Family firms, Regional Competitiveness and Productivity: A Multilevel Approach

ABSTRACT

As not all firms benefit to the same extent from regional competitiveness, this article investigates the influence of the regional context on the productivity of a sample of family and non-family manufacturing firms in Spain. Using a multilevel approach to account for the nested structure of the data, and a composite indicator of regional competitiveness, to capture the spatial endowment of tangible and intangible resources, we found family firms to be more sensitive to the regional context than non-family businesses. Cross-level interactions show that family firms achieve higher productivity gains from their location in more competitive regions than their non-family counterparts. This result is in line with our theoretical arguments postulating the unique social capital of family firms which allows them to benefit most from location advantages. Implications for regional and family business studies, as well as policymakers, are discussed.

KEYWORDS:

• Family firms
• Regional competitiveness
• Productivity
• Social capital
• Multilevel analysis
• Spain

Acknowledgement

We are grateful to the associate editor and the two anonymous referees for their constructive comments. As an early version of this paper was presented at the 13th RSAI annual conference (May 2021), the IMT research seminar (June 2021), the 6th EIASM Workshop on Family Firm Management Research (October 2021), and the Australasian Family Enterprise Research Network (AFERN) Research Seminar (April 2022), we would also like to thank all the participants for their helpful suggestions.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. Externalities are regarded as geographically-bounded resources or locational factors that reside outside firms that might lead to static (e.g. increasing return to scale) – and dynamic (e.g. higher innovativeness) advantages compared to firms elsewhere (Bellmann, Evers, & Hujer, 2018; Kitson, Martin, and Tyler 2004).

2. Adler and Kwon (2002) define social capital as ‘the goodwill available to individuals or groups. Its source lies in the structure and content of the actor’s social relations. Its effects flow from the information, influence, and solidarity it makes available to the actor’ (p. 23).

3. Similarly, the European Commission (1999, 75) states that ‘[the concept of regional competitiveness] should capture the notion that, despite the fact that there are strongly competitive and uncompetitive firms in every region, there are common features within a region which affect the competitiveness of all firms located there’.

4. The Instituto de la Empresa Familiar (IEF) is a non-profit organization owned by a hundred Spanish family firms, leaders in their respective sectors. Since its foundation in 1992, IEF is the main representative of family firms in Spain. For more information about the IEF, please refer to: www.iefamiliar.com.

5. For more information on SEPI foundation, please refer to: https://www.sepi.es/en.

6. Since ESEE is based on self-reported data, concerns related to response bias may potentially arise. However, the fact that survey participation takes place in strict anonymity and confidentiality mitigates considerably the risks associated with mis-reporting.

7. Since our focus is on the regional context, we have removed all the observations without a regional code. Few missing values have emerged in creating the other variables (all the percentages are less than 1%, except for debt on liabilities with a percentage of missing values equal to 3.8%).

8. NACE is the abbreviation for ‘Nomenclature statistique des activités économiques dans la Communauté européenne’ and represents the European standard classification of productive economic activities. Particularly, ESEE adopts the NACE Rev. 2 classification implemented in 2006. For more information on NACE classification, please refer to: https://ec.europa.eu/eurostat/web/nace-rev2.

9. NUTS stands for ‘Nomenclature of Territorial Units for Statistics’ and represents the level of territorial divisionfor statistical purposes. The Spanish territory is divided in the following levels: NUTS 1 consists of seven groups of autonomous communities (Agrupación de comunidades autónomas); NUTS 2 comprises 19 autonomous communities and cities (Comunidades y ciudades autónomas); NUTS 3 is made up of 59 provinces and islands (Provincias, Islas). However, the ESEE excludes the autonomous cities of Ceuta and Melilla, thus leaving 17 autonomous communities. For more information on NUTS classification, please refer to: https://ec.europa.eu/eurostat/web/nuts/background.

10. Given the wide time-span of our data and, hence, to consider inflation, value added and all the monetary variables have been deflated using the production price index, retrieved from the Instituto Nacional de Estadística (INE), for each industry at the NUTS 2 level.

11. Grounded on the behavioural perspective and the resource-based view of the firm, the essence approach complements the demographic approach. For an in-depth analysis of the two aforementioned approaches, please refer to Basco (2013).

12. By accounting for regional disparities in living conditions (Van Raalte et al. 2021), the regional mortality rate in general and the infant one in particular signal persistent spatial imbalances (Kibele, Klüsener, and Scholz 2015), especially in terms of the well-being of a given community and its environmental development (Fantini et al. 2006). The infant mortality rate usually displays noteable sub-national geographical variations (Rosicova et al. 2011). For these reasons, we considered information on the infant mortality rate in building our synthetic indicator of regional competitiveness.

13. With a value of 56.06% in 2002–2015, we may consider the percentage of total variance explained by the first component as satisfactorily. Indeed, it is in line with previous studies such as Fernández-Serrano, Martínez-Román, and Romero (2019) (56.15%) and Gumbau Albert (2017) (54.22%) or even higher. For instance, in Rodríguez-Pose and Crescenzi (2008) and Coshall, Charlesworth, and Page (2015), the first component explains only 43.1% and 29.39% of the total variance, respectively.

14. The KMO test value is similar to that found in previous studies dealing with regional statistics. For instance, in (Rizzi, Graziano, and Dallara 2018) the KMO test score ranges from 0.60 to 0.71. Galluzzo (2021) reported a KMO test value equal to 0.64, while in Dallara and Rizzi (2012) the score is even lower than the threshold of 0.5, amounting to 0.475.

15. The European Commission in the EU Regional Competitiveness publishes the Regional Competitiveness Index (RCI) measuring, with more than 70 comparable indicators, the major factors of competitiveness for all the NUTS-2 level regions across the European Union. The latest report is the fourth edition and refers to 2019. The previous editions refer to 2010, 2013 and 2016. Therefore, our indicator differs in a minor number of indicators and on the covered period. Considering the values of 2013, we have compared the two indicators with the Spearman rank correlation and we obtained a value equal to 0.91, demonstrating that our indicator and the RCI are highly correlated.

16. Eurostat bases the level of technological intensity on the R&D expenditure/Value added ratio. In doing so, the following groups of manufacturing activities at 2-digit level of NACE Rev. 2 classification are identified: i) High-technology, consisting of Manufacture of basic pharmaceutical products and pharmaceutical preparations (21) and Manufacture of computer, electronic and optical products (26); ii) Medium-high technology, which consists of Manufacture of chemicals and chemical products (20) and Manufacture of electrical equipment; Manufacture of machinery and equipment n.e.c.; Manufacture of motor vehicles, trailers and semi-trailers; Manufacture of other transport equipment (27 to 30); iii) Medium-low technology, which comprises Manufacture of coke and refined petroleum products (19), Manufacture of rubber and plastic products; Manufacture of other non-metallic mineral products; Manufacture of basic metals; Manufacture of fabricated metals products, except machinery and equipment (22 to 25) and repair and installation of machinery and equipment (33); iv) Low-technology, consisting of Manufacture of food products, Manufacture of beverages, Manufacture of tobacco products, Manufacture of textile, Manufacture of apparel, Manufacture of leather and related products, Manufacture of wood and wooden products, Manufacture of paper and paper products, Printing and reproduction of recorded media (10 to 18) and Manufacture of furniture and Other manufacturing (31 to 32).

17. Ecological fallacy occurs when a result obtained at an aggregate level is not confirmed after replicating the analysis on an individual basis. In this sense, micro-founded analysis is preferable since it controls for any potential aggregation bias. The atomistic fallacy represents the opposite problem: working with micro-data may lead to the absence of any link between individual-level and group-level relationships (Raspe and van Oort 2011; van Oort et al. 2012).

18. One way to overcome this problem is to estimate clustered ordinary least squares (OLS), by relaxing the assumption of independence and adjusting the error term to accommodate the lack of independence of firms within regions. However, clustered OLS leaves both the noise associated with difference between firms and noise associated, with differences between regions in the error term. On the contrary, the multilevel model goes further by allowing these two error components to be separated.

19. The basic two-level growth model, with measurement occasions at level one nested within individuals at level two, involves fitting a curve through each firm’s productivity (LP) to summarize the change in LP over the observation period. In this simplest case, a straight line is fitted, and the intercepts of the firm-specific lines are allowed to vary about the average line (Steele 2014).

20. In the multilevel approach, the sample size at any level of analysis is a key issue since the requirements of precise measurement of between-group variance impose a ‘sufficient’ number of clusters. However, there is no clear indication since some authors suggest that 20 is a sufficient number of groups (Rabe-Hesketh and Skrondal 2006), others 30 (Hox, Moerbeek, and van de Schoot 2017) or 50 (Maas and Hox 2004). Stegmueller (2016) in a Monte Carlo experiment shows that as long as more than 15 or 20 groups are available, maximum likelihood estimates and confidence interval coverage of estimated individual and macro effects are only biased to a limited extent, while cross-level interactions tend to be biased downward (p. 758).

21. Considering the so-called ‘empty’ model, that is a model without covariates, it is possible to calculate the variance partition coefficient (VPC), also known as the intraclass correlation (ICC), which represents the proportion of firms’ LP variance explained by each level and is a standardized measure of the similarity between higher-level units (Bell and Jones 2015), that is.

VPC(l) = $\frac{{\sigma }_{\mathrm{l}}^2}{{\sigma }_{\mathrm{i}}^2+{\sigma }_{\mathrm{j}}^2+{\sigma }_{\mathrm{e}}^2}$

where index l represents either i, j or e. It is worth noting, however, that while for a two-level structure where individuals are nested in groups, the exchangeability assumption, that is the correlation between a pair of responses is the same for any randomly selected pair of individuals from the same group, is plausible, on the contrary, to assume equal correlation when responses have a temporal structure is less realistic as it is probable that outcomes at consecutive times will be more highly correlated than measures that are further apart in time (Steele 2014). For example, we would expect the correlation to be higher for two consecutive years than for the first and the last year.

22. More details about the empirical specification are provided in Appendix B.

23. T has been recorded to 0, 1, 2….15, calculated as the difference between each year and the first one, 2002.

24. In a multilevel framework, the variables of the higher levels cannot vary at the lower levels: ‘By definition, a cluster-level variable must be constant within clusters’ (Schmidt-Catran, Fairbrother, and Andreß 2019, 113). In our case, in order to be a regional-level variable, the IRC value has to be the same for all firms located in a given region. Following Tojeiro-Rivero and Moreno (2019), this can be done by averaging over time our regional variable IRC. The use of the average also has the advantage of eliminating fluctuations.

25. It is worth noting that the lagging of all control variables by one year – to mitigate endogeneity concerns – results in the loss of 3,543 firm-year observations. Hence, the sample used in the regression analysis consists of 20,119 firm-year observations distributed across 3,060 unique firms from 2002 to 2015.

26. The likelihood ratio test is a statistical test of the goodness-of-fit between two models. A relatively more complex model is compared to a simpler model to see if it fits a particular dataset significantly better.

27. For space reasons, the coefficients for sector dummies are not reported in the table. Results are available upon request.

28. The Wald test for the means of firm variables shows that all of them are jointly significant in all models.

29. As suggested by Raspe and van Oort (2011), it may that the relationship between regional-level variables and firm productivity is different for certain types of firms. This heterogenous effect can be detected through interactions between variables measured at different levels of hierarchically structured data.

30. Due to the presence of a number of firms with less than three observations (30%) we do not use fixed-effects approach as the estimated group-effect is unreliable for small-sized groups while for random-effects models only the clusters must be sized with at least two observations.

31. Results are available on request.

32. We also performed a sub-sample analysis based on the family firm status. The results reveal the magnitude of the IRC coefficient is much greater for the family firm-related sample than for non-family firms (β = 0.528; p < 0.001 vs. β = 0.161; p < 0.1). Results are available on request.

33. We performed the CEM procedure with the command ‘cem’ on Stata and matched on five variables, namely Age, Size, Industry, Listed, and Foreign.

34. See footnote 24.