CHAPTER 1. INTRODUCTION
1.3. THE METHODOLOGICAL ISSUES
Methodologies applied to study firm-level learning vary in terms of the theoretical basis and the type of data used. Besides, there is quantitative and qualitative divide in the applied methodology with their respective pros and cons. The quantitative analysis is more common in the classical literatures where learning is treated as a blackbox. On the other hand, the qualitative approach is well known in analyzing systems of learning and innovation. The main concern of this section is to discuss how learning was treated and the data used.
1.3.1. MEASUREMENT OF LEARNING
In literatures, learning and innovation are closely related and improvement in innovation performance is the most common indicator of learning. However, the nature of the technological effort in most LDCs is quite different, and is based mainly on firm-level activities, which are not included in formal measures of innovation (Pietrobelli and Rabellotti, 2011). While the concept of learning is broad and generally common to any country, it is important to note the type of innovation referred to in LDCs’ contexts in which diffusion plays a major role. The type of innovation in LDCs may include minor modifications in products, processes, and organizational routines. It is, therefore difficult to differentiate between innovation and diffusion in LDCs where firms engage in continuous modification and adaptation of foreign technologies to their own contexts (Fagerberg, Srholec and Verspagen 2010). Consequently, it is not easy to find more explicit measures of learning and innovation in LDCs than in developed countries.
Likewise, the data we used in this dissertation does not have any measure of innovation that would help measure learning more explicitly. Therefore, we answered the key empirical question of whether there is learning in manufacturing firms in terms of other measures of performance. In other words, similar to the traditional approaches, learning is indirectly measured in terms of the impacts of important firm characteristics on different firm-level performances. We used survival, growth, and productivity of individual firms to measure performance. Productivity is measured in terms of both labor and total factor productivity (TFP). Higher survival, growth, and productivity of firms are assumed to be the outcome of improved learning capabilities. Any factor that would improve performance will be interpreted as a source of learning. Unlike the traditional learning-curve approach, however, the main concern, here, is on what determines heterogeneities in firm performances. According to the evolutionary perspective and models presented in Chapter 3, these heterogeneities are the manifestation of the underlying differences in the firms' learning capabilities. A clear examination of such firm-specific effects requires appropriate data and method of analysis.
1.3.2. THE DATA USED
Both micro and macro data were used in this dissertation. Micro or firm-level data were mainly used for the empirical chapters, while macro data were utilized for the chapter on the country background aimed at supporting the research questions and shedding light on national-level technological capabilities. The macro data were also used to deflate the nominal values of some variables used in firm-level analyses. These data were obtained from different sources, including national and international institutions, were utilized to achieve the objectives of the chapter. Specifically, we
used data from national institutions such as National Bank of Ethiopia (NBE), Ministry of Labor and Social Affairs (MoLSA), Central Statistical Agency (CSA), Ministry of Finance and Economic Development (MoFED), and international institutions such as International Monetary Fund (IMF), World Bank, United Nations, and African Development Bank (AfDB). Different related previous works and secondary materials were also widely consulted. The data were analyzed both qualitatively and quantitatively using descriptive statistics and graphs.
The microdata we have used for the empirical chapters of this dissertation are the most comprehensive firm-level data available for Ethiopia, which has been collected by the country's CSA through an annual survey of large and medium scale manufacturing (LMM) establishments. The survey is confined to those establishments that engage 10 persons and above and use power-driven machinery, and covers both public and private industries in all regions of the country. The survey has been the principal source of facts about the structure and function of the manufacturing industry in Ethiopia. The data have been gathered using structured questionnaire and face-to-face interviews to obtain detailed information on establishments’ year of commencing operation; major industrial groups; ownership, number of persons engaged, and employees; wages and salaries; sex; paid-up capital; gross value of production; industrial and non-industrial costs; value added; operating surplus; quantity of production; raw materials consumed; fixed assets; market of final products; investment; production capacity; and other business related aspects. CSA applied different quality assurance strategies throughout the data collection process and engaged skilled professionals at different levels.
These data have been used for different operational and policy purposes by national and international institutions. For example, Ethiopia’s Ministry of Finance and Economic Development uses the data in national income accounting and other important performance indicators such as employment and production capacities. In addition to their own data, the IMF, United Nations organizations (such as UNCTAD), and the World Bank also used the data for evaluating Ethiopia’s progress in manufacturing and structural change and in monitoring the impacts of related development aids. Several researchers have also used the data to study various aspects of firm performance in Ethiopia. The studies include Söderbom (2011); Siba and Söderbom (2011); Bigsten, Gebreeyesus, Siba, and Söderbom (2011); Seyoum, Wu, and Yang (2015); Bigsten and Gebreeyesus (2009); Bigsten, Gebreeyesus, and Söderbom (2013); Shiferaw (2007 and 2008); and Gebreeyesus (2008).
Even though microdata are available starting from 1976 until recent years, we used the data spanning 2000–2011 to minimize the impact of variability in the data in terms of structural effects and variables. The main limitation of the data is that there are no variables that directly measure innovation, R&D, and the quality of human capital of establishments. These variables would have enabled us to measure firm-
level learning more explicitly than the way it is done in this dissertation. The data were analyzed mainly using a dynamic panel data econometric called System GMM (Generalized Method of Moments) developed by Blundell and Bond (1998, 2000) to take the potential problems arising from weak exogeneity of the independent variables into account. In addition to this, we applied matching techniques and panel probit regression where necessary. Summaries of the analyses and contributions of the chapters are presented in the following sections respectively.