Related versus unrelated variety and per employee income regional disparities: A case of Polish regions

In the cognitive relatedness literature, the importance of technological, cognitive or spatial proximities of firms, and even entire sectors, has been emphasised as an important precondition for economic diversification of industries and economic development in countries, regions or agglomerations. In particular, Frenken et al. (2007) introduced related variety (RV) and unrelated variety (UV) into their empirical studies, which then became a milestone in the research on the diversified sector structure and growth nexus. In their approach, sectoral diversity is disaggregated into related and unrelated in order to distinguish between sectors where technological, cognitive or spatial proximity allows knowledge to be transferred from one sector to another (RV) and sectors where knowledge, ideas and skills are unlikely to spill over to other sectors (UV). The question of whether RV and UV can determine income per employee and, if so, whether an appropriate RV and UV structure in the regions can reduce income inequality therefore becomes significant.

The primary goal of this paper is to empirically examine the effects of related and unrelated variety on per employee income in Polish NUTS 2 regions using the GMM estimator system, and the resulting income inequality. Poland is a relatively big European country with substantial regional income disparities along the east–west line that constitute an important regional policy problem and absorb a substantial amount of EU structural funds. Due to historical reasons, western parts of Poland enjoy higher standards of living as wages are higher and unemployment rates are lower than in the eastern regions. Moreover, a large fraction of Poland's economy is concentrated in the Mazowieckie region, where the capital city – Warsaw – is located. This region has the highest wages and the lowest unemployment rate in the country.

We contribute to the literature in the following way. In particular, we identify whether the diversified sectoral structure in Polish regions, in terms of RV and UV, directly determines per employee income, and whether it contributes to regional income inequalities. We find a positive effect of related variety on per employee income and a negative effect of unrelated variety on per employee income. Finally, we perform a number of simulations to study how potential changes in RV and UV could contribute to reducing regional income inequality. Overall, research into the relationship between RV, UV and regional per employee income represents an important research gap in this area of research for Poland that this paper tries to fill.

The structure of this paper is as follows. The next section provides the review of the relevant literature. Subsequently, statistical data and the research methodology are discussed. Finally, the estimation and simulation results are reported. The last section summarises and concludes – with policy recommendations and directions for future studies.

Empirical research on the effects of agglomerations on economic growth has been conducted for nearly three decades. Formal econometric studies, initiated by Glaeser et al. (1992) and Henderson et al. (1995), highlight the effects of both static and dynamic agglomeration externalities. They distinguish between localisation externalities, which concern concentrations of firms from the same sector – the so-called Marshall-Arrow-Romer externalities

The new growth theory literature that emerged in the 1980s stressed the importance of knowledge spillovers for economic growth. In particular, Romer (1986) proposed a theoretical model, in which knowledge of other firms was an important input in the production function of an individual firm, leading to external economies of scale. However, the idea that increasing returns are external to the firm but internal to the industry was already present in Marshall (1890) and Arrow (1962).

In the MAR concept, the use of joint suppliers allows firms to reduce production and transaction costs, increase their productivity and become more competitive (Kemeny & Storper 2015). These are called static externalities because their effect on the selection of a location and on productivity is temporary. There are also dynamic localisation externalities: firms from the same sectors represent the same ‘cognitive’ community. They can benefit from sharing the same specialised knowledge and have the opportunity to learn from each other, which over time leads to a deepening of the shared knowledge and to imitation effects, and these may affect growth (Cortinovis & van Oort 2015).

These processes take place differently in the Jacobs externalities concept as they arise from the diversity of an economic environment. The diversified sectoral structure of an agglomeration allows for a much greater recombination of knowledge than in specialised areas. Such knowledge transfers from different sectors stimulate ideas and innovation. Thanks to the geographical proximity of firms from different sectors, agglomerations can generate more innovation and thus achieve faster economic growth. The urbanisation externalities also result in a more diversified market structure in both supplied goods and consumption (Glaeser & Mare 2001; Cieślik & Ghodsi 2015).

The question of which concept is more conducive to growth has been the aim of many empirical studies (Henderson et al. 1995; Glaeser et al. 1992; Combes 2000; Feldman & Audretsch 1999; Mameli et al. 2008) without providing a clear answer.

In the extensive survey of 67 previous empirical studies, Beaudry and Schiffauerova (2009) could not definitely conclude which type of externality enhances growth.

According to Frenken et al. (2007), the spread of knowledge in a region takes place in related sectors (in the technological, cognitive or spatial sense), and knowledge can easily be combined into new products and services, causing direct growth effects. In unrelated variety, knowledge is more difficult to recombine and does not produce rapid growth effects but, due to the portfolio effect, it should reduce the increase in regional unemployment. These concepts gave rise to many empirical studies on RV and UV.

The vast majority of previous empirical studies focused on the nexus between RV, UV and employment growth, or changes in the unemployment rate, rather than their effects on regional per employee incomes.

Notable exceptions include Boschma and Iammarino (2009), who used exports and imports to calculate RV and UV; Boschma et al. (2012), who studied the effects of RV and UV on value added growth; and more recently, Mewes and Broekel (2022), who studied the relationship between technological complexity and regional economic growth.

Many researchers also tested the relationship between RV, UV and productivity at the firm level.

Stavropoulos et al. (2020) examined the impact of RV and UV on labour productivity at the firm level in European regions, with the difference that companies were grouped according to high, medium and low technology regions. Their study showed that RV positively determines productivity in companies only in high-tech regions; in medium-tech regions this relationship was statistically insignificant; and in low-tech regions this relationship was negative.

In contrast to the previous studies, we focus on the effects of RV and UV on regional per employee income. We study the following hypotheses:

H1: Related variety positively affects per employee income.

The spread of knowledge from technologically, cognitively or commercially distant sectors is limited. Companies from remote sectors are characterised by a lack of complementarity in terms of factors of production and knowledge flows, which may lead to a low level of technological spillovers due to low cognitive proximity (Nooteboom et al. 2007). A high level of unrelated diversity may also result in low organisational, social and institutional proximity, which may limit the effectiveness of administrative procedures at a regional level. An unconnected and fragmented industrial structure may ultimately limit regional economies and local competition (Aarstad et al. 2016). Moreover, due to the lack of complementarity and a low level of knowledge transfer between technologically and cognitively distant sectors, no indirect impact of UV on productivity is expected.

The essence of the adopted research concept is primarily to determine relationships between RV, UV and per employee income. RV and UV can directly determine per employee income, but their effects can also be indirect. For this reason, the main determinants of per employee income are the capital–labour ratio and the level of technological advancement.

In this section, we discuss the spatial distribution of the most important variables used in our empirical study. Furthermore, we report descriptive statistics and a correlation matrix between variables. In particular, we focus on per employee income, RV and UV.

Per employee income

Per employee income was calculated by dividing regional GDP at constant 2019 prices by the number of employees in the region (in firms with more than 9 employees). Data on GDP and number of employees are from Statistics Poland – Local Data Bank. In order to illustrate regional disparities within Poland, calculated (average) values of per employee income for particular voivodeships are shown in Figure 1.

Figure 1

The highest income per employee was recorded in the Mazowieckie region, which has the largest urban agglomeration around the capital city of Warsaw, while the lowest was in Lubelskie voivodeship. Average per employee income in Mazowieckie region was almost two times higher than in Lubelskie Voivodeship. In general, among the regions of Western Poland, the average per employee income was higher than in Eastern Poland.

According to EU nomenclature, the regions of Eastern Poland include: Lubelskie, Podkarpackie, Podlaskie, Świętokrzyskie and Warmińsko-Mazurskie voivodeships.

Related variety (RV), unrelated variety (UV)

The calculations of RV and UV were done according to the methodology promoted by Frenken et al. (2007), who used the decompositional nature of entropic measures to disaggregate sectoral links into related and unrelated varieties. RV and UV were calculated for employment shares at different levels of aggregation using data from the Polish Classification of Activities. This data is compatible with the NACE classifications by sectors. UV was measured as an entropy at one-digit level, while RV was measured as the weighted sum of the entropy at two-digit level within every one-digit sector. Such levels of measurement of UV and RV are a major methodological limitation but, due to the lack of data, it was not possible to calculate UV and RV at higher disaggregation levels.

UV measures the entropy in various technologically, cognitively distant sectors, and is therefore based on the assumption that companies in sectors at this level of aggregation do not have cognitive, technological or market proximity. In general, cognitively distant sectors are not expected to have large-scale learning and knowledge diffusion. UV is calculated as: (1) UV=∑g=1GPg log2(1Pg) UV = \sum\limits_{g = 1}^G {{P_g}\,{{\log}_2}\left({{1 \over {{P_g}}}} \right)} where P_g is the ratio of employment in the one-digit sector S_g (where g = 1, …, G) to the total employment in a region (excluding those employed in agriculture). UV measures the extent to which the shares in employment are evenly distributed over distant cognitive sectors.

RV is measured as the weighted sum of entropy at two-digit level in each single-digit sector. It is assumed that sectors within this level of aggregation are linked on the basis of technological, cognitive or market proximity and can therefore effectively learn from each other and exchange complementary knowledge. RV is calculated as: (2) RV=∑g=1GPg⋅Hg RV = \sum\limits_{g = 1}^G {{P_g} \cdot {H_g}} and Hg=∑i=1Si∈SgIPiPglog2(1Pi/Pg) {H_g} = \sum\limits_{\matrix{{i = 1} \hfill \cr {{S_i} \in {S_g}} \hfill \cr}}^I {{{{P_i}} \over {{P_g}}}{{\log}_2}\left({{1 \over {{P_i}/{P_g}}}} \right)} where P_g is the share of employment in the S_g sector to the total employment in a region (excluding those employed in agriculture); P_i is the share of employment in the S_i sector (where i = 1, …, I) belonging to the same S_g sector.

RV is the degree to which employment at the single-digit level is evenly distributed over its two-digit sub-sectors. Higher values of RV correspond to a more even distribution of employment in the sub-sectors, indicating a higher level of technological relatedness in the region. RV can affect productivity and employment and thus positively contribute to regional economic growth. The results of empirical calculations of average RV and UV values for Polish regions in the years 2003–2019 are shown in Figure 2.

Figure 2

According to the concept that assumes the higher the RV level the greater the number of technologically, cognitively or market-related industries in the region, it turns out that the highest RV values were recorded in the most developed regions: Śląskie, Dolnośląskie, Mazowieckie, Małopolskie and Zachodniopomorskie, while the lowest were in Podkarpackie, Podlaskie, Warmińsko-Mazurskie, Wielkopolskie, Kujawsko-Pomorskie and Świętokrzyskie. In turn, when looking at spatial UV diversity, it turns out that the most developed Mazowieckie Voivodeship has the lowest levels of this indicator, which suggests that the service sector plays a dominant role. The higher UV values are reported for Dolnośląskie, Kujawsko-Pomorskie, Lubuskie, Pomorskie and Łódzkie voivodships.

The point of reference for our empirical study is the neoclassical framework of the Solow model with the Cobb-Douglas production function that assumes two factors of production: capital and labour. In this model, the level of per employee income depends on the stock of knowledge (which determines the level of technological advancement) and the capital–labour ratio changing over time. In our study, we endogenise the level of technological advancement by introducing variables that favour knowledge transfers. We introduce the urbanisation rate as a simple measure of the agglomeration effect, the structure of firms’ linkages based on cognitive, technological or spatial proximity, measured by related and unrelated variety, and international openness, which favours knowledge transfers from abroad. To estimate the aforementioned relationships, the following equation is used: (3) ln(yit)=α0+α1 ln(kit)+α2Urbanit+α3 RVit+α2UVit+α5Openit+εit \matrix{{\ln \left({{y_{it}}} \right) = {\alpha _0} + {\alpha _1}\,\ln \left({{k_{it}}} \right) + {\alpha _2}{Urban}_{it} + {\alpha _3}\,{RV}_{it} +} \cr {{\alpha _2}{UV}_{it} + {\alpha _5}{Open}_{it} + {\varepsilon _{it}}} \cr} where y_it is the per employee income (identified with labour productivity) in region i in year t = 2003, …, 2018; k_it–capital–labour ratio in region i in year t; Urban_it is the urbanisation rate; RV_it and UV_it are related and unrelated variety, respectively, in region i in year t; Open_it is the ratio of the sum of imports and exports to GDP in region i in year t; and ɛ_it is the error term.

Equation (3) is estimated using the Dynamic Panel Data Model based on the two-step GMM estimator system proposed by Blundell and Bond (1998) with the Windmeijer correction (2005).

The estimation results of equation (3) for all regions of Poland and regions of Western Poland are shown in Table 2 in three variants: (1) where, in addition to the other variables, it was assumed that there is only related variety (RV) in the regions between sectors, (2) without considering RV – that is, it was assumed that only UV occurs in the regions, and (3) it was assumed that both RV and UV occur in the regions.

Analysing the estimation results for all regions of Poland, it turns out that capital–labour ratio was statistically significant at the 1% level and had a positive effect on labour productivity. In this group, the variables that were introduced to endogenise the level of technological advancement (RV, UV, Urban, Open) were also statistically significant and showed the expected signs. In the group of the more developed regions of Western Poland, the capital–labour ratio was statistically significant and positively determined per employee income, and the estimated parameters in variants (1) and (3) are slightly higher than for Poland as a whole. On the other hand, RV and UV were statistically significant in determining per employee income, and the obtained signs were consistent with theoretical considerations. Moreover, the effect of UV and RV in the group of the more developed regions of Western Poland was stronger than in the group of all regions of Poland. The urbanisation rate as a variable that, in a simple way, controls the agglomeration effect was not statistically significant in any of the estimation variants. Also statistically insignificant in this group was the variable controlling international openness, which was characterised by significantly higher values and lower variability in time than in the regions of Eastern Poland.

In order to study the effects of RV and UV on regional inequality, based on the estimation results (variant 3 – all regions), we calculate the estimated per employee income in each region and the Theil index for Poland at the regional level.

Theil index was calculated according to the methodology in Fujita, Hu (2001).

Figure 3

The simulation results show that better effects for reducing regional inequality can be obtained by only increasing RV than by simultaneously increasing RV and decreasing UV by 20%. For example, in the variant in which we only increase RV by 20% in the eastern regions of Poland, we obtain lower values of the Theil index than when we simultaneously increase RV by 20% and decrease UV by 20%. This may be due to the fact that, first, the RV levels in the eastern regions of Poland are much lower than in the other regions; secondly, according to equations (1)–(2), RV is dependent on UV, and the change of UV may imply the change in RV; thirdly, the UV level in the regions of Eastern Poland is not significantly different from those of the other regions; and finally, since better effects are observed in variants when we only increase RV without decreasing UV, we limit the problem of cognitive blockade in the long run.

The main objective of the study was to show which of the factors favouring knowledge transfer by endogenising the level of technological advancement significantly determine per employee income. We found that RV had a positive impact on per employee income, regardless of the analysed group of regions, so hypothesis H1 could not be rejected. The obtained results are consistent with the studies by Saviotti and Frenken (2008) and Stavropoulos et al. (2020). The observed positive effect of RV on per employee income results from the complementarity of spreading knowledge between cognitively related sectors, which allows for knowledge to be supplemented and, thus, the physical capital is used more effectively. This is observed in the group of more developed regions of Western Poland where, in the variants of estimation in which RV was taken into account, higher parameters were obtained at the capital–labour ratio. Our study also showed a strong negative effect of UV on per employee income, so hypothesis H2 could not be rejected. This relationship is due to the low level of knowledge transfer between cognitively distant sectors.

Our results have important implications for regional policy aimed at reducing spatial income inequality. Our simulations demonstrate that regional inequality can be reduced, especially by promoting related variety in eastern regions of the country. An important limitation of the study is the high level of data aggregation used to calculate RV and UV measures (two-digit level and one-digit level, respectively), so that UV were poorly differentiated between regions. However, efforts are being made to collect data at a lower level of aggregation, as shown by the recent study of Pylak and Kogler (2021), but limited to only a single year. Therefore, in future studies data at the lower level of aggregation should be used once it becomes available.