94671 Catching Up To The Technological Frontier? Understanding Firm-Level Innovation and Productivity in Kenya Catching Up To The Technological Frontier? Understanding Firm-Level Innovation and Productivity in Kenya Xavier Cirera Innovation and Entrepreneurship, Trade and Competitiveness Global Practice, the World Bank. I would like to thank Kennedy Opala and Margaret Nyamumbo for their inputs to the paper, and Ganesh Rasagam, Ina Mogollon, Aref Adamali, Silvia Muzi, Paulo Correa, Ana Cusolito and Pluvia Zuniga for their helpful discussions. This output is part of the DfID co-funded project “Kenya Export Competitiveness and Innovation Non-Lending Technical Assistance Program”. The report is a component of the Kenya Investment Climate Assessment and part of the Kenya Investment Climate Program II. The objective of the program is to contribute to the analytical basis upon which the Government of Kenya designs policies to increase the economy’s competitiveness. TABLE OF CONTENTS Executive Summary i 1. Introduction 1 2. The Innovation Policy Framework 3 3. Data Description and Measurement of Innovation 4 3.1 The data 4 3.2 Measuring innovation 4 4. The Firm-Level Innovation Landscape in Kenya 8 4.1 How innovative are Kenyan firms 8 4.1.1 Innovation outcomes 8 4.1.2 Innovation inputs and knowledge capital investments 11 4.2 Who are the Innovators? 14 4.3 Main characteristics of the innovation process 16 5. Innovation and Firm-Level Performance 18 5.1 Methodology and data description 19 5.2 The determinants of investing in knowledge inputs 21 5.3 The innovation function 22 5.4 Innovation and productivity 24 5.5 Innovation and employment 25 6. Conclusions and Policy Implications 29 References 31 Appendix 1. Measuring intangible assets 32 Appendix 2. Innovation results by ISIC 2 digits 33 Appendix 3. Methodology 34 Appendix 4. Main Econometric Results 36 A4.1 The determinants of investing in knowledge inputs 36 A4.2 The innovation function 38 A4.3 Innovation and productivity 40 A4.4 Innovation and employment 45 Executive summary Rationale for this report these activities and investments on knowledge capital Kenya’s economy has undergone a significant process in developing countries in general, and in Kenya in of structural transformation over the last decade. particular. The aim of this report is to provide a rich Since 2002, the economy has shown an accelerating description of the nature of firm-level innovation and trend with GDP growth increasing steadily from investments in knowledge capital in Kenya, using data below 1 percent in 2002 to 7 percent in 2007. After a from more than 500 firms in various sectors, derived slowdown in GDP growth to 1.5% and 2.7% in 2008 from the most recent Enterprise Survey (2013) and and 2009 respectively, economic growth started to the linked 2014 innovation module, and to better rebound in 2010. Amidst this positive growth context, understand the link between innovation activities and in October 2013, the Kenyan Government launched productivity and employment. the Second Medium-Term Plan (MTP-2) of the Vision 2030. The aim of Kenya’s Vision 2030 is to How to measure firm-level innovation create “a globally competitive and prosperous country Innovation requires the transformation of knowledge with a high quality of life by 2030” and to shift the capital or innovation inputs, both tangible and country’s status to upper-middle income level. intangible—such as training, equipment, Research and Development (R&D) or intellectual property While the improvement in economic performance in acquisitions—into innovation outcomes; for the past decade is remarkable, there are indications instance the introduction of new products, improved that achieving such ambitious targets might be quality, new production processes, or organizational difficult, especially given the slow rate of employment changes. Firms invest in knowledge inputs in order generation. More importantly, while investment has to increase their capabilities and produce innovative continued to accelerate over the last decade, driven outcomes. As well as requiring tangible assets such as mainly by public investment, aggregate productivity technology, equipment, and the physical production growth has remained stagnant since the 1970s and facilities, innovation needs intangible assets such turned negative during the 2008-09 economic crisis as human capital, scientific and creative capital, and during the macroeconomic instability period in and organizational capital. In turn, these inputs 2011. Therefore, attaining the high rates of growth require specific innovation activities. For example, and the degree of transformation envisioned in firms invest in training in order to increase the Kenya’s Vision 2030 requires a renewed emphasis on available human capital. The combination of these boosting productivity. Only by increasing productivity inputs yield innovation outcomes in the form of can Kenyan firms become globally competitive and new or improved products and services, production generate the quantity of high quality jobs required to and delivery processes, business organization, and boost incomes and achieve shared prosperity. patented intellectual property. However, achieving the outcomes is heavily dependent on the ability of the A central element to boosting productivity is firm in question, on the specific sector and country increasing knowledge capital investments and context, and on the enabling environment and policy innovation activities at the firm-level. However, framework in place. we know very little about the nature and impact of Understanding Firm-Level Innovation and Productivity in Kenya i These innovation outcomes can impact firm method, or a new organizational method in business performance in different ways. Successful innovations practices, workplace organization or external relations.” are likely to increase firm-level productivity by The Enterprise Survey (ES) use this definition to improving the capacity to transform factors of identify innovations by directly asking firm managers production into more and better products, and by and owners whether they have implemented “new” or more efficiently creating products of higher value. “significant” changes or improvements in the last three Second, the increase in productivity is expected to years. This is problematic since “significant” is a highly increase the marginal productivity of labor, and as a subjective term, and is also self-reported. It is therefore result, increase the quality of jobs, i.e.: more productive important to supplement analysis of innovation jobs. Third, more productive firms are expected to outcomes with some measures of revealed expenditure push less productive firms out of the market, thereby on innovation inputs. The following box summarizes increasing the overall efficiency of the economy. This the main knowledge inputs and innovation outcomes will improve allocative efficiency. All this, however, measured in the innovation survey: depends on the quality of the innovation and the ability of firms to translate innovation outcomes into The Kenyan innovation policy improved performance. framework The innovation policy framework in Kenya in the A challenge of measuring innovation outcomes is the period prior to 2013 was characterized by large subjective nature of many of the questions used in the institutional fragmentation and the absence of surveys. The Oslo manual, which is the main reference a strong coordinating institution to mobilize for these type of surveys, defines innovation as “… the government efforts in the area of Science, Technology, implementation of a new or significantly improved and Innovation (STI). Vision 2030 lays out the role product (good or service), or process, a new marketing of STI in economic growth, with particular focus on priority growth sectors. The Science, Technology and Innovation in the Enterprise Survey follows a similar Innovation Act of 2013 is an attempt to improve on the methodology than the Oslo manual. First, the survey STI institutional framework, in a bid to complement measures several innovation or knowledge inputs and activities that firms carry out in order to accumulate the policy goals of Vision 2030. However, this new knowledge and capabilities that facilitates innovation. framework has not yet been implemented. These are grouped into four categories: Research and development - the source of R&D This weak institutional framework for STI is (internal vs external) as well as expenditure; translated in the lack of innovation policy support Capacity building- investments in training for instruments. Only certain programs for supporting innovation; technology transfer or intellectual property rights Equipment and technology expenditure for exist, although these tend to be small and mainly innovation; and focused on providing relevant information to firms. Purchase/licensing of inventions or other In addition, some programs that target assistance knowledge forms - expenditure on the purchase of intellectual property. for market access also support innovation via quality certification or by providing information on new Similar to the innovation surveys based on the Oslo markets and technologies. Other types of innovation Manual, the ES innovation survey measures the instruments, such as R&D tax incentives, have yet following types of innovation outcomes: to be developed. Therefore, the STI institutional Product innovations; framework in Kenya can be described as an embryonic Process innovations; policy framework. Organizational innovations; and Marketing innovations. ii Catching up to the Technological Frontier? The firm-level innovation landscape There is very little knowledge appropriation by in Kenya firms in terms of registering patents and other The analysis of the report shows some important instruments. 18.5 percent of firms apply for some stylized facts regarding firm level innovation in type of knowledge property, of which 5.5 percent Kenya. Regarding innovation outcomes and outputs, apply for a patent, but in many cases these forms the study suggests that: of knowledge appropriation are not successfully registered. Firm level innovation rates are relatively high as compared to international standards; however Regarding innovation inputs and activities: these are incremental, meaning that the degree of Investments in innovation inputs are very innovativeness is low. Specifically, 53 percent of firms concentrated in a few firms (one firm has more are product and/or process innovators, 40 percent of than 80 percent of all R&D) and their intensity is the firms introduced product innovations; and 38 similar to those in countries with the same income percent introduced process innovations. levels. The share of firms investing in innovation Innovation rates are even higher when it comes activities is similar to the average in other emerging to marketing innovation; 69 percent of firms markets and developing countries. performed marketing improvements. However, Kenyan firms appear to rely less on external sources only 11 percent of product innovators and 18 percent of process innovators introduce innovations for knowledge capital investments and innovation that are new to the national market and a mere 2 than similar countries. Further, two thirds of percent of firms are international innovators. product and process innovations are developed Organizational innovation is the weakest type of internally, suggesting little cooperation between innovation in Kenya; only 27.8 percent of firms firms and with universities. surveyed carried out organizational innovations. There is a mismatch between Kenya’s ranking In terms of sectors, innovation is more prevalent in terms of innovation outcomes and inputs. in manufacturing and hotels and restaurants, While the level of investments in innovation or than in services. The chemical sector is the most knowledge inputs are in line with those in other innovative. Organizational innovation is more similar income per capita countries, and below prevalent in Machinery and Equipment and less most advanced countries, the levels of innovation common in services. outcomes are much larger than those in advanced In terms of the size of firms, medium and large economies and many similar countries. This is firms are more innovative than small firms. indicative that the degree of “innovativeness” of Product and process innovation rates are higher in the innovation outcomes introduced is low, and younger (below 10 years old) and in older firms reinforces the idea of very incremental innovation (more than 30 years old). happening in Kenyan firms. Firms that do not participate in international markets are also less innovative. Firm-level innovation and performance There is very little correlation between product Regarding the impact of innovation activities on firm- and/or process innovators and organizational level performance and productivity, the estimation innovators. Firms do not tend to introduce results for our sample of Kenyan firms cannot find a complementary organizational innovations when statistically significant impact on productivity. Also, implementing product or process innovations, and we cannot find evidence of positive complementarities this may limit the impact of these innovations. between different types of innovation in increasing According to managers, the main reason why productivity. These results reinforce the idea that a lot innovations are implemented is to improve of the measured innovation activity is so incremental the quality attributes of existing products and that it has little impact on productivity. processes. Understanding Firm-Level Innovation and Productivity in Kenya iii Regarding employment generation, the information finance significantly constrains investments in R&D. reported in the survey suggests that innovation Third, Kenyan firms appear to have an overreliance on is likely to have a positive impact on employment internal sources for financing and executing knowledge creation. While there is a clear increase in the demand capital investments and introducing innovations. This for skilled labor resulting from product and process may indicate a deficit in the research, knowledge, and innovations, the impact on unskilled labor is more information infrastructure, as well as the absence of uncertain, although likely to be positive. Innovation cooperation with other firms and research institutions. appears to bias the relative demand for skilled labor. Fourth, the poor educational level of the available Also, contrary to some of the findings in the literature, labor force hampers the capacity of firms to transform there is little distinction between product and process knowledge capital into innovation outcomes. This innovation in terms of employment generation. reinforces the complementary role skilled labor plays in innovation and highlights the need to enhance the Overall conclusion and policy supply of skilled labor. implications Firm-level innovation activity in Kenya is higher In terms of policy implications, these results suggest than that in similar income per capita countries, three levels at which innovation policy should be particularly in product and process innovation, but focused. less so in terms of organizational innovation. It also appears to be low in innovativeness intensity, i.e.: it is At the firm level it is important to: incremental. While it is unsurprising that innovations Support the capacity of firms to convert innovation in countries far from the technology frontier are outcomes into productivity gains. Information incremental and far from radical,the question then is failures and asymmetries, where firms lack the to what extent incremental innovations contribute to resources and the understanding to gather the productivity growth. The findings of the paper suggest required information, resources and know-how that innovations do not appear to have an impact on to innovate, increase the uncertainty to innovate productivity. Furthermore, the findings suggest that due to the higher likelihood that the innovation the causal chain that connects knowledge capital outcome will fail or be commercially unviable. investments and activities (such as R&D or training) This is exacerbated by coordination failures to products, process, or organizational innovations, where the individual costs of improvements are which are then translated into increased firm very high, especially for Small and Medium productivity, breaks down in the case of Kenyan firms. Enterprises (SMEs) since the supply of advisory services is insufficient and tends to target large It is difficult to identify the factors that hinder a firms. This requires support programs that target positive linkage between innovation activities and productivity and innovation by improving the productivity, but the empirical analysis in the paper information, capabilities and management skills of suggests that there are some important obstacles to firms. Technology extension services can address innovation. First, when benchmarked internationally, these market failures and help to realize improved more innovation outcomes are produced with less organizational, managerial, and technological knowledge capital investments, which suggests low changes. These services provide information on “innovativeness” of these innovation outcomes, and managerial and production practices, and how to therefore they are likely to be insufficient to impact adopt them, in order to increase productivity and productivity. Second, and related to the first point, competitiveness. the empirical analysis suggests that a lack of access to iv Catching up to the Technological Frontier? Enhance R&D financing and cooperation among At the sector level it is imperative to: firms and academic institutions Improve the quality of the physical and human o In the presence of financial failures to fund capital infrastructure for innovation, including innovation, R&D support is likely to be research labs, as a means of improving the required to boost knowledge investments. The availability and quality of innovation services for international experience suggests that gradual firms. partial subsidies to high quality projects are more Enhance the supply of skilled labor, especially in effective than indirect support by tax exemptions. areas such as Science, Technology, Engineering, Supporting these high quality projects, in and Math (STEM) skills, which are highly conjunction with firms and university projects complementary to the introduction of innovations. (see below) can have a positive impact on the amount and quality of R&D. At the institutional level, and given the current o Support should be provided to enhance institutional vacuum regarding innovation policy, cooperation between firms, encourage private it is critical to finalize and implement the projected sector-university linkages, and remove institutional framework in the Science, Technology coordination failures by providing subsidies to and Innovation Act of 2013. This would help to high quality innovation projects that involve better coordinate and design instruments, effectively several firms and/or firms and academic diagnose and evaluate policies, and incentivize institutions. dialogue with the private sector. Understanding Firm-Level Innovation and Productivity in Kenya v SECTION ONE INTRODUCTION Kenya’s economy has undergone a significant process continued to accelerate over the last decade, driven of structural transformation over the last decade. The mainly by public investment, aggregate productivity economy showed an accelerating trend after 2002 with growth has remained stagnant since the 1970s and GDP growth increasing steadily from below 1 percent turned negative during the 2008-09 economic crisis in 2002 to 7 percent in 2007. The economy has been and during the macroeconomic instability period in hit by several shocks since 2007, starting with the 2011 (Figure 1). post-election violence in January 2008, which led to a slowdown in GDP growth to 1.5% and 2.7% in 2008 Figure 1. Productivity levels in Kenya: Value added and 2009 respectively. Nevertheless, economic growth (khs) per worker (1969-2010) started to rebound in 2010 and recent predictions 400 suggest higher growth rates during the period 2014- 350 2018, exceeding the growth rates before 2008. 300 250 Amidst this positive growth context, in October 2013, the Kenyan Government launched the Second 200 Medium-Term Plan (MTP-2) of the Vision 2030. 150 The aim of Kenya’s Vision 2030 is to create “a globally 100 competitive and prosperous country with a high quality 50 of life by 2030” and to shift the country’s status to upper-middle income level. While the improvement 0 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 in economic performance in the past decade is Total Economy Manufacturing Trade services remarkable, there are indications that achieving such ambitious targets might be difficult, especially given Source: Authors’ own elaboration from data from de Vries et al. (2013) the slow rate of employment generation. Attaining the high rates of growth and the degree Comparing Kenya with similar countries in terms of of transformation envisioned in Kenya’s Vision GDP per capita in other regions suggest some relative 2030 requires a renewed emphasis on boosting underperformance in the Kenyan economy. Kenya’s productivity. Only by increasing productivity can economic growth is driven primarily by the services Kenyan firms become globally competitive and sector and an over reliance on the domestic market. generate the quantity of high quality jobs required to This indicates a significant lack of competitiveness boost incomes and achieve shared prosperity. that prevents firms from competing in international markets.It acts to constrain the potential of the economy In order to boost productivity, there is need for broad- in terms of future growth and employment creation. based innovation and investment in knowledge This, combined with accelerating population growth, capital. Innovation is the engine of the ‘creative explains the high levels of youth unemployment in destruction’ process that spurs economic dynamism Kenya. New entrants to the labor market, who are on and transformation and is at the center of the average 20 years old, face an unemployment rate close development process (Schumpeter 1942). Innovation to 35 percent. More importantly, while investment has contributes to the twin goals of shared prosperity and Understanding Firm-Level Innovation and Productivity in Kenya 1 poverty reduction by generating productivity gains How prevalent is firm-level innovation in Kenya, that increase employment, raise wages, and improve and what sorts of firms innovate? access for the poor to products and services. Investing What are the main types of knowledge capital in innovation increases the capabilities of firms, investments that firms implement? enabling them to integrate in global value chains and compete in international markets, while facilitating What are the main determinants and barriers to the adoption of new technologies that improve labor firm-level innovation? productivity.1 What are the links between innovation activities and productivity or employment? While innovation has the potential to generate large productivity gains and significantly improve In order to answer these questions we use the most allocative efficiency2—the ability of the economy to recent enterprise survey (2013) and the linked 2014 allocate resources in more productive ways—we know innovation module. This is the most comprehensive very little about the nature and impact of firm-level innovation survey implemented in Kenya to date, as innovation activities and investments on knowledge it links a full innovation questionnaire answered by capital in developing countries in general, and in 549 firms in the manufacturing and services sectors, Kenya in particular. Firm-level innovation activities with information on the characteristics of firms.3 In such as R&D, or outcomes such as patenting, tend addition, it allows analysis of firm-level innovation in to be negligible in these countries, and there is little manufacturing as well as services. understanding regarding the impact of innovation efforts on performance or the main barriers to the The report is structured as follows. Following this adoption of innovative activities. introduction, section 2 describes Kenya’s policy framework for innovation. Section 3 defines the The aim of this report is to provide a rich description different measures of innovation, and describes the of the nature of firm-level innovation and datasets used in the analysis. Section 4 provides an investments in knowledge capital in Kenya, and the explanation of the firm level innovation landscape in link between innovation activities and productivity Kenya. Section 5 analyzes the main determinants of and employment. Specifically, we set out to answer innovation in the manufacturing and services sectors the following questions: as well as the relationship between innovation and firm performance. The last section provides conclusions and policy recommendations regarding firm-level innovation in Kenya. 1 At the aggregate level, theories of economic growth put innovation at the center of the growth process since the seminal work of Solow (1957), where economic growth is driven by technical change. This interest in innovation was reinforced by the emergence of “new growth theory” emphasizing the role of knowledge accumulation for the growth process and ‘Schumpeterian” creative destruction arising from a competitive R&D sector as the main engine of growth (Aghion and Howitt, 1992; Romer, 1986). At the micro or firm level, where innovation occurs, Klette and Kortum (2004) show how innovation activities create rich firm-level dynamics. In their model innovation increases product quality and make firms more competitive, which increases their revenue and size and forces existing firms producing old and obsolete versions of the product to exit the market. 2 Lentz and Mortensen (2008), using Danish firm-level data, find that up to 75% of productivity growth comes from reallocation of inputs to innovating firms, of which 25% is entry and exist of firms and 50% reallocation to growing innovative firms. 3 This dataset provides a larger source of information and sample than the previous 2007 enterprise survey, which had a sample of 396 firms but no comprehensive innovation questionnaire, and the 2012 innovation survey, implemented by the Ministry of Education, Science and Technology (MoEST),which had a sample of 160 firms. 2 Catching up to the Technological Frontier? SECTION TWO The Innovation Policy Framework The innovation policy framework in Kenya in the policy, and funded by the Research Fund. Its role is to period prior to 2013 was characterized by large institutionalize linkages between universities, research institutional fragmentation and the absence of institutions, the private sector, the Government and a strong coordinating institution to mobilize other actors. It also aims for the creation of science and government efforts in the area of Science, innovation parks, institutes or schools, or designated Technology, and Innovation (STI). Vision 2030 existing institutions as centers of excellence in priority lays out the role of STI in economic growth, with sectors. Finally, one of its most important roles is particular focus on priority growth sectors. The 2013 to develop and continuously benchmark national Act and the First Medium Term Plan (2008-2012) innovation standards based on international best provide rationalization guidelines for both policy and practices. institutional arrangements regarding innovation and technology, so as to align them with productivity and iii. The National Research Fund enterprise growth. The Fund aims at mobilizing resources to develop research capacity and scientific information. It The Science, Technology and Innovation Act of 2013 will also compile and maintain a national database is an attempt to improve on the STI institutional of expenditure on research and innovation–both framework, in a bid to complement the policy goals of internally and by other agencies. Vision 2030. The new STI institutional framework, under the STI Act of 2013, is comprised of three This new framework, however, has not yet been main elements: implemented and the bulk of innovation policy remains the responsibility of the MoEST. i. The National Commission for Science, Technology and Innovation (NACOSTI) This weak institutional framework for STI is The National Commission for Science, Technology translated in the lack of any significant innovation and Innovation (NACOSTI) is an autonomous policy support instruments. Only certain programs government institution, with the role of leading inter- for supporting technology transfer or intellectual agency efforts to develop policy on science, technology, property rights exist, although these tend to be small and innovation–across all levels of government. It and mainly focused on providing relevant market also assures the relevance and quality of science, information to firms. In addition, some programs technology, and innovation programs, while keeping that target assistance for market access also support track of progress in research systems. innovation via quality certification or by providing information on new markets and technologies. Other ii. The Kenya National Innovation Agency types of innovation instruments, such as R&D tax (KENIA) incentives, have yet to be developed. Therefore, the STI This Agency, also established under the Science, institutional framework in Kenya can be described as Technology and Innovation Act of 2013, is the an embryonic policy framework. implementation arm for the country’s STI agenda and Understanding Firm-Level Innovation and Productivity in Kenya 3 SECTION THREE Data Description and measurement of Innovation 3.1 The data 3.2 Measuring innovation In order to examine the innovative behavior of Innovation requires the transformation of knowledge firms in Kenya, we use the World Bank 2013 capital or innovation inputs, both tangible and Enterprise Survey (ES) and its linked innovation intangible—such as training, equipment, R&D or module. This is the most comprehensive survey on intellectual property acquisitions—into innovation innovation information carried out in Kenya to date. outcomes such as the introduction of new and It complements the first national innovation survey improved products, new production processes, conducted by the MoEST in 2012 involving 160 firms, or organizational changes (see Boxes 1 and 2). as well as a pilot innovation survey conducted in 2013 Firms invest in knowledge capital inputs in order involving 310 establishments, mainly in the services to increase their capabilities and produce innovative sector. The 2013 ES, which corresponded to the period outcomes. As well as requiring tangible assets such as of analysis 2010-2012, covered information regarding technology, equipment, and the physical production innovation activities and outcomes for 549 firms,4 facilities, innovation needs intangible assets such including micro firms and also those in the services as human capital, scientific and creative capital, and sectors. Table 1 shows the distribution of firms by organizational capital. In turn, these inputs require sector and size. The survey uses a stratified sampling specific innovation activities. Firms invest in training strategy, where firms are stratified by industry, size, in order to increase the available human capital. and location.5 Despite its small sample size, the 2013 In addition, firms invest in R&D, software, and ES improves on the previous one of 2007, which was digitalization or copyrights, patents and licenses in less sector representative. Furthermore, it is the largest order to increase their scientific or innovative capital. and most representative survey available, as it includes In the case of the creative industries, innovation innovation information. An additional advantage involves investment into developing these creative is that the survey collects substantial balance sheet assets. Finally, innovation also requires organizational data and other information regarding the investment capital through investments in marketing and climate, which enables the linkage of innovation branding, adoption of new business models, design efforts to performance and potential obstacles. and prototyping, or corporate alliances and networks. Table 1 Distribution of firms by sector and size enterprise survey 2013 (number of firms) Size Sector Large Medium Small Micro Total Computer and related activities 2 5 7 Construction 4 4 Hotels and restaurants 9 19 18 46 Manufacturing 66 99 100 15 280 Transport, storage and communications 8 9 7 1 25 Wholesale and retail 12 49 99 24 184 Total number of firms 95 178 233 40 546 Source: Author’s own elaboration from the Enterprise Survey (2013) 4 The module of the ES targeted a total of 713 firms, but only 549 of these firms are surveyed in the full innovation module. 5 Industry stratification uses four manufacturing industries (food, textiles and garments, chemicals and plastics, other manufacturing) and two service sectors (retail and other services).Regional stratification includes five regions: Central, Nyanza, Mombasa, Nairobi, and Nakuru. Finally, size stratification uses the following size definition: micro (1 to 4 employees); small (5 to 19 employees); medium (20 to 99 employees); and large (more than 99 employees). 4 Catching up to the Technological Frontier? As illustrated in Figure 2, the combination of these To measure innovation, one can focus on both inputs yield innovation outcomes in the form of measuring inputs and innovation activities, and/ new or improved products and services, production or measuring innovation outcomes. The early and delivery processes, business organization, and innovation measurement literature focused on a patented intellectual property. However, achieving specific set of innovation inputs that were easier to the outcomes is heavily dependent on the ability of quantify, for instance R&D, or the intensity of the the firm in question, on the specific sector and country technology used. These early efforts were followed context, and on the enabling environment and policy by the implementation of the Oslo Manual type of framework in place. surveys, which mainly focus on measuring innovation outcomes such as product/process improvements or These innovation outcomes can impact firm patents at the firm level. A third generation of synthetic performance in different ways. Successful innovations innovation indicators, such as the OECD STI are likely to increase firm-level productivity by scoreboard, were developed later on. These indicators improving the capacity to transform factors of combine innovation inputs and outputs/outcomes in production into more and better products, and by order to facilitate cross-country benchmarking and more efficiently creating products of higher value. comparisons. Second, the increase in productivity is expected to increase the marginal productivity of labor, and as a Innovation input indicators are often calculated at result, increase the quality of jobs, i.e.: more productive the aggregate level using different sources such as jobs.6 Third, more productive firms are expected to national accounts or by aggregating firm-level or push less productive firms out of the market, thereby sector information. On the other hand, innovation increasing the overall efficiency of the economy. This outcomes are mainly gauged by using firm-level will improve allocative efficiency. All this, however, innovation surveys. depends on the quality of the innovation and the ability of firms to translate innovation outcomes into improved performance. Figure 2. The Innovation Function INNOVATION INPUTS AND ACTIVITIES INNOVATION OUTPUTS AND OUTCOMES IMPACT TANGIBLE ASSETS • Technology • Equipment • Facilities • Improved Productivity growth products and INTANGIBLE ASSETS services • Human capital (quality of labor) - Training • Improved production • Scientific and creative capital processes More productive jobs - R&D activities and delivery - Digital capital-software and databases - Copyright, patent, licenses • Improved - Mineral exploration organization - Trade secrets Gains from factor reallocations - Movie, music and book development • Intellectual • Organizational capital property - Marketing and branding - Business models - Design and prototyping - Alliances and networks Source: Author’s own elaboration 6 The impact of innovation on the level of employment at the firm level depends on the type of innovation implemented. For example, some products innovations are likely to increase the demand for labor by expanding production lines, while on the other hand, some process innovations can reduce the demand for labor when improving the efficiency in which existing products are produced. What effect is larger is an empirical question, but most empirical studies suggest that the employment creation effect is larger. See section 5 for an empirical analysis for Kenya. Understanding Firm-Level Innovation and Productivity in Kenya 5 Box 1 Innovation Activities Innovation or knowledge inputs are activities that are associated with the development of innovation at the firm level. They are grouped into four main categories: i. Research and development – firms are surveyed about the source of R&D (internal vs external) as well as expenditure on this component; ii. Capacity building - firms report on the expenditure carried out in training for producing innovations; iii. Purchase/licensing of inventions or other knowledge forms - firms report on the purchase of inventions or intellectual property that help introducing innovations; and iv. Intellectual property - firms report on whether they applied for patents, utility models, trademarks, copyright design or registered an industrial design. A challenge of measuring innovation outcomes is using data from different sources (see framework the subjective nature of many of the questions used developed by Corrado et al. (2005) and box in in the surveys. The Oslo Manual, which is the main Appendix 1). While this approach offers a clearer reference for these types of surveys, defines innovation and broader measure of firms’ capabilities, it faces the as “…the implementation of a new or significantly challenge of being unable to obtain the information improved product (good or service), or process, a new required from existing innovation surveys. Moreover, marketing method, or a new organizational method it is important to highlight that this measure of in business practices, workplace organization or innovation assets is not equivalent to innovation external relations.” Most surveys use this definition to outcomes; inputs can be more or less efficiently used, identify innovations by directly asking firm managers and therefore, there is some uncertainty about the and owners whether they have implemented “new” type and extent of innovation outcomes that can be or “significant” changes or improvements in the last produced by firms using these knowledge assets. three years.7 This is problematic since “significant” is a highly subjective term, and is also self-reported. In general, any sound analysis of innovation activity It is therefore important to supplement analysis of should combine a focus on both knowledge capital innovation outcomes with some measures of revealed inputs and innovation outcomes. Although the expenditure on innovation inputs. ES innovation survey does not provide enough information on some of these intangible assets, it Several authors have advocated a focus on knowledge provides information on various sources of knowledge capital assets as a better measure of innovation, and in capital and innovation outcomes (see Box 1 and Box 2 recent years there has been a renewed effort to better for a detailed overview of the information available in measure and capture investments in intangible assets the ES innovation questionnaires). 7 The impact of innovation on the level of employment at the firm level depends on the type of innovation implemented. For example, some products innovations are likely to increase the demand for labor by expanding production lines, while on the other hand, some process innovations can reduce the demand for labor when improving the efficiency in which existing products are produced. What effect is larger is an empirical question, but most empirical studies suggest that the employment creation effect is larger. See section 5 for an empirical analysis for Kenya. 6 Catching up to the Technological Frontier? Box 2 Innovation Outcomes in the Enterprise Survey Similar to the innovation surveys based on the Oslo Manual, the ES innovation survey measures the following types of outcomes: i. Product innovations ii. Process innovations iii. Organizational innovations iv. Marketing innovations For each of these outcomes, the survey includes information about the: • Extent (number) of innovations introduced • Impact on the specific operational aspects of the firm • Level of automation involved in their adoption • Level of novelty of the innovation(s)–local market, national market and international market • Channels used to acquire business intelligence • Acquisition of talent as part of the innovative process • Level of collaboration in the development of the innovation(s) and the impact of the number of staff(skilled and unskilled) that the firm had. Product innovations are essentially new, redesigned, or substantially improved goods or services. In the context of the survey, there are 3 metrics used: a. New products to the firm b. Significantly improved products c. New products to the market. Process innovation is the implementation of new or significantly improved production or delivery methods. Specifically: a. Innovation methods for manufacturing products or offering services b. Innovative logistics, delivery, or distribution methods for inputs, products, or services c. Innovative supporting activity for processes, such as maintenance systems or operations for purchasing, accounting, or computing. The following are not considered to be process innovations: Minor changes or improvements; an increase in production or service capabilities through the addition of manufacturing or logistical systems which are very similar to those already in use; ceasing to use a process; simple capital replacement or extension; changes resulting purely from changes in factor prices; customization; regular seasonal and other cyclical changes; and trading of new or significantly improved products. Organizational innovation means the implementation of a new organizational method in business practices, workplace organization, or external relations. This type of innovation is grouped into: structural innovations which are meant to impact responsibilities, accountability, command lines, and information flows, as well as the number of hierarchical levels; the divisional structure of functions (research and development, production, human resources, financing, etc.) or the separation between line and support functions; and procedural innovations, which consist of changes to routines, processes, and operations of a company. Thus, these innovations change or implement new procedures and processes within the company, such as simultaneous engineering or zero buffer rules. Marketing innovations are changes made to incorporate the advances in marketing science, technology or engineering to increase the effectiveness and efficiency of marketing, in order to gain competitive advantage. Understanding Firm-Level Innovation and Productivity in Kenya 7 SECTION FOUR The Firm-Level Innovation Landscape in Kenya The evidence emerging from case studies suggests 4.1.1 Innovation outcomes that innovation, commonly seen as the work of highly educated labor in R&D intensive companies with Firm-level innovation rates in Kenya are relatively high strong ties to the scientific field, is inevitably a “first Innovation outcomes in Kenya are relatively high world” activity (Farberger et al 2010). Innovation in (Figure 3) according to the ES. 53 percent of firms developing countries, as demonstrated in Box 3 below, in the survey introduced either a product or a process is better characterized as an attempt to try out new innovation during the period 2010-1012; 39.8 percent or improved products, processes, or ways to do things of the firms introduced product innovations; and 38 (Bell and Pavitt, 1993; Kline and Rosenberg, 1986). percent introduced process innovations. Innovation This is a process of technology adoption, imitation and rates are even higher when it comes to marketing adaptation that takes place far from the technological innovation; 69.2 percent of firms performed frontier, where firms adopt incremental (as opposed marketing improvements. On the other hand, only to radical) changes (Fagerberg et al. 2010). It is also 27.8 percent carried out organizational innovations.8 a process that requires the combination of different In sum, introducing significant changes to products innovation outcomes, and modes of innovation, in and processes, and especially marketing, is relatively addition to product and processes, such as marketing common in Kenya. or organizational innovations (Bell and Pavitt, 1993). These results from the 2013 ES survey are much more 4.1 How innovative are Kenyan modest than those from the 2012 national innovation firms survey carried out by the MoEST, which indicated The objective of the rest of this section is to provide an overall innovation intensity of 89.9 percent; 70.9 a detailed description of the extent and intensity of percent of firms introduced product innovations, firm-level innovation activity in Kenya. We start by 92.4 percent introduced process innovations; and summarizing the aggregate picture arising from the 85.4 percent introduced organization and marketing innovation survey for the period 2009-2012. innovations. The results are also below, although more Box 3 An example of firm level innovation in Kenya - Fairbrooks Water Purification Fairbrooks Water Purification, a venture supported by the Kenya Climate Innovation Center (KCIC), has developed an innovative PVC-based tube system that filters water in an effective and cost-efficient way. First, ground water is pumped into a high tower; in a second phase, gravity pushes the water through a tube system that filters out bacteria, sediments, and other particles. This innovative system can produce up to 10,000 liters of clean and safe water per hour. This innovative company started three years ago and is already an inspiration for many new local entrepreneurs. The system is patented and has been sold to large hotels, restaurants, and many organizations across the country. The system costs around US$12,000 and mainly targets institutions and the service sector. However, the company is working to bring this innovation to a larger part of the population. Source: InfoDev (2015) 8 Only medium and large companies were asked about organizational innovations. 8 Catching up to the Technological Frontier? Figure 3. Innovation outcomes (% all firms) Firm innovation in Kenya (% of all firms) Firm innovation in Kenya Process innovation (% of all firms) 80 40 38.0 69.2 60 30 28.9 53.2 26.0 22.9 40 39.8 38.0 20 17.6 27.8 20 10 0 0 Product innovation Process innovation Product or process Process innovation New process to market Process-method Organizational innovation Marketing innovation Process-support activity Logistics innovation (a) Innovation outcomes (c) Process innovation outcomes Firm innovation in Kenya Product innovation (% of all firms) Firm innovation outcomes in Kenya (% of firms) 39.8 20 18.5 40 15 30 27.5 20 10 7.6 14.0 6.3 11.0 5.5 6.0 10 5 3.6 .0.17 0 0 Product innovation New product to national market International innovator Patent application Trademark application Utility model application New product for the rm % sales new product Industrial design registration Copyright application Patent, trademark, utility.. application (b) Product innovation outcomes (d) Applications for patents, trademarks Source: Enterprise Survey (2014) in line with, the ones from the 2013 pilot innovation existing products and processes, or the incorporation survey that covered mainly services, which found of products that are already produced by other firms that 62.9 percent of firms were product or process in the domestic market. Furthermore, only 1.7 percent innovators, of which 42.5 percent of firms introduced of firms (3.4 percent of product innovators) introduce products and 37.5 percent introduced processes; 73 radical innovations—innovations that are new to percent focused on organization; and 65.1 percent on international markets. Given the level of economic marketing innovations. The discrepancies are likely to development in the country, this is unsurprising, but be the result of the smaller sample sizes and sector raises questions about the degree of innovativeness compositions of the other two surveys. and the large aggregate innovation rates observed in this survey. However, innovation in Kenya is largely incremental and there are hardly any international innovators Innovations generate significant revenue Panels (b) and (c) in Figure 3 show that when In terms of firms that introduce product innovations, the only factor being considered is whether the 27.5 percent of sales are driven by these innovations. innovative product or process is new to the country, This figure is significant if the multi-product nature then innovation rates fall substantially. Only 11 of firms is considered, and the fact that the revenues percent of firms introduced products perceived to of firms tend to be diversified across various products. be new to the country and 17.6 percent of firms introduced new processes to the country. Given the Process innovations target production processes but also narrowness of Kenya’s domestic market in comparison logistics and support services to international markets, this suggests that most Regarding the type of process innovation (Figure 3, innovations mainly consist of small improvements to panel c) that is most common, most changes occur in Understanding Firm-Level Innovation and Productivity in Kenya 9 production methods, followed by support activities to most emerging markets or across the EU, and come the production process, and logistical improvements; below only a few African countries such as Zambia or 14 percent of firms introduced all three types of Uganda (see Figure 4). process innovation at the same time. Contrary to expectations, innovation rates that are There is very little knowledge appropriation by firms measured as outcome innovations at the firm level One of the main indicators of innovation outcomes are higher in developing countries, which are further is the extent to which knowledge capital investments away from the technological frontier and have lower can be commercialized and appropriated through productivity levels, than in developed countries. patents, copyrights, trademarks, etc. In line with other The reason for this finding is not clear, but there are developing countries, levels of innovation knowledge some possible explanations. The first is related to appropriation in Kenya are low. As a percentage of al the subjective nature of the questions in the survey firms, 18.5 percent apply for some type of knowledge measuring innovative outcomes and the quality of property, of which 5.5 percent apply for a patent, 7.6 the survey implementation, which in developing percent apply for a trademark, 6 percent for a utility countries could bias innovation rates upwards.10 The model, and 3.6 percent for copyright. Bearing in mind second explanation to consider when interpreting that these are applications, and that a significant these results is that the surveys refer to formal number of applications are not granted,9 knowledge companies and since developing countries often have appropriation as a result of innovation is minor. This sizeable informal economies, these indicators are perhaps indicates that the new knowledge generated less representative of the wider economy. In order to is likely to be negligible, reinforcing the idea of properly measure innovation activity therefore, we the incremental nature of existing innovations. In need to include informal firms. A third explanation addition, only 6.2 percent of firms successfully register is that developing countries are further away from the their industrial designs. technological frontier and therefore, it is less expensive to make the incremental improvements which Innovation rates in Kenya are high compared to translate in the data to higher innovation activity. A international standards final explanation is that given the extent of market Innovation rates in Kenya are high when compared to failures and challenges in the business environments international standards. They are higher than rates in in developing countries, firms need constant changes Figure 4 Benchmarking Kenya’s innovation outcomes Product and/or Process Innovation (% of all firms) 76.5 67.7 53.3 50.2 46.4 43.2 42.4 41.8 39.0 38.0 35.8 35.3 32.9 31.5 31.2 30.0 29.4 28.6 28.2 27.7 26.3 24.8 24.2 24.1 23.0 22.5 22.3 22.3 22.1 22.0 21.6 21.5 20.9 20.7 20.5 19.8 19.6 19.6 18.3 17.9 17.7 17.6 17.3 15.4 14.0 11.9 11.3 9.3 Zambia Ugan Kenya Philipines* Ghana Tanzania Israel*DRC Malaysia* Brazil* Bulgaria Estonia Netherlands Norway Spain China* Belgium Uruguay* Finland Sweden Iceland Nepal Poland Hungary EU-27 Italy Austria Germany Ireland Croatia Latvia Malta South Africa* Slovenia Portugal United Kingdom Denmark Lithuania Slovakia Czech Republic France Cyprus Turkey Serbia Romania Luxembourg Russian Federation* Egypt* Source: UNESCO statistics, CIS-2010 and Enterprise surveys. *only manufacturing firms 9 According to the Kenya Intellectual property institute (KIPI), only around 60 patents are granted per year and close to 600 have been granted in total since 2001. These are mainly granted to foreign companies since local companies find it difficult and costly to comply with the internationally harmonized application procedures. 10 In addition of the subjectivity of the question, another challenge of the innovation measure is that the information provided is completely self-reported and difficult to verify. Nonetheless, we have reviewed the written explanations of the self-reported innovations to verify that the descriptions provided are indeed innovations. 10 Catching up to the Technological Frontier? in production and management in order to survive in 4.1.2 Innovation inputs and knowledge the market–otherwise known as survival innovation. capital investments In this case, the impact of innovations on productivity in developing countries should be less visible, given the The subjective nature of the questions regarding prevalence of incremental and survival innovations. innovation outcomes as well as their self-reported nature emphasizes the need to supplement analysis Whatever the main explanation behind these high with measures of innovation inputs and knowledge rates, it is important to stress that firms in Kenya capital investments. Unless there are significant in the formal sector implement a large number of disparities in the measurement of innovation innovations, even in comparison to international outcomes across countries, or in terms of efficiency standards. However, also in line with the experience regarding which knowledge inputs are transformed in most countries, innovation rates fall considerably into innovation outcomes, comparing innovation when one considers whether the changes implemented activities across countries can help to understand the are new to the country; and almost disappear when nature of innovation outcomes. considering innovations that are new to international markets, as Figure 5 shows for a subset of developing Purchase of equipment is the main innovation input for countries. firms in Kenya Figure 6 panel (a) shows the percentage of firms Organizational innovation is the weakest type of carrying out innovation activities (inputs). For 44.2 innovation in Kenya percent of firms, the main assets for innovation are Table 2 benchmarks Kenya’s innovation outcomes investments in equipment, machinery, and software. with that of other emerging markets and developing Also 32.2 percent of firms invested in training, while countries. To facilitate the comparison with other only 4% purchased licenses or patents. Regarding surveys, we focus only on firms in the manufacturing R&D, 23.4 percent of firms invested in R&D;11 22.4 sector. While product, process and marketing percent invested mainly in intramural (in-house) innovation rates are higher than the average R&D while 4.6 percent invested in extramural, and innovation rates, organizational innovation is similar only 3.8 percent invested in both intra and extra- to the average innovation rates observed in emerging mural R&D. markets. Figure 5 Benchmarking Kenya’s product innovations by their degree of innovativeness Firm innovation in Developing countries (Product innovation % of all firms) 64.5 60.1 60 50.1 39.8 40 33.2 28.5 20 17.9 97.6 9.0 11.0 10.8 7.4 5.4 3.9 3.0 4.7 1.5 0.9 2.2 1.7 .0.5 0.6 0.7 0.4 0 Ghana Bangladesh DRC Tanzania Uganda Zambia Kenya Nepal Product innovation New product to national market International innovator Source: Enterprise survey (2014) 11 This contrasts with existing data from the census of industrial production, suggesting that only 7.9 percent of firms invest in R&D, and the services survey suggesting 2.9 percent of firms investing in R&D in services sector. Understanding Firm-Level Innovation and Productivity in Kenya 11 Table 2 Benchmarking innovation outcomes (% of firms in manufacturing sector) Product Process Product or process Organizational Marketing innovators innovators innovators innovation innovation Brazila 23.0 32.0 38.0 54.0 48.0 Chinaa 25.1 25.3 30.0 Colombiaa 4.6 20.0 13.6 10.8 Egypt a 6.0 8.3 9.3 6.2 3.6 Israela 34.2 30.9 42.4 50.6 57.9 Malaysiaa 29.5 33.3 39.0 28.1 28.0 Philippines a 38.0 44.0 50.2 58.0 50.4 Russian Federation a 8.0 5.9 11.3 4.0 3.4 South Africaa 16.8 13.1 20.9 52.6 23.3 Uruguaya 17.2 24.5 28.6 8.4 4.8 Kenya b 46.7 37.6 57.7 31.3 66.3 Tanzania b 18.4 40.8 50.7 27.2 61.5 Uganda b 56.7 60.1 78.8 46.1 88.4 Zambia b 60.7 77.2 81.3 28.2 97.1 Nepal b 17.0 34.0 41.3 19.6 92.7 Average excl. Kenya 25.4 32.1 40.1 30.5 43.8 Source: Author’s own elaboration from a UNESCO Statistics and b Enterprise Surveys. Data corresponds to different years, the last year of each country with information available. Investments in innovation inputs are highly concentrated the sample is responsible for more than 80% of R&D and not large investment.12 Figure 6 shows the average value of these activities The share of firms investing in innovation activities is as a percentage of total sales at innovative firms, similar to the average in other emerging markets and those firms that introduce some form of innovation. developing countries Investments in R&D and training come in on average at below 1 percent of sales, while the purchase of Table 3 benchmarks the share of firms investing in licenses is almost negligible. Investment in equipment, innovation activities in Kenya against other countries; machinery, and software is 3.8 percent of sales. developing countries, emerging markets and the EU average. The table suggests that Kenya ranks close In addition, the total value of investments in R&D, to the average of this group of countries in terms of training, and purchase of licenses and patents is implementing innovation or knowledge inputs. concentrated in a few large firms. A single firm in 12 This figure likely overestimates the true R&D firm effort due to the fact that some firms that claim to have invested in R&D cannot recall the actual value invested. 12 Catching up to the Technological Frontier? Table 3 Benchmarking innovation inputs (% of firms in the manufacturing sector) Acquisition Acquisition of Intramural Extramural of machinery, other external Training R&D R&D equipment and knowledge software Brazila 4.7 1.9 34.1 4.8 26.5 China a 63.3 22.1 66.0 28.1 71.5 Colombiaa 26.8 8.9 85.8 7.2 19.8 Egypta 41.3 5.5 74.3 11.0 56.9 Ghana a 42.1 14.0 80.7 15.8 86.0 Indonesiaa 48.3 5.2 39.3 21.6 37.0 Israel a 48.9 32.2 85.1 12.9 52.6 Malaysia a 42.5 15.8 64.9 29.8 50.2 Russian Federation a 18.9 20.0 64.0 12.7 18.3 South Africa a 54.1 22.4 71.2 24.8 69.6 Uruguay a 11.1 1.2 20.3 4.4 15.1 Kenya b 37.3 8.8 66.6 6.7 47.9 Tanzaniab 39.1 7.6 59.2 15.9 34.2 Ugandab 10.0 3.1 41.9 3.1 19.2 Zambia b 21.7 12.4 44.4 6.4 25.2 Nepal b 7.6 0.5 38.8 3.0 26.9 EU27 min a 8.2 5.8 25.2 2.0 8.9 EU27 max a 81.3 54.8 98.8 53.1 96.4 Average excl. Kenya 33.5 13.7 58.5 15.1 42.0 Source: Author’s own elaboration from UNESCO Statisticsa and Enterprise Surveysb Kenyan firms appear to rely less on external sources for EU countries. Kenya ranks at the bottom of these knowledge capital investments countries but at similar levels to Bulgaria. Therefore, Excluding the acquisition of equipment and for firms that invest in knowledge inputs, these machinery, Kenyan firms appear to rely less on investments appear to be in line with the level of extramural R&D and other sources of external investment predicted by GDP per capita levels. knowledge. As we will see below, this is also the case for external sources of information for innovation There is a mismatch between Kenya’s ranking in outcomes, which may indicate an insufficient supply innovation outcomes and inputs of research and knowledge services in the country, as While Kenya ranks very high in innovation outcomes, it well as a reluctance to collaborate with other firms or comes in the middle of the group in terms of innovation universities in terms of innovation. inputs and activities. This could be explained by virtue of Kenyan firms being more productive in terms of The level of investment as a share of sales is on par with innovation production, but more likely it suggests a the minimum level in the EU lower degree of innovativeness in the improvements Figure 6 panel (c) shows the average value of these introduced in product and services, than in other activities as a percentage of total sales compared with emerging markets. Understanding Firm-Level Innovation and Productivity in Kenya 13 Figure 6 Innovation activities Firm innovation activities in Kenya (% of firms) Total innovation expenditure (% sales) 44.2 Kenya 40 Bulgaria 32.2 Lativia 30 Serbia 23.1 22.4 Luxembourg 20 Romania Slovakia Norway 10 Ireland 4.6 4.0 Hungary 0 France R&D R&D intramural R&D extramural Training Spain Purchased equipment Purchased licence or patent Portugal (a) Innovation activity (% of rms) Czech Republic Poland Firm knowledge capital investments in Kenya (innovators % of sales) Malta 40 38 Italy Slovenia Lithuania 30 Netherlands Austria 20 Cyprus Belgium Germany 10 6.7 Finland 5.0 Sweden 0.6 Denmark 0 R&D % sales Training % sales Purchased equip % sales Purchased licence % sales 4.00 3.50 3.00 2.50 2.00 1.50 1.00 0.50 0 Investment (b) Innovation activities (% of sales) (c) Benchmarking innovation Source: Enterprise Survey (2014) and CIS (2010) 4.2 Who are the Innovators? digits sector classification, for sectors with more than 20 firms in the sample. The chemicals sector appears This section provides a breakdown of the aggregated to have the greatest frequency of both innovation innovation findings in terms of firm characteristics outcomes as well as outcomes such as patents and (sector and size). trademarks applications. In addition, this sector has the highest occurrence of firms carrying out Innovation is more commonplace in manufacturing and innovation inputs and the most sizeable investments hotels and restaurants, rather than in services in these knowledge capital activities. Figure 7 panel (a) shows innovation outcomes by broad sector breakdown (for sectors with more than 20 firms Organizational innovation is more commonly found in in the sample). The graph shows that product and machinery and equipment and less prevalent in services process innovation rates are similar for manufacturing Organizational innovation is a key element for and hotels and restaurants, while lower for wholesale productivity growth by virtue of improvements and retail. As expected, given the nature of their in the quality of management (Boom and Van business, marketing innovations are more common in Reenen, 2012). The table in Appendix 2 shows that the services sector. On the other hand organizational organizational innovation is more common place in innovation is more prevalent in manufacturing. machinery and equipment, as well as in the servicing of motor vehicles, and less prevalent in the services The chemical sector is the most innovative sectors. Among the manufacturing sectors, food is the The table in Appendix 2 shows the breakdown of one with the lowest rate of organizational innovation. innovation outcomes by sub-sector using the ISIC 2 14 Catching up to the Technological Frontier? Figure 7 Innovation outcomes by group Firm innovation in Kenya by sector (% of firms) Firm innovation in Kenya by age group (% of firms) 80 80 60 60 40 40 20 20 0 0 Hotels and restaurants Manufacturing Wholesale and retail Age<5 Age 5-9 Age 10-14 Age 15-19 Age 20-24 Age 25-29 Age 30-34 Age>34 Product innovation Process innovation Product or process innovation Product innovation Process innovation Product or process innovation Organizational innovation Marketing innovation Organizational innovation Marketing innovation (a) Innovation outcomes by sector (c) Innovation outcomes by age group Firm innovation in Kenya by rm size (% of rms) Firm innovation in Kenya by trade group (% of firms) 80 80 60 60 40 40 20 20 0 0 Small (<20) Medium (20-99) Large (100 and over) No trader Importer Exporter Two-way trader Product innovation Process innovation Product or process innovation Product innovation Process innovation Product or process innovation Organizational innovation Marketing innovation Organizational innovation Marketing innovation (b) Innovation outcomes by size (d) Innovation outcomes by trade group Source: Enterprise survey (2014) Patents and trademark applications are infrequent, When looking at the intensity of investments, the most particularly in textiles sizeable R&D and training investments are in the chemicals, food, and transport services sectors, and to Regarding innovation outputs such as patents, a lesser extent, in wholesale trade. R&D and training trademarks, and copyrights, these forms of knowledge investments are almost negligible in the textiles sector. appropriation are very rare in the textiles sector, Purchase of equipment is the most significant form of where knowledge appropriation largely occurs via knowledge capital across all sectors. the registration of industrial designs. Of the services sectors, transport is the one with the most innovation Medium and large firms are more innovative than outputs, mainly applications to trademarks, copyright, small firms and industrial design registration. In line with Schumperian views of innovation, R&D and training are more prevalent in medium and large firms appear to be more innovative manufacturing than in services than small firms, but innovation rates are similar More firms in manufacturing engage in R&D and between medium and large firms (Figure 7, panel training than do firms in services. However, investments b). Product and process innovation is slightly more in equipment are more common in services. This prevalent among large firms than among medium contrasts with Loof (2005) who finds greater R&D firms, while organizational innovation is more intensity in services than in manufacturing in Sweden. prevalent at medium sized firms. Understanding Firm-Level Innovation and Productivity in Kenya 15 Product and process innovation rates are higher in unlikely to also introduce organizational innovations. younger and older firms Polder et al. (2010) find positive effects of product and process innovation on productivity when combined Product and process innovation rates are higher in with organizational innovation in the example of firms that have been in existence for less than 10 the Netherlands. Thus, the lack of organizational years, and in firms that have been operational for innovation among Kenyan firms may undermine the between 30 and 34 years (Figure 7, panel c). However, potential for productivity growth that would arise organizational innovations are less prevalent in very from the complementarity between product/process young and very old firms. This suggests that different innovation and organization innovation. types of innovation hold greater relevance or feasibility at different stages of the firm’s life cycle. 4.3 Main characteristics of the Firms that do not participate in international markets innovation process are also less innovative The innovation survey provides qualitative information The productivity literature shows that productivity is regarding the description of the innovation processes. greater among two-way traders—those firms that both Here we summarize some of this information. export and import—than single exporter, importers and firms that do not trade. Similarly, Figure 7 panel Innovations aim at increasing the quality attributes of (d) shows that product and process innovation rates are existing products and processes higher among two-way traders than they are among While only 14 percent of firms introduced new importers or exporters, while rates of organizational products, innovations are aimed at increasing the innovations hold steady between importers and quality of the product. As Figure 8 panel (a) shows, two-way traders. Firms that do not participate in the perception is that in almost all cases, product international markets are less innovative. innovations improve the quality of the product. 64.5 percent of the cases introduced a new technology There is very little correlation between product and/or or design and used different inputs, and 50 percent process innovators and organizational innovators of the cases introduced a new function. The cost Figure 7 and the table in Appendix 2 suggest attribute is the focus in only 44 percent of cases. little correlation between product and/or process This suggests a link between product innovations innovation and organizational innovation. Firms and quality upgrading. that adopt product and/or process innovations are Figure 8 How innovative are new products and processes How innovative are new products? (% of product innovation firms) How innovative are new processes? (% of process innovation firms) 98.2 80 100 73.4 70.9 67.3 80 60 64.5 64.0 60 49.9 40 38.4 44 40 20 20 0 0 New function Cheaper Better quality Di erent inputs Automate manual process Adapt used technology New technology or design Adapt new technology More e cient technology (a) Product innovators (b) Process innovators Source: Enterprise survey (2014) 16 Catching up to the Technological Frontier? In terms of process innovations, firms introduce Figure 9 shows that two-thirds of product and improvements to improve the quality of the production processes innovations are fully developed in-house, processes via adapting new technology in 73.4 percent which aligns with the finding that most of the main of cases, while in 70.9 percent of cases, the aim is to innovation activities, such as R&D, are developed use more efficient technology. 67.3 percent of firms internally. 19 percent of innovations are developed look to automate manual processes while the desire to entirely by a different firm and only 15 percent of adapt used technologies is cited by only 38.3 percent innovations are developed in conjunction with other of cases as the main reason to innovate. companies, consultants or universities. This suggests very little reliance on external sources of knowledge Two thirds of product and process innovations are and cooperation. It also suggests less cooperation and developed in-house, suggesting little cooperation and reliance on external sources than has been observed in few external links other developing and emerging markets, as shown in Table 3 above. Figure 9 How was innovation developed? How was innovation developed? (% of product innovation firms) How innovation was developed (% of process innovation firms) 80 80 65 66.5 60 60 40 40 19.2 18.5 20 20 5.2 4.9 3.1 3.9 4.5 5.4 1.7 0 2.2 0.8 1.6 0.1 0 0 All in-house All other rm Cooperation domestic rm Cooperation foreign rm All in-house All other rm Cooperation domestic rm Cooperation foreign rm Domestic university Foreign university Private consulting Government Domestic university Foreign university Private consulting Government (a) Product innovation (b) Process innovation Source: Enterprise survey (2014) Understanding Firm-Level Innovation and Productivity in Kenya 17 SECTION FIVE Innovation and Firm-Level Performance Innovation is the outcome of firms’ investments in positive impact of product innovation generating knowledge capital and management decisions. The employment is larger than the displacement effect of ultimate objective of these investments is to produce process innovation, and the net effect of innovation innovations that positively impact firm performance on employment tends to be positive. Using a similar by increasing productivity, employment, sales, methodology, Hall et al. (2007) find a low but positive profits, market shares, or markups. However, there effect of product innovation on employment in Italy, is uncertainty regarding the extent to which firms and no displacement effect from process innovation. are able to convert knowledge capital investments Thus, the scarce evidence in existence suggests an into innovation outcomes and furthermore, whether overall positive impact of innovation on employment, these innovation outcomes are likely to impact but more research is needed to understand if these firm performance. Innovation is risky since it is results also hold for firms in developing countries that almost impossible to determine ex ante whether are further away from the technological frontier. the introduction of a new product, process, or organizational change will lead to an increase in sales In relation to the impact of innovation on productivity, or productivity. Hall (2011) provides a comprehensive survey of the empirical work. The survey mainly focuses on In general, most of the evidence of the impact of 16 existing empirical studies using the workhorse innovation on productivity and firm level performance empirical model, the Crepon-Douget-Mairesse has been focused on developed countries. Regarding (CDM) model (Crepon et al., 1998), implemented the connection to employment, the case study literature using firm-level data in OECD countries and a few has emphasized the possibility that innovation acts as emerging markets. Hall’s (2011) main finding is that a mechanism to reduce employment, and also, as a in general, most studies find a positive correlation force for skill bias, as it increases the relative demand between product innovation and productivity, but the for skilled labor. While this is very important, this impact of process innovation is ambiguous. According argument suffers from a real deficit of evidence, to Hall (2011), the problem with process innovation especially regarding developing countries and firm- is that it cannot be measured in the surveys beyond level information. In a recent survey of the literature, the dichotomous variable of whether or not the firm Vivarelli (2012) suggests that the more recent micro- implements process innovations.13 In general, these econometric literature tends to support a positive studies suggest that innovation has a positive impact link between technology, proxied as R&D and/or on productivity. product innovation, and employment, especially when focusing on high-tech sectors. Vivarelli (2012) also The evidence for developing countries, however, is finds significant evidence in favor of the skill-biased scarce. One relevant study is Goedhuys et al. (2008), hypothesis across different OECD countries, different which examines the main drivers of productivity economic sectors, and different types of innovation. in Tanzania. The authors do not find any link between R&D, product and process innovations, In one of the few existing micro-econometric studies, licensing of technology, or training of employees, Harrison et al. (2006) study the impact of innovation and productivity. The results suggests that Tanzanian on employment using a comparable dataset of firms firms are struggling to convert knowledge inputs into from France, Germany, Spain, and the UK. The productivity improvements due to the poor enabling authors find that product innovation has a positive environment for business, which is the main constraint impact on employment, but that process innovation on productivity according to their empirical results. has a displacing effect on employment. However, the 13 The intensity of product innovations can be approximated by the share of total revenue associated to the product innovation. 18 Catching up to the Technological Frontier? In addition, there are indications that in developing 5.1 Methodology and data countries, investments in knowledge capital are description smaller than in developed countries. For example, In order to estimate the impact of innovation activity Goni and Maloney (2014) demonstrate that on performance, we follow a logical framework based investments in R&D as a share of GDP are smaller on the one described in Figure 2, which is the basis of in developing countries than in developed countries. the CDM model (Crepon et al., 1989). The framework One potential explanation for this is the absence holds that firms invest in knowledge inputs that can of complementary factors to enable R&D, such as be transformed into innovation outcomes according to education, the quality of scientific infrastructure, and the efficiency of their innovation function. At a later the private sector, which is weaker in countries far stage, these innovation outcomes impact productivity away from the technological frontier. which is contingent on the capacity of firms to transform innovation outcomes into improvements A related strand of the literature has empirically in product quality and efficiency. As a result, the analyzed some of these complementary factors. CDM model requires the estimation of three main Polder et al. (2010) find significant complementarities components: the knowledge function; the innovation between different knowledge inputs and innovation equation; and the productivity equation. outcomes in the Netherlands. The authors find that:(i) ICT investment and usage are important When estimating these components, there are two drivers of innovation; (ii)there is a positive effect on critical choices that need to be considered. The first productivity of product and process innovation when is to define the scope of the knowledge inputs and combined with organizational innovation; and(ii) innovation outcomes to be included in the analysis. there is evidence that organizational innovation is This is often related to the availability of data in the complementary to process innovation. Miravete and innovation surveys. In most CDM applications, data is Pernias (2006) find evidence of complementarity restricted to the use of R&D for knowledge activities; between product and process innovation in Spain’s for innovation outcomes, either product and process tile industry, and Cassiman and Veugelers (2006) find innovation dummies, or the number of patents and important complementarities between internal R&D the revenue associated to new products, are used. This and external knowledge acquisitions. paper extends the traditional CDM model to include R&D as well as two other main knowledge activities: A final element is the lack of an enabling business (i) the acquisition of equipment for innovation, and environment in many developing countries and (ii) the training of workers for innovation activities. existing market failures in the supply of technical Regarding innovation outcomes, in addition to infrastructure, human capital or technology, as well as product and process innovations, the paper focuses recent evidence of low management quality in firms in on organizational innovations, since as we saw developing countries (Bloom et al, 2012). This raises earlier, some of the literature suggests important questions about the efficiency of transformation of complementarities with process innovations.14 knowledge inputs by firms into innovation outcomes, and further, into improvements in firm performance The second choice when examining these three (see Figure 2 above). innovation components relates to how the model is solved. Decisions regarding knowledge inputs and In the following section we empirically analyze this innovation outcomes can be made simultaneously question for Kenyan firms, and try to identify the and there can be feedback effects, especially between degree to which innovation activity has translated innovation outcomes and productivity. There are two into improved performance in terms of productivity main approaches in the literature on how to solve the and employment growth. The next sub-section model. Crepon et al. (1998) suggest solving the model describes the empirical methodology, while the simultaneously using Asymptotic Least Squares.15 following sub-sections describe the empirical analysis Meanwhile, Griffith et al. (2006) assume that there are regarding impact. no feedback effects and solve the model sequentially, 14 The innovation survey does not provide information on the number of patents, unless there has been a patent application, which is extremely limited and often applications are not granted. We have some information about the share of revenue associated to product innovation; however, some of the numbers are very large especially when the products are not newly introduced. For that reason we concentrated on using the usual dichotomous variables. 15 This implies solving the knowledge function, innovation equation, and productivity equation simultaneously by maximum likelihood. Understanding Firm-Level Innovation and Productivity in Kenya 19 instrumenting knowledge activities in the innovation the three innovation components sequentially. First, equation, and innovation outcomes in the productivity we focus on understanding the determinants of equation, to avoid endogeneity. investing in knowledge inputs and their intensity, proxied by R&D per worker and by the total spent The approach in this paper is closer to the one on R&D, training, and equipment for innovation advocated by Griffith et al. (2006), and solves the by worker. Second, we focus on understanding the model sequentially, although for robustness we also determinants of product, process, and organization estimate the model using a similar method to the innovation, and the role of knowledge inputs in the original CDM approach. There are however, two probability of introducing either type of innovation. main assumptions that the paper follows when solving Finally, we measure the impact of innovation on the model: (i) firms first determine the intensity of productivity, and then on employment. input choices; and (ii) choices about the different types of innovation outcomes (product, process, or Table 4 contains the definition of the variables used organization) are made simultaneously. in the analysis. To understand the determinants of knowledge input intensity, we focus on three sets The detailed methodology is described in Appendix 3. of variables. First, we examine the characteristics The following sections present the results of estimating of firms, for instance, size, age, and the existence of Table 4 Variable description Variables Description How Much Of An Obstacle is Telecommunications Knowledge intensity j Telecommunications. Index 0: not obstacles an obstacle to 4: very severe obstacle. Intramural and extramural R&D a R&D intensity This is the principal component of expenditure per worker. five government obstacle indices. Sum of R&D, equipment and How much of an obstacle is tax b Research intensity training expenditure per worker. Government rates, tax administration, business k Innovation outcomes obstacles licensing and permits, political Dummy with value 1 if any new instability and corruption. Index or significantly improved product 0: not an obstacle to 4: very severe c Product innovation obstacle. or service introduced by this establishment. How Much Of An Obstacle is Dummy with value 1 if any new inadequately educated labor force. l Education obstacle d Process innovation or significantly improved process Index 0: not an obstacle to 4: very introduced by this establishment. severe obstacle. Dummy with value 1 if Demand pull establishment make any changes Firms that are exporters and Organization m Two-way traders e in its organizational structure by importers. innovation creating, dissolving or merging any Firms that report that demand units of departments. n Demand (-) either at home or abroad has Productivity decreased. f Sales per worker Logarithm of sales per worker. o Share(t-3) Markets share three years ago. Logarithm of value added per Firm characteristics g VA per worker worker. Capital intensity defined as the log Enabling environment p Log(K/L) of the ratio of capital to labor in the How Much Of An Obstacle is firm. h Lack of finance Access To Finance. Index 0: not an q Log (lab) Log of employment. obstacle to 4: very severe obstacle. Whether the firm is at least 25% This is the principal component of r Foreign foreign owned. two indices how much of an obstacle Trade costs- s Age Age of the firm. i is transport and customs and trade obstacles Whether the firm has between 50- regimes. Index 0: not an obstacle to t Medium 4: very severe obstacle. 99 workers. u Large More than 99 workers. 20 Catching up to the Technological Frontier? foreign capital ownership. Second, as in Crepon et Finally, neither market power nor the obstacles al. (1998), we use the importance of demand factors explored—telecommunications, trade, or government in incentivizing investments in innovation inputs. obstacles—seem to have an impact on knowledge This is measured in three dimensions: (i) the firm’s capital investment intensity. This emphasizes the market share at the beginning of the period; (ii) role of external demand factors in explaining R&D whether firms are facing a reduction in their demand, intensity. and; (iii) the degree of integration in international markets, measured by a dummy with value one if the Demand factors also explain the intensity of firm exports and imports (two-way traders). Third, we investments in knowledge activities introduce the impact of various investment climate Applying a broader measure of knowledge capital measures to understand how these affect knowledge investments, including equipment and training, investments. Specifically, we look at: lack of finance; yields very similar results to the ones on R&D, albeit trade costs-obstacles; telecommunications obstacles; with two differences. Lack of finance does not have and various government regulation obstacles. We also a significant role in explaining knowledge intensity follow the literature to determine: which variables are investments when the definition is expanded and so used to explain the different innovation outcomes; to it appears less binding after machinery, equipment decide on the degree of capital intensity of the firm and training are included. Second, foreign owned as the capital per worker ratio; the quality of human firms appear to invest more in knowledge inputs capital using the degree to which firms perceive the when considering machinery, equipment and training, absence of an adequately educated labor force as an which is likely the result of the fact that they carry out obstacle; the size of the firm; and the intensity of the R&D abroad or in related firms. knowledge investments. Foreign firms appear to invest less in R&D but more on In what follows, we summarize the main findings of overall research activities the estimations. The detailed econometric description can be found in Appendix 4. Foreign owned firms appear to invest more in knowledge inputs when considering machinery, 5.2 The determinants of equipment, and training, which is likely the result of investing in knowledge the fact that they carry out R&D abroad or in related inputs firms. Table A4.1 in Appendix 4 shows the estimates of the In the case of R&D, the only variables that appear knowledge function. to explain the decision to invest are: a decreasing domestic or international demand; larger market International competition, pressure for diversification, share at the beginning of the period in 2008 (t-3); and access to finance appear to be the main determinants the perception of greater incidence of government of R&D intensity obstacles; and obstacles regarding access to finance. Two-way traders and firms that face shrinking Firms that claim to be more finance constrained are demand domestically or internationally have larger less likely to engage in R&D, while firms that have R&D intensity. Firms that feel greater financial larger sector market shares and greater incentive to constraints will invest less in knowledge inputs. diversify due to contracting demand, are more likely This suggests that international competition and to implement R&D. pressure for diversification are important predictors of R&D intensity, while access to finance is likely Finally, the results of the expanded decision to to be an important inhibitor of these investments. invest in knowledge activities in general show large Interestingly, firms with foreign ownership tend to heterogeneity when trying to explain the investment invest less in R&D, while, in line with some of the decision. Larger firms and two-way traders are more literature, size does not appear to be a consideration. likely to engage in these knowledge investments. Understanding Firm-Level Innovation and Productivity in Kenya 21 5.3 The innovation function reasons are very much related to increasing the quality attributes of the product and production processes. As suggested above, although firms invest in knowledge, there is uncertainty as to whether these Innovators are more likely to invest in investments will result in specific innovations. knowledge inputs Therefore, the second stage of understanding the relationship between innovation and performance Given these reasons for innovation, a key question is is to determine the role that knowledge capital to what extent firms invest in R&D, equipment, and investments play in producing innovation outcomes training in order to increase these quality attributes. (Table A4.2 in Appendix 4). Table A4.2 shows that in Kenya, the percentage of product, process, and/or organization innovators is Managers introduce innovations to diversify and larger in firms that have some investment in R&D upgrade products rather than replace them or in research in general, than firms that do not have any investments. However, a significant number It is important to start by analyzing the subjective of firms still carry out innovations without any reasons to innovate, as stated in the survey by the knowledge capital investments stated. For example, firms’ managers. Figure 10 shows the main reasons for 24.4 percent of firms not performing knowledge product (panel (a)) and process (panel (b)) innovation. capital investments are product or process innovators. For product innovations, the main reasons given are For these firms, the results raise questions about the to diversify existing products via increasing market degree to which some of these innovations surpass share or improving their quality, rather than replace very simple imitation when there are no investments existing products. For process innovations, the main in knowledge capabilities to develop them. Table 6 Percentage of firms engaging in knowledge capital investments by innovator group Product Process Organization Product/process Number of innovators innovators innovators innovators firms No R&D 31.58% 29.50% 28.81% 45.25% 400 R&D 75.71% 64.08% 34.02% 87.14% 142 No knowledge inputsa 18.06% 11.57% 18.75% 24.07% 216 At least one knowledge input a 56.52% 56.19% 39.44% 75.96% 210 All knowledge inputsa 76.47% 62.86% 42.86% 85.29% 70 Source: Author’s own elaboration from the Enterprise Surveys R&D, training and equipment a Figure 10 Reasons to introduce product innovations Main reasons to innovate (% of product innovation firms) Main reasons to innovate (% of process innovation firms) 100 100 93.6 94.9 89.8 88.9 88.2 83 80 74 80 66.5 61.1 60 60 50.5 42.9 39.2 40 40 31.7 31 25.8 20 20 0 0 Replace product Extent range products New market/increase share Reduce costs Increase quality products Increase production Increase exibility Compete same products Comply standards Decrease in demand Increase speed production Increase speed delivery Decrease production costs Reduce waste Comply standards (a) Product innovation (b) Process innovation Source: ibid 22 Catching up to the Technological Frontier? But a few firms with significant knowledge investment Larger and more capital intensive firms are more likely intensity do not innovate to innovate, and lack of an adequately educated work force is a major deterrent Figure 11 shows the cumulative distribution function of the intensity of knowledge capital investments. The estimates indicate there is a greater likelihood Both graphs show that the distribution of knowledge of firms that are more capital-intensive introducing capital investments for innovators does not dominate product or process innovations. Larger firms are non-innovators, which implies that some non- more likely to introduce product innovations, while innovators have greater knowledge capital investments firms with a larger perceived obstacle in the form of than similar innovators, and therefore, suggests that educational levels of the labor force, are less likely to knowledge capital investments do not always imply introduce process innovations. The most surprising innovation outcomes. result is the fact that as with the non-instrumented results, investment in knowledge activities, both In order to better measure the impact of knowledge R&D and broad knowledge inputs, is not statistically capital investments on innovation, we estimate the significant in increasing the probability of product probability of introducing an innovation (see table and process innovation. A4.2 in Appendix 4). Below, we summarize the main findings of the estimations. Organizational innovation decisions do not appear to be taken jointly with product or process innovations Knowledge investments do not appear to impact the When organizational innovation is introduced, probability of innovating significantly the decision to introduce product and process Knowledge capital investments do not appear to innovations are still correlated, but the correlations impact the probability of introducing innovations; we with the residuals of the organizational innovation find a negative sign only in association with R&D and equation are not statistically significant. This suggests organizational innovation. that organizational innovation decisions are taken independently of product and process innovations. Product and process innovation decisions are highly The results again suggest the importance of large correlated firms and capital intensity, although only for product innovation, and that the educational levels of the Estimates show a negative and statistically significant labor force can act as an obstacle for both product and correlation between the two equations, which suggest process innovation. Medium sized firms appear to be simultaneity between product and process innovation more likely to implement organizational innovations. decisions and the need to control for this correlation Again, we do not find evidence that investments by estimating both equations simultaneously. in knowledge capital are statistically significant in Figure 11 Cumulative distribution functions of knowledge capital investments by innovator group 100 100 80 80 60 60 40 40 20 20 0 0 5 10 15 5 10 15 Log (R&D expenditure per worker) Log (total knowledge expenditure per worker) Non innovators Product or process innovator Non innovators Product or process innovator (b) Distribution of research and knowledge (R&D, training and equipment) (a) Distribution of R&D expenditure per worker by innovator group expenditure per worker by innovator group Source: Author’s own elaboration from Enterprise survey (2014) Understanding Firm-Level Innovation and Productivity in Kenya 23 affecting the probability of innovation. The exception One important element to consider is the potential is a marginally significant coefficient of R&D on endogeneity of innovation outcomes to productivity, organization innovation. since firms that are more productive are potentially more likely to carry out innovations. The lack of panel Kenyan firms’ investments in knowledge and structure in the dataset does not allow the inclusion acquisition of capabilities do not necessarily translate of firm-level fixed effects that could attenuate part of into firm-level innovations this endogeneity problem. Overall, the results of the innovation equation Innovators tend to be more productive than suggest that Kenyan firms’ investments in knowledge non-innovators, although not for all productivity levels and acquisition of capabilities in the form of R&D, equipment, and training, do not necessarily translate Figure 12 shows the relative distribution of value into firm-level innovations. This result is unsurprising added per worker for both innovators and non- for R&D, since it is likely that the type of small innovators. The diagonal of the graph marks the incremental innovations do not require significant points at which the distribution would be equal for acquisition of capabilities through R&D. However, both groups at each percentile; therefore, any points it is more surprising for total knowledge capital below the diagonal suggest larger labor productivity investments, since even imitations tend to require levels for innovators. In the case of product and some form of acquisition of machinery and training process innovators, panel (a), innovators tend to be of workers. more productive except in the lower and especially, in the larger quintiles. The picture for organizational 5.4 Innovation and productivity innovators, panel (b), is also similar. Therefore, although the productivity distribution of innovators The final stage in determining the impact of innovation does not completely stochastically dominate non- on performance is to estimate the productivity innovators, productivity levels are higher for equation. For productivity measures, we use two proxies innovators at most quintiles. This could be the result of labor productivity: the logarithm of sales per worker, of innovations increasing productivity, as well as more and the logarithm of value added per worker. Although productive firms becoming innovators.16 value added per worker is a better measure of labor productivity, the existence of missing observations for material inputs in some firms significantly reduces the sample when using this variable. Figure 12 Relative cumulative distribution functions–innovators and non-innovators Relative value added per worker distribution by quintiles Relative value added per worker distribution by quintiles 100 100 Non-product or process innovators Non-organization innovators 80 80 60 60 40 40 20 20 0 0 0 20 40 60 80 100 0 20 40 60 80 100 Product or process innovators Organization innovators (a) Product and process innovators (b) Organization innovators Source: Author’s own elaboration from Enterprise survey (2014) 16 As a result, the empirical analysis should control for this potential reverse causality between innovation and productivity, by either instrumenting innovation or estimating the innovation and productivity equation simultaneously. 24 Catching up to the Technological Frontier? Innovation outcomes do not have a statistically innovation more than offsets this potential negative significant impact on productivity effect, which makes the overall effect positive; and, (ii) innovation is skill-biased and tends to increase The main result that emerges from the estimates is that the demand for skilled labor relatively more than the innovation is not statistically significant in increasing demand for unskilled labor. productivity for our sample of Kenyan firms, with or without instrumenting. Only capital intensity appears We know very little about the employment dynamics to be statistically significant in the sales per worker associated with innovation in developing countries. specifications using R&D to predict innovation. Therefore, it is important to analyze the impact of innovation on employment in these countries. This is We cannot find evidence of positive complementarities particularly important in Kenya, given the need for between types of innovation the country to absorb the large number of people Also, we introduce interactive innovation dummies entering the labor market every year. to capture complementarities between different types of innovation. However, the coefficients are not Before starting the analysis it is important to stress statistically significant. that the overall impact of innovation on employment cannot be measured only by using firm-level data. As proposed in the previous section, as a robustness Firm-level surveys allow the capture of direct impacts test we estimate stage 2 and stage 3 simultaneously of innovation on the employment levels in the firm. as a system of equations by maximum likelihood and However, there are other general equilibrium effects allowing for correlations in the innovation decisions. of innovation on employment, such as potentially Table A4.6 in the Appendix shows the results for sales changing markets shares, competition, and prices per worker. The results confirm the lack of statistically that affect the demand for labor and that can be very significant impact of innovation on productivity. difficult to identify and quantify. One needs to interpret these results with caution, The ES survey allows the exploration of the given the low number of observations and the lack of hypothesis of skill-biased labor growth based on the panel structure of the dataset. Overall the main result responses from firm managers to questions about the of the estimates is that firm level productivity appears impact of innovation. Specifically, for product and to be largely unexplained and there is no statistically process innovations, the questionnaire asks about the significant impact of innovation on productivity. This is impact each innovation type has had on skilled and at odds with most of the evidence in OECD countries unskilled workers. One caveat for the analysis is that summarized in Hall (2011), but it is consistent with the respondents only report whether employment has some evidence in other developing countries such as increased or decreased, rather than the numbers of Tanzania (Goedhuys et al., 2008). It suggests that workers, so we cannot precisely gauge the total level the capacity to transform innovation outcomes into of employment generated or destroyed as a result of increases in productivity in these countries is much the innovation. lower due to more incremental and survival type of innovation. Changes in labor associated to product and process innovation are remarkably similar 5.5 Innovation and employment The results are summarized in Figure 13. The first A final important element to analyze when looking three bars represent the share among innovators at the impact of innovation on performance is in relation to changes in skilled workers associated employment. As suggested above, most of the evidence with the particular innovation, and the last three bars regarding the impact of innovation on employment are related to unskilled labor. The percentages are in OECD countries suggests that: (i) although remarkably similar when comparing labor changes process innovation can have a negative effect on firm associated with product and process innovations. level employment, the positive impact of product Understanding Firm-Level Innovation and Productivity in Kenya 25 Figure 13 The impact of innovation on skilled/unskilled employment Impact on employment (% of product innovation firms) Impact on employment (% of process innovation firms) 51.2 52.0 60 50 54.2 45.8 51.4 46.4 40 40 30 20 16.7 17.6 20 17.5 16 10 2.2 3.0 0 0 Increase skill L Decrease skill L Zero change skill L Increase unskill L Increase skill L Decrease skill L Zero change skill L Decrease unskill L Zero change unskill L Increase unskill L Decrease unskill L Zero change unskill L (a) Product innovation (b) Process innovation Source: Enterprise survey (2014) Innovation increases skilled labor To further understand the potential differentiated impact on employment of both product and process Starting with product innovations, half of product innovation, we focus the analysis on those firms that innovators (51.4 percent) increased the number of introduced both types of innovation during the period skilled workers, while only 2.2 percent reduced the and examine what happened to net employment. number of skilled workers, and 46.4 percent did not Although this group represents 25.14 percent of change skilled labor. firms in the innovation survey, it allows us to compare the impacts of product and process innovations The impact of innovation on unskilled labor is small in simultaneously. size and uncertain in sign Regarding unskilled labor, 17.5 percent of product Table 12 tabulates the number of firms introducing innovators increased the number of unskilled workers, both product and process innovations, and the while 16 percent decreased the number, and 54.5 impact on skilled workers (upper part of the tables) percent did not change unskilled employment. The and unskilled workers (middle part of the table). The percentages for process innovation are very similar; cells in yellow denote firms where there has been an with perhaps slightly more firms showing a decrease increase in employees, either because the effects of inthe number of unskilled workers (17.7 percent) as a product and process innovations are positive at least result of process innovation. in one case or because they remained the same in the other case. The orange cells indicate where there has Although it is not possible to know the precise number been no impact on employment, according to the of workers, if we assume relatively similar size in the manager. Finally the cells in red indicate firms where variation in increases or decreases of employment, employment levels have decreased. Firms where there the results suggest three important findings: (i) the are increases in one type of innovation and decreases numbers of skilled labor seems to increase as a result in another are classified as uncertain. The last column of innovation; (ii) the numbers of unskilled labor summarizes the net impact on employment. appear to remain the same for product innovation and perhaps decrease for process innovation; and (iii) there is clear skill-bias arising from innovation in the demand for employment. 26 Catching up to the Technological Frontier? Table 12 Distribution of changes in skilled and unskilled employment for product and process innovators (number of firms) Skilled workers Remained Net impact As a result of product/process innovation Increased Decreased the same (% of firms) Increased 69 7 1 61.59% Remained the same 9 46 1 33.33% Decreased 0 3 2 4.35% Uncertain effect 0.72% Unskilled workers Remained Net impact As a result of product /process innovation Increased Decreased the same (% of firms) Increased 25 3 7 25.00% Remained the same 2 58 3 48.33% Decreased 0 4 18 20.83% Uncertain effect 5.83% Net impact All skilled and unskilled workers (% of firms) Increased 37.68% Remained the same 21.74% Decreased 6.52% Uncertain effect 34.06% Source: Author’s own elaboration from Enterprise survey (2014) The positive impact of innovation on skilled labor is Finally at the bottom of the table we add the impacts significant for highly innovative firms, and those that on skilled and unskilled workers. The ‘uncertain effect’ do product and process innovation, but is small for category is much larger now (34.1 percent), and this is unskilled labor likely due to the fact that firms show a simultaneous increase and decrease in labor, when examining The impact on skilled workers is largely positive, skilled and unskilled labor changes, for both product and only in 4.35 percent of firms has there been an and process innovations. Most firms (37.7 percent) unambiguously negative impact on skilled workers. experience an increase in net employment, while in Although we do not know the size of this employment 21.7 percent of cases, employment levels remained the reduction it is unlikely that the decrease in 4.35 same, and in 6.5 percent of cases, firms experienced percent of firms is larger than the increase in 61.6 an unambiguous negative reduction in employment. percent of firms. Therefore it is plausible to conclude Unless the firms in the uncertain effect category that innovation increases skilled employment for experienced a large employment reduction, it is likely product and process innovators. that the net impact of product and process innovations on employment is positive. Looking at unskilled labor changes, the pattern changes somehow. Employment stays the same In Table A4.7 in the Appendix, we try to estimate for most firms and the percentage of firms where econometrically the impact of innovation on total unskilled labor increased and decreased is similar at full time employment. The estimates, however, are 25 percent vs. 20.6 percent. Assuming similar trends not statistically significant. This is due to several in the increases and decreases would imply a slight reasons. First, it is possible that unaccounted factors, increase in unskilled employment associated with such as large demand shocks, impact the results, as innovation, but this is not comparable to the positive the period of analysis in part falls under a recession. change in skilled labor. Second, the dependent variable is the growth in Understanding Firm-Level Innovation and Productivity in Kenya 27 permanent workers, and it is possible that adjustments total employment, the information reported suggests in employment may occur with other types of labor such that innovation is likely to have a positive impact on as temporary or family workers. Finally, it is also possible employment creation. While there is a clear increase that innovation affects employment levels in the year of in the demand for skilled labor resulting from product their inception, rather than over the whole period. and process innovations, the impact on unskilled labor is more uncertain, although likely to be positive. Overall, likely positive impact of innovation on direct Innovation appears to bias the relative demand for employment, although skilled-bias skilled labor. Also, contrary to some of the findings Overall, and although the data does not permit in the literature, there is little distinction between conclusions regarding the impact of innovations on product and process innovation. 28 Catching up to the Technological Frontier? SECTION SIX Conclusions and Policy Implications Accelerating the process of economic development It is difficult to identify the main obstacle that hinders in Kenya and achieving the ambitious targets laid out a positive linkage to productivity, but the empirical in the Plan 2030 will require a substantial increase in analysis in the paper has some important suggestions. firm-level productivity. Central to the attainment of First, innovation outcomes in Kenya are more common this goal is increasing knowledge capital investments than in other countries with similar GDP per capita, and innovation activity. This paper has provided a but investments in knowledge inputs are similar to snapshot of the degree of firm-level innovation in that of other countries. This suggests a mismatch in Kenya as well as its links to economic performance for relative terms between inputs and outcomes, and the the period 2009-2012 with a view to better measure need for greater investment in knowledge inputs in how Kenyan firms in the manufacturing and services order to make innovation outcomes more innovative sectors can contribute to achieve this objective. and transformational. Second, and related to the first point, the empirical analysis suggests that a lack of Although the absence of panel structure in the dataset access to finance significantly holds back investment in and the small sample do not facilitate the estimation R&D. Third, there is an over reliance by Kenyan firms, of very robust statistical effects, the paper provides at least when comparing them to other countries, on some important findings. The main conclusion of the internal sources for knowledge capital investments and paper is that firm-level innovation activity in Kenya innovation sources, which may indicate the absence of appears to be high and even larger than in similar a solid research and knowledge infrastructure, as well as countries, but the extent of innovativeness is low or a lack of cooperation with other firms and institutions. very incremental. While it is expected that innovations Forth, the inadequate educational levels of the labor in countries far from the technology frontier are not force affects the capacity of firms to transform radical innovations, the question is to what extent these knowledge inputs into innovation outcomes, which incremental innovations contribute to productivity reinforces the complementary role of skilled labor growth as compared to innovations in OECD for innovation, and the need to support appropriate countries. The answer that we find in the empirical technical skillsets in the labor force. analysis is that the innovations do not have a statistically significant impact on productivity. Therefore, the Regarding the impact of innovation on employment, positive causal chain whereby knowledge inputs are the analysis of the qualitative information regarding translated into innovation outcomes, and then into labor changes associated with innovations suggests productivity, breaks down in the case of Kenyan firms. the likelihood that innovation activities have This suggests that in contrast with OECD countries, increased employment levels. However, the results some of the innovations implemented are so minor, differ significantly between skilled and unskilled or are based on imitation, to the extent that they do workers. While there is a clear increase in the demand not have a significant impact on productivity (survival for skilled labor resulting from product and process innovation).Although, similar results have been found innovations, the impact on unskilled labor is likely in other developing countries, more research is needed to be positive, but less so and more uncertain. This to better understand the nature of this incremental suggests that the increases in the demand for labor innovation. associated to innovation are skilled bias. Understanding Firm-Level Innovation and Productivity in Kenya 29 In terms of policy implications, these results o Support should be provided to enhance suggest three levels to which innovation policy cooperation between firms, encourage should be focused. private sector-university linkages and remove coordination failures by providing subsidies to At the firm level it is important to: high quality innovation projects that involve Enhance the capacity to convert innovation several firms and/or firms and academic outcomes into productivity gains. Information institutions. failures and asymmetries where firms lack resources and understanding to gather the required At the sector level it is imperative to: information, resources and know how to innovate, Improve the quality of the physical and human increase the uncertainty to innovate due to potential capital infrastructure for innovation, including failure and also affect the quality of innovations. research labs, as a means of improving the availability This is exacerbated by coordination failures where and quality of innovation services for firms. the individual costs of improvements are very high, Enhance the supply of skilled labor, especially especially for SMEs since the supply of services in areas such as STEMS, which are highly is insufficient and tends to target large firms. This complementary to the introduction of innovations. requires support programs that target productivity and innovation by improving firms’ information, At the institutional level, and given the current capabilities and management skills. Technology institutional vacuum regarding innovation policy, extension services can address these market failures it is critical to finalize and implement the projected and help to realize improved organizational, institutional framework in the Science, Technology managerial, and technological changes. These and Innovation Act of 2013. This would help to services provide information on managerial and better coordinate and design instruments, effectively production practices and how to adopt them, in diagnose and evaluate policies, and incentivize order to increase productivity and competitiveness. dialogue with the private sector. Enhance R&D financing and cooperation among firms and academic institutions This report is one of the first empirical analyses to our o In the presence of financial failures to fund knowledge that has examined innovation activity at innovation, R&D support is likely to be the firm level, and its impact on firm performance required to boost knowledge investments. in Kenya. We hope that the recent release of the The international experience suggests that innovation module of the enterprise survey and gradual partial subsidies to high quality the upcoming release of the second wave of the projects are more effective than indirect second national innovation survey by the Ministry support by tax exemptions. Supporting these of Education Science and Technology will expand high quality projects, in conjunction with research in this area and enable the tracking of firms’ firms and university projects (see below) can innovation behavior over time to better understand have a positive impact on the amount and the contribution of innovation to productivity growth quality of R&D. and shared prosperity. 30 Catching up to the Technological Frontier? References Aghion, Philippe & Howitt, Peter (1992). "A Model of Hall, B.H., Lotti, F., and Mairesse, J., 2009. “Innovation Growth through Creative Destruction," Econometrica, vol. and productivity in SMEs: Empirical evidence for Italy”. 60(2), pages 323-51, March. Small Business Economics, 33(1), pp.13-33. Bell, M., Pavitt, K. (1993). “Technological accumulation Hall, B.H., Mairesse, J., and Mohnen, P. (2010). and industrial growth: Contrasts between developed and “Measuring the returns to R&D”. In: B.H. Hall and developing countries”. Industrial Corporate Change 2, N. Rosenberg, eds. Handbook of the Economics of 157–210. Innovation. Amsterdam: Elsevier, pp.1034-1076. Bloom Nicholas & Raffaella Sadun John Van Reenen Harrison, Rupert Jordi Jaumandreu, Jacques Mairesse, (2012). "The Organization of Firms Across Countries," The and Bettina Peters (2008). “Does Innovation Stimulate Quarterly Journal of Economics, Oxford University Employment? A Firm-Level Analysis Using Comparable Press, vol. 127(4), pages 1663-1705. Micro-Data from Four European Countries NBER”. Cassiman Bruno & Reinhilde Veugelers (2006). "In Working Paper No. 14216 August 2008. Search of Complementarity in Innovation Strategy: Klette, Tor Jakob & Samuel Kortum (2004). "Innovating Internal R& D and External Knowledge Acquisition," Firms and Aggregate Innovation," Journal of Political Management Science, vol. 52(1), pages 68-82 Economy, University of Chicago Press, vol. 112(5), pages Crepon, B., Duguet, E., Mairesse, J. (1998). “Research 986-1018. and development, innovation and productivity: An Kline, S.J., Rosenberg, N. (1986). “An overview of econometric analysis at the firm level”. Economics of innovation”. In: Landau, R., Rosenberg, N. (Eds.), Innovation and New Technology 7, 115– 58. The Positive Sum Strategy: Harnessing Technology Crespi, Gustavo & Zuniga, Pluvia (2012). "Innovation for Economic Growth. National Academy Press, and Productivity: Evidence from Six Latin American Washington D.C., pp. 275–304. Countries," World Development, Elsevier, vol. 40(2), Lentz, Rasmus & Dale T. Mortensen, 2008. "An pages 273-290. Empirical Model of Growth through Product Innovation," Corrado, Carol A. & Charles R. Hulten & Daniel E. Econometrica, vol. 76(6), pages 1317-1373. Sichel (2006). "Intangible Capital and Economic Growth," Lööf, Hans (2005). “A comparative perspective on NBER Working Papers 11948, National Bureau of innovation and productivity in manufacturing and Economic Research. services” in Uwe Cantner, Elias Dinopoulos and De Vries, G.J., M.P. Timmer, and K. de Vries (2013). Robert F.Lanzillotti (eds) Entrepreneurships, the New “Structural Transformation in Africa: Static gains, dynamic Economy and Public Policy Springer Berlin Heidelberg losses.” GGDC research memorandum 136. Miravete Eugenio J. & Jose C. Pernias (2006). Fagerberg, Jan; Martin Srholec and Bart Verspagen "Innovation Complementarity and Scale of Production" (2010). “Innovation and Economic Development” in Journal of Industrial Economics, vol. 54(1), pages 1-29, B.H. Hall and N. Rosenberg, eds. Handbook of the 03.) Economics of Innovation. Amsterdam: Elsevier, pp. Romer, Paul M., "Increasing Returns and Long—Run 833-872. Growth," Journal of Political Economy 94 (October Goedhuys, M. (2007a). “Learning, product innovation, 1986): 102—1037, and firm heterogeneity in developing countries: Evidence Schumpeter, J. (1942). “Capitalism, Socialism, and from Tanzania” Industrial and Corporate Change 16, Democracy”. New York: Harper & Bros. 269–292. Solow, Robert M (1957). "Technical Change and the Goni, Edwin and William F. Maloney (2014). “Why Aggregate Production Function". Review of Economics Don't Poor Countries Do R&D?” Documento CEDE No. and Statistics (The MIT Press) 39 (3): 312–320. 2014-23.Polder et al. (2010). Vivarelli, Marco (2012). "Innovation, Employment and Griffith, R., Huergo, E., Mairesse, J., Peters, B. (2006). Skills in Advanced and Developing Countries: A Survey of “Innovation and productivity across four European the Literature," IZA Discussion Papers 6291, Institute countries”. Oxford Review of Economic Policy 22, for the Study of Labor (IZA). 483–498. Zuniga, Pluvia & Crespi, Gustavo (2013). "Innovation Greene, W. (2012). “Econometric Analysis” 7th Edition strategies and employment in Latin American firms," Prentice Hall, Cloth, 1232 pp. Structural Change and Economic Dynamics, Elsevier, Hall, B.H.(2011). “Innovation and productivity”. Nordic vol. 24(C), pages 1-17. Economic Policy Review, 2, pp.167-204. Understanding Firm-Level Innovation and Productivity in Kenya 31 Appendix 1. Measuring intangible assets The CHS framework for measuring intangible assets CHS (2005, 2009) classify firm spending on intangibles into three main categories: computerized information; innovative (scientific and creative) property; and economic competencies. They include the following sub-categories: 1. Computerized information • Computer software: own use, purchased, and customized software • Computerized databases 2. Scientific and creative property • Research and Development (R&D) in science and technology (spending for the development of new products and production processes, usually leading to a patent or license) • Mineral exploration (spending for the acquisition of new reserves) • Copyright and license costs (spending for the development of entertainment and artistic originals, usually leading to a copyright or license); and development costs in the motion picture, radio and television, sound recording, and book publishing industries • Other product development, design, and research expenses (not necessarily leading to a patent or copyright), such as new product development costs in the financial services industry, new architectural and engineering designs, and R&D in the social sciences and humanities 3. Economic competencies • Brand equity (advertising expenditures and market research for the development of brands and trademarks) • Firm-specific human capital (spending on developing workforce skills; for example, on-the-job training and tuition payments for job-related education) • Organizational capital (costs of improvement in organizational structures). Source: Corrado et al. (2005). See Dutz et al. (2012) for an application for the case of Brazil17 17 Corrado, Carol, Charles Hulten and Daniel Sichel (2005). Measuring Capital and Technology: An Expanded Framework, In Corrado, D., Haltiwanger, J. and Sichel D. (eds.), Measuring Capital in the New Economy, Studies in Income and Wealth. Vol 65, 11-45. Chicago: The University of Chicago Press. Mark A. Dutz, Sérgio Kannebley Jr., Maira Scarpelli and Siddharth Sharma (2012) “Measuring Intangible Assets in an Emerging Market Economy. An Application to Brazil” World Bank Policy Research Working Paper 6142 32 Catching up to the Technological Frontier? Appendix 2. Innovation results by ISIC 2 digits Firm innovation in Kenya by ISIC2 sector (% of firms) Firm innovation activities in Kenya (% of firms) 100 80 75 60 50 40 25 20 0 0 nt & od e tau & l ls s ts hic f tai m e ry sal ve es o od les s me y & e l ran s & ort le hic of ica tai Fo res otels ran cal sal uip ine xti les Re ole mo Se t les ts Fo xti nsp ve es tau tel em uip er Re n tor rvic i ole Te em Eq ach c Eq achin Te Wh H res Ho tor rvi Ch Tra mo Se Wh Ch M M Product innovation Process innovation Product or process innovation Organizational innovation Marketing innovation R&D R&D intramural R&D extramural Training Note: Only sectors with at elast 20 rms in the sample Purchased equipment Purchased licence or patent (a) Innovation outcomes by sector (c) Innovation activities by sector Firm innovation outputs in Kenya by sector (% of firms) Firm knowledge capital investments in Kenya (innovation inputs) by sector (% of sales of innovative enterprises) 30 15 20 10 10 5 0 0 od les ls e l ort hic of me & ran & tai sal ica Fo uip ery tau tels ve es xti mo S t ts nsp Wh s Re n le ole em c e Te rt od me & ran & ls Eq chin i hic tor les l sal res Ho tor erv tai Tra ica po uip ery tau tels ts ces t Ch Fo n les xti ve mo ole Re ns em Eq chin Ma res Ho Te Tra Wh of Ch Ma rvi Patent application Trademark application Utility model application Se Industrial design registration Copyright application Patent, trademark, utility.. application R&D % sales Training % sales Purchased equip % sales Purchased licence % sales (b) Innovation outputs by sector (d) Knowledge capital investments by sector Source: ibid Understanding Firm-Level Innovation and Productivity in Kenya 33 Appendix 3. Methodology The knowledge function The first step of the model is to specify the choice of knowledge capital investment intensity. To this end, we extent the CDM model in two main directions. First, when measuring knowledge intensity, we also include other knowledge capital investments in addition to R&D, such as equipment and training for innovation (see table 4). Second and differently from Griffith et al. (2006), who only have data on R&D activities for innovators, the enterprise survey asks the question of knowledge activities to all firms. As a result, in our dataset zero research intensity is an important outcome of knowledge capital investments that we need to incorporate this to the model. Therefore, rather than using a generalized Tobit model, often implemented in CDM model, we use a generalized Poisson estimator in order to better cater for the number of zeroes in the data. Specifically we estimate the following model (I) Where ki is knowledge intensity for firm i; xi is a vector of determinants of knowledge intensity and is a vector of estimated coefficients. We follow the literature on the determinants of knowledge activities in the Schumpeterian tradition and use as determinants variables to represent market share, diversification and demand conditions, and firm level characteristics such a size and technological opportunities. In order to avoid simultaneity between knowledge activities and market share we use firm’s market share before the introduction of any innovation activities three years ago. For diversification and demand conditions we use whether domestic or external demand is decreasing for the firm and whether the firm is a two-way trader, exporter and importer. Technological opportunities are captured by ISIS 2 digits sector dummies. In addition, we control for firm size, age and whether the firm is foreign owned. Finally, and more importantly, we extend the model and introduce variables that represent the perceptions about the business environment for the firm. Specifically, we use indices reflecting firms’ perceptions on how much of an obstacle lack of finance, trade costs, telecommunications and government policies and regulations are (see table 4 for the definition of the variables used). The innovation function The second step is to determine the innovation equation. One important element in the decision to innovate is that firms decide simultaneously what innovation outcomes to produce based on existing knowledge capital investments. As a result, one should expect some correlation between the decisions to carry out product and process innovations, and perhaps organizational innovations. In order to incorporate these correlations in the empirical estimation, we use a multivariate Probit framework, which allows us to estimate the decision to innovate in the different areas simultaneously and, therefore, correcting for potential correlation in these decisions. Specifically, we estimate m Probit equations for the probability of innovate, where m equals 3 when considering the three types of innovation. (2) Iim=1 if i*im>0 and 0 otherwise (3) 34 Catching up to the Technological Frontier? Where the log-likelihood function can be expressed as: (4) And allowing for correlation in the errors across equations, matrix Ω elements are: (5) The likelihood function depends on the multivariate standard normal distribution.18 As a determinant for the innovation equations we follow the literature and control for size and capital intensity, proxied by the ratio of capital to labor. One important input of the innovation function is the number of technical staff in the establishment that can facilitate the transformation of knowledge inputs into innovation outcomes. Given that the data on skilled labor is uncompleted we proxy skilled labor by an index of how much of an obstacle is inadequately labor force. Finally, given the potential endogeneity of knowledge capital investments in the innovation outcomes, we use the predicted values from the Poisson process in the first stage in order to instrument knowledge activities, for both R&D intensity and total research intensity. The productivity Equation The final stage to estimate the impact of innovation on firm performance is to derive the productivity equation. We approximate productivity using a Cobb-Douglas function where sales (Y) are a function of capital (K), labor (L) and innovation outcomes (H). (6) Transforming equation 6 in logarithm form and adding sector controls (Xi) we have: (7) Equation (7) can be estimated by OLS. However given the potential simultaneity between innovation outcomes and performance, we use the predicted values of the innovation outcomes as instruments and correct the standard errors by removing the mean squared error from the VCE of the second stage (Greene, 2012). Specifically we estimate by OLS the following equation: (8) Robustness The set of equations (1), (4) and (8) above are solved sequentially and instrumented at each stage following Griffith et al. (2006). This, however, assumes no major feedback effects from productivity to innovation and from innovation to knowledge capital investment. As a robustness test, we re-estimate the second and third stages, innovation functions and productivity, simultaneously by maximum likelihood and compare the results with the sequential method. 18 In order to solve the likelihood function we use the mvprobit command (Cappelari and Jenkins, 2003),. Understanding Firm-Level Innovation and Productivity in Kenya 35 Appendix 4. Main Econometric Results A4.1 The determinants of investing in knowledge inputs Table A4.1 shows the results of the first stage, the determinants of the intensity in knowledge capital investments. As shown in the methodology in Appendix 3, we use a Poisson estimator in order to account for the number of firms with zero investments. Columns (1) and (2) show the results for R&D intensity and research (sum of R&D, training and equipment) intensity. Regarding R&D intensity (column 1), two-way traders and firms that face shrinking demand domestically or internationally have larger R&D intensity. Also, firms with greater perceptions of being financially constrained invest less on knowledge inputs. This suggests that international competition and pressure for diversification are important predictors of R&D intensity, while access to finance is likely to be an important inhibitor of these investments. Interestingly, firms with foreign ownership tend to invest less on R&D, while in line with some of the literature size does not appear to matter for these investments. Finally, neither market power position nor none of the obstacles explored; telecommunications, trade or government obstacles appear to have an impact on knowledge capital investments intensity. This emphasizes the role of external demand factors in explaining R&D intensity. Looking in column (2) to a broader measure of knowledge capital investments, including equipment and training, yields very similar results to the ones on R&D, but with only two differences. First, lack of finance does not have a significant role explaining knowledge intensity investments when broadening the definition, so it appears less binding when including machinery, equipment and training. Second, foreign owned firms appear to invest more in knowledge inputs when considering machinery, equipment and training, which is likely the result of the fact that they carry out R&D abroad or in related firms. Columns (3) and (4) estimate the same specifications for robustness as (1) and (2), but this time looking at the decision (not the intensity) to invest. Thus, we use a Probit model to estimate the probability that firms incur any R&D (3) or R&D, machinery, equipment and training (4) investments. In the case of R&D, the only variables that appear to explain the decision to invest are having a decreasing domestic or international demand, larger market share at the beginning of the period in 2008(t-3), the perception of greater incidence of government obstacles and the perception of how important is access to finance as an obstacle. Firms that claim to be more finance constrained are less likely to engage in R&D, while firms that have larger sector market shares and that have more incentive to diversify due to contracting demand are more likely to implement R&D. The coefficient on government obstacles is, however, puzzling. Firms which perceive government regulations to be more of an obstacle are also more likely to engage in R&D, which could be explained by the potential endogeneity of these perceptions to firm performance. Finally, the results for the expanded decision to invest in knowledge activities in general (column (4)) show large heterogeneity when trying to explain the investment decision, since our model only captures that larger firms and two-way traders are more likely to engage in these investments. 36 Catching up to the Technological Frontier? Table A4.1 Knowledge Intensity function (1) (2) (3) (4) Research per c R&D per workera R&Db Researchb workera Two-way traders 2.2068*** 1.6164*** -0.3313 0.6213* (0.5308) (0.6021) (0.2957) (0.3506) Demand (-) 1.1358** 0.6906** 0.4990** 0.3860 (0.4602) (0.2687) (0.2168) (0.2512) Share(t-3) 0.0816 0.1294 0.1121** -0.0300 (0.1789) (0.0822) (0.0509) (0.0509) Lack finance -1.0077*** -0.2799 -0.2169*** -0.0800 (0.2323) (0.2384) (0.0839) (0.1020) Telecom_obstacle 0.1638 -0.1330 -0.0596 0.0611 (0.1788) (0.1375) (0.0738) (0.0770) Government_obstacle 0.0519 -0.1024 0.2279*** 0.1143 (0.2125) (0.1670) (0.0833) (0.0909) Trade cost-obstacle 0.0741 0.2695 0.0706 -0.0382 (0.1894) (0.1703) (0.0861) (0.0928) Foreign -2.6251** 1.4251** 0.0153 0.2597 (1.0939) (0.6933) (0.3457) (0.4803) Age 0.0223* 0.0056 0.0041 -0.0004 (0.0124) (0.0073) (0.0058) (0.0062) Medium -0.6479 0.3018 -0.1275 0.4967** (0.6092) (0.4088) (0.2450) (0.2505) Large 0.9423* 0.8550 -0.0415 1.0750*** (0.5720) (0.5436) (0.3259) (0.3833) Constant 9.5586*** 12.0140*** 0.4488 -0.4120 (1.0723) (0.8891) (0.5277) (0.5885) Observations 388 347 427 339 ISIC-2 digits dummies YES YES YES YES Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 a Poisson non-linear model estimator on the level of expenditure b Probit model on the decision to invest on R&D or research and equipment c Research and equipment defined as expenditure for knowledge activities including R&D, both intramural and extramural, training and equipment for innovation activities Understanding Firm-Level Innovation and Productivity in Kenya 37 A4.2 The innovation function In order to measure the impact of knowledge capital investments on innovation, we estimate the probability of introducing an innovation, equation (4) above. Table A4.2 shows different estimates of equation (4). The first six columns estimate individual Probit equations for product, process and organization innovations and using observed R&D and research and equipment intensity. Columns (7) to (10) show the bivariate Probit estimates allowing for correlation between product and process innovation decisions, and instrumenting knowledge intensity with the predicted values of the previous stage. Finally, columns (11) to (16) implement the multivariate framework to also include organization innovation. In general the individual equations without instruments show very little predicted value of the model. Knowledge capital investments do not appear to impact the probability of introducing innovations; we only find a negative sign associated to R&D and organizational innovation. However, these estimates are likely to be biased given the potential endogeneity between knowledge capital investments and innovation outcomes. In order to correct for this endogeneity we instrument knowledge intensity with the predicted values estimated in the previous stage. In addition, we allow for simultaneity in the decision to innovate for the different types of innovation allowing for correlation of the error terms. Estimates of equations (7) to (10) that consider only product and process innovation show a negative and statistically significant correlation between the two equations, which suggest simultaneity between product and process innovation decisions and the need to control for this correlation. The estimates also indicate that firms that are more capital intensive are more likely to introduce product or process innovations. Larger firms are more likely to introduce product innovations, while firms with a larger perceived obstacle in education of the labor force are less likely to introduce process innovations. The most surprising result is fact that similarly to the non-instrumented results, investment in knowledge activities, both R&D and broad knowledge inputs, is not statistically significant in increasing the probability of product and process innovation. Columns (11) to (16) replicate the estimations but add organizational innovation to the simultaneous equations. Adding organizational innovation implies a loss of observations since the organizational module was only implemented to firms considered medium and large in the sampling frame.19 The decision to introduce product and process innovations are still correlated, but the correlations with the residuals of the organizational innovation equation are not statistically significant, which suggests that organizational innovation decisions are done independently from product and process innovations. The results again suggest the importance of large firms and capital intensity, although only for product innovation, and education of the labor force as an obstacle for both product and process innovation. Medium sized firms appear to be more likely to implement organizational innovations. Again, we do not find evidence that investments in knowledge capital are statistically significant in affecting the probability of innovation; with the exception of a marginally significant coefficient of R&D on organization innovation. Overall, the results of the innovation equation suggest that in the case of Kenyan firms’ investments in knowledge and acquisition of capabilities in the form of R&D, equipment and training do not necessarily translate into firm level innovations. This result is not surprising for R&D, since it is likely that the type of small incremental innovations do not require significant acquisition of capabilities via R&D; but is more surprising for total knowledge capital investments, since even imitations tend to require some degree of acquisition of machinery and training of workers. 19 In the sample, however, there are small firms that were asked about organizational changes due to the fact that some firms had de facto lower employment levels than previously thought in the sampling frame. 38 Catching up to the Technological Frontier? 39 Table A4.2 Innovation function Individual regressions Bivariate probit Multivariate probit (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) product- process- organ- product- process- organ- product- process- product- process- product- process- organ- product- process- organ- R&D R&D R&D Research Research Research R&D R&D Research Research R&D R&D R&D Research Research Research Log(K/L) 0.1437* 0.1236 0.1786 0.1383 0.1121 0.1738 0.1351** 0.1980*** 0.1321** 0.1975*** 0.1358 0.1807** 0.0816 0.1324 0.1814** 0.0672 (0.0723) (0.0718) (0.1024) (0.0808) (0.0792) (0.1160) (0.0657) (0.0699) (0.0661) (0.0699) (0.0880) (0.0842) (0.0992) (0.0882) (0.0837) (0.0976) Educ_ -0.0880 -0.1749 -0.1228 -0.0685 -0.2012* -0.2062 -0.1334 -0.1997** -0.1369 -0.2022** -0.2386** -0.2609* -0.1649 -0.2525** -0.2699* -0.1654 obstacle (0.0940) (0.0908) (0.1385) (0.1057) (0.0962) (0.1391) (0.0944) (0.0919) (0.0947) (0.0919) (0.1214) (0.1413) (0.1436) (0.1214) (0.1413) (0.1437) Medium 0.3712 0.2479 0.7755* 0.2472 0.1163 1.1492** 0.3928 0.3602 0.3699 0.3462 0.6315 0.2404 0.6784* 0.6171 0.2332 0.6831* (0.2752) (0.2750) (0.3737) (0.3012) (0.2986) (0.4084) (0.2675) (0.2810) (0.2686) (0.2824) (0.3877) (0.3656) (0.4034) (0.3887) (0.3662) (0.4043) Large 0.3803 0.3601 0.5880 0.3044 0.1240 0.7848 0.8012*** 0.5023 0.6935** 0.4402 0.9273** 0.4695 0.3097 0.8086* 0.4126 0.1556 (0.2835) (0.3342) (0.3921) (0.3283) (0.3987) (0.4310) (0.3102) (0.3119) (0.3400) (0.3324) (0.4027) (0.3633) (0.4304) (0.4234) (0.3760) (0.4510) R&D 26.1994 15.4635 -0.7603*** (15.5818) (9.8886) (0.2255) Research 1.0240 1.5527 -0.1044 0.7096 0.7190 1.8906 0.7019* -0.9608 R&D_hat (0.6728) (0.5816) (3.8043) (0.3691) (0.7062) 0.4642 0.2728 0.6017 0.2797 0.2271 Research_hat (0.4044) (0.2600) (0.4537) (0.2543) (0.3050) Understanding Firm-Level Innovation and Productivity in Kenya -2.7184** -2.3407* -3.0864* -2.5650* -2.1681 -3.3033* -2.3115** -3.1834*** -2.2894** -3.1851*** -2.5337** -3.0001** -1.5791 -2.5149** -3.0222** -1.4542 Constant (1.0414) (1.0311) (1.4799) (1.1521) (1.1262) (1.6471) (0.9610) (1.0255) (0.9662) (1.0269) (1.2237) (1.2226) (1.4661) (1.2326) (1.2172) (1.4447) 0.6934*** 0.6933*** 0.6069*** 0.6030*** Rho21 (0.1720) (0.1724) (0.1880) (0.1891) 0.2297 0.2146 Rho31 (0.1829) (0.1838) -0.1977 -0.2124 Rho32 (0.1827) (0.1826) 241 243 154 207 204 130 254 254 171 171 Observations -769.1 -768.4 -455.4 -704.4 -681.8 -373.5 Log-like YES YES YES YES YES YES YES YES YES YES YES YES YES YES YES YES ISIC-2digits dummies Robust standard errors in parentheses *** p<0.001, ** p<0.01, * p<0.05 A4.3 Innovation and productivity The final stage to determine the impact of innovation on performance is to estimate equation (8). As productivity measure, we use two proxies of labor productivity: the logarithm of sales per worker and the logarithm of value added per worker. Although value added per worker is a better measure of labor productivity, the existence of missing observations for material inputs in some firms reduces significantly the sample when using this variable. In order to control for potential endogeneity of innovation decisions to firm performance when trying to estimate causality in equation (8), we instrument the different type of innovations using the predicted values of the multivariate framework described in the previous section and correct the standard errors of the regression as proposed in Greene (2012). Table A4.3 shows the results of the OLS estimates without instrumenting for comparison and Table A4.4 shows the instrumental variables estimates using the logarithm of sales per workers. Columns (1) to (4) in Table A4.4 show the results when using R&D intensity as knowledge input to predict innovation, while columns (4) to (8) uses the instruments of innovation estimated using research, training and equipment for innovation intensity to explain innovation. The main result that emerges from the table is that innovation is not statistically significant in increasing productivity for our sample of Kenyan firms, with or without instrumenting. Only column (6) in table A4.3 show statistically significant signs for a positive impact innovation on value added per worker. However, these results are likely to be biased given the endogeneity problems discussed above. When we introduce instruments and correct the standard errors in Tables A4.4 and Table A4.5, these increase significantly and only capital intensity appear to be statistically significant in the sales per worker specifications using R&D to predict innovation. Also, we introduce interactive innovation dummies to capture complementarities between different types of innovation. However, the coefficients are not statistically significant. As proposed in the previous section, as a robustness test we estimate stages 2 and stage 3 simultaneously as a system of equations by maximum likelihood and allowing for correlations in the innovation decisions. Table A4.6 shows the results for sales per worker. The results confirm the lack of statistically significant impact of innovation on productivity. One difference, however, is that when estimating the innovation equation and the productivity equation jointly, the knowledge intensity coefficients are statistically significant in explaining innovation outcomes. The estimates using research, equipment and training investment intensity also show a positive and statistically significant impact of knowledge capital investments on innovation outcomes, but no impact of innovation on productivity. One needs to interpret these results with caution, given the low number of observations and the lack of panel structure of the dataset. However, overall, the main result of the estimates is that firm-level productivity appears to be largely unexplained and there is no statistically significant impact of innovation on productivity. 40 Catching up to the Technological Frontier? Table A4.3 Productivity equation – no instruments (1) (2) (3) (4) (5) (6) (7) (8) sales per sales per sales per sales per sales per sales per sales per sales per worker- worker- worker- worker- worker- worker- worker- worker- R&D R&D R&D R&D Research Research Research Research Log(K/L) 0.3290*** 0.3299*** 0.4081*** 0.4147*** 0.1539 0.1472 0.2522** 0.2423** (0.1124) (0.1133) (0.1535) (0.1509) (0.1085) (0.1014) (0.1104) (0.1107) Log(L) 0.2131* 0.2103* 0.3482*** 0.3809*** 0.0631 -0.0036 0.3379** 0.2290 (0.1130) (0.1132) (0.1256) (0.1394) (0.1437) (0.1506) (0.1300) (0.1390) Prod inno 0.2729 0.3556 0.1784 -0.2102 0.3462 1.4266** 0.0362 0.8901* (0.2802) (0.4262) (0.3766) (0.5958) (0.4203) (0.6319) (0.3755) (0.4726) Process_inno 0.2484 0.3284 0.0849 -0.4315 0.2331 1.2825*** 0.4177 0.9011 (0.2875) (0.4270) (0.3717) (0.8309) (0.4412) (0.4416) (0.3810) (0.5822) Organ inno 0.5323 0.3323 0.6470* 0.0092 (0.3465) (0.5003) (0.3620) (0.5011) Prod*process -0.1544 0.6262 -1.9904*** -1.5143** (0.6023) (1.0476) (0.7423) (0.7623) Prod*organ -0.1423 0.7510 (0.9476) (0.7756) Process*org -0.0408 1.9237 (1.3106) (1.2475) Prod*proc*org 0.7792 -2.0019 (1.7080) (1.4699) Constant 9.2779*** 9.2938*** 7.6303*** 7.4771*** 11.4409*** 11.9913*** 8.6913*** 9.6091*** (1.6373) (1.6468) (2.0813) (2.0726) (1.4891) (1.4357) (1.6690) (1.6858) Observations 270 270 182 182 178 178 126 126 R-squared 0.2437 0.2441 0.3527 0.3586 0.1589 0.2238 0.4460 0.4829 ISIC-2 digits YES YES YES YES YES YES YES YES dummies Corrected standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Understanding Firm-Level Innovation and Productivity in Kenya 41 Table A4.4 Productivity equation – instrumented (sales per worker) (1) (2) (3) (4) (5) (6) (7) (8) Sales per Sales per Sales per Sales per Sales per Sales per Sales per Sales per worker- worker- worker- worker- worker- worker- worker- worker- R&D R&D R&D R&D Research Research Research Research Log (K/L) 0.4328** 0.4196* 0.4349*** 0.3794 0.4299** 0.4267 0.4292** 0.3902 (0.1852) (0.2426) (0.1559) (0.2958) (0.2156) (0.2726) (0.1915) (0.3579) Log (L) 0.2120 0.1991 0.1902 -0.0219 0.1291 0.0921 0.0843 -0.1115 (0.2586) (0.3608) (0.2301) (0.5047) (0.2971) (0.3905) (0.2949) (0.6158) Prod inno 2.2484 5.6757 2.8650 14.4861 3.5300 7.0117 4.7432 16.4024 (3.3423) (6.1733) (3.1705) (13.7158) (3.6970) (7.1267) (4.0873) (15.4963) Process_inno -2.8134 -1.1129 -3.0536 1.9590 -3.6382 -2.6428 -4.0233 2.1472 (3.3593) (4.9941) (2.9182) (8.7362) (3.8614) (5.2075) (3.4685) (10.0153) Organ inno -0.5815 7.4873 -4.9673 -18.1962 (1.5020) (16.8481) (6.5904) (16.6713) Prod*process -5.6822 -14.5134 -1.1916 8.8638 (5.8074) (13.6501) (2.4414) (19.9464) Prod*organ -19.5423 -22.5378 (29.4896) (33.4257) Process*org -20.1862 -24.1821 (28.3072) (32.3929) Prod*proc*org 35.0909 44.9991 (41.1207) (47.3746) Constant 7.7424*** 7.0384* 7.8599*** 6.0715 7.8743** 7.2271 8.1619*** 5.7949 (2.7880) (3.8907) (2.4044) (5.4318) (3.2472) (4.3824) (2.9935) (6.6872) Observations 255 255 255 255 255 255 255 255 R-squared 0.2698 0.2946 0.2704 0.3314 0.2742 0.2928 0.2760 0.3464 ISIC-2 digits YES YES YES YES YES YES YES YES dummies Corrected standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 42 Catching up to the Technological Frontier? Table A4.5 Productivity equation – instrumented (value added per worker) (1) (2) (3) (4) (5) (6) (7) (8) va per va per va per va per va per va per va per va per worker- worker- worker- worker- worker- worker- worker- worker- R&D R&D R&D R&D Research Research Research Research Log (K/L) 0.1853 0.1881 0.1809 0.1719 0.1863 0.1844 0.1623 0.1611 (0.2422) (0.2284) (0.2526) (0.2296) (0.2775) (0.2610) (0.3165) (0.2815) Log (L) -0.1219 -0.1445 -0.2428 -0.2043 -0.1921 -0.1762 -0.3979 -0.3512 (0.4389) (0.4312) (0.5192) (0.5073) (0.4939) (0.4939) (0.7267) (0.7106) Prod inno 4.0846 1.9006 6.7359 3.7043 5.1212 2.3905 9.4091 7.9512 (6.1594) (6.5831) (7.7282) (11.3843) (6.4799) (7.8157) (10.0681) (13.6423) Process_inno -3.8705 -5.6990 -5.0826 -3.7461 -4.7933 -5.8482 -6.2040 -4.8717 (6.4531) (7.0526) (7.3639) (8.8384) (7.1312) (7.2267) (8.8627) (10.6555) Organ inno -2.1091 -2.0674 4.4144 -0.5167 (2.7269) (13.7208) (5.8644) (12.7700) Prod*process 4.5975 1.0111 -3.5725 -1.3169 (5.0404) (9.8612) (5.2847) (16.2605) Prod*organ 3.1682 -2.3093 (22.1971) (24.3128) Process*org -5.5222 -5.6661 (23.8885) (27.4785) Prod*proc*org 5.6589 8.7958 (31.6299) (36.5690) Constant 11.5887*** 12.3021*** 12.2890*** 12.6803*** 11.7448*** 12.3742*** 13.0301*** 12.7338** (3.2668) (3.0843) (3.4363) (4.0388) (3.7672) (3.6001) (4.4971) (5.1224) Observations 172 172 172 172 172 172 172 172 R-squared 0.1502 0.1660 0.1627 0.1759 0.1606 0.1744 0.1824 0.1900 ISIC-2 digits dummies YES YES YES YES YES YES YES YES Robust standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1 Understanding Firm-Level Innovation and Productivity in Kenya 43 Table A4.6 Structural modelling - innovation and productivity (1) (2) (3) (4) (5) (6) (7) (8) process_ organ_ process_ organ_ lsales_l prod_inno lsales_l prod_inno inno inno inno inno Log (K/L) 0.2195** 0.0229 0.0188 0.0328 0.2179** 0.0229 0.0188 0.0328 (0.1016) (0.0158) (0.0168) (0.0200) (0.0989) (0.0158) (0.0168) (0.0200) Log (L) 0.0181 -0.0050 (0.1059) (0.1112) Prod inno 0.3586 -0.2209 (0.3003) (0.4726) Process_inno 0.2845 -0.0029 (0.2811) (0.4821) Organ inno 0.8691 (0.7135) Prod*process 0.1809 0.1653 (0.2895) (0.4793) Prod*organ 0.4386 (0.7345) Process*org -0.6918 (1.1689) Prod*proc*org 0.3186 (1.4281) R&D_hat 0.1231** 0.1696*** -0.1626*** 0.1231** 0.1696*** -0.1626*** (0.0510) (0.0484) (0.0548) (0.0510) (0.0484) (0.0548) Education-obstacle -0.0088 -0.0148 -0.0476 -0.0088 -0.0148 -0.0476 (0.0233) (0.0240) (0.0297) (0.0233) (0.0240) (0.0297) Medium 0.1974*** 0.1248* 0.1852* 0.1974*** 0.1248* 0.1852* (0.0733) (0.0721) (0.0957) (0.0733) (0.0721) (0.0957) Large 0.2977*** 0.2183*** 0.1809* 0.2977*** 0.2183*** 0.1809* (0.0848) (0.0845) (0.1025) (0.0848) (0.0845) (0.1025) Constant 10.6130*** -0.0688 -0.0276 -0.2152 10.8843*** -0.0688 -0.0276 -0.2152 (1.5121) (0.2270) (0.2430) (0.2913) (1.4613) (0.2270) (0.2430) (0.2913) Rho21 0.0796*** 0.0796*** (0.0128) (0.0128) Rho31 0.0158 0.0158 (0.0153) (0.0153) Rho32 -0.0139 -0.0139 (0.0151) (0.0151) Observations 255 255 255 255 255 255 255 255 ISIC-2 digits dummies YES YES YES YES YES YES YES YES Robust standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1 44 Catching up to the Technological Frontier? A4.4 Innovation and employment As a further check for the impact of innovation on employment, we use the retrospective information in the survey on full-time employment and sales and estimate a model of employment growth. We follow Harrison et al. (2008) and decompose sales growth into the share linked to new products and the share linked to old products using the share of sales attributed to product innovation. We assume that this share accounts for the growth of the entire three years period and estimate equation (9) below, where the change in permanent employment is determined by: the growth in sales of old products and product innovation (new), process innovations; and a set of controls X that include region to control for labor market conditions, whether the firm is a two-way trader, the size of the firm, the age and sector dummies. (9) Table A4.7 shows the OLS estimates of equation (9). We start estimating in column (1) the specification considering only sales growth and province and sector dummies. Surprisingly, we find that the coefficient on sales growth is not statistically significant affecting employment growth in the period. In columns (2) we add a specification with more controls: including firm characteristics such as ownership, trading status, age and size, in order to better capture the impact of sales growth. However, the coefficient is still not statically significant. Equations (3) and (4) apply the decomposition of sales growth between old and new products and also introduce process innovation, but the coefficients associated to innovations remain statistically not significant.   Understanding Firm-Level Innovation and Productivity in Kenya 45 Table A4.7 Employment and Innovation (1) (2) (3) (4) Sales growth 0.0018 0.0015 (0.0013) (0.0012) Sales growth old products 0.0015 0.0009 (0.0020) (0.0019) Sales growth new products 0.0013 0.0011 (0.0027) (0.0024) Process_inno2 0.0504 0.0419 (0.0453) (0.0474) Age -0.0024** -0.0035*** (0.0010) (0.0013) Two_way traders 0.0356 0.0163 (0.0679) (0.0778) Foreign 0.1326 0.1022 (0.0834) (0.0946) Nyanza 0.3046*** 0.2998*** 0.2677*** 0.2561*** (0.1008) (0.1061) (0.0940) (0.0974) Mombasa 0.1393*** 0.1310** 0.1310** 0.1194* (0.0522) (0.0576) (0.0571) (0.0608) Nairobi 0.1766*** 0.1649*** 0.1693*** 0.1576** (0.0525) (0.0590) (0.0603) (0.0655) Nakuru 0.1166 0.1215 0.1912 0.1976 (0.1123) (0.1161) (0.1561) (0.1586) Medium -0.0172 -0.0290 (0.0532) (0.0604) Large 0.0064 0.0446 (0.0574) (0.0661) Constant 0.0002 0.0547 -0.0117 0.0747 (0.0395) (0.0512) (0.0441) (0.0539) Observations 568 554 440 429 R-squared 0.0856 0.0981 0.0984 0.1164 ISIC-2 dummies YES YES YES YES Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1. Dependent variable change in permanent employment in the last 3 years 46 Catching up to the Technological Frontier? Design: Robert Waiharo World Bank Group Delta Center P O Box 30577-00100 Nairobi, Kenya www.worldbank.org