The study was conducted in EUCs of Ilula township in rural Iringa region in Tanzania. A custom area for each EUC was marked, enclosing an area with a high population and settlement in the township. The selection of the two EUCs was also informed by the African Urbanisation Database of Africa (Africapolis), which identifies growing towns over time (Africapolis 2019). Africapolis in a comprehensive and standardised geospatial database on cities and urbanisation in Africa; it combines demographic data, satellite and aerial imagery and other cartographic sources to generate long-term comparative and urban dynamics analyses for about 7500 agglomerations in 50 countries. Ilula (Figure 2) was purposefully selected from four EUCs whose township status was developed as they registered fast growth in business investment in rural areas (Lazaro et al. 2019). Geographically, the Ilula is in Kilolo district of rural Iringa region. The Ilula EUC comprises Ilula and Nyalumbu wards with Ilula Mwaya, Mtua, Sokoni, Matalawe, Igunga, Itunda or Isele, Ding’inayo or Ilala, Masukanzi, Itabali, Nyakilomo, Madizini and Majengo Mapya being the sub-villages.

The selection of respondents was guided by network boundaries established during fieldwork, which were product exchange ties within the predefined boundaries of the EUCs. Satellites and Google Maps were used to identify the boundaries of the EUCs, including the highly built-up areas and zones where most businesses were located, as shown in Figure 3. Using a spatial-random design, as in Brønd (2018), businesses were randomly selected within each of the predetermined categories of businesses and geocoded based on their location. The categories were formed based on the Tanzania Revenue Authority (TRA) framework of business registration. Employing name generators used in network studies, businesses were asked to name up to eight people that they frequently traded with. A total of 459 businesses in Ilula were physically surveyed through face-to-face interviews with the aid of local village and business officers.

Data were collected through a business survey, using a structured questionnaire with ethical clearance for project number 13-P02-TAN obtained from the Sokoine University of Agriculture and local authorities. All the participants were requested to participate in the survey, and their consent was obtained informally as they were presented with the clearance certificate and accompanied by village officers.

Qualitative data from EUC business support actors were collected through interviews from 23 key informants and 11 focus group discussions, to supplement the network information obtained from the questionnaire survey. Interviews were conducted with township and/or ward administrations, elders and financial institutions (including banks and other financial intermediaries, such as Savings and Credit Cooperatives [SACCOs] and VICOBA, TRA, Kilolo district offices, business associations and township trade officers).

The network data obtained from the business survey were processed and analysed using Gephi software (Bastian, Heymann & Jacomy 2009) to obtain centrality measures. Embeddedness and brokerage in the business networks were computed using degree centrality and betweenness centrality, respectively. Degree centrality measures the number of ties a node has relative to the total number of ties in the network. Embeddedness, in this study, refers to a state in which a firm has a large total network connection with a high degree centrality compared to other surrounding firms or nodes. Betweenness centrality measures how often an actor appears on the shortest path among actors in a network of businesses. This is the number of times that a node occurs along a geodesic path.

Yifan-Hu proportional and OpenOrd visualisation layout formats (Hu 2011; Martin et al. 2011) were employed to plot the embeddedness and brokerage position of actors in the product and information networks (Figure 4 and Figure 5). The qualitative data from the interviews were processed according to the different types of questions, followed by a more detailed examination of patterns and explanatory factors on business development trajectories in the EUCs.

Social networks of product exchange were studied from both relational and attribute standpoints; tie relational features in terms of embeddedness and brokerage are matched against business attribute characteristics across all business categories. In this study, brokers (businesses with high betweenness centrality) are classified based on their roles in linking businesses within and across business categories, and those who coordinate traders within and outside the EUCs. Based on their structural positions, five types of brokers were analysed. They are gatekeepers, coordinators, representatives, itinerants or consultants and liaisons (Spiro et al. 2013; Walther 2015). Gatekeepers link businesses from within the group, which in this case is within the EUC, to businesses outside the EUC. Representatives, such as gatekeepers, play a central role in linking businesses outside the EUC to those within the EUC (vertical linkage). In this study, these two factors are presumed to be central to sourcing and channelling potential (non-redundant) information and resources from new and distant actors. Coordinators are brokers who connect various types of businesses within the EUC. The information and resources shared by these brokers are vital for horizontal linkages. Liaison links businesses to actors belonging to either group. They mostly include institutions such as the government and financial institutions. Itinerants or consultants are brokers who link two or more businesses of the same kind without belonging to that kind or a cluster of businesses. Institutions are the dominant players in this category.

The role of social networks in business performance was examined using individual (owner) characteristics and business dummies as controls. Business performance was modelled using estimates of daily sales turnover and stock size measured in the Tanzanian shillings. Sales turnover and stock data were used as proxies for business performance because profits could not be obtained because of a lack of business transactional accounts. In modelling business performance, the ordinary least squares (OLS) method is commonly used because it provides a causal linear prediction of the social network effect on economic outcomes, holding other factors constant (Berrou & Combarnous 2011; Kuépié et al. 2016). The following regression was employed (Equation 1):

where: S is a vector for business performance for all categories of business, N is a vector of network indicators, which are network size measured by degree centrality and network influence measured by eigenvector centrality, all measured by α as a vector of coefficients of interest. The vector X provides other covariates and contains business owner attributes such as age, education and experience. The explicit equation of the model can be expressed as follows (Equation 2):

where: S is a vector for business performance for all categories of business, N is a vector of network indicators, which are network size measured by degree centrality and network influence measured by eigenvector centrality, all measured by α as a vector of coefficients of interest. The vector X provides other covariates and contains business owner attributes such as age, education and experience. The explicit equation of the model can be expressed as follows (Equation 2): The model diagnostics were conducted, and the results report robust standard errors for normality. Multicollinearity was tested using the variance inflation factor (VIF), and a mean VIF of 0.2 was reported, which is less than 10, indicating no multicollinearity between regressors. The model was also tested for endogeneity, and it showed that the robust score chi-square test was insignificant, which reports the factors to be exogenous, permitting for OLS regression instead of instrumental variable regression.

Before the survey, the researcher obtained necessary approvals from the Tanzania Ministry of Education, Iringa reginal commissioner, office of district commissioner, ward and township officer. The project was in conjunction with the values and principles of the ethical clearance obtained for project number 13-P02-TAN from the Research Ethics Review Committee of the Sokoine University of Agriculture on 09 Match 2017 (Ref. SUA/ADM/R.1/8/) and University of Copenhagen.

The results presented in Table 1 indicate that more than 80% of the surveyed businesses were operating as sole proprietorships. Regarding formalisation, approximately 68% of the enterprises were not registered with any formal authority, such as the Business Registration and Licensing Agency (BRELA) and TRA.

In the large informal sector, many businesses were into easy-to-enter and exit ventures, such as shops and trading (Table 2). The majority (more than 60%) of the businesses were either general shops or in trade, which were also largely unspecialised. Most of these businesses were placed in the same category to represent a group of businesses that tend to be small in size with low investment capital, employ few or no formal workers and trade small and mostly unspecialised commodities (requiring little or no trained skills). Transportation¹ (Bodaboda, Bajaj, minibuses and trucking) and food-related businesses accounted for 53% and 35% of the businesses in the service category, respectively. About 34 businesses traded food crops such as tomatoes and other cereals, followed by vendor stalls and retail shops. Food processors accounted for 31% of businesses within the category of processing and manufacturing. Timber-related businesses, such as wood or log processing and carpentry, accounted for 19% as they benefited from proximate tree plantations in Mafinga and Mufindi districts.

Another dominant manufacturing-based business was the repair shops or garages and manufacturers of bamboo baskets and tomato packing materials. These workshops included small repair shops and garages for bicycles, motorcycles (most of which were symbiotically located near hardware stores) and motor vehicles. Some of the salient and emerging businesses observed were shoemaking, tractor-trailer manufacturing and tomato and paddy and/or rice processing factories.

More than half of the stocks sold by the surveyed businesses within the EUC were purchased from within the EUC. Moreover, about 80% of the product sales went to businesses and final customers within the EUCs. Among the goods purchased within the EUC, 15% were tomatoes and food items sold in vendor stalls. A high proportion of locally purchased and sold products indicates the expansion of intra-EUC trading. This reflects the available market potential for goods produced and sold within and outside the EUCs. A local retailer stated:

Ilula EUC was also revealed as market node for goods from surrounding villages, districts and regions, most notably agricultural products including forestry products.

An interesting observation was the growth of specialised lines of business, such as mini supermarkets, hardware stores, spare parts, electronics, furniture and fixtures, motorcycles and tricycles, traded to customers in the EUCs. Other specialised industrial goods, such as hardware, gas, timber, farm inputs, clothing and processed rice, were popular commodities from the EUC traded to the surrounding sub-villages and wards. Agricultural produce, mainly tomatoes, makes up a large proportion (84%) of goods sold outside the EUC, mainly to surrounding regions such as Dar es Salaam, the Coast region and other countries such as Kenya and Uganda.

Highly embedded businesses are known to have large market power and thus act as central links to access resources such as information for other businesses in the network. In this study, businesses in the shops and trading category, mainly hardware stores, wholesale and retail traders, and tomato traders, were highly embedded in the social network of product exchange, as portrayed in Figure 4. Most of these businesses sourced goods from a few general suppliers. The highly embedded businesses were also well connected to other businesses in the service, transport, manufacturing and processing categories. Apart from businesses in the trading category, three service-based businesses and manufacturing were highly embedded, mainly because of the linkages they formed with businesses in the tomato value chains. Interconnectivity with a variety of other businesses shows their influence and is also a benefit to businesses that can be realised through these connections to further expand their product markets.

A high level of embeddedness among these businesses translates into a high level of trust among businesses. In an interview, a representative from the association of businesses and retail shops reported the following:

In the context of the brokerage of resources, the results are categorised based on four types of brokerage positions. In both EUCs, the gatekeeping position was occupied by businesses belonging to the shops and trading categories, such as wholesalers and traders of agricultural products, who were all highly connected with businesses outside the EUC. In Ilula, tomato brokers, traders, wholesalers and vendor stalls for household consumables played a gatekeeping brokerage role (tomato brokers here are identified as agents who mediate exchanges between tomato farmers and tomato traders or buyers, so the actual meaning of a broker differs from brokers as expressed in other contexts where the rest of the text like term is used such as gatekeeper, coordinator and liaison brokers). About six groups or clusters of tomato traders, wholesalers, tomato brokers, vendors stall and restaurants and/or food vendors were observed with at least one business being a major connector or broker of the cluster. As illustrated in Figure 5, brokers or gatekeepers were clustered based on the rate of product exchange recurrence. The two large actors with high normalised betweenness centrality as key brokers exerted a significant influence on the product network beyond the EUCs. It was also revealed that tomato traders and wholesale traders acted as representative brokers who played the role of linking businesses outside the EUC with those within the EUC. Information and resources from external actors or businesses linked by representative brokers engaged with tomatoes, wholesale trade, transportation and money transfer businesses are vital for business development.

Coordinator brokers – traders of wholesale-retail household consumable goods and restaurants – were observed to play leading roles in internally linking businesses in the EUC. Most businesses with medium and high levels of embeddedness are connected to many other businesses across different categories, such as retailers, transporters and service providers (e.g. local food vendors). The observation that similar businesses appear to link businesses internally reflects the low density and diversity of product networks, the implication of which is related to the lack of new information, innovation and subsequent constrained business development.

The liaison and itinerant brokerage positions are not presented in the network graphs (Figure 5) because of minimal role of institutions. The liaison brokerage position or role was played by public and private institutions such as National Microfinance Bank (NMB) and CRDB commercial banks, Mazombe Savings and Credit Cooperative Society (SACCOS), local government (Township Executive Officers office) and other government and non-governmental institutions such as the Tanzania Social Action Fund (TASAF), Bangladesh Rural Advancement Committee (BRAC), Vision Fund and Mama Bahati Foundation. Through interviews, public and private formal institutions provided knowledge and information to businesses in the EUCs through consultations and training. Itinerant or consultant brokerage positions were played by small financial intermediaries such as BRAC SACCOS. They only linked targeted business groups such as food vendors and vendor stalls.

This study employed business sales turnover and stock size as a proxy for business performance assessment. The results of the OLS regression model presented in Table 3 and Table 4 show that the structure of business networks reveals a pool of trading potential. Similarly, increasing the network connectivity of business owners increases the likelihood of business development by increasing returns. The model results revealed that social networks positively influenced EUC business performance. The size of social networks, measured by weighted degree centrality, and network influence, measured by eigenvector centrality, had a positive and significant correlation with business stock size and sales turnover. As Table 3 shows, a percentage increase in the size of the social network (measured by weighted degree centrality) in the product exchange network increases the size of stock and sales turnover by 30% and 48%. Eigenvector centrality scores suggest that being linked to actors who are also well connected (with a high weighted degree centrality) increases business sales turnover by 55%. Moreover, a percentage increase in years of business experience was observed to increase the stock size by 18% and sales turnover by 44%. Increasing capital size was also observed to significantly increase sales turnover by 60% and stock size by 3%. Businesses owned by men were also more likely to increase sales turnover and stock by 79% and 27% compared to women-owned businesses.