What Explains Latin America's Low Share of Industrial Employment?

This paper investigates the relative importance of different channels in explaining the low share of industrial employment in Latin America relative to the economies that employ a large share of the workforce in industry. Differences in domestic final consumption shares play a pivotal role and can account for 50-70 percent of the industrial share gap. The paper finds limited support for the comparative advantage hypothesis, as differences in trading patterns account for less than 15 percent of the gap. More important are the differences in sectoral linkages and wage gaps which account for more than 30 percent of the industrial employment gap individually.


Introduction
Modern economic growth is accompanied by shifts in the distribution of economic activity across the three broad sectors of the economy. As a sustained period of growth takes off, the share of agriculture in employment and output declines and this loss is picked up by industry and services. While services continue to expand, industry reaches a threshold after some time after which it starts to lose its share in the economy. These stylized facts of structural transformation have been documented since the late 1950s (Clark (1957), Chenery (1960), Kuznets (1957)). More recently, using data from multiple sources Herrendorf et al. (2014) documents these patterns of transformation for a broad set of countries. 1 Nonetheless, there exist deviations in the transformation experience of countries. In particular, Rodrik (2016) documents an intriguing aberration in the industrialization pattern of the developing countries in which the transformation started much later compared to the early industrializers. Industry in many developing countries fails to reach the thresholds achieved by the developed countries in the past. Moreover, the onset of deindustrialization from this lower peak happens at a lower level of GDP per capita. The Latin American economies are more adversely affected by this premature deindustrialization compared to the other regions of the world.
In this paper, I investigate the relative strength of different channels in accounting for the low share of industry in Latin American economies. I employ a version of the open economy multi-sector accounting framework developed in Uy et al. (2013) in which the sectoral good is an aggregate of a continuum of tradeable varieties which are in turn produced using labor and composite sectoral goods as intermediates (Eaton & Kortum (2002)). The specification links sectoral production to the production activity in other sectors as well as in other countries. The framework yields a system of equations that establishes a relationship between the sectoral share of the economic activity and various channels that are readily measured from the data and bear association with different aspects of the transformation process. The share of a sector rises with an increment in its share of the total value-added in the economy, either via an increase in domestic consumption or net exports. A sector also realizes an expansion if it sees an upturn in the intensity with which it is used as an intermediate. Moreover, sectors differ in how intensely they employ labor relative to intermediates and any rise in the intensity of labor use of a sector needs to be compensated by the transfer of labor towards it. All the above changes affect the value-added and the employment shares in the same way which equal each other. To account for the difference in the value-added and the employment share of a sector as seen in the data, I allow for labor market barriers which restrict the flow of labor thereby generating wage (or productivity) gaps across sectors.
Thus, the accounting framework links the sectoral allocation of labor to the consumption shares, the trade shares, the intermediate shares, the wage gaps, and the value-added shares of the sectors. I exploit this relationship and stack the Latin American countries against the economies that have attained a high industrial share of employment to gauge the relative success of the factors. The input-output data are an essential ingredient of the analysis which I source from the OECD database (OECD (2017)) which contains this information for many countriesincluding seven countries in Latin America, for 1995-2010/11. To construct a comparator set for these Latin American economies, I pick countries in the GGDC 10-Sector database (Timmer et al. (2015)) in which the peak industrial share during 1995-2010/11 exceeds 30 percent. Fortunately, the OECD database has information on all these comparator countries barring Mauritius. I also add China to the comparator set as its peak industrial share falls just short of the 30 percent threshold.
On average, the share of employment in the broad industrial sector in a Latin American economy is 11 percentage points lower compared to the comparator set. The counterfactual exercise reveals that more than half of this gap is accounted for by differences in consumption shares.
Cross-country differences in sectoral wage gaps and linkages are also significant and roughly account for 5 and 3.5 percentage points of the gap. 2 Differences in trade shares, on the other hand, play a muted role and can account for only a tenth of the gap in industrial employment shares. The findings for the manufacturing sub-sector which constitutes the bulk of the industrial gap, are somewhat different. Though differences in final consumption and wage gaps are still significant, they can only account for a quarter of the manufacturing gap. Differences in sectoral linkages emerge as the strongest factor and still account for a third of the employment share gap. The explanatory power of trade shares is more pronounced in manufacturing, but wanes compared to what can be explained by differences in final consumption, linkages and wage gaps.
Looking at individual countries within the Latin American sample, I find a moderately dominant role of trade shares for Colombia, Costa Rica and Mexico. The closest the differences in trade shares come in explaining the employment gap is for the manufacturing sub-sector in Costa Rica where it accounts for almost half of the 8.7 percentage points gap.
In a second quantitative exercise, I turn my attention to the experience of the Latin American economies during 1995-2010/11 in which the industrial and the manufacturing shares of employment declined by 2 and 3.5 percentage points respectively. The exercise helps in parsing determinants that dragged the employment shares down from the ones that provided tailwinds towards an expansion. Trade, arguably more essential for manufacturing, did not play a primary role in driving manufacturing shares down. It is true that changes in trading patterns allude to a contraction in manufacturing shares for five of the seven countries in the sample, but the impact is quantitatively moderate. At their most detrimental, the changes in trade shares imply a contraction of around 55 basis points in Costa Rica and Peru which is orders of magnitude smaller compared to the contractions entailed by changes in the sectoral wage gaps. At the other end, Chile and Colombia experienced an expansion in export shares, the quantitative response of which implies an employment share gain of 35 and 90 basis points respectively.
The trivial role of trade shares in explaining the changes in employment shares begs the question that whether this conclusion is specific to the Latin American economies. To this end, I conduct the same exercise for China and the United States to parallel the findings relative to what I observe for the Latin American economies. Like the Latin American economies, I find that changes in trade shares play a minor role in explaining the massive shifts in employment shares for China and US as well. In the US, changes in trading patterns imply a decline of 50 basis points in industrial and manufacturing shares which constitute 10 percent of the total slump. China's expanding share of manufacturing exports in total value-added advances employment share by 80 basis points which makes up for a quarter of the total change in the manufacturing sub-sector.
However, at the broad industrial level, changes in trade shares remain largely inconsequential.
In contrast to Latin American economies where changes in final consumption which do not uniformly explain the decline of employment shares, these changes are material in resolving the evolution of shares in China and the US. In China, they account for 85-90 percent of the entire gain in industrial and manufacturing shares.
The limited role of trade shares in accounting for the industrial employment gap needs interpretation in the light of findings in Rodrik (2016). Rodrik (2016) shows that the pace of deindustrialization is faster in non-manufactures exporters relative to manufactures exporters. Differences in comparative advantage can serve as an explanation for these differences in experiences. The comparative advantage hypothesis also proves helpful in explaining why the early-developers were able to put a larger share of the workforce in industry before they started deindustrializing. Nonetheless, the accounting exercises show the differences in the shares of value added derived from industrial exports across countries are not large enough to close the industrial gap. During 1995-2010 when global trade grew exponentially, the shifts in trade shares are not stark enough to justify the shifts in employment shares even in China and the US which are usually at the center of the debate. However, it would be naive to argue that trade plays a limited role in changing the sectoral allocation of resources. Indeed, Pierce & Schott (2016) find that the withdrawal of tariff increases on Chinese imports is causally linked to the loss of manufacturing employment in the US. The findings of this paper instead contend that any trade-induced force goes beyond the comparative advantage mechanism. Trade arguably interacts with preferences on the consumption side of the economy as households adjust their expenditure shares in response to trade. Similarly, sectoral linkages and productivity gaps might also react to trade. These avenues remain largely unexplored and need further investigation.
I conclude the quantitative analysis by highlighting some aspects of the construction subsector whose share in employment increased by 1.3 percentage points during 1995-2010/11. Specifically, the motivation is to evaluate whether construction can keep growing in the coming decades blunting the slump in manufacturing. The expansion of the sub-sector was driven by changes in sectoral wage gaps and a shift towards construction in final consumption. The continued expansion depends on whether these trends continue into the future. However, there exists a concern as to how large the sub-sector can get given that its share in most countries is close to global standards. Closing the remaining gap will yield modest expansion in most countries with Colombia and Costa Rica realizing the largest gains, partially driven by the fact that the construction sector contracted in the two countries during the 1995-2010/11 period.
The rest of the paper is organized as follows. In the next section, I outlay the accounting framework derived from the open-economy model which I use for the counterfactual analysis.
I discuss the results of the analysis in section 3 before concluding with some remarks on the relevance of the findings for policy.

Accounting Methodology
The accounting methodology that I employ in the paper essentially uses the open-economy framework developed in Uy et al. (2013). 3 In addition, I also allow wages to differ across sectors which accounts for differences in labor productivity across sectors.
There are J sectors in the economy and each sector is characterized by a unit interval of tradable varieties. A typical variety of sector j is indexed by x j ∈ [0, 1] and each variety can be produced domestically or can be imported from abroad. The varieties are used to produce a composite sector good, the quantity of which is given by Q j . The composite good has twin uses-it is either consumed domestically or is used as an intermediate in the domestic production of some variety of some sector. A variety is produced using labor and composite sectoral goods and the production technology is Cobb-Douglas in inputs. The Cobb-Douglas share of labor in the production of a particular variety x j of sector j is given by β j while the Cobb-Douglas share of the composite good from sector k used in total spending on intermediates is given by µ jk (∑ k µ jk = 1).
The household sector is endowed with a unit of labor which it supplies inelastically to the production of varieties. However, the wage rate differs across sectors which leads to differences 3 Hulten (1978) and Horvath (1998) are early examples of studies that formalize the connection between sectoral linkages and various aspects of the economy. More recently, Berlingieri (2014) and Sposi (2018) use a similar accounting framework to study structural transformation in separate settings. Also see Ngai & Samaniego (2009)  in value-added per worker across sectors. The household sector spends its wage income on the consumption of composite goods. In addition, it can lend or borrow from abroad leading to a trade surplus or deficit.
Using this framework, the value-added by sector j (V j ) can be written as where C j and N j denote the final consumption expenditure and the net exports of sector j respectively. The system of equations can be summarized in matrix notation as where the aggregate variables V j , C j and N j have been reduced to be expressed as ratios of GDP. The above equation links the value-added shares of sectors with the final consumption expenditures of the household sector as well as with the external sector. The equation also captures the association between the value-added shares and the labor shares of income (v j ) and sectoral linkages (µ kj ). There is one more step needed to convert the above relationship from value-added shares to employment shares. If w j is the wage rate in sector j, then the labor share of sector j can be recovered from the sectoral share of value added using where w j = τ j w and τ j represent wedges that create barriers which restrict the flow of labor across sectors.
Introducing time-scripts in the notation, the labor allocation L t = [l 1t , l 2t , . . . , l Jt ] at any time can be written as a function of final consumption shares C t , trade shares N t , sectoral linkages M t , wage gaps T t , and labor intensity B t where C t , N t , T t , B t are vectors of J elements and M t is the matrix of size J × J. Hence, the function F which encases equations 2 and 3 dispenses the counterfactual labor allocation when one or more variables on the right-hand side is counterfactually altered. 4

Quantitative Analysis
The framework in section 2 establishes a relationship between the sectoral distribution of the workforce and the five variables outlined before. In this section, I will use the data to perform a set of counterfactuals to highlight issues related to the industrial share of employment in the Latin American economies.

Cross-country variations in employment shares
The first exercise pertains to the stark pattern of premature deindustrialization observed in Latin America. The idea is to stack the Latin American countries against the economies that have attained a high industrial share of employment and gauge the relative importance of the aforementioned variables in explaining the gap across the two. Given that the sectoral share of employment varies systematically with development, it is desirable if the comparator countries have similar levels of economic development. My selection of comparator countries follows these two principles subject to the availability of data.
The input-output data from the OECD database (OECD, 2017) are available for the period 1995-2010/11 which defines the universe against which to compare the Latin American economies. Of more than 40 countries that feature in the GGDC 10-Sector database (Timmer et al. , 2015), the peak industrial share of employment exceeded 30 percent in 8 countries during this period. Mauritius is the only country among these 8 for which the input-output data are not available. I also include China as a comparator country as its industrial share of employment was just under 30 percent in 2011. China also merits inclusion as it is often argued that the development process in China has been different from many countries. It will be useful to see if China serves as an outlier benchmark relative to others in the present context. The table reports the peak industrial share for the seven countries in the GGDC 10-Sector database for which the share breached the 30 percent mark in any year beginning 1995. China is also included as its peak is just short of the 30 percent threshold. The last two rows report the GDP per capita for the eight countries in the year they achieve the peak share relative to the average GDP per capita of the seven Latin American economies in 1995 and 2011.  Interestingly, the cross-sectional variation in 2010/11 is much smaller. Apart from Peru and Mexico that have relatively large and small gaps respectively, both the industrial and manufacturing share gaps in the remaining five countries have converged over time. Note that most countries have lost industrial and manufacturing share in employment over the years. As such, this convergence is on the back of moderately lower losses in shares by the lagging countries.
These large gaps in industrial and manufacturing employment shares warrant explanation.
Nonetheless, as table 1 shows, the income per capita in many comparator countries is much higher compared to the Latin American economies. For example, income per capita in Japan and Singapore was 4-5 times as large compared to the average of the seven Latin American countries.
The gap in incomes converged over the next decade-and-a-half but income per capita still was orders of magnitude higher in many instances. To discipline the selection of comparator countries, To balance the total output in the counterfactual exercises, I scale the net export shares (N) by a factor such that the net lending as a share of GDP matches the net lending pattern of the counterfactual country. Similarly, when using the trade patterns from the counterfactual countries, I scale the final consumption shares (C) to retain the net lending behavior of the counterfactual country. The advantage of this adjustment is that it accepts the observed values of counterfactually changed variables. Table 2  The rows that follow report the implied gap in employment shares when I replace the specified The first row reports the actual gap in industrial/manufacturing employment share. The following rows report the implied gap remaining when the variables pertaining to the listed factor is counterfactually changed to a comparator country, keeping other variables unchanged. All figures are averages across the Latin American and the comparator set countries. All figures in percentage points.
variables of the Latin American countries with those of the comparator set. The lower the values in these rows, the higher is the contribution of the factor in explaining the aggregate gap.
On average, the share of employment in the broad industrial sector in a Latin American economy is 11 percentage points lower compared to the comparator set. The counterfactual exercise reveals that more than half of this gap is accounted for by differences in final consumption shares. Cross-country differences in sectoral wage gaps and linkages are also significant. Differ- The comparator set includes Italy and Spain which industrialized earlier together with the Asian economies in which the industrialization started much later. It is possible that some factors became more relevant for the latter set as they industrialized in a world much different from the early starters. Of singular interest are the differences in trading patterns across the two groups.
For instance, the higher industrial (and manufacturing) share in Asian economies may be driven by trade rather than domestic consumption as they engage more resources in the sector by taking advantage of foreign demand as the world became more open.
To see if the ability of factors in explaining the employment share gap responds to the exclusion of early industrialized economies, I perform the same exercise by reducing the set of comparator countries to China, Korea, Malaysia, and Taiwan. Columns (3) and (4)  Sectoral linkages remain the dominant factor driving the gap in manufacturing while differences in wage gaps come out as much less important than before.
In summary, for the aggregate industrial sector, the differences in final consumption are critical in explaining the employment share gap with differences in linkages and wage gaps also being important. Relative to these factors, trade plays a secondary role. Trade is much more important in explaining the manufacturing share gap accounting for a fifth of the gap but still plays a modest role relative to differences in linkages and final consumption. In what follows, I assess how these results seen at the regional level for Latin America relate to the seven constituent economies.

Changes in employment shares over time
In this sub-section, I turn my attention to the experience of the Latin American economies during the period under study. During the period the industrial share of employment has gone down by more than 2 percentage points. However, the decline in Chile and Costa Rica has been more severe. A couple of countries did gain some ground at the aggregate sector level but the decline in manufacturing has befallen every country in the sample. The average loss in manufacturing share is more than 3.5 percentage points with losses being exceptionally high in Chile, Costa Rica, and Peru. Analyzing inter-temporal changes is not only crucial for countries that observe significant shifts in the sectoral distribution of the workforce. The distribution may remain stable when changes in factors influencing the distribution act in opposing ways. The exercise helps in parsing determinants that dragged the employment shares down from the ones that provided tailwinds towards an industrial expansion.

Changes in trade shares in China and the United States
The trivial role of trade shares in explaining the changes in employment shares begs the question that whether this conclusion is specific to the Latin American economies. To this end, I conduct the same exercise for China and the United States to parallel the findings relative to what I observe for the Latin American economies. The employment share has risen sharply in the former, more so in the industrial sector than in manufacturing. On the other hand, employment shares in both industry and manufacturing have fallen more than 5 percentage points in the US.

Changes in consumption shares: Income-and price-effects
I conclude this sub-section by considering the diverse effect of changing patterns of final consumption in individual countries. The motivation for doing this is two-fold. From a purely accounting perspective, these changes signify large movements in employment shares for many countries. However, more importantly, the bulk of the theoretical literature on structural transformation has looked at the role of preferences and technological growth to explain changes in consumption patterns and the associated shifts in employment shares. The theory also ties to the hump-shaped behavior of the industrial sector as an economy moves up the development ladder.
In the context of the analysis here hence, it is possible to evaluate the changes in consumption shares that I observe in Latin America against the predictions of the theory.
The literature proposes two distinct explanations of why consumption patterns change over time. The first contends that household preferences are non-homothetic and as an economy grows richer, the resource allocation changes as households do not raise their expenditure proportionally across all goods (Kongsamut et al. (2001)). The second explanation hinges on differential exogenous productivity growth across sectors that alter the relative prices of goods over time (Baumol (1967), Ngai & Pissarides (2007)). Because of its association with changes in income, the first mechanism is often referred to as income-effects while the second mechanism is referred to as price-effects owing to its association with changes in relative prices. 6 Of course, it is feasible that paths of transformation differ across countries and that implied changes for Argentina and Mexico will still follow a downward trend individually. The price-effect mechanism can potentially help in resolving these deviations. Empirical analyses in different settings have found the elasticity of substitution in consumption across sectoral goods to be less than unity. This complementary association implies that the sector with relatively higher productivity growth loses its share as its price relative to other sector goods falls. Figure 6 plots the change in employment shares against the excess of labor productivity growth in the sector over the aggregate labor productivity growth. 7 For example, the industrial labor productivity growth during the period in Costa Rica outpaced the aggregate labor productivity growth by roughly 40 percent. A compelling inverse relationship between the two variables is evident with regards to the aggregate industrial sector. Changes in consumption shares imply substantial reductions in employment shares in Chile and Costa Rica where industrial productivity growth has far exceeded the growth in aggregate productivity. Concurrently, the consumption shares have shrunk in Argentina and Mexico in line with the industrial productivity growth lagging behind the aggregate. However, Peru and to some extent Brazil deviate from the broad trend when seen 6 The analysis in this paper focuses on changes in final consumption shares which is different from changes in value-added shares. In essence, the two effects are applicable when specifying either final goods or value-added production functions at the sectoral level. Nonetheless, the relative importance of the two effects is not necessarily independent of the specification. For instance, Herrendorf et al. (2013) explores the relative influence of the two effects in accounting for the transformation of the post-war US economy. 7 Labor productivity is measured as real value-added per worker using data from the GGDC 10-Sector database.
in isolation. Again, if each country follows a different path, the dominance of the income-effect can explain such deviations. Changes in the share of manufacturing bear no visible relationship with the excess of productivity growth in the sub-sector. In contrast to the industrial sector, the productivity growth in manufacturing has been higher than the productivity growth at the aggregate for each country in the sample. However, the implied change in employment share is positive for the majority. Chile and Costa Rica report an implied loss in manufacturing share on the back of a colossal relative productivity growth while the implied change is positive in Peru where the relative productivity growth in manufacturing has been the highest. On the other hand, the implied expansion in manufacturing share is highest for Argentina even though the relative productivity growth in the country has far outperformed the relative growth in Colombia and Mexico.
In conclusion, both income-and price-effects help in understanding the changes in consumption shares. Nevertheless, comparing the experience of countries reveals intriguing deviations.
It is difficult to reconcile these deviations even after employing both mechanisms jointly. An alternative is to explore if paths of transformation vary across countries, but lack of long-term data puts a natural impediment to such investigations. However, even if countries do indeed follow drastically different paths of transformation, the pertinent concern will be to determine what factors cause these divergences across countries.

Role of construction
As a final quantitative examination, I cast attention towards the construction sub-sector.

Conclusion
The anxiety that permeates from premature deindustrialization and deindustrialization, in general, is not rooted in descriptive variations observed in the data. Many observers and policymakers contend industrialization to be an engine of growth which creates relatively higher paying jobs. Indeed, there is evidence that manufacturing plays a critical role in the catch-up process as it exhibits an unconditional convergence in labor productivity unlike the other sectors of the economy (Rodrik, 2012). A fallout of this hypothesis is that policy should be placed to arrest the slide in industry. However, for the policy to be effective, it is crucial to understand what underlying factors are responsible for the divergent paths of transformation. Rodrik (2016) hints that trade can explain deindustrialization faced by many developing countries as they lose manufacturing share opening their borders because of their comparative disadvantage in the sector. In this paper, I find limited evidence of the comparative advantage hypothesis. The accounting exercises show that even if the Latin American economies counterfactually move to mimic the trade shares of countries that have a large share of their workforce in industry and manufacturing, they will realize modest gains in their shares of industrial and the manufacturing employment. More important than trade shares are the differences in sectoral linkages which account for more than 30 percent of the industrial employment gap. Integrating linkages in theories of structural transformation seem promising in getting closer to matching the data, as standard theories are not able to account for some essential trends (Buera & Kaboski, 2009).
There is growing interest in this direction with many studies investigating the role of variations in linkages across countries and time in the process of structural transformation, and economic growth in general (Berlingieri (2014), Bartelme & Gorodnichenko (2015), Fadinger et al. (2018), Sposi (2018) etc.). In a related paper, using an endogenous model of sectoral linkages, I study the distortions in intermediate markets in the same countries and find that changes in distortions can deliver quantitatively meaningful expansion of the industrial base (Sinha, 2018). 10 Comparing Latin American economies to Korea and Taiwan shows that the low growth of productivity in the region is pervasive across sectors and is pivotal in accounting for the difference in growth experience of the region compared to the Asian miracles since the 1960s. Nonetheless, differences in productivity growth in manufacturing and wholesale are especially relevant (Üngör, 2017). If the manufacturing sector in the Latin American economies was to experience the high growth in productivity as realized by Korea and Taiwan, the sector would have lost a larger share of employment during the period according to the price-effect mechanism.
Still, economic growth would be higher in such a scenario. Thus, the pertinent issue regarding policy making is to identify if the present allocation truly reflects inefficiencies. An even more important question is whether policy interventions are needed to affect the distribution of sectoral activity if indeed growth is the primary concern. While it may be possible to influence sectoral allocations, doing so may create distortions restricting the efficient flow of resources across sectors and countries, hence generating a drag on economic growth. The bars represent the actual change in employment shares. The symbols depict the counterfactual change in employment shares when the variables pertaining to a factor are changed, keeping all the other variables fixed at the 1995 levels. All figures in percentage points.

A.1 Accounting Framework
In this appendix, I outline how the two equations 2 and 3 used in the accounting framework are derived. Recall that all varieties of a sector j are tradeable and are used to produce the composite non-tradeable sectoral good Q j . In what follows, I add subscripts p, q for countries and update the notation accordingly. For example, Q jp represents the quantity of composite good of sector j in country p. Suppose that the expenditure share of sector j in country p from varieties from country q is given by π jpq . Then the receipts of all varieties of firms of sector j in country p are given by The composite good is used either for final consumption or in the domestic production of varieties in each sector. Hence, the market clearing requires where M kjp denotes the quantity of composite good j used in the production of all varieties of sector k in country p. Then, P jp M kjp is the total expenditure which can be written as where β kp is the value-added share and µ kjp is the intermediate share of sector j in the production of varieties in sector k of country p. Multiplying both sides of equation 6 by P jp and using the above relationship yields The term ∑ q =p (P kq Q kq π kqp − P kp Q kp π kpq ) is essentially the net exports of country p in sector k. Also, using C jp for the consumption expenditure P jpCjp , the following ties gross output to consumption expenditures, net exports and the share of labor and intermediates in sectoral production The payments to labor in sector j in country p are given by w jp L jp , Then, as in the case of I payments to intermediates Substituting, the above equation in 6 and recalling that value-added equals payments to labor yields Finally, denoting the value-added share of a sector j as v jp = w jp L jp ∑ k w kp L kp ≡ τ jp wL jp ∑ k τ kp wL kp , the condition that ∑ k l jp = 1 can be used to recover labor shares l jp from the value-added shares as represented in equation 3.   The first row reports the actual gap in employment shares. The following rows report the implied gap remaining when the variables pertaining to the listed factor is counterfactually changed to a comparator country, keeping other variables unchanged. Figures represent averages across the comparator set countries. All figures in percentage points.  The first row reports the actual gap in employment shares. The following rows report the implied gap remaining when the variables pertaining to the listed factor is counterfactually changed to a comparator country, keeping other variables unchanged. Figures represent averages across the comparator set countries. All figures in percentage points. The first row in each panel reports the actual change in employment shares during 1995-2010/11. The following rows indicate the counterfactual change in employment shares when the variables pertaining to a factor are changed, keeping all the other variables fixed at the 1995 levels. All figures in percentage points.