Small Business Economics

, Volume 40, Issue 4, pp 953–976

Firm ownership and productivity: a study of family and non-family SMEs

Authors

    • Department of Economics, School of Business, Technology and Sustainable DevelopmentBond University
  • Ken Moores
    • Australian Centre for Family Business, School of Business, Technology and Sustainable DevelopmentBond University
Article

DOI: 10.1007/s11187-011-9405-9

Cite this article as:
Barbera, F. & Moores, K. Small Bus Econ (2013) 40: 953. doi:10.1007/s11187-011-9405-9

Abstract

Motivated by a lack of consensus in the current literature, the objective of this paper is to reveal whether family firms are more or less productive than non-family firms. As a first step, this paper links family business research to the theoretical notion that family involvement has an effect on the factors of production from a productivity standpoint. Second, by using a Cobb–Douglas framework, we provide empirical evidence that family labour and capital indeed yield diverse output contributions compared with their non-family counterparts. In particular, family labour output contributions are significantly higher, and family capital output contributions significantly lower. Interestingly, differences in total factor productivity between family and non-family firms disappear when we allow for heterogeneous output contributions of family production inputs. These findings imply that the assumption of homogeneous labour and capital between family and non-family firms is inappropriate when estimating the production function.

Keywords

Heterogeneous input elasticityFamily firmCobb–Douglas production functionTotal factor productivity

JEL Classifications

D22D24J24M11L26

1 Introduction

The impact of ownership structure on firm performance has gained relevance since the seminal work of Berle and Means (1932). For example, see Jensen and Meckling (1976), Demsetz (1983), Fama and Jensen (1983), and Morck et al. (1988). Recently, Palia and Lichtenberg (1999) operationalized this impact by specifically focussing on the effect of managerial ownership on “productivity” and found that managerial ownership changes do drive changes in productivity. What is less established, however, is the particular impact of family ownership or, more broadly, family involvement on the productivity of the firm. Measuring this impact would in turn necessitate a comparison of family and non-family firms.

Despite more contemporary investigations into financial performance differences between family and non-family firms (Anderson and Reeb 2003; Lee 2006; Miller et al. 2007; Sciascia and Mazzola 2008), to date only a modest amount of analysis has been devoted to determining the specific effects of family involvement on the fundamental drivers of these performance differences, for example productivity. A review of previous studies reveals that while a consistent and significant relationship between family involvement and a firm’s productivity has been found, there is no consensus about the direction of this relationship. Granted, different definitions of a family firm, time periods, measures of productivity, methodologies, and data sets cause these results to vary; however, the inconsistencies beg the following fundamental question: does family involvement have a positive or negative impact on firm productivity?

To answer this question, it is important to first determine if there are differences in how family and non-family firms produce. Further to this, it is important to determine if family involvement affects the output contribution of production inputs, namely labour and capital. Despite the importance of unique resources and capabilities on firm productivity (Penrose 1959), one of the more curious aspects of previous investigations is that those utilising a Cobb–Douglas framework have assumed fixed factor elasticities in the production process for both family and non-family firms. In other words, it is assumed that labour and capital output contributions for both types of firm are homogeneous. However, on the basis of the established differences highlighted in the literature reviewed in this paper, such an assumption excludes the impact family involvement may have on the firm’s production process; thus, the purpose of this research is twofold:
  1. 1)

    to link the family business research to the theoretical notion that family involvement has an effect on the factors of production from a productivity standpoint; and

     
  2. 2)

    to quantify these productivity differences by use of an estimation technique which enables the economist to interpret, in a traditional manner, producer behaviour and the productivity of inputs which they use.

     

As a starting point, the section “The nature of family firms” reviews the family business literature which substantiates potential differences in the capital and labour inputs of family firms, and how such differences could affect specific factor output contributions and the overall total factor productivity of the firm. Based on this review, we formulate testable hypotheses. In the next section we outline our “Methodology”. The data and analytical findings are then presented in the sections “Data: the business longitudinal survey” and “Empirical results”, respectively. Finally our conclusions are discussed in the last section.

2 The nature of family firms

Considering that family run businesses are the prevalent form of business among OECD economies,1 we are very interested in understanding their production process; however, despite family businesses being the focus of study for many years, the persisting challenge facing researchers is defining what exactly a family business is. In an attempt to clarify this issue, two different conceptual approaches have been established in the literature.

Following the work of Berle and Means (1932), the first approach focuses on a structure-based classification. For example, family firms have been defined as those which are either owned, controlled, and/or managed by a family unit. Such a definition allows for a wide range of “family firms”, because the extent of family ownership, control and management can differ among individual firms, and studies have shown that different amounts of family involvement do matter empirically (Villalonga and Amit 2006; Miller et al. 2007; Sciascia and Mazzola 2008).

In fact, some researchers have come to realize that the components of family involvement do not necessarily determine whether a firm is a family firm, because the structure-based approach does not account for the possibility of intraorganisational aspirations within the firm to either increase or reduce the degree of family-based relatedness (Litz 1995). Thus, when attempting to narrow the definition of a family firm, an intention, or essence-based, approach can be useful (Chua et al. 1999). For example, the intangible desire of the family unit to transfer ownership, through succession, within the family is considered to be a unique characteristic of family firms.2

On the basis of these approaches, a family firm is one in which the family will exert some strategic control over the firm’s resources and processes. As outlined in Table 1, previous studies which have investigated the effect of such control on productivity3 have all, curiously, assumed that the labour and capital output contributions of both family and non-family firms are equal4; however, this assumption does not account for the control that the family may exert over the firm’s production process.
Table 1

Previous investigations of the effects of family involvement on firm productivity

Author(s)

Study time period(s)

Data source and sample size

Measure of productivity and methodology

Findings

Kirchhoff and Kirchhoff (1987)

1978–1982

“University of Minnesota Data Base” data on 702 small businesses located in Minnesota, Ohio, Oregon, and Washington

Partial measure of productivity comparing sales per employee between family firms which use paid and unpaid family labour

Positive and significant correlation between productivity and the use of family labour, both paid and unpaid

McConaughy et al. (1998)

1986–1988

COMPUSTAT data on 219 publicly traded firms

Partial measures of productivity such as sales per employee and total asset turnover. Matched-pairs method to compare family and non-family firms

Founding family (and descendant)-controlled firms are more efficient than non-family firms. Younger founder-controlled firms are more efficient than older ones

Wall (1998)

1994

Firm-level survey data on 506 privately held companies in Western New York

Cobb–Douglas production function using industry as a proxy for capital intensity and including an intercept dummy variable for family business. Factor elasticities are assumed to be equal for both family and non-family firms

From a “macro” perspective, family firms contribute less per firm to the examined regional economy than non-family firms. This is based on a lower level of sales generated by family firms

Bosworth and Loundes (2002)

1994–1995

1997–1998

Australian Bureau of Statistics’ “Business Longitudinal Survey” of 4354 small to medium-sized Australian firms

Cobb–Douglas production function controlling for technology, human resources and organizational characteristics, including family ownership. Factor elasticities are assumed to be equal for both family and non-family firms

Focussing on the interaction of “discretionary” investments, innovation, productivity, and profitability, family firms are incidentally found to be significantly less productive than non-family firms

Barth et al. (2005)

1996

Firm-level survey data among 438 firms associated with the “Confederation of Norwegian Business and Industry”

Cobb–Douglas production function including intercept dummy variables for family owned and family managed firms. Factor elasticities are assumed to be equal for both family and non-family firms

Family-owned firms are less productive than non-family firms. This productivity gap can be explained by a management regime in that family owned and managed firms are significantly less productive

Martikainen et al. (2009)

1992–1999

S&P 500 firm data on 159 manufacturing firms. Source list originally compiled and classified by Anderson and Reeb (2003)

Cobb–Douglas production function including an intercept dummy variable for family business. Factor elasticities are tested for invariance and found to be equal for both family and non-family firms

Production technologies between family and non-family firms are found to be the same; however, on the basis of a positive and significant intercept dummy variable, family firms are found to be more efficient in their production than comparable non-family firms

Understanding family concerns and preferences are crucial for understanding family business behaviour (Ward 1988; Harris et al. 1994; Nelly and Rodríguez 2008); thus we acknowledge that different motives may drive differences in behaviour between family and non-family firms. As a result, long-term objectives akin to the continuity of the business, preservation of financial strength, and maintenance of family control may have greater precedence than immediate profits or other short term objectives. Such objectives may in turn manifest themselves in the production process. More specifically, to meet production demands, family firms may utilise labour and capital differently from non-family firms.

Further to the question of utilisation, the output contribution of both family firm labour and capital, when considered as heterogeneous production factors, may also be different from those of non-family firms. In other words, the notion that family firm labour and capital inputs yield exactly the same contribution toward output as non-family firm labour and capital may be flawed, and thus treating them as equal theoretically and empirically could be a mistake. On the basis of the literature to date, it is possible that family involvement may have both positive and negative effects on production inputs.

For example, family firms tend to avoid external debt and prefer to use internal financial resources instead. Manifestations of this behaviour have consistently been observed in that family firms have been found to have significantly lower leverage than non-family firms (Dreux 1990; Gallo and Vilaseca 1996; Anderson et al. 2003; Villalonga and Amit 2006). The rationale for such behaviour lies in the fact that inside equity holders of family firms typically have undiversified portfolios and intend to pass the firm on to their descendants, and are thus less willing to subject the firm to future cash-flow risks that result from financing via debt (McMahon and Stanger 1995).

It is for these reasons that Anderson and Reeb (2003) contend that inside family business equity holders are a unique class of shareholders. More specifically, there is strong identification by inside owners between the family and the business (Gallo and Vilaseca 1996), and family business owners, unlike owners of other companies, have to satisfy the current and future needs of family members in addition to the needs of the business (Dreux 1990). Because physical capital investments can represent large upfront costs and, once acquired, may yield irregular and uncertain returns over time, they represent a risk to the firm; and with the family’s wealth so closely tied to the firm’s future, it might become difficult for inside family owners to support such risk-taking activities (Agrawal and Nagarajan 1990; Zahra 2005; Gómez-Mejía et al. 2007). Ward (1988) outlines how such family considerations can limit the strategic aggressiveness of family firms; and Morck et al. (2000) also recognise the distaste for risk displayed by family-owned firms, arguing that they may be excessively risk averse, even to the point where they forego profitable expansion strategies, for example investment in physical capital.

Differences in the capital intensity of family-firm production alone would not necessarily deem the capital owned by family firms more or less productive; however, a reluctance to invest in physical capital may also extend to investments in capital-improving innovations. Further to this notion, Morck and Yeung (2003) consider situations in which family firms may actively suppress capital-improving innovation to protect their already established wealth. Suppression of capital-enhancing innovation may in turn reduce the output contribution of capital during family-firm production.

Other studies, however, portray family firms as more likely to invest in capital enhancing innovations. For example, Zellweger (2007) finds that family firms have a longer time horizon than non-family firms, because they have a longer CEO tenure and strive for long-term independence and succession within the family. Because of these extended horizons and the long-term presence of family owners, family-owned firms may be more likely to invest in longer-term value-maximizing projects. The lengthy presence of family owners may also result in superior knowledge of the firm’s technology, which could induce improvements in productivity (Martikainen et al. 2009); thus there is the question of whether family-firm capital contributes more or less to output than non-family-firm capital. To shed further light on this issue, it is not necessarily sufficient to simply determine whether or not family firms enhance their capital via innovation. Because it is the efficient use of capital that ultimately will affect output contribution, perhaps more central to the question is the issue of understanding how existing capital, enhanced or otherwise, is actually being utilised by family firms.

Family owners differ from non family shareholders in that the latter obtain only monetary benefits of control whereas family owners also obtain non-pecuniary benefits, for example the satisfaction of transferring the firm to descendants and, more importantly, use of capital and consumption of amenities by the family at the expense of firm profits (Demsetz and Lehn 1985). More specifically, Demsetz (1983) argues that combining ownership and control leads to such owners choosing non-pecuniary consumption, thereby drawing scarce resources away from profitable projects.

Especially when managerial ownership is large, as in many family firms, there is ample evidence of managers pursuing private benefits when their control of the firm becomes “entrenched” (Stulz 1988; Shleifer and Vishny 1997; Claessens et al. 2002). Such benefits can take the form of advantageous dividend treatment (DeAngelo and DeAngelo 2000), the preservation of “socioemotional” wealth5 (Gómez-Mejía et al. 2007), or shirking or other on-the-job consumption (Hoopes and Miller 2006); thus the capital of family firms may be used for both productive and non-productive purposes. Such self serving behaviour may reduce the share of capital in the family firm’s production process.

Despite the potential differences, the literature to date has largely assumed that capital output contributions between family and non-family firms are the same. To verify this assumption, we test the following null hypothesis.

H1

There is no difference between the output contribution of capital in family and non-family firms.

Based on the discussion thus far, we acknowledge that any differences in physical capital intensity and production share between family and non-family firms may have implications for other production factor intensities and shares also, for example labour. In fact, there is theoretical and empirical evidence demonstrating that family firms utilise labour in the production process differently from non-family firms, and that such labour yields unique contributions to output.

A reduction in capital intensity may compel family firms to increase the utilisation of labour to meet production demand. Alternatively, family firms may prefer labour in the production process, not because they are deprived of capital, but because the use of labour may be particularly advantageous for family firms. Evidence for such an advantage is consistently found throughout the family business literature.

By assessing the human and organizational resources related to employees, Habbershon and Williams (1999) suggest that family firms manage and socialise their employees better than non-family firms, potentially leading to a competitive advantage and improved performance. Such an advantage may be based on findings that family firms have a unique working environment that promotes employee dedication and commitment (Ward 1988). Other cultural attributes pertaining to employees of family firms include a shared sense of identity, better communication with greater privacy, and emotional involvement among co-workers (Tagiuri and Davis 1996). Greater employee loyalty and trust would serve to provide the family firm with access to employees who would potentially contribute more to output than otherwise (Rutherford et al. 2005; Bertrand and Schoar 2006).

In turn, more trusting and loyal employees can be an advantage in economizing transaction costs (Williamson 1996). More specifically, the expense of auditing employee performance and employee/employer contractual protection costs, for example union representation, may be substantially less for family firms (Ouchi 1980).

Additional costs associated with labour may also be less for family firms than for their non-family counterparts. For example, although we recognise that agency costs for family firms are greater than zero,6 based on recent evidence provided by Chrisman et al. (2004), family involvement has been found to reduce overall agency problems. Empirically, lower agency costs in family firms have been shown by Moskowitz and Levering (1993) who find that family firms have lower recruitment costs, lower human resource costs, and are said to be more effective than other companies in labour-intensive businesses. Further labour cost savings can be realised because family members employed in the firm usually contribute more unpaid hours than paid employees (Benedict 1968; Rosenblatt 1985) and can potentially be paid lower efficiency wages based on a less binding non-shirking condition (De Paola and Scoppa 2009).

In addition to less costly labour, family firms may enjoy more flexible labour arrangements. Because family firms encourage informal, adaptive, and flexible work practices for their employees (Goffee and Scase 1985), those employees in turn may provide greater versatility in the production process. More specifically, family firm employees have greater occupational mobility within the firm (Becker 1974) and are, often, trained in all aspects of the business, spanning various departments and roles (Aronoff and Ward 1995; Fiegener et al. 1996). This attention to training develops through a lifetime of learning experiences inside the business leading to greater firm-specific employee knowledge, skills, and capabilities; a phenomenon which may not occur in non-family firms. Bhattacharya and Ravikumar (2001) refer to such tacit knowledge as the firm’s “special business skill” and a distinguishing characteristic of family firms. Such labour resources may be recruited to, or diverted from, specific production tasks with greater ease, further enhancing the flexibility of labour in the production process.

Despite the potential advantages, there are also some limitations on family firm labour. For example, family firms which hire employees who are also members of the family unit face a restricted labour pool from which to obtain qualified and capable talent (Anderson et al. 2003). In addition, it is recognised that there are further limits to family firm labour if family members are hired on the basis of nepotism, birth order, or gender rather than merit (Dyer 1988).

As is the case with capital, and despite the abovementioned differences, the assumption that labour output contributions are the same for family and non-family firms is prevalent in the literature. To test this assumption, our second hypothesis is as follows:

H2

There is no difference between the output contribution of labour in family and non-family firms.

The stylised facts presented thus far lead us to an interesting question: how does family involvement affect the total productivity of the firm? That is, considering the potential bearing that family involvement may have on input factors such as labour and capital, are there any differences in total factor productivity for family firms? Previous studies have suggested that there are (Wall 1998; Bosworth and Loundes 2002; Barth et al. 2005; Martikainen et al. 2009); however, those studies have assumed fixed labour and capital output contributions for family and non-family firms. Allowing for input heterogeneity, we formulate the following third and final hypothesis:

H3

Accounting for different input shares, there is no difference between total factor productivity of family and non-family firms.

Referring back to Table 1, previous studies have yet to address H3 in its specific form. If it is rejected, we are also interested in the direction of the difference in productivity between family and non-family firms; thus, secondary to H3 we wish to know if family firms are more or less productive than non-family firms. As in previous studies, we employ a Cobb–Douglas production function to test these hypotheses, but with the additional specification of varying production factor shares between family and non-family firms. This model and the data used are discussed in the next section(s) of this paper.

3 Methodology

A production function expresses the maximum product obtainable from the input combination at the current state of technical knowledge (Carlson 1909); thus, in order to test this paper’s proposed hypotheses, we respecify the standard Cobb–Douglas production function.

Our specification extends beyond what has already been analysed in previous research, because we allow for varying factor elasticities, or in other words different output contributions of labour and capital inputs, among family and non-family firms. Because the Cobb–Douglas framework measures output as a function of production inputs, we can observe the effect of a family firm on the production process, namely on the output contribution of labour and capital inputs and the total factor productivity of the firm.

It is well established in the economic literature that estimates of production function parameters in cross section analysis are subject to bias resulting from the exclusion of a variable which measures management (Mundlak 1961). This is mainly because of the difficulty of obtaining such a variable. More recently, it has become apparent that classifying family firms as a specific type of firm can be useful in solving this measurement problem. That said, in all other previous studies investigating family firm productivity using a Cobb–Douglas framework (Wall 1998; Bosworth and Loundes 2002; Barth et al. 2005; Martikainen et al. 2009), the standard log transformed Cobb–Douglas production function has been re-specified to include a family business intercept variable, so that:
$$ \ln (Y_{i} ) = \ln (A_{ij} ) + \alpha \ln (L_{i} ) + \beta \ln (K_{i} )\quad \left( {\alpha , \beta } \right) > 0; \ j = 1, 2 $$
(1)
Where Y is a measure of homogenous total output of the ith firm, L and K are measures of homogenous labour and capital inputs, respectively,7 and A, otherwise known as total factor productivity, is a constant for all qualitative forces which contribute to output yet are not represented in the quantitative measures of labour and capital. As specified in Eq. 1, total factor productivity differences between family and non-family firm types, denoted by j, are accounted for in the intercept; however, α and β, which represent the computed proportionate share of labour and capital in the total product, respectively, are assumed to be fixed for both firm types. Thus, the log transformation of the standard Cobb–Douglas production function for family and non-family firms differs only by the constant A in Eq. 1). As argued in the section “The nature of family firms”, fixing factor elasticities omits the effect a family may have on the output contribution of the firm’s labour and capital, and could result in biassed estimates of total factor productivity.

While such an assumption may be appealing when comparing total factor productivity for one firm with that for another, it does not shed any light on which qualitative forces are accountable or which inputs, labour or capital, are affected. Further, assuming equal factor elasticities may be particularly problematic, because the log transformation form of the Cobb–Douglas model implies that the factor elasticities for both labour and capital represent the percentage change in output with regard to labour (and holding capital constant) or capital (and holding labour constant), or in other words, the output contribution of labour and capital. The consequences of this assumption being violated when using a Cobb–Douglas framework are that actual differences in labour and capital output contribution, if present, are captured and aggregated in the total factor productivity constant8; thus the chance of finding a statistically significant difference in total productivity between family and non-family firms is enhanced when fixing the factor elasticities for both groups (Zellner et al. 1966).

To address this problem, we respecify Eq. 1, so that
$$ \ln (Y_{i} ) = \ln (A_{ij} ) + \alpha_{j} \ln (L_{i} ) + \beta_{j} \ln (K_{i} ) + \theta X_{i} + e_{i} \quad \left( {\alpha , \beta } \right) > 0; \ j = 1, 2. $$
(2)
It can be seen in Eq. 2 that we allow for heterogeneous labour and capital output contributions among the two types of firm, denoted by j. As a result the qualitative differences that one may find when using Eq. 1 are now permitted to be distributed among αj and βj. X represents a vector of control variables, for example the firm’s industry and age. It is also important methodologically to resolve any simultaneous equation bias that may be present in our sample data by using a two-staged least-squares, with instruments, regression technique (Ramsey 1969).9

Estimating the parameters specified in Eq. 2 enables us to test our proposed hypotheses. For example, H1 and H2 can be tested by observing whether there are any significant differences between family and non-family firms for estimated values of α and β, respectively. H3 can be tested by observing whether there are any significant differences between family and non-family firms for the estimated value of A. The direction of any found differences is also interesting, as a positive (negative) value of A for family firms would indicate an additive (subtractive) difference in total factor productivity specifically for family firms relative to their non-family firm counterparts, considering different capital and labour output contributions. To perform such tests, we must first have a reliable data source, which is discussed next.

4 Data: the Business Longitudinal Survey

The Australian Bureau of Statistics” “Business Longitudinal Survey” (BLS) was designed to provide information on the growth and performance of privately held Australian small and medium-sized enterprises (SME), i.e. fewer than 200 employees. The BLS is the longitudinal component of several waves of the “Business Growth and Performance Survey”. As such, the structure of the data includes not only a cross-sectional component, but also a longitudinal aspect for the years 1994–1995 to 1997–1998 inclusive.10 The BLS contains potential information for many areas of research, including industrial relations, business, finance, and economics (Hawke 2000); however, for the purpose of our analysis, and considering the neo-classical theory of production, some narrowing of the data was required.

For example, based on the standard Cobb–Douglas production function, only those firms which reported positive values for our measures of output and inputs were included. Furthermore, to test our results over time, and to eliminate any selection or attrition bias, this study is exclusively focussed on production output and inputs for those firms which participated in each year of the study. Consequently, firms which did not participate in each wave of the BLS, from 1994 to 1998, were excluded. This treatment has reduced our sample to 3,364 firms per year. The sub-sample, classified by industry and year, can be seen in Table 2.
Table 2

BLS data sub-sample

BLS defined industry

Firms in 1994–1995 sub-sample

Firms in 1995–1996 sub-sample

Firms in 1996–1997 sub-sample

Firms in 1997–1998 sub-sample

Average proportions by industry from 1995–1998 (%)

Family firms in sub-sample

Average proportion of family firms from 1995–1998 (%)

Mining

27

26

26

27

0.79

7

24.53

Manufacturing

1372

1374

1369

1358

40.67

735

53.72

Construction

185

189

190

190

5.6

138

73.08

Wholesale trade

564

561

562

565

16.74

296

52.53

Retail trade

348

353

354

353

10.46

215

61.15

Accommodation, cafes, and restaurants

120

120

121

123

3.60

59

48.55

Transport and storage

126

124

123

123

3.69

72

57.86

Finance and insurance

65

66

67

69

1.98

35

52.43

Property and business services

429

424

425

428

12.68

149

34.94

Cultural and recreational services

61

61

61

61

1.81

16

26.23

Personal and other services

67

66

66

67

1.98

37

56.02

Total

3364

3364

3364

3364

100

1758

52.26

Although the same firms are analysed in each year, their industry classification may change from one year to the next. This results in a slight variation in the total number of firms in each industry over time. Despite this, on average it can be said that manufacturing firms represent approximately 40% of all firms sampled, followed by nearly 17% representation of wholesale trade firms and 13% representation of property and business service firms; whereas less than 1% of our sample is represented by mining firms, which is understandable considering we are focussing on small to medium sized enterprises, i.e. firms with less than 200 employees.

Most importantly, for the purpose of this study, the BLS includes information on the extent of family involvement in each of the firms included in our sample, which, with other relevant variables, is discussed further in the next section.

4.1 Variable treatment

Referring to Eq. 2, where Y is a measure of total physical output, and considering that the BLS does not offer data on “output” per se, an index number for value added (VA) is used as a proxy for total output. Such an index follows Kenneth Arrow’s (1974) generally accepted “real value added” measure and is constructed by taking sales plus the change in inventories less purchases of intermediate inputs and other operating expenses. Although from a purely theoretical standpoint we would rather use actual output, in terms of number of units produced, the value-added index enables us to analyse those firms which do not necessarily have a tangible output, for example services rendered. Furthermore, the value-added index has been found to accurately measure the dependent variable in the production function that explains value added in terms of tangible and intangible primary factors, for example labour and capital, and as such the function is independent of non-primary inputs (Sato 1976).

Similar to our treatment of the dependent variable, the independent variables specified in Eq. 2 deviate slightly from the theoretical notion of primary production inputs in the sense that they are derived from the BLS data. Here we briefly discuss our proxies and operationalization of these variables.

4.1.1 Labour input

Rather than the number of labour hours worked as a measure of labour input, the number of full-time equivalent (FTE) workers employed in the firm is used. This figure is found via the sum of full-time workers and full-time equivalent part-time workers. Full-time equivalent part-time workers are found via the product of the number of part-time employees for each individual firm and a full-time equivalent ratio. The equivalent ratio is the Australian Bureau of Statistics’ estimate of average hours worked by part-time non-managerial employees per week in time t compared with full-time employees for all firms.11 Because our sampled firms utilise both part-time and full-time labour, the transformation of number of workers to FTE workers is essential to obtain a standardised, comparable measure of labour.

4.1.2 Capital input

The difficulty in measuring capital and then applying such a measurement in a production function framework has been the cause of much controversy over the years. Joan Robinson’s (1953) now famous critique shed light on these problems, sparking the so-called Cambridge–Cambridge controversies. We attempt to alleviate some of these problems by viewing a firm’s stock of capital as the specific list of all the goods in existence at any given moment. As far the BLS data are concerned, we measure capital as the value of all assets in the firm’s possession. Our measure of capital is analysed in a cross-sectional framework, so changing capital stock over time does not come into consideration because, in the short-run, the supply of concrete capital goods is assumed not to alter. Furthermore, the BLS contains data on the net value of total assets, so depreciation is factored into the measurement. Finally, we consider the notion that capital is heavily user-dependent and heterogeneous by controlling, in our analysis, for industry and for the age of the firm.

4.1.3 Family involvement

To determine the interaction between family involvement and a firm’s production inputs, we first need a measure of family involvement. As discussed, a family firm can be defined by using both a structure and essence-based approach. The BLS data contain unique information in this regard, because the following questions, asked of all businesses which participated, were included in the survey.
  1. 1.

    Do you consider the business to be a family business? Yes/No.

     
  2. 2.
    If yes, why do you consider this a family business? Family member are:
    1. i.

      Working directors or proprietors. Yes/No.

       
    2. ii.

      Employed in the business. Yes/No.

       
    3. iii.

      Not working, but contribute to decisions. Yes/No.

       
    4. iv.

      Business acquired from parents. Yes/No.

       
    5. v.

      Close working relationship between management and staff. Yes/No.

       
    6. vi.

      Other. Yes/No.

       
     
On the basis of the question listed above, a family firm is defined as one for which the answer “yes“ was given to question 1. Furthermore, considering one’s business as a family firm could be because of one or more of the reasons listed under question 2. This list captures the structural definition of a family firm, i.e. options i to iii, and the essence of family firms, i.e. options iv and v. It is important to note that the options listed under question 2 are not mutually exclusive and thus identifying different “types” of family firms within these options does not lead to practical classifications12; thus, for the purpose of our analysis and given Eq. 2, we only consider the overarching question 1 and, on the basis of a standard Hausman test, use elements of question 2 as instrument variables in our econometric analysis as to control for simultaneous equation bias. Of the 3,364 firms included in our sample (Table 2), 1,758, or 52%, are regarded as family firms and 1,606, or 48%, are not. Table 3 outlines our proxies and operationalization of the dependent and independent variables using the BLS data.
Table 3

BLS proxies for dependent and independent variables

Concept

Proxy variable

Operationalizationa

Production output

Value added

VAit = Salesit + closing inventoryit − opening inventoryit − purchasesit

Labour input

Full-time equivalent employees

FTEit = Full-time employeesit + part-time employeesit × equivalent ratiot

Capital input

Total assets

Capitalit = Total liabilitiesit + total equityit

Family business

Structure and essence based definition

Do you consider the business to be a family business?

Yes = 1; No = 0

Capital intensity

Capital to Labour ratio

K/Lit = Capital inputit/labour inputit

Type of labour

Part-time ratio

PTit = Part-time employeesit/full-time employeesit

Labour cost

Wage ratio

WRit = Total wage expenseit/total expensesit

ai denotes an individual firm in time period t

Table 3 also reports additional variables, namely the capital to labour, part-time, and wage ratios, which offer greater insight into the production of family firms, but are not directly included in Eqs. 1 and 2. Despite this, the wage ratio is used as an alternative to labour input in the section “Robustness checks”.

4.1.4 Control variables

As specified in Eq. 2, we control for heterogeneity across sectors by including both slope and intercept industry dummies. This is important, because the distribution of family firms across sectors, outlined in Table 2, is uneven. Also seen in Table 2, eleven Australian and New Zealand Standard Industrial Classifications (ANZSIC) were incorporated, leading to twenty additional estimated parameters.13 Further heterogeneity across the life cycles of different firms is also controlled for by including an age variable in our specification.

4.2 Summary statistics

Beginning with Mendershausen’s (1938) early criticisms, it is well known that there are usually high correlations among the “independent” variables used by Cobb and Douglas. We can see in Table 4 that, although the relationship between labour and capital is quite weak, the BLS data are not immune to this common issue; however, because perfect multicollinearity is not present, we can still claim that our estimates will remain unbiased. Furthermore, given our stated hypotheses, the presence of multicollinearity in Eq. 2 should not hinder our ability to observe any potential family firm differences among input contributions and total factor productivity.
Table 4

Correlation table

 

Value added (95)

Value added (96)

Value added (97)

Value added (98)

Labour (95)

Labour (96)

Labour (97)

Labour (98)

Capital (95)

Capital (96)

Capital (97)

Capital (98)

Value added (95)

1.000

 

 

 

 

 

 

 

 

 

 

 

t statistic

 

 

 

 

 

 

 

 

 

 

 

Value added (96)

0.922

1.000

 

 

 

 

 

 

 

 

 

 

t statistic

138.271

 

 

 

 

 

 

 

 

 

 

Value added (97)

0.827

0.925

1.000

 

 

 

 

 

 

 

 

 

t statistic

85.153

141.605

 

 

 

 

 

 

 

 

 

Value added (98)

0.900

0.930

0.885

1.000

 

 

 

 

 

 

 

 

t statistic

119.635

146.709

110.160

 

 

 

 

 

 

 

 

Labour (95)

0.588

0.556

0.561

0.564

1.000

 

 

 

 

 

 

 

t statistic

42.200

38.759

39.319

39.592

 

 

 

 

 

 

 

Labour (96)

0.587

0.568

0.574

0.586

0.958

1.000

 

 

 

 

 

 

t statistic

42.012

40.029

40.641

41.927

193.736

 

 

 

 

 

 

Labour (97)

0.562

0.548

0.571

0.578

0.925

0.957

1.000

 

 

 

 

 

t statistic

39.370

38.011

40.310

41.053

141.261

191.821

 

 

 

 

 

Labour (98)

0.514

0.517

0.555

0.563

0.884

0.914

0.937

1.000

 

 

 

 

t statistic

34.711

34.986

38.704

39.464

109.610

130.193

154.897

 

 

 

 

Capital (95)

0.675

0.764

0.704

0.769

0.334

0.343

0.324

0.295

1.000

 

 

 

t statistic

53.009

68.719

57.449

69.660

20.559

21.174

19.829

17.924

 

 

 

Capital (96)

0.702

0.775

0.703

0.785

0.323

0.337

0.317

0.284

0.983

1.000

 

 

t statistic

57.094

71.088

57.339

73.386

19.785

20.744

19.350

17.187

312.093

 

 

Capital (97)

0.657

0.685

0.581

0.727

0.300

0.323

0.303

0.262

0.928

0.954

1.000

 

t statistic

50.500

54.525

41.435

61.340

18.232

19.763

18.460

15.715

144.384

184.418

 

Capital (98)

0.645

0.699

0.617

0.745

0.300

0.322

0.308

0.278

0.947

0.950

0.971

1.000

t statistic

48.917

56.740

45.473

64.676

18.205

19.750

18.785

16.779

171.053

176.482

234.291

Moving on to other characteristics of the data, using the labour and capital input proxies defined in Table 3, and the capital to labour, part-time and wage ratios, we summarize the distribution among the full sample, family firms, and non-family firms in Table 4. Table 5, outlines any differences among these variables within our sampled industries between family and non-family firms and reports the standard t test results for significance.
Table 5

Summary statistics for BLS sub-sample (1995–1998)

Variable (year)a

Panel A: full sample (N = 3,364)

Panel B: family firms (N = 1,730)

Panel C: non-family firms (N = 1,634)

25th percentile

Median

Mean

75th percentile

25th percentile

Median

Mean

75th percentile

25th percentile

Median

Mean

75th percentile

Total labour (95)

4.85

13

26.31

35.84

4

9.97

21.04

28

6

18

31.89

45.43

Total labour (96)

4.85

13

26.49

35.82

4

10

20.96

28.35

6

18

32.35

46

Total labour (97)

4.85

13.28

26.56

36

4

10

21.18

28.96

6.28

18.25

32.27

45

Total labour (98)

4.85

13.68

27.59

38

4

10

22.08

30

6

18.48

33.42

46

Total capital (95)

181

755

4607.98

3204

142.25

542

2536.09

1949.75

245.75

1147

6801.61

5044

Total capital (96)

205.75

814

4971.44

3405.5

166.25

584.5

2609.95

2067

265.25

1238.5

7471.66

5375.75

Total capital (97)

218.75

860.5

5251.04

3547

169

602.5

2841.39

2223.25

291.25

1292

7802.27

5559.5

Total capital (98)

220.75

890.5

5414

3639.5

167.25

642

3084.28

2306.5

300.5

1333.5

7880.6

5922

Capital to labour ratio (95)

27.46

59.19

128.11

119.77

25.56

53.82

91.94

98.73

29.8

66.49

166.4

144.61

Capital to labour ratio (96)

30.53

61.77

135

126.62

29.01

56.12

97.1

106.63

33.38

74.69

175.14

154.2

Capital to labour ratio (97)

32.37

66.35

149.25

132.37

30.02

58.47

106.77

111.28

35.92

75.93

194.23

165.36

Capital to labour ratio (98)

31.95

66.42

172.26

136.03

30.02

59.29

116.13

116.03

34.44

78.22

231.68

164.18

Part-time employee ratio (95)

0.00%

4.55%

17.03%

25.00%

0.00%

5.00%

18.27%

30.29%

0.00%

4.17%

15.71%

22.22%

Part-time employee ratio (96)

0.00%

5.56%

17.81%

27.57%

0.00%

5.88%

18.73%

33.33%

0.00%

5.23%

16.83%

25.00%

Part-time employee ratio (97)

0.00%

2.50%

15.85%

21.96%

0.00%

2.33%

16.83%

25.00%

0.00%

2.55%

14.81%

18.83%

Part-time employee ratio (98)

0.00%

1.96%

16.43%

25.00%

0.00%

2.40%

17.88%

27.78%

0.00%

1.66%

14.90%

19.48%

Wage to total expense ratio (95)

11.94%

21.55%

25.26%

34.74%

11.91%

21.50%

24.54%

34.07%

12.05%

21.60%

26.02%

36.24%

Wage to total expense ratio (96)

12.13%

21.99%

25.20%

34.58%

11.91%

22.01%

24.53%

33.59%

12.20%

21.90%

25.91%

35.98%

Wage to total expense ratio (97)

11.85%

21.82%

24.97%

34.66%

11.88%

21.87%

24.52%

33.33%

11.73%

21.78%

25.44%

36.09%

Wage to total expense ratio (98)

12.06%

22.43%

25.59%

35.71%

12.04%

22.41%

24.99%

33.99%

12.12%

22.44%

26.22%

37.35%

aLabour is measured in FTE and capital is measured in $000

Using total assets and total number of FTE employees as indications of firm size, or capital and labour inputs as defined in Table 3, we can see that family firms are generally smaller than non-family firms. That is, on average, consistent throughout the entire sample period, family firms employ fewer workers and own less capital than non-family firms. To control for any potential bias because of an industry effect (Table 2), we can observe in Table 6 that this size difference is significant and is valid across nearly all industries in the form of either less capital, less labour, or both.14 The finding that family firms are generally smaller is consistent with previous studies which have observed that, as firm size decreases, family ownership becomes more common (Anderson et al. 2003).
Table 6

Industry comparison of summary statistics between family and non-family firms (average for 95–98)

BLS defined industry

Mean difference in family capital

Mean difference in family labour

Mean difference in family PT ratio

Mean difference in family KL ratio

Mean difference in family wage ratio

Mining

−56715.61

−47.85***

3.10

−1.83

0.183*

Manufacturing

−4470.81***

−12.47***

2.49***

−57.25***

0.024***

Construction

457.30

−4.38

2.91

−13.49

0.006

Wholesale trade

−6277.34***

−10.79***

2.60**

−100.45***

0.003

Retail trade

−1321.63***

−6.79**

−1.57

−36.97*

0.014

Accommodation, cafes, and restaurants

−1685.96

−53.39

43.83

−9.43

−0.167

Transport and storage

−10271.41

−15.31**

2.54

−118.04

−0.031

Finance and insurance

−7206.24***

−21.82***

12.91**

−81.52

0.055

Property and business services

1227.29

−14.47***

8.63***

39.64

−0.028

Cultural and recreational services

−10514.24

−16.06*

5.51

−332.86

−0.026

Personal and other services

202.23

5.90

12.56***

6.44

−0.049

All industries

−4719.18***

−11.27***

2.84***

−85.33***

−0.011*

* Significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level

The capital to labour ratio is defined as capital input over our standardised labour input and measures the capital intensity of any given firm. Table 5, Panel B indicates that family firms are less capital-intensive than non-family firms. This result is consistent throughout the entire sample period and, when broken down by industry, is most prevalent in the manufacturing, wholesale trade, and retail trade industries.

Lower capital intensity in theory would diminish labour productivity, because the use of tools and machinery can make labour more effective.15 Despite this, the capital to labour ratio alone only offers some insight into productivity and is more a measure of how production is undertaken, i.e. allocation of labour and capital inputs, rather than how effective said inputs contribute to output. Nevertheless, we report the capital to labour ratio here to obtain a better understanding of our specified production function in the “Methodology” section.

A “part-time ratio” is defined as the proportion of part-time workers utilised, and has been calculated as the total number of part-time workers over the number of all workers employed in any given firm.16 From this we can observe what type of labour, part or full-time, family firms prefer. Table 5, Panel B indicates that, across all years, family firms utilise more part-time workers than non-family firms. This result is significant in nearly half of all industries included in the BLS sample. One explanation could be that family firms initially associate offspring with the firm through family discussions or part-time employment (Stavrou 1999). These results may also underpin the flexible labour arrangements enjoyed by family firms, as discussed in the section “The nature of family firms”.

Given a significant preference for part-time labour and hiring, on average, fewer employees, family firms do not necessarily enjoy cheaper labour in terms of wage costs as a proportion of total expenses. Although the results vary across industry and are for the most part insignificant, we do observe some significant differences in the manufacturing and mining industries, where the wage ratio is respectively 2 and 18% higher. The wage ratio as an indicator of labour cost is a very crude measure as it accounts for total labour expenses across all levels of employment, and thus may be subject to bias; however, following Barth et al. (2005), such a measure may be useful as an alternative measure of labour input as far as the production function is concerned.

On the basis of these summary statistics, it is sufficient to say that there are observable differences between family-firm production factors relative to their non-family counterparts; however, it is still unknown whether these differences translate into unique output contributions of family labour and capital. To address this issue, using the variables outlined in the section “Data: The Business Longitudinal Survey”, we proceed to estimate our model specified in the section “Methodology”, the results of which are discussed next.

5 Empirical results

It can be seen from Eq. 2 that we allow for heterogeneous inputs for both family and non-family firms, which is the main difference between our analysis and previous research (Table 1). As mentioned, we also control for firm industry and age. In this section, the empirical results for each of our stated hypotheses are presented.

As a first step, and replicating analysis in the literature (Wall 1998; Bosworth and Loundes 2002; Barth et al. 2005; Martikainen et al. 2009), we estimate the coefficients for Eq. 1, which assumes homogeneous output contributions of labour and capital, and accounts for management bias in a family business intercept dummy alone. These results are listed in Table 7.
Table 7

Family-firm production function with homogeneous input

 

 

Equation 1 (OLS)

Equation 1 (2SLS)

1995

1996

1997

1998

1995

1996

1997

1998

Intercept

2.796***

(0.152)

2.668***

(0.139)

2.723***

(0.147)

2.719***

(0.129)

2.795***

(0.152)

2.672***

(0.139)

2.722***

(0.147)

2.721***

(0.129)

Family firm

−0.124***

(0.021)

−0.128***

(0.020)

−0.114***

(0.020)

−0.139***

(0.020)

−0.123***

(0.022)

−0.132***

(0.021)

−0.113***

(0.021)

−0.141***

(0.021)

ln Labour

0.709***

(0.076)

0.678***

(0.073)

0.649***

(0.081)

0.629***

(0.083)

0.709***

(0.076)

0.678***

(0.073)

0.649***

(0.081)

0.629***

(0.083)

ln Capital

0.329***

(0.037)

0.364***

(0.039)

0.370***

(0.045)

0.383***

(0.039)

0.328***

(0.037)

0.364***

(0.039)

0.370***

(0.045)

0.382***

(0.039)

R2 adj

0.854

0.873

0.862

0.871

0.854

0.873

0.862

0.871

N

3364

3364

3364

3364

3364

3364

3364

3364

Equation (1) is estimated as ln (Yi) = ln (Aij) + αln (Li) + βln (Ki) + ei, where j = 1 for non-family firms and 2 for family firms. The estimate of A2 for family firms (labelled Family firm) measures the difference between family firm and non-family firm total factor productivity and thus indicates whether family firms are more or less productive. Both an ordinary least-squares (OLS) technique (left column) and a two-staged least-squares (2SLS) technique (right column) were implemented. Instrument variable for Family firm is question 2, subsection i of the BLS. White-corrected standard errors are reported in parentheses. Level of significance: *** 1%; ** 5%; * 10%

Consistent with Barth et al. (2005), Bosworth and Loundes (2002) and Wall (1998), we find that family firms are significantly less productive, from a total factor productivity standpoint, than non-family firms. This productivity gap is identified as the Family firm intercept in Table 7 and ranges from as little as 11% in 1997 to nearly 14% in 1998. Also consistent with previous findings, this gap is enlarged when endogeneity problems are addressed using the two-staged least-squares technique (right column).

The problem with these results is that the labour and capital production inputs are assumed to yield the same output contributions for both firm types. As we have argued in this paper, such an assumption ignores the effect the family may have on its labour and capital inputs. In fact, we have yet to establish whether this total factor productivity gap remains when we account for heterogeneous production inputs between family and non-family firms. To test this and our remaining hypotheses, we estimate Eq. 2, the results of which are listed in Table 8. Both the ordinary least-squares (left column) and two-staged least-squares estimates (right column) are presented.
Table 8

Family-firm production function with heterogeneous inputs

 

Equation 2 (OLS)

Equation 2 (2SLS)

1995

1996

1997

1998

1995

1996

1997

1998

Intercept

2.557***

(0.139)

2.343***

(0.134)

2.602***

(0.140)

2.583***

(0.134)

2.557***

(0.140)

2.362***

(0.135)

2.637***

(0.141)

2.614***

(0.135)

Family firm

0.019

(0.076)

0.012

(0.074)

−0.061

(0.077)

0.022

(0.075)

0.019

(0.082)

−0.024

(0.082)

−0.128

(0.085)

−0.039

(0.083)

ln Labour

0.757***

(0.055)

0.798***

(0.052)

0.835***

(0.055)

0.681***

(0.052)

0.753***

(0.056)

0.802***

(0.052)

0.843***

(0.055)

0.689***

(0.052)

ln Capital

0.308***

(0.031)

0.343***

(0.030)

0.281***

(0.032)

0.347***

(0.030)

0.309***

(0.032)

0.339***

(0.030)

0.274***

(0.032)

0.339***

(0.030)

ln Family labour

0.141***

(0.029)

0.108***

(0.027)

0.077***

(0.027)

0.083***

(0.026)

0.152***

(0.031)

0.100***

(0.029)

0.062**

(0.029)

0.064***

(0.027)

ln Family capital

−0.075***

(0.018)

−0.059***

(0.017)

−0.035**

(0.018)

−0.051***

(0.017)

−0.079***

(0.019)

−0.051***

(0.019)

−0.019

(0.019)

−0.035**

(0.018)

R2 adj

0.865

0.884

0.877

0.883

0.865

0.884

0.878

0.883

N

3364

3364

3364

3364

3364

3364

3364

3364

Equation 2 is estimated as ln (Yi) = ln (Aij) + αjln (Li) + βjln (Ki) + ei. Where j = 1 for non-family firms and 2 for family firms. Estimates of A2 (labelled Family firm) measures the difference between family and non-family firms from a total factor productivity standpoint and estimates of α2 (labelled Family labour) and β2 (labelled Family capital) measure the output contribution differences of family labour and capital respectively. Dummy intercept and slope variables were also included to control for firm industry and age; for brevity these results are not listed in Table 8. Both an ordinary least-squares (OLS) technique (left column) and a two-staged least-squares (2SLS) technique (right column) was implemented. Instrument variable for Family firm is question 2, subsection i of the BLS. White-corrected standard errors are reported in parentheses. Level of significance: *** 1%; ** 5%; * 10%

On the basis of the results listed in Table 8, we find that differences in the output contribution of family-firm capital (denoted ln Family capital) are significant relative to their non-family counterparts. As a semi-log, the negative coefficient indicates that family capital contributes less to output than non-family-firm capital by approximately 2–8%, depending on time period and estimation technique. This estimate can be interpreted as for all capital utilised, family-firm capital contributes less to total output than the benchmark non-family-firm capital (denoted ln Capital). This finding refutes H1. Furthermore, this relationship is consistent throughout time with the exception of 1997 in the two-staged least-squares technique. Despite this, the coefficient is still negative and it is worth noting that the ordinary least-squares technique estimates a significant and negative coefficient for family-firm capital in 1997.

Also apparent in Table 8 is the positive and consistent interaction between family firms and labour employed in the production process. Depending on the time period and estimation technique, we find differences in the output contribution of family firm labour (denoted ln Family labour) are significant in that it is greater than the output contribution of non-family firm labour; this difference ranges from as little as 6% to as large as 15%. This estimate can be interpreted as for all labour employed, family firm labour contributes more to total output than the benchmark non-family firm labour (denoted ln Labour). This result refutes H2 and is consistent across all time periods and remains after correcting for potential simultaneous equation bias.

Moving on to H3, interestingly, the intercept representing family firm total factor productivity differences (denoted Family firm) has become statistically insignificant when heterogeneous inputs are specified in the Cobb–Douglas production function. That is, family firm labour and capital are found to yield significantly different output contributions and, when these differences are accounted for, total factor productivity differences disappear. In this sense, H3 is confirmed in that there are no unquantifiable productivity differences between family and non-family firms when heterogeneous inputs have been considered. This leads us to believe that previous investigations suffer from omitted variable bias in that they do not consider the family differences in output elasticity of labour and capital.

Finally, to test if there is a time effect not being captured in the single equation estimation of Eq. 2, we also estimate the group of annual production functions simultaneously across all years within a system. The results obtained are outlined in Table 9.
Table 9

Family firm and total factor productivity with heterogeneous inputs (system estimates)

 

Simultaneous estimate (2SLS)

H0: βi,95 = βi,96 = βi,97 = βi,98

1995

1996

1997

1998

Chi-squared value, Prob.

Intercept

2.689***

(0.140)

2.366***

(0.134)

2.633***

(0.140)

2.603***

(0.134)

 

Family firm

−0.006

(0.076)

−0.092

(0.074)

−0.121

(0.077)

−0.018

(0.075)

1.329, 0.722

 

ln Labour

0.761***

(0.055)

0.856***

(0.052)

0.842***

(0.055)

0.687***

(0.052)

 

ln Capital

0.314***

(0.031)

0.320***

(0.030)

0.274***

(0.032)

0.341***

(0.030)

 

ln Family labour

0.146***

(0.029)

0.081**

(0.027)

0.064**

(0.027)

0.069**

(0.026)

4.852, 0.183

 

ln Family capital

−0.071***

(0.018)

−0.032

(0.017)

−0.021

(0.018)

−0.040**

(0.017)

3.475, 0.324

 

R2 adj

0.865

0.884

0.877

0.883

 

N

13456

   

 

System estimation of Eq. 2 is estimated as ln (Yit) = ln (Aijt) + αjln (Lit) + βjln (Kit) + eit. Where j = 1 for non-family firms and 2 for family firms, i = (1,…,3364) and t = (1,…,4). Estimates of A2 (labelled Family firm) for family firms measure the difference between family and non-family firms from a total factor productivity standpoint and estimates of α2 (labelled Family labour) and β2 (labelled Family capital) measure the output contribution differences of family labour and capital respectively. Dummy intercept and slope variables were also included to control for firm industry and age; for brevity these results are not listed in Table 9. A two-staged least-squares technique was implemented. Instrument variable for Family firm is question 2, subsection i of the BLS. White-corrected standard errors are reported in parentheses. Level of significance: *** 1%; ** 5%; * 10%

As seen in Table 9, the simultaneous coefficient estimates yield slightly different values compared with Table 8, because the system method takes into account the cross-equation error correlations when estimating parameters. This makes the model more sensitive to specification error. In fact, the estimates for Family-firm capital in 1996 and 1997 are now insignificant; however, it is worth reporting here the p-values of 0.123 and 0.272 respectively. Despite this the sign and magnitude are in accordance with our single-equation estimates.

To check the stability of these results over time, we employ a Wald coefficient test by imposing the following restriction: βi,95 = βi,96 = βi,97 = βi,98, where i = the family firm intercept and interaction effects. The chi-squared values and probabilities presented in the extreme right column of Table 9 indicate that we do not reject the possibility that these values are equal across time. In other words, we do not find any significant time effect. Based on the analysis performed in this section, we have refuted our hypotheses proposed in the section “The nature of family firms” with the interesting exception of H3. Next we test whether these results are valid when using alternative specifications to Eq. 2.

5.1 Robustness checks

In Table 10 we report some alternative specifications to Eq. 2. Keeping with our focus, only differences in family firm total productivity and production input contributions are reported.
Table 10

Output contribution effects for family firms under alternative specifications

Alternative specifications

N = 3364

Family firm intercept

Family labour

Family capital

R2 adj

N

Proportion of family firms (%)

Alternative measure of output

      

 Dependent variable: ln (Sales) 95

0.036

(0.083)

0.140***

(0.031)

−0.080***

(0.019)

0.869

3364

51.43

 Dependent variable: ln (Sales) 96

0.066

(0.083)

0.103***

(0.030)

−0.067***

(0.019)

0.883

3364

51.43

 Dependent variable: ln (Sales) 97

0.041

(0.083)

0.090***

(0.029)

−0.057***

(0.019)

0.885

3364

51.43

 Dependent variable: ln (Sales) 98

0.135

(0.083)

0.127***

(0.028)

−0.086***

(0.019)

0.884

3364

51.43

Alternative measures of labour input

      

 Wage ratio (95)

−0.039

(0.094)

0.012

(0.066)

−0.007

(0.0141)

0.787

3364

51.43

 Wage ratio (96)

−0.120

(0.090)

0.017

(0.014)

−0.009

(0.0142)

0.812

3364

51.43

 Wage ratio (97)

−0.169

(0.092)

0.001

(0.012)

−0.007

(0.014)

0.804

3364

51.43

 Wage ratio (98)

−0.158

(0.091)

0.001

(0.012)

−0.003

(0.014)

0.810

3364

51.43

Sub samples

      

 10 employees or fewer (95)

0.132

(0.138)

0.174***

(0.070)

−0.107***

(0.029)

0.545

1504

58.98

 10 employees or fewer (96)

0.352

(0.139)

0.069

(0.066)

−0.117***

(0.029)

0.608

1475

58.85

 10 employees or fewer (97)

0.204

(0.141)

0.141***

(0.063)

−0.103***

(0.029)

0.592

1471

59.41

 10 employees or fewer (98)

0.008

(0.136)

0.134**

(0.061)

−0.078***

(0.028)

0.602

1478

59.74

Between 10 and 100 employees (95)

0.123

(0.156)

−0.039

(0.050)

−0.009

(0.024)

0.713

1724

46.87

Between 10 and 100 employees (96)

0.123

(0.146)

−0.059

(0.046)

−0.004

(0.022)

0.751

1749

47.22

Between 10 and 100 employees (97)

−0.164

(0.156)

−0.079

(0.049)

0.045

(0.023)

0.724

1747

46.82

Between 10 and 100 employees (98)

0.239

(0.150)

−0.032

(0.048)

−0.033

(0.023)

0.752

1709

46.63

More than 100 employees (95)

1.399

(2.922)

0.319

(0.625)

−0.317**

(0.092)

0.519

136

25.73

More than 100 employees (96)

2.934

(2.649)

−0.645

(0.589)

0.004

(0.085)

0.578

140

25.71

More than 100 employees (97)

0.5911

(2.869)

−0.219

(0.607)

0.033

(0.075)

0.554

146

26.03

More than 100 employees (98)

−0.531

(2.131)

0.052

(0.454)

0.014

(0.067)

0.580

177

28.24

13 employees or fewer (95)

0.051

(0.128)

0.179***

(0.061)

−0.090***

(0.027)

0.597

1690

58.34

13 employees or fewer (96)

0.263

(0.125)

0.081

(0.056)

−0.101***

(0.026)

0.658

1686

58.48

13 employees or fewer (97)

0.110

(0.128)

0.129***

(0.054)

−0.080***

(0.027)

0.641

1674

59.08

13 employees or fewer (98)

0.049

(0.125)

0.104**

(0.053)

−0.065***

(0.026)

0.643

1666

59.06

More than 13 employees (95)

0.233

(0.162)

0.022*

(0.051)

−0.052***

(0.024)

0.739

1674

44.44

More than 13 employees (96)

0.244

(0.157)

−0.062

(0.047)

−0.016*

(0.024)

0.757

1678

44.34

More than 13 employees (97)

−0.089

(0.168)

−0.071

(0.049)

0.031

(0.025)

0.739

1690

43.85

More than 13 employees (98)

0.352

(0.0154)

−0.039

(0.046)

−0.042*

(0.023)

0.775

1698

43.93

White-corrected standard errors are reported in parentheses. Level of significance: *** 1%; ** 5%; * 10%

First we regress Eq. 2 using sales as the dependent variable. This procedure is consistent with Palia and Lichtenberg (1999) and Barth et al. (2005), who believe total sales to be an acceptable gauge of production output. As seen in Table 10, the relationships found in Table 8 are not affected by this alternative treatment. That is, the output contributions of family labour (capital) are still found to be greater (less) than their non-family counterparts. Furthermore, the family firm total factor productivity intercept remains insignificant once output contribution heterogeneity is accounted for. However, because the value of intermediate goods is omitted, we prefer the analysis presented in Table 8.

Next we address the issue of potential differences in labour quality by using the wage ratio, as defined in Table 3, as an alternative measure of labour input. The validity of such a measure depends heavily on the assumption that wage rates equate to the marginal productivity of labour. In Table 10 we can see that although no statistical significance is found, the direction of the family labour and capital relationships are maintained. More importantly, the family firm total factor productivity intercept remains insignificant. We still prefer the measure of labour input specified in Eq. 2, because labour quality differences between ownership structure are, in fact, already accounted for when we allow for labour output contribution heterogeneity. Furthermore, as Bath et al. (2005) report, if wages are related to productivity, they will be correlated with the error term and thus introduce a bias to the estimators of the production function.

To address the potential for our results being driven by underlying heterogeneity rather than differences in ownership structure, we narrow our sample to three sub-samples depending on the number of workers employed. Doing so would indicate whether the effects presented in Table 8 are occurring across all firms, irrespective of size. Arbitrarily, we classify firms with up to 10, between 10 and 100, and more than 100 employees. Not surprisingly, we find the family effects are strongest in firms with up to 10 employees, where nearly 60% of all family firms in our sample are concentrated. As family firm concentration falls, as is the case for firms which employ between 10 and 100 workers, differences in production factor output contribution become insignificant, yet the relationships are maintained. In firms with more than 100 employees, a sub-sample consisting of only 25% family firms, the relationships presented in Table 8 break down.

To check if this result is because of an insufficient number of family firm observations as firm size increases, we break the sample in half and report the results of Eq. 2. On the basis of the BLS data, approximately half of all firms have 13 employees or fewer. Utilising this classification, we can see that the family firm effect is significant in the smaller half of the sample across nearly all years. In the larger half of the sample the family effect is more prevalent in diminished capital output contribution, but less so in enhanced labour output contribution. This result suggests there may be a size effect in that larger family firms do not necessarily enjoy greater output contributions from their labour inputs as smaller family firms do. Despite this, and consistent with our results, when input heterogeneity is accounted for, there is no difference in total factor productivity between family and non-family firms in either classification or across any alternative specification.

As a final measure of robustness, we estimate Eq. 2 in a panel framework. Not only will such a framework effectively control for any unobserved heterogeneity, but it utilizes all the available data in our BLS sample. In the context of our analysis, we estimate a random-effects, rather than fixed-effects, model because:

  1. 1

    the family ownership constant we are concerned with in this analysis cannot be directly estimated by use of a fixed-effect model approach, because only the within variance, variance over time, is considered and the between variance, variance across firms, is disregarded;

     
  2. 2

    a fixed-effects model will not be able to identify the effect of time-invariant variables17 (Baltagi 2001); and

     
  3. 3

    as is the case in our BLS sample, we assume that the firm-specific intercept values are randomly drawn from a larger population of firms.

     
As a result, we utilize a random effects model, which, in the context of this paper, can be expressed as:
$$ \ln (Y_{it} ) = \ln (A_{ijt} ) + \alpha_{j} \ln (L_{it} ) + \beta_{j} \ln (K_{it} ) + u_{i} + \varepsilon_{it} \quad \left( {\alpha , \beta } \right) > 0;\ j = 1, 2. $$
(3)
where j = 1, 2 distinguishes family or non-family ownership, i = (1…3364), and t = (1…4). The individual differences in the intercept values (i.e. the unobserved heterogeneity) of each firm are now reflected in the error term which is separated into two components:
  1. 1

    the unobserved random disturbance, ui, for the ith firm and constant through time; and

     
  2. 2

    the combined time series and cross-section disturbance, εit.

     
The results obtained from Eq. 3 are listed in left panel of Table 11. Heteroskedasticity, which is likely to occur in panel data, is accounted for by using panel-corrected standard error (PCSE) methodology (Beck and Katz 1995) to estimate the coefficients in the analysis. To test if a random effects model is appropriate, we use the Breush and Pagan lagrange multiplier test, which tests the null hypothesis that there are no random effects. Listed in Table 11, with a p-value of 0.00, we reject the null hypothesis and conclude there are, in fact, random effects. Also listed in the right panel of Table 11 are the Hausman–Taylor estimates for Eq. 3.18
Table 11

Panel estimation of the family-firm production function with heterogeneous inputs

 

Equation 3 (OLS)

Equation 3 (2SLS)

Coefficient

Std error

Coefficient

Std error

Intercept

1.275***

0.062

1.270***

0.058

Family firm

0.037

0.023

0.032

0.035

ln Labour

0.802***

0.063

0.800***

0.068

ln Capital

0.240***

0.038

0.243***

0.037

ln Family labour

0.074***

0.016

0.068***

0.024

ln Family capital

−0.064***

0.010

−0.059***

0.017

R2 adj

0.700

 

0.701

 

N

13456

 

13456

 

BPLM test p

0.00

 

0.00

 

Panel estimation of Eq. 3 is estimated as ln (Yit) = ln (Aijt) + αjln (Lit) + βjln (Kit) + ui + εit, where j = 1 for non-family firms and 2 for family firms, i = (1,…,3364), and t = (1,…,4). Estimates of A2 (labelled Family firm) for family firms measure the difference between family and non-family firms from a total factor productivity standpoint and estimates of α2 (labelled Family labour) and β2 (labelled Family capital) measure the output contribution differences of family labour and capital respectively. Dummy slope variables were also included to control for firm industry and age; for brevity these results are not listed in Table 9. In the right panel, a two-staged least-squares technique was implemented. Instrument variable for Family firm is question 2, subsection i of the BLS. Pane-corrected standard errors are reported in parentheses. Level of significance: *** 1%; ** 5%; * 10%

As seen in Table 11, our OLS estimates reported in Table 8 are valid. That is, there is a positive (negative) interaction with family ownership and the output contribution of labour (capital). Differences in the output contribution of family-firm capital (denoted ln Family capital) in both panels is significant and show a negative difference in output contribution of 5–6% depending on the estimation procedure. Differences in family labour output contributions (denoted ln Family labour) are also significant in both panels and show a positive difference in output contribution of 6–7% depending on the estimation procedure. As with our other alternative specifications, after allowing for input heterogeneity in a panel framework, no significant total factor productivity differences between family or non-family firms are found.

6 Conclusion

Despite numerous investigations into the effect of firm ownership on performance, very little analysis has focussed specifically on the productivity of firms; even less research has been dedicated to studying the effect of family involvement on productivity. Those few studies which have tackled the issue fail, unfortunately, to reach a consensus on the direction of the relationship. Different time periods and estimation techniques contribute to the dilemma; however, the curious assumption of homogeneous output contribution of production inputs for both family and non-family firms has consistently been made in examinations which estimate total factor productivity via a Cobb–Douglas production function. This assumption, as argued in this paper, is perhaps missing family firm differences in the efficient use of labour and capital and may lead to omission bias.

Although a review of the family business literature reveals that family involvement does affect the output contribution of labour and capital, the evidence is primarily anecdotal. In this paper, we have linked this evidence to the theoretical notion of heterogeneous production factors for family firms and show empirically that family labour and capital are indeed diverse in that they yield significantly different output contributions to the firm; As far as we are aware, this is the first study which separates the labour and capital components of the family production function in this manner.

In particular, we have found that family firm labour contributes significantly more, and family capital significantly less, to output than for comparable non-family firms. We have also found that these effects, in terms of labour contribution, weaken as the firm gets larger. Contrary to previous studies, this study has shown that when we account for heterogeneous production inputs all previously found unquantifiable differences in total factor productivity disappear. This result leads us to believe that previously found qualitative productivity differences between family and non-family firms can be better explained by attributing those differences to the output contributions embodied in the firm’s heterogeneous production factors.

Despite offering a theoretical explanation for different output contributions between family and non-family firms on the basis of relatively contemporary family business research, it is important to note that the production function technique merely estimates the presence of the phenomenon and does not offer any insight into the definitive cause(s) of the relationships found. Future research should preferably explore these causes in greater depth. This study has investigated the effect of family involvement on Australian SMEs from 1995 to 1998, so we cannot claim to have definitively established whether family firms are more or less productive than non-family firms; however, we can claim that the assumption of homogeneous labour and capital shares between family and non-family firms is inappropriate. Moreover, on the basis of the unique characteristics of family firms, heterogeneous production inputs do matter empirically. Therefore, if we account for the role for family involvement and allow for unequal factor elasticities, perhaps we could better understand the differences in production strategy, planning, and other important productivity drivers between family and non-family firms. These results hopefully shed further light on the unique attributes of family firms and bring us closer to understanding the specific economic impact family involvement may have on a firm level.

Footnotes
1

Burns and Whitehouse (1996) report that 85% of businesses in the European Union and 90% of businesses in the United States are family controlled. It is also generally recognized that family businesses are critical to entrepreneurship and socioeconomic development and industrialization in unstable, low income, or transitional economies.

 
2

Handler (1994) describes the issue of succession as the most important issue that all family firms face; Chua et al. (2003) found that succession is the number one concern of family firms; and Ward (1987) goes so far as to define all family firms specifically as those that will be “passed on for the family’s next generation to manage and control”.

 
3

Some studies listed in Table 1 have concentrated on the partial productivity of family firms in that they focus on the ratio of output to a single input factor, usually labour; however partial analysis only provides a general indication of total factor productivity, because it fails to consider tradeoffs between other input factors.

 
4

One notable exception is Martikainen et al. (2009) who tested whether factor elasticities (namely the coefficient estimates for both labour and capital) are invariant across both family and non-family firms. They found that, for their sample of 159 manufacturing firms, there is no such variance in elasticity, and proceeded to test differences in productivity using fixed factor elasticities for both family and non-family firms.

 
5

According to Gómez-Mejía et al.(2007), “the socioemotional wealth of family firms comes in a variety of related forms, including the ability to exercise authority… the perpetuation of family values through the business… the preservation of the family dynasty… the conservation of the family firm's social capital… the fulfilment of family obligations based on blood ties rather than on strict criteria of competence… and the opportunity to be altruistic to family members. Losing this socioemotional wealth implies lost intimacy, reduced status, and failure to meet the family's expectations”.

 
6

If both principal and agent have the same interests, there is no conflict of interest and no “agency problem” (Berle and Means 1932; Ross 1973); thus by virtue of their intra-familial altruistic element, family firms should be exempt from agency problems (Becker 1974; Jensen and Meckling 1976; Parsons et al. 1986; Eisenhardt 1989; Daily and Dollinger 1992). However, more recent investigations have looked into other types of agency problem that may be specific to family firms (Morck and Yeung 2003; Chrisman et al. 2004).

 
7

In their analysis, Cobb and Douglas (1928) investigate production in manufacturing firms and, as a result, land is excluded as a factor of production.

 
8

In the log transformed Cobb–Douglas production function, the value of the constant coefficient is independent of labour and capital. This assumption has been made to ignore the qualitative effects of any force for which there is no quantitative data. The coefficient is thus made a “catch-all” for the effects of such forces (Cobb and Douglas 1928).

 
9

An important consideration is the simultaneous equation bias that may arise when specifying management variables in the production function (Hoch 1958). In the case of the added family firm variable, we may find that productivity depends on whether the firm is a family firm and whether the firm is a family firm depends on productivity. For example, whether a firm remains in the control, management, and ownership of the family may be endogenously determined by the performance of the firm. Poorly performing family firms may resort to outside management as a potential remedy and, on the other hand, families may be less inclined to relinquish ownership, management, or control of a highly performing firm (Demsetz and Lehn 1985; Demsetz and Villalonga 2001). If correlations between the error term and independent variables exist, coefficient estimates of Eqs. 1 and 2 may end up being biassed and therefore inconsistent, because it is assumed that independent variables are in fact independent or exogenous.

 
10

The BLS samples were drawn from the ABS Business Register, with 8745 business units being selected for inclusion in the 1994–1995 survey. For the 1995–1996 survey, 4,948 of the original selections for the 1994–1995 survey were selected, and this was supplemented by 572 new business units added to the ABS Business Register during 1995–1996. The sample for the 1996–1997 survey included 4,541 businesses which were previously sampled, and an additional sample of 529 new businesses from the 1995–1996 interrogation of the Business Register, and 551 new businesses from the 1996–1997 interrogation of the Business Register.

 
11

The equivalent ratio is simply calculated as average part-time hours per week divided by average full-time hours per week for all non-managerial employees. This information is from the Australian Bureau of Statistics’ “Employee Earnings and Hours, Australia” report as of 1998 and from the previously known “Earnings and Hours of Employees, Distribution and Composition, Australia” report.

 
12

Of all family firms responding to question 2, 34.91% selected i only; 27.45% selected both i and ii; 11.79% selected i, ii and v; 4.39% selected i and v; 3.18% selected i, ii, iv and v; and 3.18% selected i, ii and iv. On this basis, and out of 64 possible permutations, nearly 95% of all family firms at least selected i, which is understandable, because we would expect small to medium sized family firms to have a more operational classification; however, not excluding these, approximately 37% also selected iv and v, which is associated with the essence-based classification of a family firm.

 
13

“Personal and other services” was excluded and used as the benchmark industry.

 
14

Notable exceptions are the construction, accommodation, and personal services industries.

 
15

For a discussion on the relationship between capital intensity and labour productivity, see Wolff (1991).

 
16

Note that the measure of total workers, as the denominator in the part-time ratio, has not been converted to FTE workers and is simply reported as the total number of all employees in any given firm.

 
17

Considering that our sub-sample of the BLS is relatively small (i.e. the number of cross-sectional subjects, N = 3364, is greater than the number of time periods, T = 4), the family ownership dummy specified in Eq. 2 is, in fact, constant for each family firm across the entire period under analysis.

 
18

An important assumption of the random effects model is that the unobserved random disturbance, ui, is uncorrelated with the individual regressors in Eq. 3. As is the case with many panels, the Hausman test has revealed that the coefficients estimated by the efficient random effects estimator are not the same as those estimated by the consistent fixed effects estimator. In such cases a fixed effect model would be preferred; for reasons already stated, however, and to directly estimate the family firm intercept, we require a random effects approach. To overcome this issue, the Hausman–Taylor random effects procedure, with instruments, can be used to control for endogeneity (Hausman and Taylor 1981).

 

Acknowledgments

The authors wish to thank the Grand Valley State University’s Family Owned Business Institute for their generous funding of this study. Further acknowledgements are extended to Dr. Gulasekaran for his econometric expertise and Dr. Khalid and Dr. Craig for their support in developing this paper.

Copyright information

© Springer Science+Business Media, LLC. 2011