Policy Research Working Paper 8992 Manufacturing in Structural Change in Africa Pierre Nguimkeu Albert G. Zeufack Africa Region Office of the Chief Economist August 2019 Policy Research Working Paper 8992 Abstract This paper investigates the scale, causes, and timing of signif- countries have begun to deindustrialize. Moreover, the icant episodes of industrialization and deindustrialization study documents meaningful heterogeneity across Sub-Sa- in Sub-Saharan Africa. Recent studies have argued that the haran Africa subregions, with the Southern region being turning point of manufacturing output and employment the only subregion to have witnessed deindustrialization. shares tends to occur prematurely in this region. The anal- However, this deindustrialization of the Southern subregion ysis is performed using panel data methods for fractional does not appear to be occurring prematurely. The study responses and data from a variety of sources for a panel of also explores the potential role of the Dutch disease and 41 African countries. The results overwhelmingly do not resource curse hypotheses in understanding Sub-Saharan support the common finding that Sub-Saharan African Africa’s manufacturing experience in resource rich countries. This paper is a product of the Office of the Chief Economist, Africa Region, commissioned as part of the ‘Industrialization for Jobs in Africa’ regional study. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at nnguimkeu@gsu.edu; azeufack@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Manufacturing in Structural Change in Africa∗ Pierre Nguimkeu Albert G. Zeufack Georgia State University The World Bank Group ∗ We thank the World Bank’s Office of the Chief Economist for Africa (AFRCE) for funding this research. We thank Augustine Denteh for excellent research assistance. We thank partici- pant of the World Bank’s Midterm Workshop on “Industrializing for Jobs in Africa” for useful comments.The content is solely the responsibility of the authors. Corresponding author - Pierre Nguimkeu: Department of Economics, Andrew Young School of Policy Studies, Georgia State University, P.O. Box 3992, Atlanta, GA 30302-3992, e-mail: nnguimkeu@gsu.edu. Albert Zeu- fack: Office of the Chief Economist, Africa Region (AFRCE), The World Bank Group, e-mail: azeufack@worldbank.org. 1 Introduction It is commonly agreed among economists that industrialization is the cornerstone of economic development (Kaldor 1966, 1967). In advanced economies, industri- alization enabled sustained productivity growth as well as economic convergence among both the western and European countries such as the United States and Great Britain as well as non-western countries like Japan, China, South Korea and Taiwan (Rodrik 2009). Defined as the share of manufacturing in economy-wide output or employment, the level of industrialization is known to typically follow an inverted U-shaped path over the course of development, where the turning point corresponds to the threshold at which manufacturing exhausts its function as the main driver of economic growth (Rodrik 2012). Recent studies have documented that the turning point of manufacturing output and employment shares tends to occur “prematurely” in certain developing countries. In this vein, deindustrial- ization is said to have occurred prematurely if manufacturing shrinks at levels of income that are much lower than those at which the advanced economies started to deindustrialize. While Rodrik (2016) documents premature industrialization in sub-Saharan Africa (SSA) and Latin America, he concludes that high-income coun- tries have not suffered the same fate. However, given that Rodrik (2016)’s sample includes a limited number of sub-Saharan African countries, it is quite possible that these conclusions do not apply to many or even most countries of the region, or that they are missing some of the specific characteristics of individual countries. In this paper, we study cross-country patterns and trends in the share of man- ufacturing in national output and employment among sub-Saharan African coun- tries. We investigate the extent to which countries differ in terms of the timing, causes, and consequences of industrialization or “deindustrialization.” Indeed, the fact that development policies are always framed, implemented and evaluated at the subregional and/or country levels suggests that the questions raised in this paper should also be addressed at finer geographic (aggregate) levels. Moreover, since countries are heterogeneous in terms of regional trade clusters, availability of skills, and natural resource endowment, it is important to exploit such richness and nuances to draw conclusions accordingly. More specifically, the analysis in this paper is organized around the following questions. (i) What are the main cross country patterns and trends in the share of manufacturing in national out- put and employment? Do these patterns suggest a process of industrialization or de-industrialization in specific countries, in specific sub-regions, or at the re- gional level? (ii) What are the main determinants of the observed outcomes of de-industrialization in these locations, and what are their quantitative effects? (ii) Are there other channels that might have led to the observed (de)industrialization patterns? This means, for example, looking at whether there are evidence on the 1 implication of the Dutch disease and resource-curse hypotheses to the outcome of industrialization in resource rich countries. Manufacturing has been touted as a key driver for economic development, espe- cially for developing countries (Kuznets 1953, Matsuyama 1992). Jones & Olken (2008) examines the allocation of resources underlying economic growth, docu- menting that movements into and out of the labor force in manufacturing tend to coincide with accelerations and decelerations in the growth rate. In examining the constraints for sustained growth in Africa based on a comparison to benchmark countries (e.g., Indonesia and Thailand), Johnson, Ostry & Subramanian (2007) finds that, in almost all cases, these benchmark countries have escaped poverty through manufacturing exports, and posits that the manufacturing sector may be crucial for sustained growth in Africa. A small but growing literature documents the manufacturing experience of developed and developing countries with partic- ular emphasis on the evolution of manufacturing output and employment shares. Dasgupta & Singh (2007) examines data for a set of 14 developing countries for the period 1986-2000, and finds evidence of deindustrialization, that is, manufac- turing tends to monotonically increase with per capita income until it peaks and begins to decline.1 They also note that such deindustrialization appears to occur at lower levels of income than developed countries. Similar findings have emerged for Indian states, where Amirapu & Subramanian (2015) document premature deindustrialization and reports that by 2010, manufacturing employment shares peaked at about 45 percent of the corresponding 1988 turning point value. To our knowledge, the only study examining deindustrialization in SSA is Rodrik (2016), which uses a larger set of countries (albeit only 11 African countries are included) and comes to a similar premature deindustrialization for Sub-Saharan Africa and Latin America. Thus, there is a dearth of literature focused on sub-Saharan Africa’s manu- facturing experience, and those that do have utilized a limited number of SSA countries due to data inadmissibility. This paper attempts to fill this void by as- sembling data on 41 sub-Saharan African countries from multiple sources including the Groningen Growth and Development Center 10-sector Database, the Maddi- son Project Database, the World Development Indicators, and the International Labor Organization’s ILOSTAT database. Our data spans 1960-2016, which is the longest panel data on output and employment shares on Sub-Saharan African countries to our knowledge. If sub-Saharan African countries are experiencing prematurely deindustrialization, this has potentially negative ramifications for the region’s economic development. 1 See also Dasgupta & Singh (2005). 2 Our main dependent variable of interest at any level of aggregation is the level of industrialization defined as the share of manufacturing in economy wide GDP or in economy wide aggregate employment. To adjust for possible resource boom, we also consider the manufacturing’s percentage contribution to non-oil GDP for the restricted set of oil-exporting countries. Given that these variables are bounded (i.e. these are fractional response variables) we perform our analysis using both the linear fixed-effects panel data model - which is subject to pitfalls similar to linear probability models - as well as the “fixed-effects fractional logit” model recently developed by Hardin, Hardin, Hilbe & Hilbe (2007). Our results overwhelmingly do not support the finding that SSA countries have began to deindustrialize prematurely. Moreover, we document meaningful heterogeneity across SSA subregions. The Southern region is found to be the only subregion to have witnessed deindustrialization. However, we are unable to confirm that this deindustrialization in the Southern region of SSA is premature. The rest of the paper is organized as follows. Section 2 presents the data. Section 3 discusses the relevant methodological issues. Section 4 presents the main results. Section 5 explores geographic heterogeneity. Section 6 considers potential hypotheses and Section 7 concludes. 2 Data and Descriptive Analysis This section describes the data sources used for the analysis and provides summary statistics as well as trend analysis of our main indicators. 2.1 Data Sources and Measurements The data for this study come from a variety of sources. The data for manufacturing valued added and manufacturing employment come from the Groningen Growth and Development Center 10-sector Database (Timmer, de Vries & de Vries 2015). The GGDC Database collects time series data on value added and employment for ten broad sectors of the economy for 42 countries (both developed and developing) spanning 1950 through 2012.2 The advantage of the GGDC is that it provides a long-run series of sectoral productivity that is internationally comparable across a broad range of countries. However, the GGDC’s Africa Sector Database (ASD) only includes eleven sub-Saharan African countries.3 Thus, for other sub-Saharan African countries, we collect manufacturing value added and GDP time series from 2 This period of data availability is as of September, 2017. 3 The Sub-Saharan African countries in the ASD are Botswana, Ethiopia, Ghana, Kenya, Malawi, Mauritius, Nigeria, Senegal, South Africa, Tanzania, and Zambia. 3 the World Bank’s World Development Indicators (WDI). The income and popu- lation data for all countries come from the 2018 version of the Maddison Project Database, which provides historical income statistics for 169 countries through 2016 (Bolt, Robert Inklaar & van Zanden 2018). The income figures are at con- stant 2011 US dollars. Our main analysis sample is an unbalanced panel of 41 sub-Saharan African countries over the period 1960 through 2016. We also contrast our main results with the analysis of the evolution of manu- facturing in non-oil GDP series for a smaller subset of countries. For 38 countries in our main sample (excluding Equitorial Guinea, Sudan, and Swaziland), we col- lect data on real growth rates of non-oil GDP, oil revenues as a percent of GDP, and real effective exchange rates from the International Monetary Fund (IMF)’s Africa Regional Economic Outlook (AFRREO) Fall 2018 database for the pe- riod 2004 through 2016. The AFRREO Fall 2018 database is in turn based on the statistical appendix of the Regional Economic Outlook (Sub-Saharan Africa) of the Fall 2018 World Economic Outlook (WEO).4 We utilized the real growth rates of non-oil GDP data series to recover estimates of the manufacturing output shares in non-oil GDP. Finally, for the purpose of testing the implications of the Dutch Disease for countries with significant oil-exporting and agricultural sectors, we collect data on the monthly minimum wages from the International Labor Or- ganization’s ILOSTAT Database for Cameroon, Ghana, Nigeria, and South Africa. Three measures of industrialization are commonly employed in the literature, namely the share of manufacturing valued added in GDP at constant prices (RMVA), the share of manufacturing valued added in GDP at current prices (CMVA), and the share of manufacturing employment in economy-wide employment (EMP). We focus on RMVA and EMP for our substantive statistical analysis since the trend in the share of manufacturing valued added in GDP at current prices derives from movements in both price and quantity that are indistinguishable to the researcher. While the manufacturing share at current prices conflates price and quantity, the real manufacturing output share adjusts for general movements in prices over time, and thus, is more reliable for understanding structural changes in industrialization and the economy. The real manufacturing output shares are at constant 2005 national prices. Also, since WDI does not include data on manufacturing employ- ment, our employment analysis is limited to the aforementioned eleven countries in the GGDC’s Africa Sector Database. To account for possible natural resource boom, we also examine the trends and patterns of the share of manufacturing value added in non-oil GDP (RMVA-NOIL). This could be important in separating the 4 As of January 2019, the AFRREO Fall 2018 Database is accessible at https://www.imf. org/external/datamapper/datasets/AFRREO. 4 trends in manufacturing output shares that may arise from the sheer increase in GDP due to the oil boom as discussed in Fardmanesh (1991). Put differently, by subtracting oil’s contribution from GDP, it is possible to isolate the manufacturing sector’s true trajectory as a country develops. 2.2 Trends and Patterns at Country Levels We explore the trends and patterns of these industrialization measures across Sub-Saharan Africa countries, alongside other factors. We begin our analysis by summarizing manufacturing and employment shares for the sub-Saharan African countries for which data are available. The Appendix Tables A.1 and A.2 present the means of the share of manufacturing value added in GDP (constant prices) and the manufacturing employment shares, respectively, for each country in our sample across the six decades for which data are available. As one would expect, these tables reveal significant heterogeneity in the trends of all three measures with some countries experiencing increasing, decreasing, or stable trends while others witnessed a mixed combination of trends across the decades. By eyeballing the means in Table A.1, a majority of seventeen countries appear to have a relatively stable share of manufacturing value added in GDP. Ten countries appear to have experienced declines while seven countries show an increasing trend in RMVA over the decades. Also, five countries seem to be consistent with an inverted U-shaped trend while the remaining two countries show an initial rising trend that stabilized in latter decades. The manufacturing employment shares exhibit similar hetero- geneity in trends (Table A.2). To motivate our analysis, we depict the evolution of deindustrialization with respect to per capita income. Figures 1 and 4 display scatter plots of the shares of manufacturing output in GDP and non-oil GDP, respectively, as functions of GDP per capita (with fitted quadratic trends). While the fitted quadratic trend in manufacturing output shares in Figure 1 shows a weak semblance of an inverted U-shaped relationship, the scatter plot for manufacturing output shares in non-oil GDP in Figure 2 is suggestive of a rather U-shaped relationship with respect to per capita income. While this exercise is purely descriptive, it is also suggestive of plausibly different trajectories for manufacturing trends depending on whether one examines GDP versus non-oil GDP. In particular, this contrast is more obvious for our four countries with significant oil-exporting sectors (Cameroon, Ghana, Nigeria, and South Africa) displayed in Figure 3. Mechanically, the manufacturing output shares in GDP versus non-oil GDP mirror each other but Figure 3 shows that while the former is stable for Cameroon, it is increasing for Nigeria, and decreasing for Ghana and South Africa. Again, the goal here in examining the time series in Figure 3 is to illustrate the point that manufacturing output shares 5 Figure 1: Manufacturing output shares (% GDP) and GDP per capita Figure 2: Manufacturing output shares (% non-oil GDP) and GDP per capita can potentially take a different trajectory when we take into account oil resources. Turning our attention to manufacturing employment shares, Figure 4 depicts an increasing trend with the level of per capita income. Again, these observations are descriptive and only serve to foreshadow our main statistical analysis. 6 (a) Cameroon (b) Ghana (c) Nigeria (d) South Africa Figure 3: Evolution of manufacturing output shares (% GDP and % non-oil GDP) for selected African countries (2004-2016) 2.3 Trends and Patterns at Subregional Levels We now present regional summaries of the share of manufacturing valued added in GDP in Table 1. For the remainder of this paper, subregional analysis are only performed for RMVA (and not EMP) due to the limited availability of employ- ment shares data as noted previously. We follow the geographic SSA subregions of the African Union, namely West Africa, East Africa, Central Africa, and Southern Africa (for the manufacturing output shares only).5 Table 1 reveals a more con- sistent pattern at the subregional level; that is, the share of manufacturing value added in GDP (constant prices) appears to follow an inverted U-shape. For West Africa, RMVA rises from 0.08 to 0.12 and declines to 0.07 by the 2010s. The other 5 See Table A.3 for a list of sub-Saharan African countries in our sample and their respective geographic subregional groups. 7 Figure 4: Manufacturing employment shares vs GDP per capita regions follow a similar pattern with East Africa rising from 0.07 through 0.09 to 0.08; Central Africa moving from 0.16 to a high of 0.25 before declining to 0.09; and Southern Africa increasing from 0.11 through 0.15 to 0.14. For all countries, Table 1 unsurprisingly shows that the same inverted U-shape persists with RMVA beginning from 0.10 in the 1960s, reaching a high of 0.15 in the 1970s and declining afterwards to 0.10 in the 2010s. These observations are generally reflected in Figure 5, which depicts the scatter plots of the shares of manufacturing output as a function of GDP per capita (with fitted quadratic trends) for the SSA subregional groupings. Except for Central Africa, the fitted quadratic trend in Figure 5 highlights an inverted U-shaped rela- tionship between RMVA and per capita income at the subregional level. Obviously, these eyeball tests do not accurately depict the underlying, structural aggregate trends in manufacturing output and employment shares in SSA and we conduct formal statistical tests momentarily. 8 Table 1: Regional variation in Manufacturing Value Added (% of GDP) 1960s 1970s 1980s 1990s 2000s 2010s West Africa 0.08 0.12 0.11 0.11 0.09 0.07 Observations 24 63 75 99 121 82 Number of countries 3 7 8 10 14 14 East Africa 0.07 0.09 0.09 0.09 0.09 0.08 Observations 30 62 88 90 90 38 Number of countries 4 7 9 9 9 9 Central Africa 0.16 0.25 0.17 0.13 0.1 0.09 Observations 12 32 50 53 76 61 Number of countries 3 4 5 6 9 9 Southern Africa 0.11 0.13 0.14 0.14 0.15 0.14 Observations 26 69 80 89 90 40 Number of countries 5 7 8 9 9 9 Sub-Saharan Africa (All SSA) 0.11 0.15 0.13 0.12 0.11 0.10 Observations 92 226 293 331 377 221 Number of countries 15 25 30 34 41 41 The table reports SSA regional mean shares of manufacturing output shares in GDP (constant prices) for each south-Saharan African country by decade (e.g., 1960-1969, etc) based on our main sample of 41 SSA countries. Each column represents decade-specific mean shares followed (on the corresponding rows) by the number of countries in each region as well as the number of years for which data are available. See Table A.3 for a list of the SSA countries belonging to each of the four subregional groups in our sample. 9 (a) West (b) East 10 (c) Central (d) Southern Figure 5: Subregional scatter Plot of manufacturing employment shares as a function of GDP per capita 3 Methodological Issues The previous section explored the trends and patterns of industrialization or deindustrialization by eyeballing the trends in the shares of manufacturing value added and employment in GDP and economy-wide employment. This section pursues various statistical approaches to provide rigorous evidence of industrial- ization and/or (premature) deindustrialization in Sub-Saharan Africa and studies its determinants. Several economic phenomena are non-monotonic and commonly follow either a U-shaped or an inversed U-shaped relationships. While some of these relationships are grounded in theory, others tend to the subject of much em- pirical investigation. Classic examples of inverted U-shapes in economics include the Laffer Curve (the relationship between the rate of taxation and the level of government revenue) and the Kuznets Curve (the relationship between income and inequality). The nature of the relationship between the shares of manufacturing in GDP (or aggregate employment) and per capita income is theoretically ambiguous and is thus an empirical question. To investigate non-monotonic relationships, it is standard to include some poly- nomial (typically quadratic terms) in the vector of covariates within a linear regres- sion framework. Since we are interested in the trend of manufacturing relative to income, we follow Rodrik (2016) by including the logarithm of population (P OP ) and per capita GDP (GDP P C ) as well as their respective quadratic terms in our vector of control variables. Thus, the basic econometric framework can be described by Yit = β0 + β1 log (GDP P C )it + β2 log (GDP P C )2 it + 2 β3 log (P OP )it + β4 log (P OP )it + Xit γ + αi + εit , (1) where Yit denotes any of our dependent variables of interest for country i at time t, Xit is a set of country-specific, time-varying explanatory variables (which may include time trends), αi are country fixed effects, and εit is the idiosyncratic error term. There are a number of econometric challenges to investigating whether manu- facturing follows an inverted U-shape or not in equation (1). The first challenge concerns the appropriate approach for testing for the presence of an inverted U- shaped relationship. The most common approach in the literature is to specify a regression equation such as equation (1) and conclude based on the sign and statistical significance of the coefficients estimates of β1 and β2 . For instance, if both coefficients are statistically significant and βˆ1 > 0 while β ˆ2 < 0, then one 11 concludes an inverted U-shape exists (provided the estimated extremum is within the data range). However, Lind & Mehlum (2010) argues painstakingly that the aforementioned procedure is a weak and deeply flawed test for U-shaped or in- verted U-shaped relationships. They suggest a new test based on prior work by Sasabuchi (1980), which we employ in this study as further described below. The second challenge in estimating the parameters of equation (1) is that the de- pendent variable of interest is bounded (i.e., it is a fractional response variable), thus, requiring an appropriate functional form specification. In fact, using linear models for fractional outcome a is subject to the same drawbacks and pitfalls as those related to linear probability models (LPM) for binary outcome. Finally, the fact that we have repeated measures on each country over time (longitudinal data) poses a peculiar difficulty for estimating fractional response models. While the panel structure of the data can be exploited to address unobserved heterogene- ity in explaining the evolution of manufacturing shares in output and employment, the unbalanced nature of the panel poses special (often ignored) difficulties when the dependent variable is fractional as in this study. Panel data methods suitable for explaining fractional responses in the context of panel data is an active area of ongoing research. Papke & Wooldridge (1996)’s seminal work proposed direct methods for explaining the conditional mean of a fractional response using quasi- maximum likelihood estimation methods in cross section data. Such methods have come to be known as “fractional logit” models and have been applied widely. Papke & Wooldridge (2008) extends these methods to balanced panel data contexts by employing the Chamberlain-Mundlak device (Mundlak 1978, Chamberlain 1980) to handle the unobserved effects. Although an important advancement in the literature, Papke & Wooldridge (2008)’s work is limited to balanced panel data scenarios as they remark that, “[t]he nonlinear models we apply are difficult to extend to unbalanced panel data-a topic for future research.” In this paper, we employ the “fixed effects fractional logit” model (Wagner 2003, Hardin et al. 2007, Hausman & Leonard 1997). In particular, if we denote the logistic CDF with Λ[.], then the conditional expectation of the manufacturing shares takes the form E[Yit |Xit , αi ] = Λ β0 + β1 log (GDP P C )it + β2 log (GDP P C )2 it + 2 β3 log (P OP )it + β4 log (P OP )it + Xit γ + αi . (2) As noted by Hausman & Leonard (1997), this specification is not subject to the incidental parameters problem if we assume that the cross sectional units (i.e., countries) are fixed while the number of years per country goes to infinity, which 12 seems natural in our case.6 Because the logistic CDF is strictly monotonic, the regression coefficients give the direction of the marginal effects. Also, notice that the predicted manufacturing shares based on equation (2) are constrained to the unit interval. 4 Deindustrialization in Sub-Saharan Africa? This section presents the regression results based on equation (2) for our measures of deindustrialization. Tables 2 and 3 show the coefficients estimates as well as the estimated average marginal effects for real manufacturing output shares and manufacturing employment shares, respectively. Panels A and B of Table 2 fur- ther break down these results by the full sample and low-exporting SSA countries, respectively. We classify countries as low-exporting countries if across their respec- tive years of available data, their average shares of manufacturing in total volume of exports exceeds 50 percent. For this subgroup analysis, the excluded high- exporting countries in our sample are Botswana, Lesotho, Swaziland, Cape Verde, and Mauritius. It is important to examine results by this dichotomy because the premature deindustrialization results for sub-Saharan Africa in Rodrik (2016) de- pend crucially on whether Mauritius, a high-exporting country is included in the sample or not. Rodrik (2016) remarks that “[f]inally, the estimates for sub-Saharan Africa depend heavily on whether Mauritius – a strong manufactures exporter – is included in the sample or not. Without Mauritius in the sample, sub-Saharan African countries emerge as large losers on all three measures of industrialization.” We argue that SSA’s manufacturing experience is more nuanced than whether a single country is included or not and varies meaningfully across countries and ge- ographic subregions. For both Panels A and B in Table 2, the coefficient estimates for β1 and β2 are statistically significant and suggest that RMVA is inversely U-shaped based on their signs following the conventional test. The turning points occur at around $7,500 (constant 2011 US dollars) and the Lind & Mehlum (2010) test confirms an inverted U-shaped relationship. For comparison, the turning point at which manufacturing output shares falls for Rodrik (2016)’s full sample is above $70,000 (1990 US dollars), which is admittedly implausible as it falls outside the data range in his paper. The corresponding turning point for Indian and African countries is reported to be around $700 Rodrik (2016). Also, using data on 135 economies, Felipe, Mehta & Rhee (2014) finds evidence of deindustrialization, with manufac- turing output shares declining after about $2000 (constant 2000 US dollars). Thus, 6 Also, see Wooldridge (2018) for a more recent discussion of Chamberlain-Mundlak style strategies for dealing with unobserved heterogeneity in unbalanced panel data. 13 Figure 6: Predicted manufacturing shares (constant prices) as a function of GDP per capita (in 2011 dollars) while there are discrepancies regarding the turning point at which manufacturing output shares falls, the results suggest SSA has deindustrialized. However, the results in Panel B when we exclude the high-exporting countries depicts a weaker U-shaped relationship. Although the conventional test still sug- gests an inverted U-shape, the Lind & Mehlum (2010) test indicates a much weaker relationship; the slope at the maximum GDP per capita is positive and statistically not significant; the overall Lind & Mehlum (2010) U test is also not statistically significant. Also, RMVA now peaks at about $30,000 (which is much closer to the maximum value in the sample). As depicted in Figure 6, which plots the predicted RMVA using the estimates from Table 2, the predicted RMVA shows an inverted U-shape for the full sample while that of the low-exporting countries shows a much weaker relationship. Put together, although we find evidence that SSA has deindustrialized on the basis of manufacturing output shares, this finding may be driven by a smaller subset of countries (we return to this point in the next section). Finally, the manufacturing employment shares results in Table 3 are not con- sistent with an inverted U-shape. The coefficient estimate on the quadratic term in GDP per capita in Table 3 is positive and statistically significant at the 10% sig- nificance level. Again, the Lind & Mehlum (2010) test trivially rejects an inverted U-shape and we do not report the results. Thus, we find conflicting evidence re- 14 Figure 7: Predicted employment shares as a function of GDP per capita (in 2011 dollars) garding output and employment shares, with the former showing weak evidence for deindustrialization while the latter showing none. Felipe et al. (2014) also notes the differential results for output and employment shares in their sample. Our results highlight one important point, that is, the evidence regarding premature deindustrialization (in output and employment shares) in the literature may not be as robust as previously understood. Overall, the results so far demonstrate that unlike previous studies, we do not find an overwhelming evidence that manufacturing in sub-Saharan Africa follows an inverted U-shape, except for the full sample using RMVA. Even in the case of the RMVA finding, our results contrast with Rodrik (2016) who documents an inverted U-shape relationship for sub-Saharan Africa only after excluding Mauritius. We find that excluding high-exporting countries works in the opposite way as the U-shape relationship weakens (or vanishes per Lind & Mehlum (2010)’s test). Our interpretation is that these findings are indicative of important geographic variation in deindustrialization and we explore this further in the next section at the SSA subregional level. 15 Table 2: Manufacturing value added as percent of GDP (constant prices) regressions Panel A: All SSA Panel B: Low-exporting countries Coefficients Marginal Effects Coefficients Marginal Effects Panel A: Regression Estimates Population (logarithm) 2.827** -0.039 3.122** -0.011 (1.148) (0.031) (1.437) (0.035) Population squared (logarithm) -0.104*** -0.102** (0.040) (0.049) GDP per capita (logarithm) 3.576*** 0.042*** 2.328*** 0.055*** (0.899) (0.007) (0.859) (0.011) GDP per capita squared (logarithm) -0.200*** -0.113** (0.057) (0.053) 16 Intercept -36.523*** -36.887*** (9.767) (12.278) Panel B: Lind & Mehlum (2010) U Test Slope at minimum GDP per capita (logarithm) 1.229*** 0.958*** (0.244) (0.238) Slope at maximum GDP per capita (logarithm) -0.732** 0.043 (0.366) (0.257) Lind & Mehlum (2010) Test statistic 2.180 0.31 [0.015] [0.377] Observations 1540 1540 1293 1293 Number of countries 41 41 36 36 Country Fixed Effects Yes Yes Yes Yes Time Fixed Effects Yes Yes Yes Yes Standard errors in parentheses and p-values in square brackets. Results based on fixed effects fractional logistic regressions in equation (2). ∗p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗p < 0.010 Table 3: Manufacturing share of total employment regressions Coefficients Marginal Effects Population squared (logarithm) 2.390 0.130*** (2.855) (0.040) Population (logarithm) -0.011 (0.096) GDP per capita (logarithm) -2.177 0.041*** (1.553) (0.013) GDP per capita squared (logarithm) 0.170* (0.095) Intercept -27.114 (22.918) Observations 523 523 Number of countries 11 11 Country Fixed Effects Yes Yes Time Fixed Effects Yes Yes Standard errors in parentheses. Results based on fixed effects fractional logistic re- gressions in equation (2). ∗p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗p < 0.010 5 Deindustrialization at the SSA subregional lev- els This section explores deindustrialization by subregional groups. This is important for two reasons. First, there is no reason to expect that countries in different regions share the same deindustrialization experience and geographic differences across SSA may mask any evidence of (premature) deindustrialization. Second, examining the evolution of manufacturing for different regions may shed light on why SSA countries may have undergone deindustrialization. We perform subre- gional analysis for only the manufacturing output shares since we only have a small number of countries with data on manufacturing employment shares. Nonetheless, this is fitting for our purposes as it is also for manufacturing output shares that we find some evidence of deindustrialization in the previous section. The subregional group results are presented in Table 4, with the predicted manufacturing output shares plotted in Figure 8. These results highlight signifi- cant regional differences. From Table 4, the coefficient estimate on the quadratic term in GDP per capita is only statistically significant for Southern SSA. Figure 8 graphically sums up this finding: only Southern SSA appears to have deindus- 17 trialization in Sub-Saharan Africa and maybe the group of countries driving our findings in the previous section. The inverted U-shape of manufacturing output shares is also confirmed by the Lind & Mehlum (2010) test for Southern SSA. Af- ter peaking at around 20 percent, the estimated turning point for manufacturing output shares occurs at GDP per capita level of $5,800. Thus, it turns out that only the exploratory observations for Southern SSA in Figure 5 hold up in formal statistical analysis. For West Africa, the predicted RMVA values are flat in Figure 8. Both East and Central Africa have increasing predicted RMVA values. This confirms our conjecture in the previous section that deindustrialization in output shares is not the experience of the vast majority of sub-Saharan African countries. Given that we find robust deindustrialization in Southern SSA, we ask whether this deindustrialization can be considered to be premature. We do this by examin- ing how the turning point in manufacturing shares changes over time by temporally dividing our sample at the 1990 year mark. Mechanically, this amounts to includ- ing interactions of a post-1990 dummy variable with the logarithm of GDP per capita (and its quadratic term) in equation (1).7 Table 5 presents the results of this analysis for Southern SSA, and shows that the observed deindustrialization does not appear to be occurring prematurely. In other words, while Southern SSA has deindustrialized, this phenomenon is not occurring at lower levels of per capita income. The estimated coefficients on both interaction terms in Table 5 are not statistically significant, suggesting that the turning points for manufacturing output shares in Southern SSA have not changed over time. This is graphically illustrated in Figure 9, where the turning points for the pre- and post-1990 period are essentially the same. 7 The results are not sensitive to other breakpoints such as 1995. 18 Figure 8: Predicted manufacturing shares (constant prices) as a function of GDP per capita by subregional blocks (in 2011 dollars) Figure 9: Predicted manufacturing shares (constant prices) as a function of GDP per capita (in 2011 dollars) 19 Table 4: SSA Region Regressions: Manufacturing value added as percent of GDP (constant prices) West East Central Southern Coefficients Marginal Coefficients Marginal Coefficients Marginal Coefficients Marginal Effects Effects Effects Effects Population (logarithm) 0.777 0.100* 1.749 0.083 9.353*** -0.146 4.178*** -0.148*** (1.491) (0.057) (3.170) (0.058) (2.458) (0.108) (0.779) (0.053) Population squared (logarithm) 0.013 -0.022 -0.340*** -0.177*** (0.061) (0.089) (0.054) (0.034) 20 GDP per capita (logarithm) 0.762 -0.009 -0.177 0.036*** -0.084 0.055*** 3.960*** 0.029* (1.914) (0.016) (1.823) (0.012) (1.395) (0.009) (1.088) (0.017) GDP per capita squared (logarithm) -0.058 0.040 0.045 -0.229*** (0.122) (0.118) (0.092) (0.071) Intercept -17.989 -22.421 -68.173** -45.201*** (15.324) (28.636) (26.858) (6.886) Observations 464 464 398 398 284 284 394 394 Standard errors in parentheses. Results based on fixed effects fractional logistic regressions in equation (2). Sub-regions based on the African Union geographic classification of SSA countries. See Table A.3 for additional details. ∗p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗p < 0.010 Table 5: Manufacturing output shares (% GDP) re- gressions with post-1990 indicator - Southern SSA Coefficients Population (logarithm) 4.384*** (1.409) Population squared (logarithm) -0.184*** (0.044) GDP per capita (logarithm) 5.689* (3.011) GDP per capita squared (logarithm) -0.333* (0.196) GDP per capita (logarithm) × post1990=1 -2.185 (2.399) GDP per capita squared (logarithm) × post1990=1 0.132 (0.152) Constant -53.967** (21.526) Observations 394 Standard errors in parentheses. Results based on fixed effects fractional logistic regressions in equation (2) with a post-1990 dummy variable for Southern SSA. See Table A.3 for additional details on subregional grouping. ∗p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗p < 0.010 6 Robustness and other considerations Our main finding so far is that we do not find robust evidence for deindustrial- ization in SSA. While manufacturing employment shares do not appear to follow an inverted U-shaped relationship, manufacturing output shares tends to be in- versely U-shaped. Our results suggest that the RMVA result is driven by Southern SSA countries who appear to have deindustrilized when examining manufacturing output shares. This section investigates two specification issues in an attempt to understand the nature of SSA’s manufacturing experience. First, we augment the set of covariates beyond the population size and per capita income in equation (2). In particular, we include other potential correlates of deindustrialization such as the levels of remittances and foreign direct investments (FDI) (Lartey, Mandelman & Acosta 2012). Second, we also conduct our analysis by limiting the sample to countries for which we have non-oil GDP data. Recall that we have a subsample of SSA countries for which we have non-oil GDP data spanning 2004-2016 (see Section 2 for details). Our objective is to examine whether there other channels 21 that might have led to the observed (de)industrialization patterns. Spefically, we look at whether resource booms such as oil have induced sustained real exchange rate appreciation that could be linked to slower or less industrialization in re- source rich countries. Finally, we also test implications of the Dutch disease and resource-curse hypotheses to the outcome of SSA’s deindustrialization, although this analysis is limited to only four countries with significant oil-exporting sectors. Table 6 presents the regression results based on equation (2) with and without the additional covariates specified above in Columns (1) and (2), respectively. The results in Table 6 suggest that personal remittances may be an important corre- late of deindustrialization in sub-Saharan Africa. In particular, higher personal remittances are associated with lower manufacturing output shares in SSA. Table 6: Manufacturing output shares regressions (addi- tional covariates) (1) (2) Population (logarithm) 2.803*** 2.848*** (0.965) (0.965) Population squared (logarithm) -0.099*** -0.092*** (0.035) (0.034) GDP per capita (logarithm) 2.528** 2.047** (1.169) (1.001) GDP per capita squared (logarithm) -0.145** -0.117* (0.074) (0.064) Foreign direct investment, net inflows (% of GDP) -0.002 (0.002) Personal remittances, received (% of GDP) -0.003*** (0.001) Observations 1031 1031 Standard errors in parentheses. Results based on fixed effects fractional logistic regressions in equation (2). ∗p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗p < 0.010 Table 7 presents results for the non-oil GDP sample. For an apples-to-apples comparison, we estimate the same regression specification on the non-oil GDP sample but using first, the usual manufacturing output shares in GDP (RMVA) and second, the share of manufacturing value added in non-oil GDP (RMVA- NOIL). As shown in Table 7, the results are qualitatively the same regardless of the deindustrialization measure used in terms of statistical significance. Although 22 the coefficient estimates on the quadratic terms in both columns are statistically insignificant, the signs on the GDP per capita and its quadratic terms are flipped. Given the caveats of our reduced sample size for the non-oil GDP case, the op- posite signs are suggestive of different trajectories for the manufacturing output shares in GDP versus non-oil GDP. This also suggests that studies on SSA’s dein- dustrialization should take into account the potential role for the natural resource endowment. Table 7: Manufacturing output shares regressions for non-oil GDP sample Dependent variable RMVA RMVA-NOIL Population (logarithm) 3.774* -5.298*** (2.031) (2.038) Population squared (logarithm) -0.098* 0.148** (0.059) (0.061) GDP per capita (logarithm) -1.160 0.638 (0.983) (0.840) GDP per capita squared (logarithm) 0.071 -0.044 (0.058) (0.052) Observations 368 368 Standard errors in parentheses. Results based on fixed effects fractional lo- gistic regressions in equation (2) for the non-oil GDP sample. Column RMVA presents results using the manufacturing output shares in GDP as dependent variable while Column RMVA-NOIL is based on the share of manufacturing value added in non-oil GDP. ∗p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗p < 0.010 Finally, we focus on four countries in sub-Saharan Africa with significant oil- exporting sectors for which we have data to test the Dutch disease - Cameroon, Ghana, Nigeria and South Africa. To do so, we specify an equation similar to (2) but use as our dependent variable, the share of manufacturing value added in non-oil GDP. We use the same set of explanatory variables, but add oil revenues (% GDP), the real effective exchange rate and the monthly minimum wage. The monthly minimum wage variable serves as a proxy for the wage in domestic oil industries. In this specification, the presence of the Dutch disease can be inferred from the coefficient on the monthly minimum wage variable. From Table 8, the sign of marginal effect of the monthly minimum wage variable is negative (although not statistically significant), providing supporting evidence for the Dutch disease. Given the caveats of a limited size for this analysis, the results in Table 8 suggest that future studies should account for the role of the Dutch disease in studying 23 Table 8: The Dutch disease hypothesis: manufacturing output shares (% non-oil GDP) regressions Marginal Effects Population (logarithm) -0.0851*** (0.0314) GDP per capita 0.0872*** (0.0253) Real effective exchange rate 0.0004 (0.0003) Oil rents (% of GDP) 0.0017** (0.0007) Monthly minimum wage (2011 US dollars) -0.0000735 (0.0002) Observations 27 Marginal Effects are reported. Standard errors in parentheses. Results based on fixed effects fractional logistic regressions in equation (2) for Cameroon, Ghana, Nigeria, and South Africa. ∗p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗p < 0.010 sub-Saharan Africa’s manufacturing experience. 7 Concluding Remarks This paper investigates the cross-country patterns and trends in the share of man- ufacturing in national output and employment among sub-Saharan Africa. Ro- drik (2016) claims that several developing countries including those from sub- Saharan Africa are prematurely deindustrializing because manufacturing has be- gun to shrink at levels of income that are much lower than those at which the advanced economies started to deindustrialize. We study the extent to which African countries differ regarding the scale, timing and causes of “deindustrializa- tion” using recent panel data methods for fractional responses. We use two main measures of industrialization, namely the share of manufacturing valued added in GDP at constant prices and the share of manufacturing employment in economy- wide employment. Our data come from a variety of sources and comprises of an unbalanced panel of 41 African countries. We also explore potential explanations or correlates of deindustrialization such as natural resource endowment. Our key finding is that deindustrialization does not appear to be the common experience of the majority of Sub-Saharan African countries. Only the Southern SSA subregion appears to have deindustrialized over the period under study. We do not, however, find evidence that this deindustrialization of the southern subregion has occured 24 prematurely. We also uncover meaningful geographic variation in manufacturing experience across sub-Saharan Africa, and a potential role of the Dutch disease in understanding SSA’s manufacturing experience. Although alaysis of the latter hinges on the sample size of oil-exporting countries and the availability of key vari- ables such as non-oil GDP, which are limited in this study, it suggests that future research should account for the role of the Dutch disease in studying Sub-Saharan Africa’s manufacturing experience. 25 References Amirapu, A. & Subramanian, A. (2015), ‘Manufacturing or services? an indian illustration of a development dilemma’, Center for Global Development Working Paper No. 408 . Bolt, J., Robert Inklaar, H. d. J. & van Zanden, J. L. (2018), Maddison Project Database, Version 2018, Technical report, Rebasing ‘Maddison’: new income comparisons and the shape of long-run economic development. Mad- dison Project Working paper 10 (Available at https://www.rug.nl/ggdc/ historicaldevelopment/maddison/research). Chamberlain, G. (1980), ‘Analysis of covariance with qualitative data’, The Review of Economic Studies 47(1), 225–238. Dasgupta, S. & Singh, A. (2005), ‘Will services be the new engine of indian eco- nomic growth?’, Development and Change 36(6), 1035–1057. Dasgupta, S. & Singh, A. (2007), Manufacturing, Services and Premature Deindus- trialization in Developing Countries: A Kaldorian Analysis, Palgrave Macmillan UK, London, pp. 435–454. Fardmanesh, M. (1991), ‘Dutch disease economics and oil syndrome: An empirical study’, World Development 19(6), 711–717. Felipe, J., Mehta, A. & Rhee, C. (2014), ‘Manufacturing matters. . . but it’s the jobs that count’, Cambridge Journal of Economics . Hardin, J. W., Hardin, J. W., Hilbe, J. M. & Hilbe, J. (2007), Generalized linear models and extensions, Stata press. Hausman, J. A. & Leonard, G. K. (1997), ‘Superstars in the national basketball as- sociation: Economic value and policy’, Journal of Labor Economics 15(4), 586– 624. Johnson, S., Ostry, J. D. & Subramanian, A. (2007), The prospects for sustained growth in africa: Benchmarking the constraints, Working Paper 13120, National Bureau of Economic Research. URL: http://www.nber.org/papers/w13120 Jones, B. F. & Olken, B. A. (2008), ‘The anatomy of start-stop growth’, The Review of Economics and Statistics 90(3), 582–587. Kaldor, N. (1966), Causes of the slow rate of economic growth of the United King- dom: an inaugural lecture, Cambridge University Press. 26 Kaldor, N. (1967), Strategic Factors in Economic Development, Frank W. Pierce memorial lectureship and conference series, New York State School of Industrial and Labor Relations, Cornell University. URL: https://books.google.com/books?id= IgeAAAAMAAJ Kuznets, S. (1953), ‘International differences in income levels: reflections on their causes’, Economic Development and Cultural Change 2(1), 3–26. Lartey, E. K., Mandelman, F. S. & Acosta, P. A. (2012), ‘Remittances, exchange rate regimes and the dutch disease: A panel data analysis’, Review of Interna- tional Economics 20(2), 377–395. Lind, J. T. & Mehlum, H. (2010), ‘With or without u? the appropriate test for a u-shaped relationship’, Oxford bulletin of economics and statistics 72(1), 109– 118. Matsuyama, K. (1992), ‘Agricultural productivity, comparative advantage, and economic growth’, Journal of economic theory 58(2), 317–334. Mundlak, Y. (1978), ‘On the pooling of time series and cross section data’, Econo- metrica: journal of the Econometric Society pp. 69–85. Papke, L. E. & Wooldridge, J. M. (1996), ‘Econometric methods for fractional re- sponse variables with an application to 401 (k) plan participation rates’, Journal of applied econometrics 11(6), 619–632. Papke, L. E. & Wooldridge, J. M. (2008), ‘Panel data methods for fractional re- sponse variables with an application to test pass rates’, Journal of Econometrics 145(1-2), 121–133. Rodrik, D. (2009), ‘Growth after the crisis’, Globalization and Growth 125, 126. Rodrik, D. (2012), ‘Unconditional convergence in manufacturing’, The Quarterly Journal of Economics 128(1), 165–204. Rodrik, D. (2016), ‘Premature deindustrialization’, Journal of Economic Growth 21(1), 1–33. Sasabuchi, S. (1980), ‘A test of a multivariate normal mean with composite hy- potheses determined by linear inequalities’, Biometrika 67(2), 429–439. Timmer, M., de Vries, G. J. & de Vries, K. (2015), Patterns of structural change in developing countries, in ‘Routledge handbook of industry and development’, Routledge, pp. 79–97. 27 Wagner, J. (2003), ‘Unobserved firm heterogeneity and the size-exports nexus: Evidence from german panel data’, Review of World Economics 139(1), 161– 172. Wooldridge, J. M. (2018), ‘Correlated random effects models with unbalanced panels’, Journal of Econometrics . 28 Table A.1: Manufacturing Value Added (constant prices, % of GDP) Country 1960s 1970s 1980s 1990s 2000s 2010s Burundi 0.17 0.12 0.09 Number of years 3 10 7 Benin 0.22 0.18 0.22 0.19 0.13 Number of years 9 10 10 10 7 Burkina Faso 0.14 0.13 0.12 0.1 0.07 Number of years 10 10 10 10 7 Botswana 0.01 0.08 0.07 0.07 0.07 0.08 Number of years 6 10 10 10 10 1 Central African Republic 0.02 0.04 0.08 0.07 0.06 0.07 Number of years 5 10 10 10 10 6 Cote d’Ivoire 0.13 0.12 Number of years 2 7 Cameroon 0.1 0.11 0.15 0.15 0.14 0.14 Number of years 5 10 10 10 10 7 Congo, Dem. Rep. 0.68 0.64 0.57 0.37 0.24 0.16 Number of years 2 10 10 10 10 7 Congo, Rep. 0.03 0.03 0.03 0.03 0.04 Number of years 2 10 10 10 7 Comoros 0.04 0.05 0.04 0.05 Number of years 10 10 10 5 Cabo Verde 0.11 0.06 0.06 Number of years 9 10 7 Ethiopia 0.03 0.03 0.05 0.05 0.05 0.05 Number of years 9 10 10 10 10 3 Gabon 0.01 0.01 0.02 0.04 Number of years 10 10 10 6 Ghana 0.15 0.17 0.12 0.12 0.1 0.09 Number of years 10 10 10 10 10 2 The Gambia 0.05 0.05 0.06 0.06 0.06 0.05 Number of years 4 10 10 10 10 7 Guinea-Bissau 0.12 0.11 Number of years 5 7 Equatorial Giunea 0.15 0.21 Number of years 4 7 Kenya 0.07 0.09 0.13 0.13 0.12 0.12 Number of years 6 10 10 10 10 2 Liberia 0.04 0.04 Number of years 10 7 Lesotho 0.03 0.05 0.08 0.14 0.11 Number of years 10 10 10 10 7 Mozambique 0.09 0.13 0.09 Number of years 9 10 7 Mauritania 0.15 0.14 0.11 0.08 Number of years 5 10 10 7 29 Table A1 cont.: Manufacturing Value Added (con- stant prices, % of GDP) Country 1960s 1970s 1980s 1990s 2000s 2010s Mauritius 0.18 0.22 0.25 0.21 0.18 Number of years 10 10 10 10 3 Malawi 0.1 0.11 0.11 0.11 0.1 0.1 Number of years 4 10 10 10 10 1 Namibia 0.12 0.13 0.13 0.11 Number of years 10 10 10 7 Niger 0.05 0.06 Number of years 4 7 Nigeria 0.02 0.02 0.04 0.04 0.03 0.04 Number of years 10 10 10 10 10 2 Rwanda 0.1 0.09 0.09 0.07 0.07 0.06 Number of years 5 10 10 10 10 7 Sudan 0.04 0.06 0.06 0.05 0.06 Number of years 10 10 10 10 2 Senegal 0.15 0.16 0.17 0.16 0.15 Number of years 10 10 10 10 1 Sierra Leone 0.04 0.03 0.02 Number of years 10 10 7 Sao Tome and Principe 0.08 0.08 Number of years 9 7 Swaziland 0.2 0.23 0.33 0.32 0.31 Number of years 9 10 10 10 7 Seychelles 0.05 0.06 0.09 0.11 0.08 Number of years 2 10 10 10 7 Chad 0.01 0.01 Number of years 3 7 Togo 0.07 0.06 0.07 0.08 0.08 Number of years 4 10 10 10 7 Tanzania 0.1 0.12 0.09 0.08 0.09 0.1 Number of years 10 10 10 10 10 2 Uganda 0.05 0.07 0.09 0.08 Number of years 8 10 10 7 South Africa 0.17 0.22 0.23 0.21 0.2 0.18 Number of years 10 10 10 10 10 2 Zambia 0.1 0.13 0.14 0.12 0.11 0.09 Number of years 5 10 10 10 10 1 Zimbabwe 0.18 0.17 0.17 0.16 0.12 0.09 Number of years 1 10 10 10 10 7 The table reports mean shares of manufacturing valued added in GDP (constant prices) for each south-Saharan African country by decade (e.g., 1960-1969, etc) based on our main sample of 41 SSA countries. Each column represents decade-specific mean shares followed (on the corresponding row) by the number of years for which data are available. 30 Table A.2: Manufacturing employment (% of total employment) 1960s 1970s 1980s 1990s 2000s 2010s Botswana 0.01 0.02 0.03 0.06 0.06 0.06 Observations 6 10 10 10 10 1 Ethiopia 0.02 0.02 0.02 0.02 0.04 0.07 Observations 9 10 10 10 10 2 Ghana 0.1 0.13 0.12 0.12 0.11 0.11 Observations 10 10 10 10 10 2 Kenya 0.04 0.04 0.04 0.07 0.12 0.13 Observations 1 10 10 10 10 2 Mauritius 0.17 0.26 0.3 0.24 0.19 Observations 10 10 10 10 2 Malawi 0.03 0.03 0.03 0.03 0.03 0.04 Observations 4 10 10 10 10 1 Nigeria 0.06 0.07 0.05 0.04 0.04 0.04 Observations 10 10 10 10 10 2 Senegal 0.06 0.06 0.07 0.09 0.1 Observations 10 10 10 10 1 Tanzania 0.01 0.02 0.01 0.01 0.02 0.03 Observations 10 10 10 10 10 2 South Africa 0.11 0.14 0.16 0.13 0.13 0.12 Observations 10 10 10 10 10 2 Zimbabwe 0.02 0.04 0.04 0.03 0.03 0.03 Observations 5 10 10 10 10 1 The table reports mean shares of manufacturing employment shares in total employment for each south-Saharan African country by decade (e.g., 1960-1969, etc) based on our employment sample 11 SSA countries. Each column represents decade-specific mean shares followed (on the corre- sponding row) by the number of years for which data are available. 31 Table A.3: Geographic subregions of Sub-Saharan Africa Dependent Variable Country Manufacturing output Manufacturing shares (constant prices) employment shares West Benin 1971 - 2016 Burkina Faso 1970 - 2016 Cape Verde 1991 - 2016 Cote d’Ivoire 2008 - 2016 Gambia 1966 - 2016 Ghana 1960 - 2011 1960 - 2010 Guinea-Bissau 2005 - 2016 Liberia 2000 - 2016 Mauritania 1985 - 2016 Niger 2006 - 2016 Nigeria 1960 - 2011 1960 - 2011 Senegal 1970 - 2010 1970 - 2010 Sierra Leone 1990 - 2016 Togo 1976-2016 East Comoros 1980 - 2014 Ethiopia 1961 - 2012 1961 - 2010 Kenya 1964 - 2011 1969 - 2010 Mauritius 1970 - 2012 1970 - 2010 Rwanda 1965 - 2016 Seychelles 1978 - 2016 Sudan 1970 - 2011 Tanzania 1960 - 2011 1960 - 2010 Uganda 1982 - 2016 Central Burundi 1971 - 2016 Cameroon 1965 - 2016 Central African Republic 1965 - 2015 Chad 2007 - 2016 Congo Republic 1978 - 2016 Democratic Republic of 1968 - 2016 Congo (DR Congo) Equitorial Guinea 2006 - 2016 Gabon 1908 - 2015 Sao Tome and Principe 2001 - 2016 32 Table A.3 cont.: Geographic subregions of Sub-Saharan Africa Dependent Variable Country Manufacturing output Manufacturing output shares (constant prices) shares (current prices) Southern Botswana 1964 - 2010 1964 - 2010 Lesotho 1970 - 2016 Malawi 1966 - 2010 1966 - 2010 Mozambique 1991 - 2016 Namibia 1980 - 2016 South Africa 1960 - 2011 1960 - 2010 Swaziland 1971 - 2016 Zambia 1965 - 2010 1965 - 2010 Zimbabwe 1969 - 2016 The table presents the sub-Saharan African countries (by geographic grouping) in our sample and indicates the years of data availability for our measures of industrialization. 33