Policy Research Working Paper 9657 Firm-Level Adoption of Technologies in Senegal Xavier Cirera Diego Comin Marcio Cruz Kyung Min Lee Finance, Competitiveness and Innovation Global Practice May 2021 Policy Research Working Paper 9657 Abstract Technology is key to boost productivity and generate more formal status. Second, most firms still rely on pre-digital and better quality jobs in Senegal. This paper uses a novel technologies to perform general business functions, such as approach to measure technology adoption at the firm level business administration, production planning, supply chain and applies it to a representative sample of firms in Senegal. management, marketing, sales, and payment. Third, most It provides new measures of technology adoption at the firms, including large and formal firms, still rely on manual firm level, which identify the purposes for which technol- methods or manually operated machines to perform critical ogies are used and analyzes some of the key barriers to pro duction tasks that are sector specific, such as harvest- improving technology adoption at the firm level in Sene- ing in agriculture or packaging in food processing. The gal. First, the adoption of general-purpose information and paper presents evidence of three main challenges to improve communications technologies, such as computers, the inter- technology adoption: access to finance, information, and net, and cloud computing for business purpose, is low but knowledge (firm capabilities), and access to markets and very heterogeneous and positively associated with size and competition. This paper is a product of the Finance, Competitiveness and Innovation Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at marciocruz@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Firm-Level Adoption of Technologies in Senegal∗ Xavier Cirera Diego Comin Marcio Cruz Kyung Min Lee World Bank Dartmouth College World Bank World Bank ∗ The authors would like to thank Mark Dutz, Carlos Rodr´ ıguez-Castel´an, Laurent Corthay, Meriem Ait Ali Slimane, Laurent Gonnet, Nathan M. Belete, Consolate Rusagara, Denis Medvedev, and Idrissa Diabira (Director General, ADEPME) for their useful comments. We are very thankful for the inputs provided by Antonio Soares Martins-Neto. We also thank Insa Sadio (Head of the Bureau of Business Statistics at ANSD) and the ANSD team for their collaboration on providing the sampling frame for conducting the Firm-level Adoption of Technology survey in Senegal. Financial support from the Competitive Industries and Innovation Program (CIIP), the Korea World Bank Group Partnership Facility (KWPF), and the Facility for Investment Climate Advisory Services (FIAS) is gratefully acknowledged. 1 Introduction Technology is the key driver of productivity differences across countries (Comin and Hobijn, 2010; Comin and Mestieri, 2018a; Easterly and Levine, 2001; Hall and Jones, 1999). Some estimates suggest that cross-country differences in the timing of adoption of new technologies account for at least a quarter of per capita income disparities (Comin and Hobijn, 2010). Despite the importance of technology in explaining productivity and income differences and the fact that technology is a buzz word for many policymakers, good measures of the use of technology at the firm level are scarce and confined to a few technologies and sectors. This data gap is especially large in developing countries, which makes it challenging to examine whether firms in developing countries use better technologies that already exist and why. Therefore, a deeper understanding of the meaning of new technologies, their use, and the obstacles to adoption requires appropriate tools and methodologies to measure their diffusion. This paper describes the results of the “Firm-level Adoption of Technology” (FAT) survey implemented in Senegal.1 It identifies some of the key obstacles for the adoption of new technologies faced by those firms. The paper describes with large granularity the level of technology use across firms and sectors, the relationship between technology use and performance, and provides some cross-country benchmark. The FAT survey developed by Cirera, Comin, and Cruz 2020 is an innovative tool to measure the use of technology at a granular level. The survey implemented in Senegal includes a nationally representative random sample of about 1,800 firms with 5 or more employees, both formal and informal, in agriculture, industry and services from the latest establishment census for Senegal.2 In the paper, we analyze the adoption and use of technologies in Senegal through three key angles: (1) standard measures of technology related to general purpose technologies; (2) use of technologies applied to general business support functions; and (3) use of sector- specific technologies. The standard firm-level measures of technologies refer to “traditional” measures of general-purpose technology (GPT) adoption, which enable firms to apply more technologies in non-specific tasks. It includes the access and usage of GPTs such as electricity, phone, computers, internet, cloud computer and digital platforms. We define general business support functions (GBF) as those tasks necessary in any firm, regardless of the sectors they are in, such as business administration, production planning, sales or payments methods. The sector specific business functions (SSBF) are those applied for business functions that are industry specific (e.g. land preparation in agriculture industries, or input testing in the 1 This paper is part of a series of studies analyzing country-specific data from the FAT survey. Cirera a, in Brazil. et al. (2021b) and Cirera et al. (2021a) describes the results for Vietnam and the state of Cear´ Cirera et al. (2020) provides a cross-country analysis and provides further details about the FAT survey. 2 See section Appendix C for more details about sampling and data in Senegal. 2 food processing industry) and that often refer to sector-specific production processes. For each general and sector-specific business functions, we measure the technologies used (extensive margin ) and the most frequently used (intensive margin ). The distinction between these margins provides a very granular measure of technology adoption to inform policy decisions. For example, when asked about the adoption of particularly technology such as digital payment methods, many firms may report that they are already using it (extensive margin ), but very few of them may be using this technology intensively (intensive margin ). By disentangling the differences between the extensive and intensive margins we can have more clarity to inform policy, such as if lack of adoption this is a problem of infrastructure, or if this is more related to lack of capabilities of firms. To our knowledge, this is the first study providing a comprehensive analysis of technology adoption at the firm-level using a standard methodology across different sectors and based on nationally representative data covering formal and informal firms in Senegal.3 To advance some of the key results, most firms in Senegal are performing general business functions with pre-digital technologies. On a scale from 1 to 5, where 1 represents the most basic technology to perform a general business function, 2 is the usual threshold to use computers (e.g. using a computer with standard software for production planning) and 5, the most sophisticated technology, usually advanced digital technologies, firms in Senegal score on average 1.92 on the extensive margin and 1.29 on the intensive margin.4 These results suggest that on average firms are not yet adopting basic digital technologies to perform general business functions. For sector-specific technologies, firms in Senegal are also far from the frontier. On a scale from 1 to 5, the sector-specific technology index is 1.59 for the extensive margin, and 1.27 for the intensive margin. Senegalese firms are less likely to adopt advanced sector-specific technologies, compared to general technologies. In addition, the intensive margin being close to 1 implies that even if firms adopt slightly advanced technologies, they do not fully use such technologies and remain using basic technologies in their production processes. Senegal faces relevant challenges to improve the adoption of technologies by local firms. Our analysis highlights three groups of these challenges, based on perceived obstacles for the firms, as well as the facts associated with these obstacles: i) access to finance; ii) lack of capabilities; and iii) access to markets and uncertainty. The paper is structured as follows. Section 2 describes the level of adoption using stan- dard measures of general-purpose technologies, such as access to electricity and Internet, 3 The coverage of technology questions in standard firm level datasets is almost negligible, even for high- income economies. 4 The FAT index is described by Cirera et al. (2020) 3 without identify the specific purpose of use. Section 3 describes the new measures of tech- nologies developed using the FAT survey, and provides a conceptual framework for general and sector specific business functions. Section 4 analyzes the level of technology adoption and use for general business functions. Section 5 analyzes the level of technology adoption and use for sector specific technologies. Section 6 describes some of the key barriers and drivers to technology adoption and use. Sections 7 analyzes the relationship between technology use and employment. The last section concludes. 2 Standard Measures of Technologies: Electricity, ICT, and Industry 4.0 Standard measures of general purpose technologies (GPTs) can be linked to different stages of industrial revolutions, following the reference period in which these technologies became available. We organize the information on adoption and use of GPTs in three types according to the period when they were originated and production processes changed: Industry 2.0, 3.0, and 4.0. Industry 2.0 encompasses electricity and generators, which are technologies from the 1880s. Industry 3.0 refers to the ICT revolution, including mobile phone, computer, and Internet. These technologies became available over the 1970-1980 period.5 Industry 4.0 refers to technologies that in most cases have some digital component, but higher level of autonomy and connection of information across different devices and machines to perform tasks. Among the technologies usually associated with Industry 4.0 are the Internet of Things, Big data analytics and artificial intelligence, 3D printing, advanced robotics, and cloud computing. 6 2.1 Industry 2.0: Electricity A large share of firms in Senegal has access to electricity; yet power outages are common across the country. The availability and reliability of electricity and telecommunications services can be decisive in a firm’s decision to adopt a new technology. Table 1 shows the descriptive statistics of access to electricity, power outages, and alternative source of electricity. The first two columns show the average and standard deviation. The means are further divided by formality and size groups. Formality was defined based on whether 5 Comin and Mestieri (2018b) present the reference year of invention for these technologies: electricity (1882), PCs (1973), cellphones (1973), and Internet (1983). 6 Nayyar and Hallward-Driemeier (2018) provide further discussion on the emergency of Industry 4.0. Although some of these technologies, such as AI, were already available since the 1960s, they have been increasingly available in recent years. 4 firms are in formal or informal sectors. Three firm size groups are defined by the number of employees: small (5-19), medium (20-99), and large (100+). The variation in access is not significant across size groups in general, but there is a significant variation in the use of generators where larger firms have much higher levels of adoption. This finding suggests that firms in Senegal across different scale are facing similar challenges associated with infrastructure for electricity; but what seems to differ mostly is their capacity to access an alternative source of electricity such as their own generators. Among those firms that have or share generators, the large majority relies on fuel. Although a large share of medium and large firms uses generators, very few use alternative sources of energy such as solar or wind-power. Of firms in Senegal that own or share a generator, 85% use “fuel” as the main source of energy; followed by solar power, with only 9% of the generators. Table 1: Industry 2.0: Access and quality of electricity (percentage) Technology Mean Std. Dev. Small Medium Large Formal Informal Having Electricity 79.6% 0.40 80.1% 75.5% 82.7% 98.1% 74.3% Power Outage 79.8% 0.40 78.7% 83.5% 88.6% 84.6% 78.0% Having Generator 24.4% 0.43 19.1% 43.5% 64.0% 62.9% 13.4% Energy: Solar Power 8.7% 0.28 12.6% 2.5% 0.0% 3.6% 15.5% Energy: Fuel 85.2% 0.36 80.5% 94.9% 89.8% 89.3% 79.6% Energy: Wind Power 0.3% 0.06 0.5% 0.0% 0.0% 0.0% 0.7% 2.2 Industry 3.0: ICT While a large share of firms has access to mobile phones, other basic technologies, such as computers or smartphones and tablets are not widely available. Table 2 shows the summary statistics on general purpose technologies. While there is a clear correlation between the size of firms and access to landline phones, most of the firms use mobile phones regardless of their size. About 88%-90% of firms use mobile phones for business purposes. This is consistent with previous findings and measures on the diffusion of fixed-line telephones vis-` a-vis mobile orkegren (2019)). However, the same pattern is phones in several countries in Africa (e.g. Bj¨ not observed with other digital enablers such as computers, or even smartphones and tablets. 5 Table 2: Access to ICT Technology Mean Std. Dev. Small Medium Large Formal Informal Having Telephone 32.1% 0.47 25.9% 55.2% 76.5% 86.7% 16.6% Having Mobile Phone 88.8% 0.32 88.9% 88.4% 89.8% 88.3% 89.0% Having Computer 35.9% 0.48 29.9% 58.4% 77.0% 93.0% 19.6% Having Smartphone 30.0% 0.46 28.4% 36.2% 40.5% 35.6% 28.4% Having Internet 34.1% 0.47 28.1% 57.8% 72.8% 87.2% 19.0% Type: Dial Up Internet 12.2% 33 15.8% 6.4% 1.1% 2.5% 25.2% Type: DSL Internet 62.9% 48 55.3% 75.2% 86.7% 84.5% 34.0% Type: Wireless Internet 12.3% 33 12.6% 12.8% 7.8% 10.4% 14.8% Type: BPL Internet 1.8% 13 2.6% 0.0% 1.3% 0.6% 3.5% Type: Other 10.8% 0.31 13.7% 5.6% 3.1% 2.0% 22.5% Acquisition of software 7.1% 25.7 5.0% 13.0% 30.1% 25.1% 2.0% Number of Telephones 1.02 6.55 0.42 1.77 11.17 3.57 0.28 Number of Mobile Phones 3.89 13.14 2.17 7.14 31.89 9.30 2.41 Number of Computer 2.68 24.36 0.77 5.46 33.48 10.73 0.32 Number of Smartphone 0.92 11.24 0.47 1.59 8.36 2.63 0.44 The divergence in the adoption of computers and smartphones or tablets for business purposes is clearer when considering the intensive margin. Larger firms have a significantly larger number of devices, which is consistent with their scale. On average, small firms have less than 1 computer per firm (0.77), while medium firms have about 5.6 computers per firm, and large firms have about 33.5 computers, either desktop or laptop, per firm. Figure 1: Share of Firms with Internet, Own Website and Social Media There is also a clear and positive association between access to internet and firm size. 6 Around 34% of establishments with more than 5 employees in Senegal have access to internet, while internet is widely available among large firms (about 75% of them have access to internet). Yet, it is striking to see that about one-quarter of large firms report that they do not have access to internet and only 19% of informal firms. In terms of quality of the fixed internet access; the majority of firms rely on DSL. About 12% of firms still rely on dial up. Figure 1 shows the use of internet in terms of having own website and use of social media. While these uses is more extended in large firms, 57% have their own website and 41% use social media, use among smaller firms is still very incipient. 2.3 Industry 4.0 Most of Senegalese firms are not benefiting from Industry 4.0 technologies yet.7 When looking into the adoption of advanced digital technologies embedded into the production process, which usually provides higher level of automation to perform a task, we observe that a very small share of firms in Senegal have been using these technologies. Among the so-called industry 4.0, the most diffused technology in Senegal is cloud computing, which is used by less than 5% of firms. Other advanced and more autonomous technologies, such as AI, robots and 3D-printers for manufacturing, or precision agriculture are used for less than 1% of Senegalese firms. Figure 2: Adoption of Industry 4.0 Technologies 7 Industry 4.0 is characterized by new technologies such as robotics and artificial intelligence (AI) with high autonomy. 7 Figure 3 shows the estimated probability of having adopted the different technologies for formal and informal firms. For Industry 2.0 technologies, the main difference between formal and informal firms is in the use of generators; a formal firm has 50% probability of using a generator while an informal firm has 20% probability. For Industry 3.0 technologies there are significant differences in the use of computers and access to internet, with informal firms having less than 25% probability of using them.8 Figure 3: Summary of General Purpose Technology Adoption in Senegal Note: Robots refer to formal firms. Despite the novelty of some of the measures described in this section, an important limitation is that we do not identify the purpose for which these technologies are being used. 8 We followed the formal definition of ANSD. The ANSD uses a definition of formality based on the accounting system used by the firms. According to their definition, formal firms are those with an accounting system that is compatible with the West African Accounting System (SYSCOA). Only formal firms reported the use of robots. 8 For example, we do not know what are the tasks for which computers or Internet are used or how are these technologies being used in production. To address this issue, the next section explores and measures the purpose for which technologies adopted by firms are used. 3 New Measures of Technology Adoption and Use: Link- ing Technologies to Business Functions To identify the purpose for which a technology is used by the firm, we link the informa- tion on the use of technology with specific business functions. We follow the methodology proposed by Cirera, Comin, and Cruz (2020) and split business functions in two groups: i) General Business Functions, which are common tasks that apply to all firms (e.g. business administration, sales, payment, quality control); and ii) Sector Specific Business Functions, which varies across each sector and is usually more related to core production functions. The General Business Support Functions (GBFs) are commonly available across all firms, irrespective of the industries they are in. Therefore, they provide good comparison across firms, sectors and countries. The FAT survey identifies the purpose for which a given tech- nology is being applied. Figure 4 describes the key GBFs covered by the survey and the technologies associated with them: 1) Business Administration; 2) Production Planning; 3) Sourcing and Procurement; 4) Marketing and Customer Information; 5) Sales; 6) Methods of Payment; and 7) Quality Control. The technologies associated with most business functions follow a ladder of sophistication that goes from the most basic (e.g. handwritten process for production planning) to the most sophisticated level (e.g. Enterprise Resource Planning (ERP) systems for production planning). The Sector Specific Business Functions (SSBFs) are tasks that are associated with the core production or service provision activity and vary across sectors. The FAT survey in Senegal has specific sets of questionnaires for 8 sectors: i) Agriculture (Crops, Fruits, and Vegetables); ii) Agriculture (livestock); iii) Food Processing; iv) Wearing apparel; v) Retail and Wholesale; vi) Land Transportation; vii) Finance; and viii) Health. Among those, the survey was stratified for and provides a representative sample for firms in agriculture, food processing, wearing apparel, and retail. The survey asks information on more than 300 technologies associated to almost 50 business functions. To analyze the level of technology adoption and use in a more systematic way, we convert the information for each business function into a technology index. The index, described by Cirera et al. (2020), varies between 1 and 5, where 1 stands for the most basic level of technology been used and 5 reflects the most sophisticated level been used. 9 Figure 4: General Business Functions With the help of experts for each industry, we assigned a rank to the technologies in each business function according to their sophistication. We construct two basic indices: i) The extensive margin, and ii) The intensive margin. The extensive margin identifies if the firm is adopting a technology to perform a given task. This is based on a yes or no question for the adoption of a technology to perform a specific task. The intensive margin is based on the most used technology to perform this task.9 Table 3 compare the different technology indices for the State of Cear´ a, Vietnam, and 10 Senegal. The results suggest that firms in Senegal have on average relatively low level of technology adoption. For Senegal, the average index for GBFs is 1.92 in the extensive margin and 1.29 in the intensive margins, while the average index for SSBFs are 1.59 in the extensive margin and 1.27 in the intensive margins.11 The results suggests that Senegal stands far way of the frontier, both at the extensive and the intensive margins. As expected, 9 For example, if a firm performs administrative processes associated with HR, financing, and accounting through handwritten processes and computers with standard software, the extensive margin index equals 2. In this case, the maximum value (5) is attributed to a firm using Enterprise Resource Planning (ERP) system, which was identified as the technological frontier to perform this task. Because this firm uses two different methods to perform this task, we ask what is the most frequently used method. If handwritten, the intensive margin index equals (1). If computer with standard software, the intensive margin equals (2). Figure B2 in the appendix describes an example of the index in the extensive and intensive margins for one general business function (left) and one sector specific function (right), following a vertical quality ladder. 10 a in Brazil and Vietnam have At the time of this report, in addition to Senegal, only the state of Cear´ been completed. Bangladesh is also completed but only includes some manufacturing sectors. 11 The index oscillates between 1 for manual technologies and 5 for frontier technologies, with 3 as the middle index. 10 the distance to the frontier is higher at the intensive margin. Overall, the gap between Senegal and the state of Cear´a, one of the poorest states in Brazil, is between 36% and 16% across the different indices. The results suggest that firms in Senegal are not only far from the technological frontier, but they are also far from adopting technologies that have been widely available and adopted by firms in other developing or emerging regions. Table 3: Cross-Country Difference in Technology General Business Function Specific Business Function Extensive Intensive Extensive Intensive Average 2.67 1.90 2.30 1.66 a (Brazil) Cear´ 3.35 2.49 2.75 1.92 Vietnam 2.75 1.92 2.55 1.80 Senegal 1.92 1.29 1.59 1.27 Gap: BR - SN 1.43 1.20 1.16 0.65 Relative Gap** 36% 30% 29% 16% Source: Cirera et al. (2020) Note: Average is the average of Brazil, Vietnam, and Senegal. Relative gap is the difference be- tween Brazil and Senegal relative to the maximum technology gap of 4 ((BrazilSenegal)/Maximum Gap(4)).Technology measures are weighted by the sampling weights. Despite the differences on the average across countries, an important finding of our anal- ysis in Senegal is the heterogeneity across firms. Figure 5 presents the distribution of the technology index for GBFs and SSBFs in Senegal across firms. The right-skewed distribution shows that a large share of firms in Senegal do rely on basic technologies to perform either GBFs or SSBFs. Yet, those firms with higher level of technology measured by the index we propose also have better performance in terms of productivity, measured by value added per worker. 12 While these results do not suggest any causal relationship between technology and performance, they are consistent with previous literature. Although further investigation is needed to determine the causal relationship between technology adoption and performance, available evidence suggest that adopting technology pays off. Easterly and Levine (2001) and Comin and Hobijn (2010) and Comin and Mestieri (2018a) show that technology is a key driver of productivity differences across countries. Kwon and Stoneman (1995) show this relationship for firms in manufacturing and an extensive literature on agriculture has shown the impact of technology adoption on farm productivity. Sections 4 and 5 provide more details of the heterogeneity of adoption for general and sector specific functions. 12 The elasticity of the technology index with respect to value added per worker is 1.2 for the intensive margin GBFs. Table C2 in the appendix provides the full results for these estimates. 11 Figure 5: Technology Adoption – Firm-level Distribution (a) General Business Function (b) Sector-Specific Business Function Note: Lines represent Kernel densities. Vertical dotted lines show the averages. Figure 6: Firm-level Tech Adoption Index and value added per worker (a) Extensive GBF Tech Index (b) Intensive GBF Tech Index (c) Extensive SSGBF Tech Index (d) Intensive SSBF Tech Index 12 4 Technology Use in General Business Functions (GBF) With the exception of payment methods and sales, the average adoption across firms is low for most general business functions in Senegal. Figure 7 describes the average indices for the intensive and extensive margins of adoption for each business function. There are important differences between the extensive and the intensive margins. Even though some firms are adopting more sophisticated technologies in a given business function (e.g. payment methods), these are not the most used technologies. The gap between the extensive and intensive margins vary across business functions. While in payment methods firms use relatively advanced technologies (e.g. digital payment methods), they rely mostly on simpler technologies. Figure 7: General Business Functions in Senegal - Extensive and Intensive Margins Figure 8 provides the level of adoption on the extensive and intensive margins for the general business functions. The results suggest that the large majority of firms still rely on the most basic technologies to perform these tasks (77% rely mostly on handwritten methods of business administration, 91% of firms rely mostly on face-to-face chat to obtain consumer’s information used for marketing purpose and product development, 98.3% rely mostly on establishment’s premises or phone, email, representatives for sales; 95% rely mostly 13 on cash and checks as the most used payment methods). Digital payment is the most diffused (advanced) digital technology in the extensive margin; But 95% of firms still rely mostly on cash or checks (intensive margin). Figure 8: Share of firms using technologies applied to General business functions (a) Business administration processes related (b) Production or service operations plan- to account, finance, and HR ning (c) Customer information for marketing and product development (d) Sales Methods (e) Payment Methods (f) Quality control inspection 4.1 Heterogeneity across formal status, size, and sector Large and formal firms are more likely to use more advanced technologies for general business functions (Figure 9). The gap between formal and informal firms in Senegal is large. This is 14 in part explained by the composition of those firms, given that they are smaller and less likely to be in manufacturing, but difference is still significant and robust even after controlling for size and other characteristics. The same applies to firms based on size. Large firms dominate medium and small firms over the adoption of advanced technologies for all GBFs. Yet, the gap between small and large firms is smaller than the gap between formal and informal firms. Moreover, the gap is larger for business functions that do not involve interaction outside the firm, such as business administration and production planning. Large and formal companies are close to the frontier in the case of Business Administration, Production Planning, and Sourcing and Procurement. Figure 9: General Business Functions in Senegal (a) Extensive GBF by formality (b) Intensive GBF by formality (c) Extensive GBF by size group (d) Intensive GBF by size group Manufacturing and services are the sectors with highest adoption index at both the extensive and intensive margins. Looking across sectors, Figure 10 shows the difference in technology adoption indices by aggregated sector for the average level of adoption - average between general and sector specific business functions - general business functions only and sector specific business function only (see next section); for both the intensive and extensive 15 margins. Looking at the intensive margin - mostly used technology - services is the sector using more sophisticated technology followed by industry and then agriculture. However, the agriculture sector has a higher index when it comes to sector specific business functions at both the intensive and extensive margin. Figure 10: General Business Functions in Senegal - Heterogeneity 5 Sector-Specific Business Functions The sector specific business functions reflect the level of technologies that are specifically related to core production processes or service provisions of each sector. Overall, we observe significant heterogeneity in the level of technology used across business functions within firms in different sectors. 5.1 Agriculture In agriculture (crops, fruits, and vegetables) the business functions with more advanced technologies are those related to land preparation and irrigation, both in the extensive and intensive margin. These results suggest that firms on average are using technologies beyond manual operation for those tasks, such as animal aided or tractors for land preparation. In the intensive margin, 62% of farms rely on manual plowing with the use of simple tools. Similarly, for irrigation, the most often used method are either “ no additional irrigation” (33%) or manual irrigation with watering cans (52%). Also, almost the totality of farms (about 97%) 16 rely on the manual application of organic herbicide as the most used method for weeding and pest control. The small gap between the extensive and the intensive margins of technology in irrigation suggest that this is a business function in which firms adopt intensively the most advance technologies they use. This pattern tends to be more commonly observed across sector specific functions. Figure 11: Agriculture: Crops and Livestock (a) Crops - SSBF (b) Livestock - SSBF Yet, for tasks related to packaging, storage, harvesting, and weeding and pest manage- ment, the level of technology is below 2 on the extensive margin, and under 1.5 in the intensive margin. Overall, these results suggest that farms in Senegal are still relying mostly on manual operations to perform these tasks. This fact suggests heterogeneity in the level of technology used across business functions within firms. Figure 12 provides an example of a small agriculture establishment of 1 acre and 8 workers located around Dakar adopting a drip irrigation, which is a relatively more advanced irrigation system, while relying on “precarious facilities, with products totally or partially exposed to sun, rain, and wind,” which is the most basic option for storage. The results suggest that this picture, taken over the pilots of the FAT survey, is a typical example of the average agriculture establishment, particularly small establishments, in Senegal. For livestock, Figure 11 shows adoption of more sophisticated technologies than for crops; especially for animal health. technologies used for transport and nutrition are less sophis- ticated but not primitive. On the other hand, herd management and monitoring is done manually; which is likely to result in low productivity of livestock. 17 Figure 12: Picture of technologies used for Irrigation and storage in Senegal (a) Irrigation (b) Storage Note: Pictures taken over the pilot of the FAT survey in Senegal in a farm of 1 acre with 8 workers. 5.2 Manufacturing In the food processing sector, the index varies between 2.6 for mix, blend, and cook or food storage and 1.6 for input testing in the extensive margin. These values are smaller in the intensive margin, varying from 2.1 for mix, blend, and cook to 1.1 for input testing. Overall, the indices in the extensive margins suggest that the average firm has been adopting tech- nologies with the support of machines, although many of them are still manually operated. Although almost 30% of the establishments rely on the review of supplier testing, the large majority, about 69% of the establishments, rely on “human sensory” methods, which is most basic procedure available to perform this task. For mixing, blending, and cooking, firms are using machines, but mostly manually operated. 73% of firms use “manually-operated machine” in the extensive margin, and about half of the firms (54.2%) use it in the intensive margin. For anti-bacterial processes, the average firm will be adopting between “wash or soaking and thermal methods,” but almost 83% still rely on minimal-processing or wash or soaking. For packaging, most firms (74%) still use manual procedures as the more often used technologies; while for storage, about 87% relies on the most basic technologies (minimal protection or closed building). 18 Figure 13: Manufacturing: Food Processing and Apparel (a) Food Processing - SSBF (b) Wearing Apparel - SSBF For wearing apparel, the majority of firms still rely on manual design and manual cutting, machine manually operated for joining parts, and manual ironing as the most used technolo- gies. Sewing is the business function with the highest index, for which a large share of firms is adopting “machine manually operated” (almost 80%) or “semi-auto sewing machine” about 22%). About 68% of firms used one of these processes as the most use technologies for “sewing.” On the other hand, design and cutting are the business functions for which most firms still rely on manual process. 91% of firms use manual cutting in the intensive margin - manual cutting (72%) or manually operated machines (19%) -, while almost all firms rely only on manual design; without adopting digital technologies such as digital 2-D or CAD. 5.3 Services In retail, on average, firms are still relying mostly on manual technologies for customer ser- vices, pricing, merchandising, inventory, and advertisement. About 20% of firms are using social media for customer services, but only 4% of them use this as the most frequently used technologies. In the intensive margin, almost 90% of firms provide the services at the premise (80%) or by phone (13%). In other business functions, such as pricing strategies, 60% uses “manual cost” as the most frequently used technologies; 80% relies on manual selection as the most used technology for merchandising 62% are relying mostly on handwritten record for inventory; and almost 62% relies mostly on paper based communication or radio, bill- boards, and TV as the most frequent technologies for advertisement. Digital technologies are being relatively more relevant in the extensive margin for advertisement, where more than 40% of firms also use e-mail or mobile phone and social media. These technologies 19 combined represent almost 40% of the most frequently used technologies for advertisement. For inventory, 32% of the firms are using computer databases in the extensive margin, and 25% in the intensive margin. The results for land transport services are similar to retail in terms of the us of tech- nologies and with an index at the intensive margin below 2 in all SSBFs. With some more sophisticated use of technologies in terms of performance management, in some cases, with the support of computers, but primarily using manual technologies for the different SSBFs. Figure 14: Services: Retail and Wholesale and Land Transport (a) Retail and Wholesale - SSBF (b) Land Transport - SSBF 5.4 Cross-sector differences in sector-specific technologies Comparing across sectors, on average agriculture sectors, especially livestock, and food pro- cessing firms use more advanced sector specific technologies than in other sectors. Although sector specific functions are not directly comparable, our technology index allows to use sim- ilar scales in terms of the distance to each relative frontier. Figure 15 shows the differences within and between sectors in SSBFs, and show some important facts about technology adoption in the country. First, it suggests that on average the agriculture sector is using the most advanced sector specific technologies in Senegal for the extensive and intensive margin. Second, the within sector variance is larger for Food Processing and Wholesale and Retail. In the intensive margin, Livestock firms are using more sophisticated technologies, followed by Food proecssing and Wearing Apparel companies. Interestingly, the variance within the Livestock sector is also the largest, indicating very different technologies across companies. 20 Figure 15: SSBF - Sector Comparison (a) Extensive SSBF (b) Intensive SSBF 6 Barriers to Technology Adoption Why is that firms do not upgrade their technologies? A critical question for policy is what are the main barriers that constrain the adoption of more sophisticated technologies among firms. The survey provides some detailed information to respond to this question. 6.1 Perceived barriers to adoption If adopting technologies is associated with greater productivity, as Figure 6 and the literature suggests, why is that firms do not adopt them? The survey asks firms about their top three obstacles to adopt technology. Figure 16 describes the share of firms reporting obstacles by firm size group. The most common obstacles for all types of firms are lack of capabilities (about 70%) and the second obstacle is lack of finance (for over 60%). These obstacles contrast with other countries where lack of demand and uncertainty dominate obstacles, and suggest that policy instruments that combine technical assistance could be effective in supporting the diffusion and adoption of technologies in Senegal. 21 Figure 16: Perceived Obstacles for Adopting Technology by Firm’s Size We estimate linear regressions to analyze the statistical association between the level of technology adoption and observed obstacles, while controlling for the size of the firms, formality, sector, and region. First, the results indicate that technology adoption for general or sector specific business functions, as well as at the extensive or intensive margins, are strongly and positively associated with size and negatively associated with informality. The results also suggest that technology adoption for both general and specific business functions are negatively associated with the lack of capabilities. The unexplained part of the technology index is large, which is likely to be related with the often poor quality of these perceived obstacles. 22 Table 4: Technology adoption is associated with access to knowledge and information VARIABLES GBF Ext GBF Int SBF Ext SBF Int Lack of capabilities -0.035** -0.004 -0.064** -0.043* (0.016) (0.011) (0.027) (0.022) Government regulations 0.006 -0.004 -0.005 -0.037* (0.019) (0.012) (0.029) (0.021) Lack of finance 0.023 -0.009 0.068** 0.045** (0.016) (0.010) (0.027) (0.020) Lack of demand and uncertainty 0.008 0.017* 0.002 -0.010 (0.015) (0.010) (0.023) (0.018) Poor Infrastructure 0.027 0.008 0.064 0.018 (0.025) (0.014) (0.040) (0.029) Other 0.078*** 0.023* 0.030 0.052* (0.018) (0.013) (0.031) (0.027) Ln (Employment 2018) 0.055*** 0.057*** 0.045*** 0.042*** (0.008) (0.006) (0.014) (0.011) Informality -0.300*** -0.253*** -0.309*** -0.170*** (0.020) (0.016) (0.034) (0.026) Constant 0.620*** 0.265*** 0.867*** 0.373*** (0.050) (0.034) (0.072) (0.061) Observations 1,778 1,786 1,074 1,071 R-squared 0.460 0.442 0.325 0.262 Sector FE YES YES YES YES Region FE YES YES YES YES Note: Robust standard errors in parentheses. *** p < 0.01, ** p< 0.05, * p <0.1 For policymakers, the information on the factual elements that determine lack of adoption is critical to define policy priorities; including areas that deserve further experimentation with impact evaluations that would provide more rigorous evidence of interventions. Yet, perceived obstacles do not necessarily imply that are the most relevant issues faced by the firms. Firms do not know what they do not know. Thus, we now investigate these obstacles based on factual information associated with them. We group them into three main groups: i) Financial constraints; ii) Information and knowledge; and iii) Access to markets and competition, which is associated with uncertainty of demand and consumers’ preference. 6.2 Financial constraints Previous studies suggest that an inefficient financial system may reduce the firm-level tech- nology adoption within a country even if a technology is more profitable. For example, by studying a model of establishment dynamics with a producer-level data, Midrigan and Xu (2014) found that financial friction distort firm entry and technology adoption decisions, which results in lower level of aggregate productivity. Cole et al. (2016) provides a dynamic 23 state model to explain that the efficiency of the financial system with available technologies determines which technologies are adopted by firms across countries. Similarly, other stud- ies also found suggestive evidence that the improvement of local financial systems affects firm-level technology adoption in the Russian Federation (Bircan and De Haas, 2019) or in agriculture in Ethiopia (Abate et al., 2016). Figure 17: Loans for Purchasing Machines/Software and Interest Rates (a) Tech Adoption on Loans (b) Loans on Size (c) Tech Adoption on Interest Rate (d) Interest Rate on Sectors Note: Panel (a) and (c) provide the coefficients and 95% confidence intervals from regressions. Each technology measure is regressed on a dummy for taking loans to purchase machine/software and interest rates, respectively, while controlling for formality, sector, size, and regions. Panel (b) show the predicted probability of getting loans by size groups and confidence intervals from the Probit regression with controlling for other baseline characteristics. Panel (d) presents the predicted interest rates by sectors from the linear regression with controlling for other baseline characteristics. All estimates are weighted by sampling and country weights. Figure 17 panels (a) and (c) present the predictions of our measures of financial access - whether firms took loans and what interest rate - on the index of technology and on the financial access variables - panel (b) and (d) - by firm type. Taking a loan for purchasing a machine is significantly associated with between 0.04 higher intensive index for GBFs and 0.19 for extensive SSBF index. In the case of the interest rates paid on existing loans, a one 24 unit increase in the interest rate paid decreases the level of GBFs between 0.011 (INT) and 0.021 (EXT). The coefficients were statistically significant at the 90% of confidence interval. The results, however, are not statistically different from zero for SSBFs. In sum, the access to finance is correlated with the adoption and use of more sophisticated technologies (both GBF and SSBF), but the cost of finance is only associated with the GBF technologies. Figure 17 panels (b) and (d) describe the prediction for the two variables by firm size and sector, and controlling for other observables. Small firms have less than 20% probabilities of having a loan, compared to large firms that have 30% probability. By sector, and controlling for other factors, livestock and land transportation firms appear to face larger interest rates on their loans, but the differences are not huge across sectors. 6.3 Firm capabilities Management quality and skills An important source of firm capabilities is knowledge associated with human capital of managers and workers (Caselli and Coleman, 2001; Riddell and Song, 2017; Comin and Hobijn, 2004). Figure 18 shows the results of a similar exercise than before showing the predictions associated to human capital measures. Panels (a) and (c) focus on the correlation between the human capital of managers and workers and technology use. Having a manager that has studied at least a BA does not increase the technology index very much, between 0.1 and 0.06. However, if the manager studies abroad, it more than doubles the impact of studying a BA; with an increase between 0.29 and 0.19. Similarly, the effect of having a larger percentage of workers with vocational or secondary education does little to increase the sophistication of technology use; in various cases with no effect or correlation different from zero. The impact, albeit very small is more visible with increases in the percentages of workers with university education. When looking at the incidence of these human capital characteristics across types of firms, having managers that have studied abroad is uncommon, and more probable in formal firms with 20% probability and larger firms with 18%. Similarly, the probability of having more workers with college degrees is larger in formal and larger firms as expected. Lack of capabilities, manifested in the education of workers and managers is likely to be undermining adoption and use of technologies especially in informal and small firms. 25 Figure 18: Human Capital (a) Technology and Top Managers (b) Top Managers and Formality/Size (c) Technology and Workers (d) Workers and Formality/Size Note: Panel (a) and (c) provide the coefficients and 95% confidence intervals from regressions. Each technology measure is regressed on a dummy for top managers’ education (e.g, BA+ and study abroad) and the percent of workers with different education levels (e.g., secondary school, vocational training, and college degree), respectively, while controlling for formality, sector, size, and regions. Panel (b) show the predicted probability of having top managers with BA+ or studying abroad by formality and size with confidence intervals from the Probit regressions controlling for other baseline characteristics. Panel (d) presents the predicted percent of workers with different education by formality and size from the linear regressions controlling for other baseline characteristics. All estimates are weighted by sampling and country weights. The use of formal incentives and performance monitoring is positively associated with technology adoption. Innovation and the adoption of technologies are often driven by workers when they are incentivized to do so. The recent introduction of the World Management Survey (WMS), initiated by Bloom and van Reenen (2007, 2010), has permitted a quantum leap in the comparative quantitative analysis of management practices and their implications for productivity and innovation. The FAT survey allows us to compare the relationship between firms’ managerial capabilities and technology adoption. The questionnaire ask if the firm makes use of formal incentives and the number of performance indicators it uses. We use these two measures and compare then with the GBFs index. Figure 19 shows that firms 26 using formal incentives for workers have a higher index for both the extensive and intensive margins. Panel B also suggests that firms with more performance monitoring indicators are using more advanced technologies. Although the correlations are not large, the results highlight the importance of management as complementary to technology adoption. Figure 19: Management Capabilities and Technology Adoption (a) Technology and Formal Incentives (b) Technology and Performance Monitoring Note: Panel (a) and (b) provide the coefficients and 95% confidence intervals from regressions. Each technology measure is regressed on a dummy for providing formal incentives and performance indicators, respectively, while controlling for formality, sector, size, and regions. All estimates are weighted by sampling and country weights. Awareness, information and overconfidence Flows of information and skills with MNEs and other firms that can facilitate adoption are larger among formal and larger firms. These tend to be geographically closer to other large firms producing similar products or providing similar services (Foster and Rosenzweig, 1995; Bandiera and Rasul, 2006; Conley and Udry, 2010), and doing business with those firms as well as with other multinational firms (Alipranti et al., 2015). Figure 20 presents some potential sources of information about technology by formality and firm size. Compared to informal firms, a larger share of formal firms have similar products or services within 50 km (35%) and supplier or buyers within 50 km (37%), and business relationship with multinational corporate (31%). Panel B also shows that the share of having potential source of information rapidly grows as firm size increases. Considering informal and smaller firms adopted less advanced technologies, these results suggest that they may face more restrictions to get information. Formal and larger firms are also more likely to have CEOs or managers with previous experience in other large firms (30%) that can facilitate diffusion. Assuming that these man- agers are exposed to more firms with more advanced technologies, they become an important 27 source of information on technology adoption. The gap of having CEOs or managers with experience in other larger firms between formal and informal firms or small and large firms are slightly lower than the gap of other sources, but the shares of formal and large firms with CEOs or managers with previous experience in other large firms are more than twice of informal and small firms, respectively. More importantly, as shown in panel (a) the sources of information that are more correlated with larger technology indices are the links to MNEs as suppliers and buyers and the experience of CEOs with larger firms. Figure 20: Awareness and information (a) Technology and information (b) Information and formality Note: Panel (a) provides the coefficients and 95% confidence intervals from regressions. Each technology measure is regressed on a dummy for providing formal incentives and performance indicators, respectively, while controlling for formality, sector, size, and regions. Panel (b) shows the predicted probability of each awareness variable on formality from the Probit regressions with controlling for other baseline characteristics. All estimates are weighted by sampling and country weights. Looking at other sources of external information, the use of external consultants is very low in Senegal, but formal and larger firms are more likely to have access to external consul- tants on technology issues, such as the adoption of new machines or software. Shin (2006) found that getting an external consultant plays an important role in adopting IT technologies by small businesses, particularly when CEOs or managers do not have technical expertise. Comin et al. (2016) also show that a company may seek advice from public organizations with prior experience in the technology. Less than 10% of firms in Senegal uses external consultant for purchasing machines or software (Figure 21). In addition, the type of consul- tants varies significantly. About 39% of firms rely on other local companies as a source of technical assistance. Meanwhile, only 11.8% of companies have worked with universities for external consultancy, suggesting the need to improve university-firms collaboration. Also, only 5.4% worked with government agencies. Among those firms that have not used external consultants, the most common reason reported for not accessing external consultants is “no need” (62%). However, the results also show that 9.3% find it too costly. 28 Figure 21: Access to external consultants A final important element for non-adopting and using more sophisticated and new tech- nologies is willingness to do so. For example, if one believes that they are already adopting more sophisticated technologies in relative terms, it is unlikely that business will adopt. Then the question is whether firms are aware of their actual technology gap. To address this question, we map their self-assessment of their technological level with the actual mea- surement index in the survey. The FAT survey asks for a self-assessment of technology from 1 to 10 (here re-scaled to 1 to 5) comparing the respondent’s firm with firms within the country and with global technology leaders in their sector. 13 Figure 22 shows the predicted self-assessment of technology by sector specific technology adoption index with 95% of the confidence interval. The 45 degrees line shows the point where self-assessed and actual coincide. Panels (a) and (b) compare with national firms, while panels (c) and (d) to global leaders. Interestingly for most firms there is overconfidence (upper triangle) since they perception is larger than their actual level of technology. This can be a constrain to adoption and use of more sophisticated technologies. Logically this overconfidence is larger when comparing with national firms, which on average is common in firms below 3.2 index in SSBFs. 13 We ask the self-assessment question before any of the technology adoption questions to prevent any bias in the self-assessment from potential framing. 29 Figure 22: Association Between Self-Assessment and Technology Adoption (a) In relation to other firms in the country (b) In relation to other firms in the country (c) In relation to the most advanced firms in (d) In relation to the most advanced firms in the world the world Note: Red line shows the quadratic fit with 95% confidence interval. Each technology measure is regressed on firms self-assessment with respect to other firms in the country (panels (a) and (b)) and the most advanced firms in the world (panels (c) and (d)), while controlling for sector, size, and regions. 6.4 Access to international markets and competition A third important driver of technology adoption in Senegal is access to markets and compe- tition in the domestic market. Figure 23 shows the main reasons to adopt new technology. When asked about the main reason to adopt new technologies 59% of firms report “compe- tition in the domestic market” as a key driver, followed by depreciation and replacement. This finding is consistent with the previous studies that competition may affect firm-level technologies (Milliou and Petrakis, 2011). Although self-reported, the top driver of adoption is consistent with previous literature emphasizing the importance of competition, not only in the internal market, but also in the external market. 30 Figure 23: Main reason to adopt new technologies Access to international markets has large effects on productivity via competition and learning, and these channels can also result in the use of more sophisticated technologies. Figure 24 panel (a) shows the relationship between trade status and the technology index. Two-way traders are associated with between 0.25-0.62 more sophisticated technologies than non-traders. This technology premium is less apparent for one-way traders, which only export or import. International activities, therefore, have a significant correlation with technology used, but for two-way traders. But of course are large firms that are more likely to export and import, which is consistent with the behavior observed across firms around the world (Comin and Hobijn, 2004; Hobday, 1994; Rasiah and Gachino, 2005). While only 1.5% of small firms export, up to 36% of large firms are exporting their goods. This implies that this international trade premium is concentrated mainly in larger firms. Figure 26 shows the index for the different GBFs differential between trading groups. Aside from payment methods, exporting companies use more advanced technologies for gen- eral business functions. In the intensive margin, the gap is particularly large for Business Administration tasks. Interestingly, at the intensive margin there are almost no differences on average on the sophistication of technologies between one-way traders and non-traders. 31 Figure 24: Association between exporters/importers and technology adoption (a) Technology and export/import (b) Export/import and size Note: Panel (a) provides the coefficients and 95% confidence intervals from regressions. Each technology measure is regressed on exporter/importer dummies, respectively, while controlling for formality, sector, size, and regions. Panel (b) shows the predicted probability of exporter/importer status on size from the Probit regressions with controlling for other baseline characteristics. All estimates are weighted by sampling and country weights. Figure 25: Exporters (a) Extensive Margin (b) Intensive Margin We also repeat the exercise of decomposing the technology index between domestic and foreign owned companies, those with more than 50% foreign ownership. Interestingly and despite the previous results of the larger technology indices of firms with managers that have worked in MNEs, the differences between domestic and foreign owned companies are small. Yet, foreign firms use more sophisticated technologies, especially for business administration, production planning and sourcing. 32 Figure 26: Domestic and foreign owned companies (a) Extensive Margin (b) Intensive Margin 6.5 Access to government support A small number of firms are aware of government support programs and even a lower number benefit from this support. Informal firms have a 17% probability of being aware of govern- ment programs but 3% likely to benefit from these programs. The probabilities for formal forms are 12% and 4%. Interestingly these probabilities are much larger for larger firms that are 32% likely to be aware of government support and 21% to benefit. The probabilities are much lower for small and medium firms. These results suggest that it is important to disseminate the information about government support programs to facilitate adoption, especially among medium and small size firms; although in general firms are very unaware about any potential support. 33 Figure 27: Awareness of government program and subsidy (a) Formality (b) Size Note: Panel (a) shows the predicted probability of the awareness of government program or subsidy by formality with confidence intervals from the Probit regressions controlling for other baseline character- istics. Panel (b) presents the predicted percent of benefits from government program or subsidy by size from the Probit regressions controlling for other baseline characteristics. All estimates are weighted by sampling and country weights. 6.6 Summary on barriers and drivers As a final exercise, we add the elements discussed in a regression framework to analyze which variables are more correlated with technology use. Table 5 shows the results for all four technology indices and controlling for size, sector, region and informality. We use only factual variables. The models for the GBFS have better explanatory power, especially for intensive margin. Human capital variables for managers and workers are all positively correlated with higher technology indices. Formality and firm size are also correlated with larger indices, and also use of external consultants. We do not find significance, however, on the finance variable or the access to links to GVCs and distance or agglomeration with other firms. These results emphasize capabilities as a key driver of technology adoption. The specifications, however, perform very poorly in explaining SSBFs, and only manager education and use of external consultants are more positively associated to the index. However, agglomeration in the same sector in this case is an important predictor of technology use; which suggests that this agglomeration/externalities may be more important for sector specific technology. 34 Table 5: Technology adoption is associated with access to knowledge and information VARIABLES GBF Ext GBF Int SBF Ext SBF Int Loan for Machine 0.024 -0.014 0.049 0.028 (0.023) (0.014) (0.035) (0.029) Similar Products in 50km 0.039* 0.010 0.093** 0.056* (0.023) (0.016) (0.039) (0.032) Supplier and Buyers in 50km 0.049** 0.008 0.042 -0.002 (0.023) (0.015) (0.044) (0.034) Supplier or Buyer MNCs 0.076*** 0.033 0.060 0.072 (0.027) (0.020) (0.052) (0.045) Manager Experience in Large Firms 0.038 0.040** 0.081* 0.049 (0.024) (0.015) (0.042) (0.036) Use of External Consultant 0.093*** 0.088*** 0.081 0.083* (0.028) (0.024) (0.049) (0.043) Benefit from Government Support -0.071* 0.069** -0.015 0.023 (0.038) (0.033) (0.060) (0.056) Manager with University or More 0.137*** 0.113*** 0.116** 0.101** (0.033) (0.026) (0.053) (0.043) % of Worker with College 0.003*** 0.003*** 0.003** 0.000 (0.001) (0.001) (0.001) (0.001) Ln (Employment 2018) 0.041*** 0.050*** 0.032* 0.027* (0.011) (0.009) (0.017) (0.016) Informality -0.133*** -0.124*** -0.122** -0.049 (0.030) (0.024) (0.055) (0.045) Constant 0.461*** 0.132*** 0.649*** 0.221*** (0.056) (0.041) (0.093) (0.080) Observations 1,111 1,111 677 674 R-squared 0.494 0.509 0.348 0.290 Sector FE YES YES YES YES Region FE YES YES YES YES Note: Robust standard errors in parentheses. *** p < 0.01, ** p< 0.05, * p <0.1 35 7 Technology and Employment The relationship between technology adoption and employment has gained significant trac- tion in the last decade with the emergence of advanced labor-saving technologies and the evidence in more advanced countries on job polarization (Autor et al., 2006; Acemoglu and Autor, 2011). When asked about how firms adjust their labor to the adoption of new tech- nologies through the acquisition of a new machine, equipment, or software, more than 78% of firms suggest that they do not change the number of workers, and about 27% suggest that they offer some training to current workers (Figure 28). Only a small number of firms, about 2%, report a reduction in the number of workers as a mechanism of adjustment for the acquisition of new technologies, which is a much smaller share than the number of firms that report an increase in the number of workers with same skills (3.8%) or hire more workers with more qualified (6.1%). Figure 28: How firms self-report their adjustments to new technologies? Our results suggest that firms with better technology grow more. Table 6 shows a positive and statistically significant association between employment growth (between 2016 and 2018) and technology adoption, across different measures of technology and even after controlling for the initial size of the firm, their age, sector, region, foreign ownership, and exporting status. Although these results do not infer a causal relationship, they are in line with other findings in the literature suggesting that firms with better technologies tend to be more productive and benefit from opportunities to expand. The correlation between firms’ growth and the level of technology is more robust for general business functions at the intensive margin. 36 Table 6: Employment growth and tech adoption (Extensive and Intensive Margins) (1) (2) (3) (4) (5) (6) ABF Ext 0.099*** 0.099*** (0.019) (0.020) GBF Ext 0.067*** 0.073*** (0.014) (0.016) SBF Ext 0.073*** 0.068*** (0.017) (0.017) Ln (Employment 2016) -0.092*** -0.098*** -0.093*** -0.104*** -0.090*** -0.095*** (0.017) (0.017) (0.014) (0.015) (0.018) (0.017) Observations 1,041 970 1,710 1,616 1,041 970 R-squared 0.072 0.096 0.069 0.098 0.067 0.090 Firm characteristic NO YES NO YES NO YES Sector FE NO YES NO YES NO YES Region FE NO YES NO YES NO YES ABF Int 0.128*** 0.109*** (0.032) (0.036) GBF Int 0.149*** 0.142*** (0.025) (0.028) SBF Int 0.051** 0.037 (0.024) (0.025) Ln (Employment 2016) -0.090*** -0.096*** -0.102*** -0.109*** -0.082*** -0.091*** (0.018) (0.017) (0.015) (0.016) (0.017) (0.017) Observations 1,038 967 1,711 1,617 1,038 967 R-squared 0.059 0.083 0.076 0.101 0.049 0.076 Firm characteristic NO YES NO YES NO YES Sector FE NO YES NO YES NO YES Region FE NO YES NO YES NO YES Note: Robust standard errors in parentheses. *** p < 0.01, ** p< 0.05, * p <0.1 When analyzing the association between technology adoption and growth for specific general business functions, we observe a positive and statistically significant association for all functions in the intensive margin. Figure 29 shows that this association is larger and more precisely estimated for the intensive use of more advanced digital technologies applied to business administration and production planning (e.g. Specialized Software and ERP systems). One important question for the impact of technology on employment is how adoption of more sophisticated technologies affects the skill composition towards skilled workers; the skill bias technological change hypothesis. To investigate this relationship, we analyze the correlation between the technology index and changes in the skill composition of the firm based on existing occupations in 2016 and 2018. We use as a proxy for high-skill intensity the share of high-skilled (CEOs and managers, professionals, and technicians) on total workers, 37 Figure 29: General Business Functions and Job Growth Note: The figure provides the coefficients and 95% confidence intervals from regressions. Job growth is regressed on each specific general business function at the intensive margin, while controlling for sector, size, and regions. which also include low-skilled (clerks, production, and service workers) occupations. We then take the difference of this share between 2016 and 2018 and use it as a dependent variable. Table 7 shows a negative association between changes in the skill intensity and the level of technology, controlling for the initial size of the firm. This association does not infer a causal relationship between technology and skill intensity, but it has an important implication suggesting that on average firms with higher level of technologies generated more jobs and increased the share of unskilled workers in their payroll.14 14 This does not necessarily mean that these technologies are unskilled-biased, given that the results could be driven by the growth effect. Yet, evidence in the literature suggests that technologies such as online platforms used for export sales can lead to reduction in the wage skill premia Cruz M (2020). 38 Table 7: Change in the share of high-skill occupations and tech adoption (1) (2) (3) (4) (5) (6) ABF Int -0.056* -0.061 (0.032) (0.038) GBF Int -0.059** -0.069** (0.027) (0.029) SBF Int -0.041* -0.043* (0.022) (0.022) Ln (Employment 2016) 0.014 0.016 0.037 0.040 0.012 0.015 (0.011) (0.011) (0.024) (0.028) (0.010) (0.010) Constant 0.064* 0.104* 0.008 0.040 0.049* 0.089* (0.039) (0.058) (0.044) (0.056) (0.029) (0.049) Observations 1,038 967 1,711 1,617 1,038 967 R-squared 0.004 0.014 0.010 0.020 0.004 0.014 Firm characteristic NO YES NO YES NO YES Sector FE NO YES NO YES NO YES Region FE NO YES NO YES NO YES Note: Robust standard errors in parentheses. *** p < 0.01, ** p< 0.05, * p <0.1 8 Concluding Remarks This paper has provided a granular view of technology use and adoption in Senegal that can be summarized in three stylized facts. First, the adoption of general-purpose ICT technologies, such as computer, Internet, and cloud computing for business purposes is low, but very heterogeneous and positively associated with size and formal status. Second, most firms still rely on pre-digital technologies to perform general business functions, such as business administration, production planning, supply chain management, marketing, sales, and payment. Third, most firms, including large and formal, still rely on manual methods, or manually operated machines to perform critical production tasks that are sector specific, such as harvesting in agriculture, or packaging in food processing. The analysis also suggests that promoting technological upgrade of existing firms (formal and informal with 5+ employees) should be a key priority for Senegal. Senegalese firms of all sizes are far from the technological frontier, across all business functions. Upgrading technologies is critical to generate better firms, which are needed for better jobs. The Recensement G´ eral des Entreprises (RGE), which was the last establishment census led en´ by the Agence Nationale de Statistique et de la D´ emographie suggests that formal and informal firms with 5 or more employees cover almost half of the workers and about 80% of the total sales measured by the RGE data, despite the fact that they represent only 6% of the establishments. These firms are very important for the economy of Senegal, but still, the large majority, including those that are large, formal, and located in Dakar, present low 39 indices of technology adoption. Our analysis suggests at least three areas for which policy interventions in Senegal should focus to diffuse the adoption of more advanced technologies: i) Improving access to finance; ii) Improving access to information and knowledge on technologies (firm capabilities); and iii) Improving access to markets. Senegal has relevant government agencies, such as the Agence de D´ eveloppement et d’Encadrement des Petites et Moyennes Entreprises (ADEPME) and the D´ egation de l’entrepreneuriat rapide (DER) conducting several programs to support el´ businesses, which include some of these areas. It is critical that these programs are coordi- nated in order to prioritize specific technologies where the gaps are larger. Clear inter-agency coordination around budgets and priorities that matches the challenges faced by Senegalese firms described above are critical to adopt more effective technologies. Also, it is important that these programs follow clear monitoring and evaluation mechanisms. 40 References Abate, G. T., S. Rashid, C. Borzaga, and K. Getnet (2016). Rural Finance and Agricultural Technology Adoption in Ethiopia: Does the Institutional Design of Lending Organizations Matter? World Development 84 (C), 235–253. Acemoglu, D. and D. Autor (2011). Skills, tasks and technologies: Implications for employ- ment and earnings. In Handbook of Labor Economics, Volume 4. Amsterdam: Elsevier- North, pp. 1043–1171. Aghion, P. and P. Howitt (1992, March). A Model of Growth through Creative Destruction. Econometrica 60 (2), 323–351. Alipranti, M., C. Milliou, and E. Petrakis (2015). On vertical relations and the timing of technology adoption. Journal of Economic Behavior & Organization 120 (C), 117–129. Autor, D. H., L. F. Katz, and M. S. Kearney (2006, May). The polarization of the u.s. labor market. American Economic Review 96 (2), 189–194. Bandiera, O. and I. Rasul (2006). Social networks and technology adoption in northern mozambique. Economic Journal 116 (514), 869–902. Bircan, C. and R. De Haas (2019, 06). The Limits of Lending? Banks and Technology Adoption across Russia. The Review of Financial Studies 33 (2), 536–609. orkegren, D. (2019). The Adoption of Network Goods: Evidence from the Spread of Mobile Bj¨ Phones in Rwanda. Review of Economic Studies 86 (3), 1033–1060. Bloom, N. and J. van Reenen (2007). Measuring and Explaining Management Practices Across Firms and Countries. The Quarterly Journal of Economics 122 (4), 1351–1408. Bloom, N. and J. van Reenen (2010). Why do management practices differ across firms and countries? Journal of Economic Perspectives 24 (1), 203–24. Caselli, F. and W. J. Coleman (2001, May). Cross-country technology diffusion: The case of computers. American Economic Review 91 (2), 328–335. Cirera, X., D. Comin, and M. Cruz (2020). A new approach to measure technology adoption at the firm-level. World Bank. Cirera, X., D. Comin, M. Cruz, and K. Lee (2020). Technology within and across firms. 41 Cirera, X., D. Comin, M. Cruz, K. M. Lee, and A. Soares Martins-Neto (2021a). Firm-level a in brazil. Policy Research Working Paper, n. technology adoption in the state of cear´ 9568 . Cirera, X., D. Comin, M. Cruz, K. M. Lee, and A. Soares Martins-Neto (2021b). Firm-level technology adoption in vietnam. Policy Research Working Paper, n. 9567 . Cole, H. L., J. Greenwood, and J. M. Sanchez (2016). Why doesn’t technology flow from rich to poor countries? Econometrica 84 (4), 1477–1521. Comin, D. and B. Hobijn (2004, January). Cross-country technology adoption: making the theories face the facts. Journal of Monetary Economics 51 (1), 39–83. Comin, D. and B. Hobijn (2010, December). An exploration of technology diffusion. Amer- ican Economic Review 100 (5), 2031–59. Comin, D. and M. Mestieri (2018a, July). If technology has arrived everywhere, why has income diverged? American Economic Journal: Macroeconomics 10 (3), 137–78. Comin, D. and M. Mestieri (2018b, July). If technology has arrived everywhere, why has income diverged? American Economic Journal: Macroeconomics 10 (3), 137–78. Comin, D., G. Trumbull, and K. Yang (2016, December). Fraunhofer: Innovation in Ger- many. In DRIVERS OF COMPETITIVENESS, World Scientific Book Chapters, Chap- ter 17, pp. 409–444. World Scientific Publishing Co. Pte. Ltd. Conley, T. G. and C. R. Udry (2010, March). Learning about a new technology: Pineapple in ghana. American Economic Review 100 (1), 35–69. Cruz M, Milet E, O. M. (2020, May). Online exports and the skilled-unskilled wage gap. PLoS ONE 15 (5). Easterly, W. and R. Levine (2001). It’s not factor accumulation: Stylized facts and growth models. The World Bank Economic Review 15 (2), 177–219. Foster, A. D. and M. R. Rosenzweig (1995, December). Learning by Doing and Learning from Others: Human Capital and Technical Change in Agriculture. Journal of Political Economy 103 (6), 1176–1209. Hall, R. E. and C. I. Jones (1999). Why do Some Countries Produce So Much More Output Per Worker than Others? The Quarterly Journal of Economics 114 (1), 83–116. 42 Hobday, M. (1994, April). Export-led Technology Development in the Four Dragons: The Case of Electronics. Development and Change 25 (2), 333–361. Kwon, M. J. and P. Stoneman (1995). The impact of technology adoption on firm produc- tivity. Economics of Innovation and New Technology 3 (3-4), 219–234. Midrigan, V. and D. Y. Xu (2014, February). Finance and misallocation: Evidence from plant-level data. American Economic Review 104 (2), 422–58. Milliou, C. and E. Petrakis (2011, September). Timing of technology adoption and product market competition. International Journal of Industrial Organization 29 (5), 513–523. Nayyar, G. and M. Hallward-Driemeier (2018). Trouble in the making? the future of manufacturing-led development. The World Bank . Rasiah, R. and G. Gachino (2005). Are Foreign Firms More Productive and Export- and Technology-intensive than Local Firms in Kenyan Manufacturing? Oxford Development Studies 33 (2), 211–227. Riddell, W. C. and X. Song (2017, October). The Role of Education in Technology Use and Adoption: Evidence from the Canadian Workplace and Employee Survey. ILR Re- view 70 (5), 1219–1253. Romer, P. M. (1990, October). Endogenous Technological Change. Journal of Political Economy 98 (5), 71–102. Shin, I. (2006, April). Adoption of Enterprise Application Software and Firm Performance. Small Business Economics 26 (3), 241–256. 43 Appendix A The structure of the survey Cirera et al. (2020) provide more details about the methodology and data collection of the survey.The FAT is organized in five modules: • Module A– Collects general information about the characteristics of the establishment. • Module B – Covers the technologies used in eight generic business functions. • Module C – Covers the use of technologies for functions that are specific to each of ten industry and services sectors • Module D – Includes questions about the drivers and barriers for technology adoption. • Module E – Collects information on employment, balance sheet and performance, which allow us to compute labor productivity and other measures at the company level. Modules B and C collect the information to measure technology adoption, while the other modules collect information on firm characteristics, performance and variables that can provide information on the barriers and drivers of technology adoption. The survey differentiates between general business functions that all firms conduct regard- less of the sector where they operate (e.g. businesses administration related human resources and finance, production planning, sourcing and procurement, sales, method of payment) and sector specific functions/production processes that are relevant only for companies in a given sector (e.g., food refrigeration in food processing, or sewing in apparel). Information about technologies used in the former is collected in module B, while information on sector-specific technologies is collected in module C. To design modules B and C, the survey draws upon the knowledge of experts in produc- tion and technology in various fields and sectors. These experts provided their knowledge on: i) what are the key general and sector-specific business functions, ii) what are the different technologies used to conduct the main tasks in each function, and iii) how are the different technologies related, both in terms of their sophistication and the degree of substitutability between them. These key businesses functions and technologies identified in modules B and C were validated by sector specialists. 44 Figure A1: Firm-Level Adoption of Technology (FAT) Conceptual Framework B The technology index A full description of the indices can be found in Cirera et al. (2020). Let’s consider a function f with Nf possible technologies. Based on the experts’ assessment we order the technologies in a function according to their sophistication, and assign them a rank ri ∈ 1, 2, ..., Rf . Because several technologies may have the same sophistication, the highest rank in a function Rf ≤ Nf .15 Combining the technology rankings with the information collected by the FAT survey on the technologies used by a firm, we construct two indices of technology at the business function level. Intensive The first index reflects the sophistication of the most widely used technology in a business function. The intensive index of a firm j in a business function f is computed as IN T IN T rf,j Tf,j =1+4∗ (1) Rf IN T where rf,j is the sophistication rank of the technology identified by the firm as being most widely used for the business function, and Rf is the maximum technology rank in the function. Note that we have scaled this index so that it is between 1 and 5. 15 In a small number of business functions, the technologies covered are used in various subgroups of tasks. For example, in the body pressing and welding functions of the automotive sector, the survey differentiates between technologies used for pressing skin panels, pressing structural components and welding the main body. In cases like this, we construct ranks of technologies for each subgroup of tasks within the business function, and then aggregate the resulting indices by taking simple averages across the tasks groups. 45 Extensive The second index we construct measures the sophistication of the array of technologies used to conduct a business function, and we call it EXT (an abbreviation of extensive). In contrast with the intensive, the extensive does not reflect how much each technology is used but it reflects the sophistication of all the technologies adopted and used in production, rather than just the most relevant one. To measure the sophistication of the range of technologies, we must first understand the degree of substitutability between the technologies in the business function. Figure B1 illustrates four possible structures we encounter in the business functions covered by FAT and that differ in the substitutability between their technologies. Panel A depicts a quality ladder or vertical structure (Aghion and Howitt, 1992). In quality ladders there is no productivity gain from using technologies below M AX the maximum sophistication rank employed in the firm, rf,j . Therefore, the sophistication M AX of the technologies employed in business functions with a quality ladder structure is rf,j . Figure B1: Different technology sophistication structures Nf ... 2 ... 1 1 2 Nf (a) Quality Ladder (b) Horizontal Relationship N1,f ¯f +1 r ... Nf N2,f ¯f r ... ... ... 2 2 2 1 1 1 (c) Hybrid I (d) Hybrid II (Tree) The technologies in other business functions may have a horizontal relationship (Romer, 1990), depicted in panel B. In horizontal structures, the use of less sophisticated technologies facilitates the fulfillment of the tasks in the function even conditional on using more sophis- ticated technologies. For example, in marketing the use of less sophisticated technologies such as face-to-face communications may allow firms to reach some customers that may not be reachable by more sophisticated technologies such as customer relationship management 46 (CRM) software. The sophistication of the array of technologies used in horizontal structures is measured by the fraction of the possible technologies in the function that the firm uses. Figure B2 shows an example for business and administration processes and for storage in Agriculture. Figure B2: Technology Adoption Index: Example 47 C Sampling and implementation The FAT survey was implemented in Senegal between August 2019 and February 2020. The sample is nationally representative and includes 1,776 establishments with five or more employees, randomly selected from the 2016 Recensement G´ en´eral des Entreprises (RGE), emographie (ANSD). The universe provided by the Agence Nationale de Statistique et de la D´ includes establishments with 5 or more employees in agriculture, manufacturing, and services. The sample was stratified by formality status (formal and informal according to the ANSD definition), region (Dakar, Diourbel, Kaolack, Kolda, Saint-Louis, Thi` es, and Ziguinchor), size (small: 5–19; medium: 20–99; and large: 100+ employees), and sector (agriculture, food processing, wearing apparel, wholesale retail, land transport, finance, health services, other manufacturing, and other services). The survey was implemented by Kantar Public through face to face interviews using computer-assisted personal interviews For Senegal, our universe includes 9,631 establishments with 5 or more employees. We collected data for 1,786 establishments randomly selected from the RGE-ANSD census. Ta- ble B1 and Table B2 provide the information on the distribution of firms in the population and the sample for Senegal. Table B1: Population Distribution, Senegal Agri- Food Wearing Other Wholesale Other Total Region Size culture Processing Apparel Manuf. & Retail Services Region Dakar Small 72 273 809 859 1126 979 4930 Medium 9 75 19 114 125 281 Large 9 22 0 48 26 84 Diourbel Small 18 84 182 204 214 80 816 Medium 1 9 1 7 8 5 Large 1 1 0 0 0 1 Kaolack Small 26 36 242 175 91 50 820 Medium 50 12 3 18 63 26 Large 11 1 0 0 8 8 Kolda Small 480 28 74 87 64 51 819 Medium 21 1 1 1 4 6 Large 1 0 0 0 0 0 St. Louis Small 125 43 60 116 96 70 688 Medium 65 3 1 5 21 31 Large 41 2 0 1 4 4 Thies Small 26 66 229 237 292 217 1207 Medium 2 14 4 4 33 60 Large 6 3 0 1 5 8 Ziguinchor Small 50 15 32 74 46 98 351 Medium 11 1 0 0 7 12 Large 1 1 0 0 1 2 Total 1026 690 1657 1951 2234 2073 9631 48 Table B2: Sample Distribution, Senegal Agri- Food Wearing Other Wholesale Other Total Region Size culture Processing Apparel Manuf. & Retail Services Region Dakar Small 14 48 102 136 160 222 993 Medium 6 40 8 52 21 89 Large 5 16 0 27 17 30 Diourbel Small 2 14 15 22 25 7 102 Medium 1 4 0 4 2 4 Large 1 0 0 0 0 1 Kaolack Small 3 5 23 19 13 15 133 Medium 4 3 2 1 7 16 Large 9 1 0 0 6 6 Kolda Small 54 10 9 11 9 10 124 Medium 8 1 0 1 4 6 Large 1 0 0 0 0 0 St. Louis Small 22 11 7 17 8 7 142 Medium 10 3 1 4 5 11 Large 27 2 0 1 3 3 Thies Small 3 9 22 31 26 34 162 Medium 1 5 1 0 3 14 Large 4 2 0 0 4 3 Ziguinchor Small 11 14 8 18 15 28 130 Medium 11 1 0 0 7 12 Large 1 1 0 0 1 2 Total 198 190 198 344 336 520 1786 49 D Additional tables and Figures Table C1: Tech adoption and firm’s characteristics VARIABLES GBF Ext GBF Int SBF Ext SBF Int Manufacturing 0.076** -0.004 -0.252*** -0.099** (0.035) (0.017) (0.037) (0.039) Services 0.094*** 0.023 -0.422*** -0.277*** (0.036) (0.018) (0.039) (0.039) Informality -0.259*** -0.243*** -0.344*** -0.186*** (0.026) (0.019) (0.034) (0.027) Firm age (6-10) 0.067* 0.045*** 0.115*** 0.060* (0.035) (0.017) (0.044) (0.033) Firm age (11-15) 0.055 0.022 0.005 0.001 (0.036) (0.016) (0.041) (0.029) Firm age (15+) 0.023 0.018 -0.002 0.005 (0.033) (0.014) (0.039) (0.028) Multinationals 0.043 0.122*** 0.025 0.102* (0.044) (0.046) (0.053) (0.057) Exporting 0.104*** 0.042*** 0.109*** 0.038 (0.026) (0.015) (0.033) (0.024) Ln (Employment 2018) 0.013 0.024*** 0.034** 0.021* (0.010) (0.007) (0.014) (0.012) Constant 0.689*** 0.309*** 0.998*** 0.521*** (0.058) (0.031) (0.069) (0.062) Observations 966 971 937 934 R-squared 0.403 0.380 0.316 0.221 Region FE YES YES YES YES Note: Robust standard errors in parentheses. *** p < 0.01, ** p< 0.05, * p <0.1 50 Table C2: Firm-level Tech Adoption Index and value added per worker VARIABLES (1) (2) (3) (4) GBF Ext 1.466*** (0.387) GBF Int 3.082*** (0.458) SBF Ext 0.972*** (0.357) SBF Int 0.968** (0.383) Informality -1.643*** -1.282*** -1.729*** -1.858*** (0.269) (0.264) (0.281) (0.262) Ln (Employment 2018) -0.403*** -0.444*** -0.342*** -0.326** (0.122) (0.117) (0.129) (0.133) Constant 7.895*** 7.864*** 7.713*** 8.184*** (0.578) (0.533) (0.675) (0.609) Observations 463 463 451 451 R-squared 0.467 0.493 0.457 0.450 Sector FE YES YES YES YES Region FE YES YES YES YES Note: Robust standard errors in parentheses. *** p < 0.01, ** p< 0.05, * p <0.1 Figure C1: SSBF - Predicted Values (a) Extensive SSBF (b) Intensive SSBF Note: Panel (a) and (b) provide the coefficients and 95% confidence intervals from regressions, while controlling for size and regions. All estimates are weighted by sampling and country weights. 51 Table C3: General Business Functions heterogeneity VARIABLES GBF Ext GBF Int Medium 0.150*** 0.141*** (0.045) (0.025) Large 0.417*** 0.393*** (0.073) (0.056) Informality -0.670*** -0.414*** (0.043) (0.025) Manufacturing 0.173*** 0.050** (0.050) (0.022) Services 0.168*** 0.054** (0.052) (0.023) Diourbel 0.260*** -0.001 (0.070) (0.023) Kaolack -0.524*** -0.095*** (0.039) (0.021) Kolda -0.477*** -0.065*** (0.047) (0.019) Saint Louis -0.372*** -0.040 (0.057) (0.027) Thies -0.207*** -0.070*** (0.046) (0.022) Ziguinchor -0.349*** -0.059** (0.058) (0.026) Constant 2.376*** 1.557*** (0.064) (0.031) Observations 1,778 1,786 R-squared 0.409 0.394 Note: Robust standard errors in parentheses. *** p < 0.01, ** p< 0.05, * p <0.1 52 Table C4: Sector Specific Functions heterogeneity VARIABLES SSBF Ext SSBF Int Medium 0.137* 0.132*** (0.070) (0.046) Large 0.465*** 0.241*** (0.106) (0.082) Informality -0.584*** -0.254*** (0.066) (0.040) Livestock 0.739*** 0.617*** (0.120) (0.154) Food Processing -0.057 0.015 (0.083) (0.060) Apparel -0.317*** -0.021 (0.065) (0.054) Wholesale or retail -0.522*** -0.256*** (0.066) (0.053) Financial services -0.064 -0.340*** (0.202) (0.101) Land transport -0.412*** -0.245*** (0.116) (0.076) Health services -0.149 0.429*** (0.143) (0.114) Constant 2.607*** 1.662*** (0.086) (0.061) Observations 1,074 1,071 R-squared 0.305 0.230 Region YES YES Note: Robust standard errors in parentheses. *** p < 0.01, ** p< 0.05, * p <0.1 53