received considerable attention from researchers and policy-makers around the world for reasons identified by Turok [1, p. 29] as follows:

There is considerable interest within the field of small firms policy and research in the identification of features that distinguish firms which grow from those that stand still or fail. This is thought important if more selective small firms policies are to be developed. Identifying distinctive features of more and less successful firms may also provide insights into the factors influencing small firm development and hence improve understanding of the growth process. 2

Gibb & Davies [2] give a fuller account of the research and policy imperatives for ‘picking winners’ amongst

SMEs world-wide.

This is a frequently investigated domain for researchers in a variety of business-related fields. However, in many respects, knowledge acquisition has not been cumulative and there is much that is yet to be settled. In their review of the relevant literature, O’Farrell & Hitchens [3, p. 1380] conclude that ‘At present an adequate explanatory framework within which to analyse the growth of the small owner-managed manufacturing enterprise has not been developed’. On the basis of their review, Gibb & Davies [2, p. 26] are of the opinion that

‘The production of such a theory and explanation in the near future is unlikely’. The review of Holmes &

Zimmer [4, p. 97] expresses the belief that ‘an operational framework that distinguishes growth from non-growth small businesses does not exist’.

This paper reports on the pilot stage of a proposed research effort to derive, characterise and employ an empirically-based development taxonomy for SMEs in the manufacturing sector using panel data recently made available from Australia’s Business Longitudinal Survey. The initial objective is to seek for possibly stable pathways for manufacturing SMEs that, to some helpful degree, reflect their achievement of business growth and development. The paper proceeds as follows. After briefly considering some background literature that encourages this investigation, the research method is outlined. Thereafter, the findings of this stage of the research are presented, followed by tentative conclusions and recommendations that will shape the direction and specifics of further inquiry.

Background Literature

For many decades it has been very common amongst writers in the area to view SME growth as a series of phases or stages of development through which the business may pass in an enterprise life-cycle. Having its origins in the literature of economics [5, 6, 7, 8], reliance on this paradigm in the SME literature is most frequently claimed to date back to Steinmetz [9]. In an often cited book of readings on the organisational life- cycle, Kimberly & Miles [10, p. ix] draw attention to:

. . . the cyclical quality of organizational existence. Organizations are born, grow, and decline. Sometimes they reawaken, and sometimes they disappear.

This quotation invokes a biological metaphor for business organisations which has been the source of much controversy in the literature of economics, business and sociology [6, 10]. 3

Before presenting the findings of their own empirical research, Hanks et al. [11] critically review virtually all significant prior writing and research on the enterprise life-cycle construct. Commenting on wide differences in the specifics of prior stages of growth models (particularly inclusion of from 3 to 10 stages),

Hanks et al. [11, pp. 11-12] observe that:

In recent years, a few empirical studies of the organization life cycle have emerged, providing important contributions to life-cycle theory [12, 13, 14, 15]. However, most of these studies have defined growth stages a priori, using existing conceptualizations. The lack of specificity and empirical rigour in these typologies may account for unexpected intrastage variance found in some analyses. . . . It may be possible to address some of these difficulties by deriving taxonomic rather than typological models . . .

Hanks et al. [11] see the strength of a taxonomic approach to identifying and specifying stages in an enterprise life-cycle model as deriving from use of multivariate analysis of empirical data to reveal common patterns and relationships in the data. They acknowledge only Smith et al. [15] as having previously employed a taxonomic approach to developing an enterprise life-cycle model, but note that that research had a very small sample size and various other weaknesses. The taxonomic stages of growth model subsequently described by Hanks et al.

[11] is represented in Figure 1.

After conducting his own critical appraisal of recent research in the field, the present author [16] believes that some reliance can be placed on both the broad approach and the general features of the Hanks et al. [11] stages of growth model. As well as overcoming concerns that such models are frequently not empirically based, it can also be claimed to at least partially answer the most prevalent objections to this type of model that have appeared in the relevant literature [3]. Importantly, this model uniquely incorporates two disengagement (or arrested development) configurations that are frequently observed amongst SMEs – the life-style business and the business electing for capped growth. Furthermore, while predominantly focused upon stages of growth, the model is sympathetic to, or at least not inconsistent with, an alternative gestalts of growth perspective that has recently received some support in the literature [17].

The present research does not aspire to directly testing the validity of the Hanks et al. [11] model with new data. Rather, it responds to a suggestion for further research made at the close of the Hanks et al. [11, p. 24] paper:

. . . longitudinal studies of the organization life cycle that trace changing organizational configurations over time are needed. Although the cross-sectional approach taken in this study is suggestive of life-cycle stages, it is impossible to differentiate between configurations representative of life-cycle stages and those suggestive of firms simply choosing to do business in different ways. Both historical and repeated measures designs would provide important insights into patterns of organization growth. 4

Where attempts have been made to empirically validate stages of growth models, this has hitherto been done with relatively small and narrowly defined samples, and with cross-sectional data. The need for larger and more representative samples in SME research is difficult to dispute [18]. Furthermore, a strong argument can be made that longitudinal data are inherently appropriate to conceptualising growth and development of businesses over time [18]. The availability of data from Australia’s Business Longitudinal Survey now provides an opportunity to meet such research demands.

Research Method

The panel data employed in this research are drawn from the Business Longitudinal Survey (BLS) conducted by the Australian Bureau of Statistics (ABS) on behalf of the federal government over the financial years 1994-95,

1995-96, 1996-97 and 1997-98. Costing in excess of $4 million, the BLS was designed to provide information on the growth and performance of Australian employing businesses, and to identify selected economic and structural characteristics of these businesses.

The ABS Business Register was used as the population frame for the survey, with approximately 13,000 business units being selected for inclusion in the 1994-95 mailing of questionnaires. For the 1995-96 survey, a sub-sample of the original selections for 1994-95 was chosen, and this was supplemented with a sample of new business units added to the Business Register during 1995-96. The sample for the 1996-97 survey was again in two parts. The first formed the longitudinal or continuing part of the sample, comprising all those remaining live businesses from the 1995-96 survey. The second part comprised a sample of new business units added to the

Business Register during 1996-97. A similar procedure was followed for the 1997-98 survey. Approximately

6,400 business units were surveyed in each of 1995-96, 1996-97 and 1997-98.

All business units in the Australian economy were included within the scope of the BLS except for the following:

 Non-employing businesses.

 All government enterprises.

 Businesses classified to the following Australian and New Zealand Standard Industrial Classification

(ANZSIC) Divisions:

A – Agriculture, forestry and fishing D – Electricity, gas and water supply J – Communication services M – Government administration and defence 5

N – Education O – Health and community services

ANZSIC Sub-Divisions 96 Other services and 97 Private households employing staff, and ANZSIC

Groups 921 Libraries, 922 Museums and 923 Parks and gardens, were also excluded from the BLS.

The BLS did not employ completely random samples. The original population (for 1994-95) was stratified by industry and business size, with equal probability sampling methods being employed within strata. Further stratification by innovation status, exporting status and growth status took place for the 1995-96 survey. The

ABS has calculated a system of weights, reflecting the sample fractions used for each stratum, that can be used to estimate population parameters from the BLS data.

Data collection in the BLS was achieved through self-administered, structured questionnaires containing essentially closed questions. Copies of the questionnaires used in each of the four annual collections can be obtained from the ABS. The questionnaires were piloted prior to their first use, and were then progressively refined in the light of experience after each collection. As well as on-going questions, each questionnaire also included once-off questions dealing with certain matters of policy interest to the federal government at the time of the collections. Various imputation techniques, including matching with other data files available to the ABS, were employed to deal with any missing data. Because information collected in the BLS was sought under the authority of the Census and Statistics Act 1905, and thus provision of appropriate responses to the mailed questionnaires could be legally enforced by the Australian Statistician, response rates were very high by conventional research standards – typically exceeding 90 per cent.

The specific BLS data used in this pilot study are included in a Confidentialised Unit Record File (CURF) for the first 3 annual collections, released by the ABS on CD-ROM on 2 July, 1999. This CURF contains data on

9,230 business units employing fewer than 200 persons – broadly representing SMEs in the Australian context.

Restricted industrial classification detail, no geographical indicators, presentation of enterprise age in ranges, and omission of certain data items obtained in the BLS all help to maintain the confidentiality of unit records.

Furthermore, all financial variables have been subject to perturbation – a process in which values are slightly varied to provide further confidentiality protection.

This research is concerned only with the manufacturing sector of the BLS CURF. There are two reasons for this. First, over the last few decades, the performance of the Australian manufacturing sector has been a major preoccupation of policy-makers and government departments dealing with industry and trade. The sector has been characterised as non-competitive by international standards, and it is considered to have failed in 6

countering Australia’s growing trade imbalance with the rest of the world [19]. Over 99 per cent of all businesses in the Australian manufacturing sector are SMEs according to generally accepted definitions [20].

The second reason for considering only the manufacturing sector is that is highly probable that cross-industry differences in the nature of business activities, typical employment per business, capital intensity, etc. could confound findings relating to SME development patterns, and to SME growth and performance more generally.

Such influences are, to a reasonable extent, controlled for by examining a single (albeit broadly defined) industry. There are 3,331 manufacturing SMEs in the BLS CURF, representing approximately 36 per cent of businesses in the file.

Additional focus is provided to this research by considering only manufacturing SMEs legally organised

as proprietary companies. There are a number of reasons for this further narrowing of the unit of analysis.

First, as Freedman & Godwin [21, p. 234] indicate, a particular concern with proprietary companies is not

uncommon amongst SME researchers world-wide:

It would appear that, in so far as the issue is considered at all, the limited liability company is of more interest to the small business research community than are unincorporated firms; . . . limited liability companies and entrepreneurship have become equated, or at least associated.

Second, the primary concern in this research is with SME growth and development, and it is more likely that these will be evident in businesses legally organised as proprietary companies [21, 22, 23, 24, 25]. Third, the research ultimately aspires to comparing possible groupings of manufacturing SMEs in terms of key financial performance measures. This becomes problematic if the study sample contains both incorporated and unincorporated businesses because of the customary procedural difference in accounting for owners’ wages which are not separately reported in the BLS data.

Proprietary companies are currently of considerable policy significance in Australia. The First Corporate

Law Simplification Act 1995 provides for establishment of one member/director companies where previously a minimum of two had been required. An outcome of this, and other recent reforms, is a likely reduction in the proportion of Australian SMEs legally organised as sole proprietorships or partnerships. Incorporation has become a more feasible option. Thus, this research focuses on the proprietary company as an increasingly more predominant form of legal organisation for SMEs in Australia. As a consequence, the population for the research is unlikely to be diminished unacceptably if interest is restricted to proprietary companies. There are 2,374 manufacturing SMEs legally organised as proprietary companies in the BLS CURF, representing approximately

71 per cent of manufacturing SMEs in the file. 7

The principal analytical procedure used in this research is exploratory cluster analysis (as employed by

Hanks et al. [11]). This is a multivariate statistical technique for developing meaningful clusters or groupings of cases. The aim is to objectively classify cases into a small number of mutually exclusive groups on the basis of similarities amongst values for certain clustering variables selected by the researcher. The groups should exhibit high internal (within-cluster) homogeneity and high external (between-cluster) heterogeneity. The groups are not predefined, but are derived from the ‘natural’ structure of the data. Subsequent profiling using characteristic or demographic variables facilitates interpretation of the nature of the groups.

As Hair et al. [26, p. 435] point out, ‘Cluster analysis is not a statistical technique where parameters for a sample are assessed as possibly representative of a population’. Rather, cluster analysis is typically an exploratory technique that requires minimal distributional assumptions regarding clustering variables.

Nevertheless, issues such as the representativeness of the sample, differences in scale amongst clustering variables, the undue influence of outliers, and multicollinearity amongst clustering variables are important considerations.

Research Findings

On the basis of prior research in the area, particularly Hanks et al. [11], the following key clustering variables were selected for use from the BLS CURF for the first 3 annual collections (1994-95, 1995-96, 1996-97):

 Enterprise age – as measured by an ordinal variable with 5 categories (less than 2 years, 2 to 5 years, 5

to 10 years, 10 to 20 years, and more than 20 years). Cluster analysis normally requires clustering

variables to be measured on at least an interval scale. However, for reasons of confidentiality, such an

age measure is not to be available in BLS CURFs. This problem will be ameliorated in subsequent

stages of this research, because the 4-year BLS CURF to be released early in 2000 will include

enterprise age measured in 2-yearly intervals up to 30 years.

 Enterprise size – as measured by total employment and annual sales. To avoid problems of

multicollinearity amongst clustering variables, total assets was not used as a size measure during

cluster analysis. However, it was used in ex post characterisation of clusters.

 Enterprise growth rate – as measured by annual employment growth and annual sales growth. It was

not possible to estimate annual growth rate in total assets for the first collection (1994-95), and so this

variable was not used in the cluster analyses. In order not to exclude new starts in the first collection, 8

employment and sales growth estimates were calculated for all collections as follows (1994-95 growth

in employment used as an example):

Employment growth, 1994-95 = (1995 Employment – 1994 Employment) x 100 per cent 1995 Employment

While acknowledged to be unusual, this form of growth measure was used by Hanks et al. [11] for the

same reason. Clearly, caution should be exercised when interpreting such measures.

In order to deal with differences in scale amongst the clustering variables, and with the undue influence of outliers, all these variables were first standardised and then observations 5 or more standard deviations from the mean value were removed. Given the mixed use of both point-of-time and flow variables in the cluster analyses, it was not possible to use the system of sample weights included in the BLS CURF. This undoubtedly introduces some bias to the samples employed; but, given the exploratory aspirations of the study and the size and coverage of the samples, this is not considered to be a serious problem.

Because of the heavy computational and storage demands on personal computer memory, agglomerative hierarchical cluster analysis (that is, the CLUSTER procedure in SPSS for Windows) is not recommended for use with samples containing 200 or more cases [26, 27]. Since all samples employed in this study exceeded

1,000 cases, with the largest exceeding 2,000 cases, the alternative k-means cluster analysis was primarily used

(that is, the QUICK CLUSTER procedure in SPSS for Windows). This non-hierarchical procedure is based on nearest centroid sorting with squared Euclidean distance being the similarity measure.

The analysis began by examining data from the 1994-95 BLS collection including all operating businesses in that collection (n=2,162 after removing outliers). Sub-samples were subjected to agglomerative hierarchical cluster analysis to get a feel for the likely number of clusters needed to represent the data. For the reason given in the previous paragraph, this procedure was extremely slow; but eventually a 7 cluster solution seemed to be indicated. The full sample was then subjected to k-means cluster analysis using randomly selected seed points, with 7 clusters being pre-specified. Since Hanks et al. [11] had arrived at a 6 cluster solution, this possibility was also examined. Ultimately, the 7 cluster solution was chosen as potentially being the more useful and more amenable to interpretation. This solution is represented in Figure 2. Possible descriptors for some pathways between the clusters are suggested in the figure. The ANOVA significance figures in the first panel of

Figure 6 suggest that all clustering variables do differ between clusters in the solution. In the first panel of Figure

7, the results of Kruskal-Wallis one-way analysis of variance tests lead to rejection of null hypotheses that 9

median values for all characterisation variables (enterprise age, total employment, annual sales, total assets, annual employment growth, and annual sales growth) do not differ between clusters in the solution.

Subsequently, a similar analysis was undertaken examining data from the 1994-95 BLS collection including only businesses operating in all 3 collections (that is, the first collection for businesses constituting the on-going longitudinal panel; n=1,189 after removing outliers). Again, agglomerative hierarchical cluster analysis seemed to indicate a 7 cluster solution. However, a 6 cluster solution was ultimately chosen as potentially being the more useful and more amenable to interpretation. This solution is represented in Figure 3. The ANOVA significance figures in the second panel of Figure 6 suggest that all clustering variables do differ between clusters in the solution. In the second panel of Figure 7, the results of Kruskal-Wallis one-way analysis of variance tests lead to rejection of null hypotheses that median values for all characterisation variables do not differ between clusters in the solution.

Finally, similar analyses were undertaken examining data from the 1995-96 and 1996-97 BLS collections including only businesses operating in all 3 collections (that is, the second and third collections for businesses constituting the on-going longitudinal panel; n=1,186 and n=1,186 after removing outliers). For the 1995-96 collection, agglomerative hierarchical cluster analysis seemed to indicate a 7 cluster solution. However, a 6 cluster solution was ultimately chosen as potentially being the more useful and more amenable to interpretation.

For the 1996-97 collection, agglomerative hierarchical cluster analysis seemed to indicate a 5 cluster solution.

However, a 4 cluster solution was ultimately chosen as potentially being the more useful and more amenable to interpretation. These solutions are represented in Figures 4 and 5. For the 1995-96 collection, the ANOVA significance figures in the third panel of Figure 6 suggest that all clustering variables do differ between clusters in the solution. For the 1996-97 collection, the ANOVA significance figures in the fourth panel of Figure 6 suggest that clustering variables other than annual sales growth do differ between clusters in the solution. In the third and fourth panels of Figure 7, the results of Kruskal-Wallis one-way analysis of variance tests lead to rejection of null hypotheses that median values for all characterisation variables do not differ between clusters in the solutions.

Conclusions and Recommendations

Bearing in mind the exploratory nature of this pilot study, the acknowledged limitations of the research method used, and that it has not yet been possible to employ the full 4-year BLS CURF with its improved enterprise age measure, the following tentative conclusions might nevertheless be reached at this stage: 10

 While not corresponding in all respects, it would appear that certain features of the first BLS 1994-95

enterprise life-cycle model (n=2,162) match those of the Hanks et al. [11] model reasonably well. These

include a discernible stage-wise development pattern (see below), approximately the same number of

stages, and the apparent existence of disengagement stages such as life-style and capped growth.

Furthermore, the pace of SME development – viewed over 20 or so years – seems similar in the two

models. Not unexpectedly, the enterprise age and size benchmarks for stages vary somewhat. In judging

these models against each other, it should be borne in mind that the BLS model is derived from a much

larger and broadly more representative manufacturing SME sample.

 Support for a stage-wise SME development pattern is gained by comparing the second BLS 1994-95

model (n=1,189) with the models for 1995-96 (n=1,186) and 1996-97 (n=1,186) which are derived for

exactly the same businesses. As time elapses, fewer stages are evident, but certain development pathways

– subsistence, capped growth and continued growth in particular – seem to persist. Clusters at the left of

Figures 3, 4 and 5 tend to disappear and there appears to be a rightward drift in the models towards

roughly the same range of configurations. Moreover, the stability of the ultimate configurations seems to

be suggested by the diminished significance of differences in employment and sales growth measures

between clusters – but not so for enterprise size measures – as the businesses mature.

Thus, further scholarly inquiry certainly seems to be encouraged by this preliminary multivariate analysis of empirical data derived from Australia’s BLS to reveal common development patterns amongst manufacturing

Pending informed feedback on this paper, the following recommendations are made to shape the direction and specifics of further research effort to derive, characterise and employ an empirically-based development taxonomy for Australian manufacturing SMEs legally organised as proprietary companies:

 To facilitate interpretation of the development models produced, use of the unusual employment and sales

growth measures underpinning this study should be reconsidered – even if this means a reduction in

sample sizes because of missing growth observations.

 Further inquiry should employ the 4-year BLS CURF to be released early in 2000. First, this will include

an improved enterprise age measure that may enhance the validity and specificity of development models

produced. Second, an extra year of observations for the on-going longitudinal panel should make these

models more dependable and informative. 11

 Further inquiry should seek to employ truly longitudinal methods of analysis, rather than repeated cross-

sectional methods. There is no readily available longitudinal counterpart to cluster analysis, other than

relaxing independence requirements and using pooled data from the various collections. However, there

are longitudinal methods for assessing the significance of differences between clusters in successive

collections. Other, less common, possibilities for longitudinal analysis should also be explored.

Acknowledgments

The permission of the Australian Statistician to use confidentialised data from the federal government’s Business

Longitudinal Survey, and to publish findings based on analysis of that data, is gratefully acknowledged.

References

[1] Turok, I., ‘Which small firms grow?’, in L.G. Davies & A.A. Gibb eds Recent Research in

Entrepreneurship, Avebury, England, 1991, pp. 29-44.

[2] Gibb, A.A. & Davies, L.G., ‘In pursuit of frameworks for the development of growth models of the small

business’, International Small Business Journal, 1990, vol. 9, no. 1, pp. 15-31.

[3] O’Farrell, P.N. & Hitchens, D.M.W.N., ‘Alternative theories of small-firm growth: a critical review’,

Environment and Planning A, 1988, vol. 20, no. 2, pp. 1365-1383.

[4] Holmes, S. & Zimmer, I., ‘The nature of the small firm: understanding the motivations of growth and non-

growth oriented owners’, Australian Journal of Management, 1994, vol. 19, no. 1, pp. 97-120.

[5] Marshall, A., Principles of Economics, Macmillan, London, England, 1890.

[6] Penrose, E.T., ‘Biological analogies in the theory of the firm’, American Economic Review, 1952, vol. 42,

no. 5, pp. 804-819.

[7] Penrose, E.T., The Theory of the Growth of the Firm, John Wiley & Son, New York, New York, 1959.

[8] Rostow, W.W., The Stages of Economic Growth, Cambridge University Press, Cambridge, England, 1960.

[9] Steinmetz, L.L., ‘Critical stages of small business growth: when they occur and how to survive them’,

Business Horizons, 1969, vol. 12, no. 1, pp. 29-34.

[10] Kimberly, J.R. & Miles, R.H., ‘Preface’, in J.R. Kimberly, R.H. Miles & Associates eds The

Organizational Life Cycle: Issues in the Creation, Transformation, and Decline of Organizations, Jossey-

Bass, San Francisco, California, 1980, pp. ix-xiii. 12

[11] Hanks, S.H., Watson, C.J., Jansen, E. & Chandler, G.N., ‘Tightening the life-cycle construct: a taxonomic

study of growth stage configurations in high-technology organizations’, Entrepreneurship Theory and

Practice, 1993, vol. 18, no. 2, pp. 5-29.

[12] Kazanjian, R.K., ‘Relation of dominant problems to stages of growth in technology-based new ventures’,

Academy of Management Journal, 1988, vol. 31, no. 2, pp. 257-279.

[13] Kazanjian, R.K. & Drazin, R., ‘A stage-contingent model of design and growth for technology based

ventures’, Journal of Business Venturing, 1990, vol. 5, no. 3, pp. 137-150.

[14] Miller, D. & Friesen, P.H., ‘A longitudinal study of the corporate life cycle’, Management Science, 1984,

vol. 30, no. 10, pp. 1161-1183.

[15] Smith, K.G., Mitchell, T.R. & Summer, C.E., ‘Top level management priorities in different stages of the

organizational life cycle’, Academy of Management Journal, 1985, vol. 28, no. 4, pp. 799-820.

[16] McMahon, R.G.P., ‘Stage models of SME growth reconsidered’, Small Enterprise Research: The Journal

of SEAANZ, 1998, vol. 6, no. 2, pp. 20-35.

[17] Kazanjian, R.K. & Drazin, R., ‘An empirical test of stage of growth progression model’, Management

Science, 1989, vol. 35, no. 12, pp. 1489-1503.

[18] McMahon, R.G.P., ‘Recent SME research: a critical review’, Small Enterprise Research: The Journal of

SEAANZ, 1999, vol. 7, no. 1, pp. 68-75.

[19] Pappas, Carter, Evans and Koop/Telesis, The Global Challenge: Australian Manufacturing in the 1990s,

Australian Manufacturing Council, Melbourne, Victoria, 1990.

[20] Australian Bureau of Statistics, Small Business in Australia 1995, Australian Government Publishing

Service, Canberra, Australian Capital Territory, 1996.

[21] Freedman, J. & Godwin, M., ‘Incorporating the micro business: perceptions and misperceptions’, in A.

Hughes & D.J. Storey eds Finance and the Small Firm, Routledge, London, England, 1994, pp. 232-283.

[22] Hakim, C., ‘Identifying fast growth small firms’, Employment Gazette, 1989, vol. 97, no. 1, pp. 29-41.

[23] Gray, C., ‘Growth orientation and the small firm’, in K.Caley, E. Chell, F. Chittenden & L. Mason eds

Small Enterprise Development, Paul Chapman Publishing, London, England, 1992, pp. 59-71.

[24] Hughes, A. & Storey, D.J., ‘Introduction: financing small firms’, in A. Hughes & D.J. Storey eds Finance

and the Small Firm, Routledge, London, England, 1994, pp. 1-17.

[25] Yellow Pages Australia, A Special Report on Small Business Growth Aspirations and the Role of Exports,

Small Business Index, Melbourne, Victoria, 1995. 13

[26] Hair, J.F., Anderson, R.E., Tatham, R.L. & Black, W.C., Multivariate Data Analysis, 4th edn, Prentice-

Hall, Englewood Cliffs, New Jersey, 1995.

[27] Norusis, M.J. & SPSS Inc., SPSS for Windows Professional Statistics, Release 5.0, SPSS Inc., Chicago,

Illinois, 1992. 14

Figure 1: Hanks et al. (11) Enterprise Life-Cycle Model

DEVELOPMENT DISENGAGEMENT STAGES STAGES

START-UP Mean number of employees: 6.46 persons Mean annual sales revenues: US$0.27 million LIFE-STYLE Mean age: 4.29 years Mean number of employees: 7.00 persons Mean annual sales revenues: US$0.41 million EXPANSION Mean age: 18.71 years Mean number of employees: 23.64 persons Mean annual sales revenues: US$1.40 million Mean age: 7.36 years CAPPED GROWTH Mean number of employees: 24.65 persons Mean annual sales revenues: US$2.05 million MATURITY Mean age: 12.65 years Mean number of employees: 62.76 persons Mean annual sales revenues: US$3.71 million Mean age: 6.66 years

DIVERSIFICATION Mean number of employees: 495.40 persons Mean annual sales revenues: US$45.76 million Mean age: 16.20 years 15

Figure 2: Business Longitudinal Survey 1994-95 Enterprise Life-Cycle Model 1 (n=2,162)

ENTERPRISE AGE (YEARS)

CLUSTER 2 (n=384) CLUSTER 7 (n=441) CLUSTER 5 (n=713) Enterprise age mode: 2 to 5 years Enterprise age mode: 5 to 10 years Enterprise age mode: 10 to 20 years Total employment mean: 10.2 persons SUBSISTENCE?CE Total employment mean: 12.3 persons Total employment mean: 14.7 persons Sales mean: $1.1 million p.a. Sales mean: $1.5 million p.a. Sales mean: $1.7 million p.a. Total assets mean: $0.8 million Total assets mean: $0.8 million Total assets mean: $1.1 million Employment growth mean: 23.4% p.a. Employment growth mean: -3.9% p.a. Employment growth mean: 1.1% p.a. Sales growth mean: 39.5% p.a. Sales growth mean: 8.1% p.a. Sales growth mean: 10.1% p.a.

LIFE-STYLE? CLUSTER 6 (n=330) Enterprise age mode: more than 20 years Total employment mean: 53.4 persons Sales mean: $7.3 million p.a. CLUSTER 4 (n=168) Total assets mean: $4.5 million EARLY GROWTH? Enterprise age mode: 5 to 10 years Employment growth mean: 5.1% p.a. Total employment mean: 57.8 persons Sales growth mean: 16.1% p.a. Sales mean: $9.3 million p.a. Total assets mean: $6.4 million Employment growth mean: 21.4% p.a. Sales growth mean: 35.8% p.a.

CLUSTER 3 (n=109) CAPPED GROWTH? Enterprise age mode: more than 20 years Total employment mean: 114.4 persons Sales mean: $16.2 million p.a. Total assets mean: $11.6 million Employment growth mean: 6.0% p.a. Sales growth mean: 22.0% p.a.

CLUSTER 1 (n=47) CONTINUED GROWTH? Enterprise age mode: more than 20 years Total employment mean: 116.2 persons Sales mean: $39.7 million p.a. Total assets mean: $32.3 million ENTERPRISE SIZE Employment growth mean: 8.2% p.a. (EMPLOYMENT, SALES, TOTAL ASSETS) Sales growth mean: 27.2% p.a. 16

Figure 3: Business Longitudinal Survey 1994-95 Enterprise Life-Cycle Model 2 (n=1,189) 17

CLUSTER 6 (n=138) CLUSTER 1 (n=361) Enterprise age mode: 2 to 5 years Enterprise age mode: 5 to 10 years Total employment mean: 16.5 persons SUBSISTENCE?CE Total employment mean: 17.2 persons CLUSTER 2 (n=446) Sales mean: $2.2 million p.a. Sales mean: $2.3 million p.a. Enterprise age mode: 10 to 20 years Total assets mean: $1.9 million LIFE-STYLE? Total assets mean: $1.5 million Total employment mean: 24.2 persons Employment growth mean: 39.9% p.a. Employment growth mean: 2.0% p.a. Sales mean: $3.2 million p.a. Sales growth mean: 77.4% p.a. Sales growth mean: 7.2% p.a. Total assets mean: $2.1 million Employment growth mean: 1.4% p.a. Sales growth mean: 7.1% p.a.

CLUSTER 5 (n=95) Enterprise age mode: 5 to 10 years EARLY GROWTH? Total employment mean: 78.7 persons Sales mean: $12.6 million p.a. CAPPED GROWTH? Total assets mean: $7.6 million CLUSTER 1 (n=47) Employment growth mean: 17.8% p.a. Enterprise age mode: more than 20 years Sales growth mean: 52.3% p.a. Total employment mean: 97.5 persons Sales mean: $13.8 million p.a. Total assets mean: $9.8 million Employment growth mean: 4.9% p.a. Sales growth mean: 6.3% p.a.

CLUSTER 3 (n=38) CONTINUED GROWTH? Enterprise age mode: more than 20 years Total employment mean: 113.2 persons Sales mean: $44.3 million p.a. Total assets mean: $39.7 million Employment growth mean: 10.3% p.a. Sales growth mean: 26.3% p.a. ENTERPRISE SIZE (EMPLOYMENT, SALES, TOTAL ASSETS) 18

Figure 4: Business Longitudinal Survey 1995-96 Enterprise Life-Cycle Model (n=1,186) 19

CLUSTER 1 (n=62) Enterprise age mode: 5 to 10 years SUBSISTENCE?CE Total employment mean: 14.8 persons Sales mean: $2.5 million p.a. Total assets mean: $1.4 million Employment growth mean: -101.1% p.a. Sales growth mean: -38.0% p.a. CLUSTER 5 (n=396) Enterprise age mode: 10 to 20 years Total employment mean: 18.0 persons Sales mean: $2.5 million p.a. Total assets mean: $1.6 million CLUSTER 6 (n=414) Enterprise age mode: 5 to 10 years Employment growth mean: 0.0% p.a. Total employment mean: 19.0 persons Sales growth mean: 0.0% p.a. Sales mean: $2.8 million p.a. Total assets mean: $2.0 million LIFE-STYLE? Employment growth mean: 1.9% p.a. Sales growth mean: 5.3% p.a. CLUSTER 4 (n=179) Enterprise age mode: more than 20 years CAPPED GROWTH? Total employment mean: 61.4 persons Sales mean: $9.0 million p.a. Total assets mean: $6.2 million Employment growth mean: 0.5% p.a. Sales growth mean: 5.1% p.a.

CLUSTER 3 (n=89) CLUSTER 2 (n=46) Enterprise age mode: 5 to 10 years CONTINUED GROWTH? Enterprise age mode: more than 20 years Total employment mean: 114.0 persons Total employment mean: 100.3 persons Sales mean: $17.0 million p.a. Sales mean: $46.5 million p.a. Total assets mean: $11.8 million Total assets mean: $34.9 million Employment growth mean: 5.9% p.a. Employment growth mean: -0.5% p.a. Sales growth mean: 6.9% p.a. Sales growth mean: 7.5% p.a.

ENTERPRISE SIZE (EMPLOYMENT, SALES, TOTAL ASSETS) 20

Figure 5: Business Longitudinal Survey 1996-97 Enterprise Life-Cycle Model (n=1,186) 21

CLUSTER 1 (n=404) CLUSTER 4 (n=519) Enterprise age mode: 5 to 10 years Enterprise age mode: 10 to 20 years Total employment mean: 18.4 persons Total employment mean: 20.0 persons SUBSISTENCE?CE Sales mean: $2.7 million p.a. Sales mean: $2.8 million p.a. Total assets mean: $1.9 million Total assets mean: $1.9 million Employment growth mean: -3.5% p.a. Employment growth mean: -8.1% p.a. Sales growth mean: -854.6% p.a. Sales growth mean: -571.4% p.a.

CLUSTER 3 (n=194) CAPPED GROWTH? Enterprise age mode: more than 20 years Total employment mean: 77.0 persons Sales mean: $12.7 million p.a. Total assets mean: $8.6 million Employment growth mean: 0.3% p.a. Sales growth mean: 3.0% p.a.

CLUSTER 2 (n=69) CONTINUED GROWTH? Enterprise age mode: more than 20 years Total employment mean: 128.9 persons Sales mean: $38.7 million p.a. Total assets mean: $30.1 million Employment growth mean: 1.8% p.a. Sales growth mean: -0.5% p.a.

ENTERPRISE SIZE (EMPLOYMENT, SALES, TOTAL ASSETS) 22 18 18

Figure 6: K-MEANS CLUSTER ANALYSIS ANOVAs

Cluster Error Annual Standardised Mean Mean Univariate Collection Variable Square df Square df F Ratio Significance

Z Score Enterprise Age 289.685 6 0.195 2155 1482.592 0.000

Z Score Total Employment 150.683 6 0.096 2155 1568.038 0.000 1994-95 Z Score (n=2,162) Annual Sales 272.760 6 0.179 2155 1519.836 0.000

Z Score Annual Employment 8.743 6 0.226 2155 38.763 0.000 Growth

Z Score Annual Sales 5.735 6 0.153 2155 37.458 0.000 Growth

Z Score Enterprise Age 152.820 5 0.359 1183 425.630 0.000

Z Score Enterprise Age 153.925 5 0.349 1180 441.147 0.000 Z Score Total Employment 170.969 5 0.245 1180 698.192 0.000 1995-96 Z Score (n=1,186) Annual Sales 136.111 5 0.128 1180 1062.720 0.000 Z Score Annual Employment 41.106 5 0.213 1180 193.210 0.000 Growth Z Score Annual Sales 6.696 5 0.231 1180 28.970 0.000 Growth Figure 6 (Continued): K-MEANS CLUSTER ANALYSIS ANOVAs

Cluster Error Annual Standardised Mean Mean Univariate Collection Variable Square df Square df F Ratio Significance

Z Score 19

Enterprise Age 258.362 3 0.345 1182 748.927 0.000

Figure 7: KRUSKAL-WALLIS TESTS FOR CLUSTER CHARACTERISATION VARIABLES

Annual Annual Annual Test Enterprise Total Annual Total Employment Sales Collection Details Age Employment Sales Assets Growth Growth

Kruskal- 20

Wallis 1994-95 Statistic 1750.806 1365.673 1210.178 1058.357 150.282 183.789 (n=2,162) df 6 6 6 6 6 6 Significance 0.000 0.000 0.000 0.000 0.000 0.000

Kruskal- Wallis 1994-95 Statistic 804.243 595.215 525.983 472.751 137.886 387.468 (n=1,189) df 5 5 5 5 5 5 Significance 0.000 0.000 0.000 0.000 0.000 0.000 Kruskal- Wallis 1995-96 Statistic 809.915 660.335 562.352 514.641 184.616 62.218 (n=1,186) df 5 5 5 5 5 5 Significance 0.000 0.000 0.000 0.000 0.000 0.000

Kruskal- Wallis 1996-97 Statistic 788.827 602.650 535.058 481.634 11.557 10.274 (n=1,186) df 3 3 3 3 3 3 Significance 0.000 0.000 0.000 0.000 0.009 0.016