Effects of innovation on employment in Latin America

Abstract

This study examines the impact of process and product innovation on employment growth and composition in Argentina, Chile, Costa Rica, and Uruguay using micro data from innovation surveys. Employment growth is related to process innovations and to the growth of sales separately due to innovative and unchanged products. Results show that compensation effects are pervasive and that the introduction of new products is associated with employment growth at the firm level. No evidence of displacement effects due to the introduction of process innovations was observed. With respect to the impact of innovation on employment composition, there is scant evidence of a skill bias, although the product innovation is more complementary to skilled than to unskilled labor.

1. Introduction

Innovation is widely considered to be a primary source of economic growth, and policies to encourage firm-level innovation are growing in importance on the agenda in most Latin American countries. But innovation alone may not be sufficient to generate employment. And for countries facing labor market problems, persistent poverty, and inequality,¹ employment generation is probably the main route out of poverty and the most efficient way to reduce inequality. Thus, the effects of innovation on employment are of particular interest.

The relationship between innovation and employment is complex. Innovation could trigger direct (mainly firm-level), partial, and general equilibrium effects on employment, and across all these levels the relationship between these variables depends on many different transmission mechanisms, feedback loops, and institutional factors (Pianta, 2005; Vivarelli, 2011). Recent evidence regarding the firm-level relationship between innovation and employment in developed economies indicates that whether and how innovation creates new jobs depends first and foremost on the type of innovation (Harrison et al., 2014; Hall et al., 2008; Lachenmaier and Rottmann, 2011) and the sector (Greenhalgh et al., 2001; Coad and Rao, 2011; Bogliacino et al., 2011). In addition, the effects of innovation on innovators’ employment depend on the state of the technology that determines the extent to which innovation improves productivity and the demand conditions that induce different compensating effects.² At the sector level, innovation can also trigger indirect effects, including the competitive redistribution of outputs and jobs from low to high innovation-intensive firms, job losses due to the exit of non-innovative firms, and job creation from innovative spin-offs. For instance, Greenan and Guellec (2000) find that although innovating firms create more jobs than non-innovating ones, the reverse is true at the sectoral level. Finally, general equilibrium effects clearly emerge when the interactions between different markets are considered. Indeed, how fast innovators can meet increased demand depends in part on how fast complementary inputs produced by other industries can be supplied. Innovation can also affect employment through complementarities in consumption goods and increased variety or better quality of intermediate inputs. Finally, new products could lead to completely new economic activities (Doms et al., 1995; Klette and Forre, 1998; Spiezia and Vivarelli, 2002; Pianta, 2005; Harrison et al., 2014).

The evidence on the relationship between innovation and employment for Latin America is scant where the very idiosyncratic nature of innovation means that the abovementioned findings cannot be simply extrapolated to this region.³ Indeed, for Latin American firms, the acquisition of technological knowledge from abroad through contacts, trade, collaborations, and joint ventures with industrialized countries is very relevant (Katz, 1987). Technological change in developed countries might respond to different objectives, incentives, and factor endowments as well as go in different directions from technological change in developing countries. Innovations borrowed from developed countries may not be fully adaptable to developing country contexts and may produce different effects on employment than locally developed innovations. Thus, it is not only that Latin American firms might produce different types of innovations (based on imitation of the best-practice frontier rather than being the first to introduce world-class innovations) but also that the very nature of the innovation process is different. Consequently, the effects of innovation on employment generation in this region might be quite different.

Furthermore, in Latin America, there are important structural features that might lead to different outcomes of innovation on employment. In the first place, the firm population is strongly dominated by small- and medium-sized enterprises (SMEs), accounting for a sizable proportion of employment and value-added. The innovation process is very different in SMEs than in large firms. Indeed, innovation in SMEs is strongly dominated by informal search routines and learning from already available knowledge and technologies, while in large firms, innovation processes are more systematic and tend to be formalized in R&D labs (Baldwin, 1997). Thus, the typical business innovation strategy observed in Latin America is quite different from that which is dominant in frontier economies. Second, Latin America’s production structure is heavily dominated by the manufacturing of commodities and low technologically intensive goods. To the extent that in these sectors the dominant innovation strategy is more related to process than product innovations, the expected effects of innovation on employment might be different.⁴

This article aims at closing the evidence gap on the effects of innovation on employment growth at the firm level in Latin America by using innovation surveys for four Latin American countries: Argentina, Chile, Costa Rica, and Uruguay. Specifically, this article will highlight the existing relationship between innovation outputs and employment growth and its effects on skill composition, taking into account size and sector differences. Firm-level data enable the innovation process to be accounted for and related to the firm’s employment trends.⁵

The rest of this article is organized as follows. Section 2 presents the relationship between innovation and employment. Special emphasis is placed on explaining potential identification problems and the need to implement instrumental variable (IV) estimation techniques to obtain consistent results. Section 3 describes the sources of the data used and presents the main characteristics of the firms’ behavior in the four countries under study. Section 4 presents the results on employment growth and uses these results to decompose the different effects of innovation on employment growth. Section 6 presents the relationship between innovation output and employment composition in terms of skills. Section 7 offers conclusions.

2. Relationship between innovation and employment generation

2.1 Type of innovation and compensation mechanisms

Recent evidence on the firm-level relationship between innovation and employment in developed economies indicates that whether and how innovation creates new jobs depends first and foremost on the type of innovation (Harrison et al., 2014).⁶ Specifically, the effects of innovation on employment (quantity) depend on the relative intensity of the displacement and compensation effects that it might induce. The introduction of new processes is generally driven by labor cost considerations and tends to reduce labor (i.e., displacement). At the same time, the introduction of new products or services may replace or add to the list of existing products or services with different effects on employment generation (see Figure 1).⁷ Overall, the economic theory states that technological unemployment is a temporary circumstance, which can be automatically compensated through various mechanisms such as: (i) additional employment in the production of machines and capital goods, (ii) decreases in prices resulting from lower production costs on account of technological innovations, (iii) new investments channeling the extra profits caused by technological change, (iv) decreases in wages resulting from price adjustment mechanisms, (v) increases in income resulting from redistribution of gains from innovation, and (vi) new products created using new technologies. It is important to state that the compensation mechanism via decrease in prices, to work properly, needs to counterbalance the reduction in aggregate demand associated with workers dismissal. A few conditions should be met to secure this effectiveness: significant price elasticity for those products affected by the price reduction, relevance of these products on the consumption bundle, and non-oligopolistic market structures. Limited validity of such conditions can result in unchanged—or even reduced—aggregate demand. A similar framework would also apply to the wage reduction channel. However, this mechanism is additionally based on the principle of factor substitution. Even new products may displace older products and so imply a weaker impact in terms of job-creation.⁸

Figure 1.

Employment effects of innovation.

Source: Adapted from Harrison et al. (2014).

In particular, low demand and capital/labor substitution elasticities, attrition, pessimistic expectations, and delays in investment decisions may involve that compensation can only be partial (Meschi et al., 2016). Organizational innovation is frequently an indispensable complement to the adoption of new technologies critically affecting the productivity and employment consequences of technological innovation, especially Information and communications technologies (ICT) (Black and Lynch, 2004). Evangelista and Vezzani (2012) enrich previous analyses by studying the role of organizational innovation in firm-level employment dynamics. By exploiting cross-sectional Community Innovation Survey (CIS), IV data for a selected number of European countries show that process innovation does not show any direct negative effect unless combined with organizational innovation in manufacturing.

Overall, the empirical literature on the subject has pointed out that product innovation tends to be labor friendly, while process innovation reveals to be labor-saving; moreover, the job-creating effect of innovation is far more obvious in high-tech sectors and new services rather than in low-tech manufacturing and traditional services (for recent studies, see Bogliacino and Pianta, 2010; Lachenmaier and Rottman, 2011; Bogliciano, Piva and Vivarelli, 2012; Bogliacino et al., 2012).

It should be noted that the discourse on compensation mechanisms and their functioning has often taken place within the context of developed countries. Therefore, the validity of the compensation theory becomes more questionable in developing countries contexts, where process innovations dominate product innovations and where mature manufacturing sectors and traditional services represent the bulk of their economic structure. However, product innovations may reveal their labor-friendly nature in developed countries as well (see Mitra and Jha, 2015).

Harrison et al. (2014) show that to untangle the employment-creating versus displacing effect of innovation, a distinction between product and process innovation is useful. This research will take the same starting point. In the basic model, two types of products are distinguished: the production of existing products and the production of new products. The change in employment is then decomposed into the part due to the increased efficiency in production of old products (which could be related to process and organizational innovations) and the part due to the introduction of new products (product innovations). Hence, it is possible to capture the relative extent of the expansion and displacement effect of innovation on employment, as follows.

We assume that a firm can produce two types of products: “old products” and “new products.” Outputs of old and new products at time t are denoted $Y_{1t}$ and $Y_{2t}$⁠, respectively. We observe firms at two points in time, at the beginning (t = 1) and at the end of the period (t = 2). We also assume that each type of product is produced with an identical separable technology production function, with constant returns to scale in capital and labor. Each production technology has an associated efficiency parameter $-{\theta }_{it}$– that change over time. New products can be produced with higher or lower efficiency than old products, and the firm can influence the efficiency of production of either product through investments in process innovation. The firm’s cost function at time t can then be written as follows:

$$C(w_{1t},w_{2t},Y_{1t},Y_{2t},{\theta }_{1t},{\theta }_{2t})=c(w_{1t})\frac{Y_{it}}{{\theta }_{1t}{e}^{\eta +{\omega }_{it}}}+c(w_{2t})\frac{Y_{2t}}{{\theta }_{2t}{e}^{\eta +{\omega }_{2t}}}$$

(1)

Where $c(w)$ is a function of input prices. In the production of each good, we assume that firm productivity levels are affected by unobservable firm-specific fixed effects (η) and idiosyncratic shocks (ω’s). The η captures all of the factors that make the firm more productive but remain constant over time (superior innovation management skills, location, etc.), while the ω’s captures time varying shocks that affect production (energy shocks, labor disputes, unexpected organizational problems, etc.). According to the Shephard’s Lemma, the conditional demand for labor in the production of each product is as follows:

$$L_{it}=c_L(w_{it})\frac{Y_{it}}{{\theta }_{1t}{e}^{\eta +{\omega }_{it}}}$$

(2)

where $c_L(w)$ is the derivative of $c(w)$ with respect to wages. Under the assumption that $c_L(w)$ remains constant over the period and that it is the same for old and new products,⁹ the growth rate of employment at the firm level is given by the growth rate of employment allocated to the production of old products plus the growth rate of the employment allocated to the production of new products. Given that the production of new products at the beginning of the period is nil (⁠$Y_{21}=0$⁠), we can approximate the employment growth decomposition as follows:¹⁰

$$l=\frac{\Delta L}{L}=-\left(\frac{{\theta }_{12}-{\theta }_{11}}{{\theta }_{11}}\right)+\left(\frac{Y_{12}-Y_{11}}{Y_{11}}\right)=\frac{{\theta }_{11}}{{\theta }_{22}}\frac{Y_{22}}{Y_{11}}-\left({\omega }_{12}-{\omega }_{11}\right)$$

(3)

This expression says that employment growth is the result of the change in efficiency in the production process for the old products, the rate of change in the production of these products, and the expansion attributable to the new products. The increase in the efficiency of the old products’ production process is expected to be larger for firms introducing process innovations related to the old products (firms that introduce process innovations only, according to the surveys). On the other hand, the effect of product innovation on employment growth depends on the difference in efficiency between the production processes for the old and the new products. If the new products are produced more efficiently than the old products, then this ratio is less than 1 and employment does not grow at the same pace as the growth of output accounted for by new products. Equation (3) suggests the following regression to estimate the effects of innovation on employment:

$$l={\alpha }_0+{\alpha }_{1}d+y_1+\beta y_2+\upsilon$$

(4)

where l is total employment growth, y₁ is the real growth in sales of old products, y₂ is the real growth in sales of new products (product innovations), and d captures the introduction of process innovations in the production of old products. The error term, ν, captured the productivity shocks In general, one should expect that while innovations in the production processes of old products tend to displace employment, product innovations tend to create employment (unless new products substitute for old products, and the production efficiency of new products is higher than that of the old products).¹¹

It is clear from the preceding discussion that the effects of innovation on employment depend on the type of innovation carried out by the firms. Given that the type of innovation can differ considerably across sectors, it is natural to assume that the effect of innovation on employment can also differ by sector. In addition, labor market regulations can have different effects depending on firm size. Large firms can circumvent labor rigidities by outsourcing part of their work, while this is more difficult for small firms. On the other hand, in the Latin American context, small firms are also more informal. Thus, in principle, labor regulations might be less binding on them. This heterogeneity can have relevant policy implications, and it is therefore important to pay attention to it. Henceforth, throughout the article we explore the heterogeneity of the impact of innovation on employment by size, estimating equation (4) separately for small firms and low- and high-tech sectors.¹²

2.2 Identification issues, causality, and measurement errors

Identification estimation of equation (4) can be affected by two different problems: the potential endogeneity of the innovation variables and the measurement error problem generated by using nominal sales rather than real sales among the regressors. With regard to the endogeneity problem, consistent estimation of equation (4) relies on the lack of correlation between the variables representing process and product innovations and the error term. Innovations are the result of investment decisions (R&D, e.g.), which have to be decided by the firms in advance. These decisions depend on firm’s productivity, which can be characterized as an unobservable made of two components: firm attributes that are mainly constant over time (such as managerial skills or the η’s in our previous notation) and productivity shocks (the ω’s). Hence, if innovation investments are correlated with firm productivity, innovation outputs will be as well. This will lead, in turn, to innovation outputs being endogenous, creating a serious problem of identification.

Given that equation (4) is specified as a real growth rate, it is expected that the influence of firm-specific fixed effects is already removed from the error term. The correlation between innovation outputs and productivity shocks [they remain in the error term of equation (4)] depends on the exact timing of investment decisions. If investment decisions are made in advance of productivity shocks (e.g., because there is a “time-to-build” period between when investments decisions are made and actual innovations materialize), innovation variables in equation (4) will not be correlated with the error term, and equation (4) could be estimated by Ordinary Least Squares (OLS) methods.¹³ If, on the other hand, investment decisions are made at the same time as productivity shocks are observed, innovation outputs might become endogenous in equation (4).

In this case, it is worth exploring the potential direction of this bias. If process innovation (d) is positively correlated with the productivity shock of old products in the second period (ω₁₂), the fact that this shock enters the error term in equation (4) preceded by a negative sign means that the correlation between process innovation and the error term will be actually negative. So, OLS results will tend to overestimate any displacement effect or to underestimate any compensating effect of process innovation. On the other hand, for product innovation, we expect a negative correlation between this variable and the error term as well because while (ω₁₁) shows up in the error term with a positive sign, (Y₁₁) shows up in the denominator of this equation. That means that estimating equation (4) using OLS will underestimate the true impact of product innovation on employment growth. In summary, the OLS-estimated impacts of innovation on employment growth should be interpreted as the “lower bounds” of the true relationship among these two variables.

The identification of the true relationship will depend on the availability of instruments correlated with the innovation variables and uncorrelated with the error term. Although the innovation surveys provide interesting information that can be used as instruments, the majority of them are actually more suitable for the identification of product than process innovation, which is a relatively more idiosyncratic outcome. It is important to consider that the majority of the firms in the sample that report having introduced a product innovation have done so in combination with a process innovation (a phenomenon known as “co-innovation” in the literature). In the empirical implementation, these firms will be considered product innovators. The proportion of firms introducing process innovation only is fairly small, so even when the results for this variable are downward biased, the actual impact of this problem on employment growth is expected to be of second order. Thus, in the empirical implementation, the focus will be on getting reliable estimates for product innovation, maintaining the assumption that process-only innovation is fairly exogenous.¹⁴

A second possible source of endogeneity is the presence of measurement error. Ideally, in equation (4) we would like to have the real production growth of old (y₁) and new products (y₂). Instead we must replace these two variables with nominal sales growth rates (g₁ and g₂) because we do not have firm-level prices. For both products, nominal sales growth rates can be approximately decomposed into two terms: real growth rates and price changes. In other words, we have: g₁ = y₁ + π₁ for old products and g₂ = y₂ + π₂ for new products. Substituting these two expressions in equation (4) and moving the nominal sales growth rates of old products to the left hand side, we have:

$$l-g_1={\alpha }_0+{\alpha }_{1}d+\beta g_2+\left(-{\pi }_1-\beta {\pi }_2+\upsilon \right)$$

(5)

Hence, the growth in prices of old and new products is left in the error term, and correlation between growth in prices of new products π₂ and g₂ can create an additional bias for product innovation, As before, this will also be an attenuation bias in the estimation of β when estimated using the OLS method. To deal with this measurement error problem, we follow Harrison et al. (2014), and we use IVs that are correlated with real growth in the production of new products but uncorrelated with its nominal growth.

According to Harrison et al. (2014), the use of nominal sales growth rates will also affect the interpretation of the results for process innovation. The price growth rates of old products can itself be affected by the efficiency impact of process innovation, to the extent that these efficiency gains are expressed in the prices. In other words, it is possible that π₁ = π₀ + γα₁d, where γ is a pass-through parameter that exists in the interval [0, 1]. So, by replacing this in equation (5) we get:

$$l-g_1={\alpha }_0+\left(1-\gamma \right){\alpha }_{1}d+\beta g_2+\left(-{\pi }_1-\beta {\pi }_2+\upsilon \right)$$

(5.1)

In the absence of firm-level data, true displacement effect of process innovation might also be underestimated. The severity this problem will depend on the size of the pass-through effects. If these effects are large, with γ∼1, we might end up getting a non-significant effect of process innovation on employment growth. To partially correct for this problem, we follow the strategy developed by Harrison et al. (2014), which consists of approximating firm-level prices (π₁) by using industry-level deflators (π). Thus, the estimate is:

$$l-(g_1-\pi )={\alpha }_0+{\alpha }_{1}d+\beta g_2+\left(-\left({\pi }_1-\pi \right)-\beta {\pi }_2+\upsilon \right)$$

(6)

If firm-level prices do not deviate much from industry-level deflators (π∼π₁), we might be able to obtain more consistent estimations of the displacement effect of process innovation of employment.

2.3 Innovation and employment quality

It is also important to investigate the qualitative effect of technological change on different categories of workers. The basic premise here is that innovations are skill-biased and, therefore, its impact may be different for skilled and unskilled workers. If innovation is skilled-biased, as several empirical and theoretical studies argue (see Acemoglu, 1998; Card and DiNardo, 2002), innovation tends to replace tasks traditionally carried out by unskilled workers with new jobs demanding skilled workers. Hence, higher innovation could be associated with lower employment growth for unskilled workers and higher employment growth for skilled workers.

Indeed, literature, focusing on the complementarity between technological change and skilled labor, has put forward the so-called “skill biased technological change” hypothesis (SBTC) which supports the view that new technologies require suitable skills to be implemented effectively and efficiently. Consequently, the availability of qualified workers may act as a constraint, which limits the adoption and diffusion of new technologies, on the one hand, and the pursuit of full employment, on the other hand. This is the so called “human resource constraint” (Amendola and Gaffard, 1988). In other words, in the presence of labor-saving and skill-biased process innovation, the scarcity of skilled labor can easily generate unemployment among unskilled workers, unless proper retraining policies are put in place. The concept of SBTC, first developed by Griliches (1969) and Welch (1970), is based on the hypothesis of capital-skill complementarity, and suggests that employers’ increased demand for skilled workers is driven by new technologies that are penetrating into modernized industries, and which only workers with a higher level of skill can operate (see Machin, 2003; Piva and Vivarelli, 2009).

The literature on SBTC remains mainly empirical, where many studies indicate that SBTC has gained momentum during the past three decades due to the surge in information technology and diffusion in computers (Pianta, 2005). The first to explore SBTC empirically were Berman et al. (1994) who analyzed manufacturing in the United States during the 1980s and related the employment shift in favor of skilled workers to the investments in computers and R&D.

Autor et al. (1998) extended Berman et al. (1994) over a period spanning almost half a century and included non-manufacturing sectors. Autor, Katz, and Krueger confirmed the complementary relationship between investment in computers and the skill structure. They show that the spread of computer technology in the United States since 1970 can in fact explain as much as 30–50% of the increase in the growth rate of relative demand for skilled labor.

Empirical studies supporting SBTC were conducted for several other OECD countries, such as, for example, UK (see Machin 1996, Haskel and Heden, 1999), France (see Goux and Maurin, 2000; Mairesse et al., 2001), Germany (Falk and Seim, 1999; Falk, 2001), Italy (Casavola et al., 1996; Piva and Vivarelli, 2004; Piva et al., 2005, 2006; Baccini and Cioni, 2010), and Spain (see Aguirregabiria and Alonso-Borrego, 2001; Luque, 2005). Additionally, Machin and Van Reenen (1998) provide evidence of SBTC in a cross-country study on seven OECD. Overall, empirical studies outside North America reveal results that are generally consistent with the SBTC hypothesis, albeit sometimes less strikingly.

Regarding developing countries, technology (both new and used) imports and FDI inflows may generate technological spillovers in favor of domestic firms which can absorb the new imported technologies that may involve productivity gains that can be harmful to employment levels. In particular, the role of imported machinery and equipment, implying labor-saving process innovation, can drastically decrease domestic demand for labor. In this respect, there is growing evidence (although still mixed) on the diffusion of SBTC. Feenstra and Hanson (1999) and Hanson and Harrison (1999) found a positive effect of FDI and technology acquisition on the hiring of more skilled workers in Mexico. Görg and Strobl (2002) revealed that the acquisition of foreign machinery raised the relative demand for skilled labor of manufacturing firms in Ghana over the 1990s. Giovannetti and Menezes-Filho (2006) and Fajnzylber and Fernandes (2009) found similar results in a cross-section of Brazilian firms. For Chile, while Pavcnik (2003) failed to find a significant relationship between foreign technology and the demand for skilled workers, Fuentes and Gilchrist (2005) find a robust positive association between those two.

To analyze the effect of innovation on employment composition, we follow the earlier approach and use a variation of equation (6) for assessing the innovation impact on employment quality. Specifically, we split the growth rate of employment in both skilled (l_s) and unskilled workers (l_us).

$$l_{it}^s-(g_{1it}-\pi )={\alpha }_0^s+{\alpha }_1^s{d}_{it}+{\beta }^s{g}_{2it}+{\epsilon }_{it}$$

(7)

$$l_{it}^{us}-(g_{1it}-\pi )={\alpha }_0^{us}+{\alpha }_1^{us}{d}_{it}+{\beta }^{us}{g}_{2it}+{\eta }_{it}$$

(8)

The dependent variable is employment growth (minus old product real sales growth) for both types of workers: skilled and unskilled. In doing so, ls can be estimated as the rate of growth of the sum of employees with technical and professional education, while lus is the rate of growth of the sum of employees with only basic education or less. In short, through equations (7) and (8), we can assess the extent to which innovation, both in process and product separately, affects employment generation when we consider employment quality and not only total employment. Once again, we will use IVs as discussed before to address the identification problem related to correlation between d and g₂ and the error term.

3. Data sources

The models just described are run for four Latin American countries with a focus on the manufacturing industry. Innovation surveys used were Argentina (2010–2012), Chile (1995, 1998, 2001, 2005, and 2007), Costa Rica (2006–2007), and Uruguay (1998–2000, 2001–2003, 2004–2006, and 2007–2009). The findings presented here are the results of a collaborative project in which a team of researchers from each of these countries implemented a common empirical model. A series of national studies have been conducted in parallel by local researchers to fully exploit the richness of each individual survey.¹⁵

In the case of Argentina we use data from Encuesta Nacional de Dinámica del Empleo y la Innovación (ENDEI), Nation Survey of Firm Dynamics and Innovation (2010–2012) conducted jointly by the Ministry of Science, Technology and Productive Innovation (MINCYT) and the Ministry of Labor and Employment. The survey is based on a statistically representative sample of manufacturing collected information from 3433 firms.

In the Chilean case, there are several waves of the innovation survey available for studying the issue in question. We use the innovations surveys carried out in 1995, 1998, 2001, 2005, and 2007. This information is complemented by firm-specific information obtained from the Annual Survey of Manufactures (ENIA). This link between the two sources of information is relevant given that innovation surveys tend to have limited information on firm characteristics.

For Costa Rica, the main source of data used in the study is the Costa Rican Innovation Survey for the years 2006–2007. This survey is based on a statistically representative sample of the manufacturing, energy, and telecommunications sectors. According to the official data of the National Institute of Statistics and Census (INEC), these sectors comprised a total of 2285 firms. In the case of the 2006–2007 survey, the INEC provided a sample of 566 firms distributed over all sectors. Using this sample, it was possible to obtain complete responses from 376 firms. After eliminating firms from energy and telecommunications and any manufacturing firms with fewer than 10 employees, we ended up with a sample of 208 firms.

Finally, the data on Uruguay were derived from four waves of Manufacturing Innovation Surveys: 1998–2000, 2001–2003, 2004–2006, and 2007–2009, as well as the annual Economic Activity Surveys (EAS) for the period 1998–2007. The innovation survey data are collected by the National Bureau of Statistics (INE) in parallel with the EAS (same sample and statistical framework). In the case of the innovation survey, all firms with more than 49 workers are included. Units with 20–49 employees and with fewer than 19 workers are selected using simple random sampling within each economic sector at the ISIC two-digit level up to 2005. Since then, random strata are defined as those units with fewer than 50 workers within each economic sector at the ISIC four-digit level.

When comparing results across countries, we need to bear in mind that business, economic, and policy environments in Latin America differ between countries and generally diverge from OECD countries; thus, in principle, the results are not expected to be similar to those reported in previous studies. Finally, as this is an analysis of the manufacturing industry, which represents a small share of the total economy in some countries (IDB, 2010; Tacsir, 2011), the results apply only to this industry. We acknowledge, however, that innovation is relatively more important in manufacturing than in service industries (Crespi and Zuñiga, 2012).

3.1 Descriptive statistics

Innovation surveys contain detailed information on the firms’ characteristics, innovation activity, and employment—both the number of employees and employment composition by education and length of the labor contracts. Importantly, they contain detailed information on the composition of sales, which allows us to compute the percentage of sales corresponding to new products and from this the nominal growth rate of new products (g₂).¹⁶

Firms were classified in mutually exclusive categories according to their innovation status: product innovators, process only innovators, and non-innovators. Product innovators are firms that have introduced product innovations. Process-only innovators are firms that have introduced process innovations or organizational change innovations excluding product innovators. Non-innovators are firms not classified as product or process innovators. Following Harrison et al. (2014), we classify firms that have introduced both product and process innovations as product innovators. The implicit assumption is that product and process innovators are more similar to product innovators than to process-only innovators.

Table 1 presents the share of innovative firms, employment growth, sales growth, and labor productivity in the four countries under study. Table 1 shows the high proportion of innovative firms in the region. In fact, the proportion of innovative firms ranges from 51.9% (in Uruguay) to 78% (in Costa Rica). Among them, more than half of the innovators have introduced product innovations.

Table 1.

Process and product innovators, growth of employment and sales

Size	Argentina		Chile		Costa Rica		Uruguay
	All firms	Small	All firms	Small	All firms	Small	All firms	Small
Number of observations	3433	1533	2049	652	208	119	2532	1353
Distribution of firms (%)
Non-innovators (no process or product innovations)	37.3	49.1	42.6	67.5	22.0	29.4	48.1	62.2
Process only innovators (non-product innovators)	8.7	5.6	4.0	3.7	4.0	5.9	19.4	14.0
Product innovators	53.9	45.3	53.4	28.8	74.0	64.7	32.5	23.7
Number of employees at the beginning of (each) survey	76.1	16.1	214.5	26.1	182.0	25.7	91.2	26.2
Foreign ownership (10% or more) (%)	9.0	1.2	12.5	5.8	14.9	6.7	13.2	6.2
Located in the capital of the country (%)	NA	N.A.	52.0	48.2	57.7	63.9	81.0	76.7
Employment growth (%) (yearly rate)
All firms	2.2	2.2	−0.2	1.4	3.3	3.6	−0.7	−3.7
Non-innovators (no process or product innovations)	−2.5	−3.0	0.8	1.5	3.5	3.7	−3.4	−5.3
Process only innovators (non-product innovators)	2.4	7.5	2.1	4.5	7.4	5.4	1.7	−1.5
Product innovators	5.5	7.2	−0.5	1.8	3.0	3.3	1.8	−1.0
Sales growth (%) (nominal growth) (yearly rate)a
All firms	22.7	23.5	6.5	5.3	23.7	20.0	5.5	3.6
Non-innovators (no process or product innovations)	21.3	21.8	2.9	3.1	27.3	23.1	1.7	1.2
Process only innovators (non-product innovators)	21.8	24.5	7.1	7.6	11.7	12.8	9.6	9.4
Product innovators	23.8	25.1	8.5	8.5	23.7	19.3	8.7	6.4
of which:
Old products	11.3	8.4	−5.7	−5.7	−54.9	−46.1	−21.2	−25.1
New products	40.5	40.8	14.2	14.2	78.6	66.1	29.9	31.5
Labor productivity growth (%)a(yearly rate)
All firms	20.6	21.2	6.7	3.9	20.5	16.5	6.2	7.3
Non-innovators (no process or product innovations)	24.0	24.9	2.1	1.6	23.8	19.4	5.1	6.5
Process only innovators (non-product innovators)	19.6	16.9	4.9	3.1	4.3	7.4	7.9	10.9
Product innovators	18.4	21.3	9.0	6.6	20.4	16.0	6.9	7.4
Prices growth (%)b
All firms	10.9	11.2	5.0	3.6	14.3	13.5	6.8	7.7
Non-innovators (no process or product innovations)	10.9	11.0	3.8	1.6	14.1	14.1	6.8	7.4
Process only innovators (non-product innovators)	10.7	10.5	5.0	3.3	11.8	9.7	6.8	9.0
Product innovators	11.1	11.4	5.6	6.4	14.6	13.6	6.8	7.7

Source: Authors' elaboration based on country studies.

Notes: Argentina (AR)-Innovation Survey 2012–2012; Chile (CH): pooled regressions for the innovation surveys 1995, 1998, 2001, and 2007; Costa Rica (CR): Innovation survey 2006–2007 Uruguay: pooled regressions for the surveys 1998–2000, 2001–2003, 2004–2006, and 2007–2009. Product innovators are firms that have introduced product innovations. Process-only innovators are firms that have introduced process innovations or organizational change innovations excluding product innovators. Non-innovators are firms not classified as product or process innovators. Sample: Firms with information in all the relevant variables for the empirical analysis.

a

Sales growth for each type of firm is the average of the variable g, and averages for old and new products are the averages of variables g₁ and g₂, respectively, and

b

prices computed for a set of industries and assigned to firms according to their activity.

Despite the difference in overall performance across countries, it is evident that innovators perform better than non-innovators in terms of employment creation. Although less clear, a similar situation arises in the case of sales. Innovators exhibit better sales performance with the exception of Costa Rica, where process innovators show smaller growth rates than non-innovators (partially due to a small growth rate in prices).

In summary, the results in Table 1 suggest that employment growth by innovators is higher than by non-innovators.¹⁷ The results for product and process innovation are remarkably similar, and there is no strong a priori preliminary evidence that process innovation is particularly harmful for employment growth. Thus, it seems that compensating effects for process innovations are prevalent in the model. With the exception of product innovators in Argentina, we observe that the sales growth rates of old products are always negative but more than compensated for by the growth in sales of new products. In the following sections, we explore further the relative impacts of process and product innovations by estimating several variants of the model outlined in the previous section.

4. Econometric results of innovation on employment

In this section, we present several results of the effects of innovation on employment growth at the firm level. We begin by presenting the results of the OLS followed by the IV estimation of equation (6). The section closes with a decomposition exercise as in Harrison et al. (2014). Overall, we analyze the extent to which these effects are different for all manufacturing firms, small firms, and low- and high-tech industries.

4.1 Innovation and employment growth estimates

Columns 1––4 in Table 2 show the OLS results for the innovation–employment model as outlined in equation (6) for the case of all manufacturing firms. The panel to the right in the same Table exhibits the results for the same model for the case of small and medium enterprises (SME). SME’s are defined in the four countries as enterprises with less than 50 employees. For each country and sample, Table 2 shows the results after controlling for additional variables to those presented in equation (6), namely, foreign ownership and whether the firm is located in the capital region of the country.

Table 2.

Employment growth. All manufacturing and small manufacturing firms. Dependent variable: l - (g₁ - Π)–OLS estimation

Sector	All firms in manufacturing				Manufacturing small firms
Regression	AR-OLS	CH-OLS	CR-OLS	UY-OLS	AR-OLS	CH-OLS	CR-OLS	UY-OLS
Constant	−6,495***	1.997**	−1.616	2.662***	2.739	3.136**	−0.845	1.757**
	(0.809)	(0.825)	(5.241)	(0.555)	(1.685)	(1.326)	(6.650)	(0.775)
Process only innovator (d)	−2,751	−2.780**	8.175	−4.002***	−2.489	−3.346	5.726	−4.127**
	(1.978)	−1.275	(6.539)	(1.06)	(2.425)	(2.717)	(8.770)	(1.686)
Sales growth due to new products (g2)	0.654***	0.833***	0.887***	0.853***	0.963***	0.706***	0.932***	0.826***
	(0.063)	(0.034)	(0.042)	−0.018	(0.03)	(0.084)	(0.059)	(0.028)
Located in the capital	NA	−0.13	0.950	NA	3.990*	3.064	10.083	NA
		(1.231)	(5.161)		(1.9)	(4.309)	(8.525)
Foreign owned (10% or more)	2.281	−0.275	6.672*	1.655	−3.441	−2.318	2.112	−3.048
	(1.523)	(0.889)	(3.884)	(1.181)	(3.682)	(1.718)	(5.949)	(2.51)
Two-digit industry dummies	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Time dummies	No	Yes	No	Yes	No	Yes	No	Yes
R-squared	0.170	0.27	0.632	0.441	0.785	0.163	0.615	0.369
Number of firms	3433	2049	208	2532	1415	652	119	1353

Source: Authors' elaboration based on country studies.

Notes: Product innovators are firms that have introduced product innovations. Process innovators are firms that have introduced process innovations or organizational change innovations. Product-only innovators are firms that are product innovators but not process innovators. Process-only innovators are firms that are process innovators but not product innovators. Product or process innovators are firms that are product innovators or process innovators. Product and process innovators are firms that are both product innovators and process innovators.

Robust standard errors in parentheses. Significance level: *** 1%, ** 5%, and * 10%. In the case of Uruguay and Argentina was not possible to differentiate whether the firm was located in the capital region.

As described earlier, a negative coefficient in the process innovation only indicator represents an additional increase in productivity in the production of old products. Similar to Harrison et al (2014), Table 2 shows that process innovation either does not have a significant effect on employment or it has a negative one. The negative effect is only found in Chile and Uruguay for the whole sample, and in the latter country when analyzing small firms. The OLS results of the innovation–employment model show consistently that product innovation has a positive and significant effect on employment. The estimated coefficient on g₂ is close to 1, which indicates no important differences in efficiency in the production of old and new products.

With respect to the control variables, while location has almost no effect on employment growth, foreign ownership induces employment growth in Costa Rica. In general, there are no differences between the results for the whole sample and those for small firms.¹⁸

These results are in line with those of Harrison et al (2014). However, as discussed above, endogeneity (due to omitted price growth or correlation with the non-technological productivity shocks) is likely to produce a downward bias in this coefficient, overstating the productivity gains associated with the production of new products. Thus, OLS results might produce lower-bound results. To control for this problem, we use IV techniques. Ideally, we should have used the same set of instruments in each country; however, differences in how the surveys are designed at the country level made this impossible. Thus, we use the following set of country-specific instruments. For Argentina, we use whether the firm has some knowledge (but is not necessarily a user) of public support programs for innovation. This variable is more related to the coverage and diffusion of the public support system than to the actual innovation activities carried out by the firms. For Chile, we use innovation obstacles averaged at the regional levels but across all sectors in the same region, controlling for the strength of the regional innovation system where the firm operates. In both Costa Rica and Uruguay, we use the increase in the range of goods. This is the same preferred instrument as in Harrison et al. (2014). In both countries, this variable is coded as 0 if innovation is not relevant for the range of goods and services produced, 1 if the impact of innovation on the range is low, 2 if it is medium, and 3 if it is high. This is a variable that is clearly related to product innovation. However, to be a valid instrument it needs to have no correlation with the growth rates of the prices of new products relative to the prices of old products. Two other related questions in the surveys ask the firms about the impact of innovation on increased market share and on the improved quality of goods. So, given this we believe that the increase in the range of products is related to an increase in demand for reasons other than changes in product prices and quality. Thus, we expect this instrument to be uncorrelated with changes in the price of new products compared to old products. In further robustness checks, we also test for the validity of the different instruments.

As in Harrison et al (2014), the most notable result is that the IV estimates of the coefficient on g₂ (Table 3) are higher than the OLS estimates, which is consistent, as expected, with the correction of the downward biases related to endogeneity and price mismeasurement. Although the coefficient greater than 1 found in three countries would offer evidence that new products are produced less efficiently than old products, we find (with the exception of small firms in Chile) this evidence to be tenuous, given that the estimate is not statistically different from 1. In the case of Argentina, on the contrary, the coefficients suggest that new products are produced more efficiently than old products. To summarize, there is no evidence of a displacement effect on employment after a product innovation, only a creation effect due to demand enlargement. The results show that process innovation has only negative effects on employment in the case of the whole sample in Uruguay (and a positive effect only in Costa Rica).

Table 3.

Employment growth. All manufacturing and small manufacturing firm-dependent variable: l - (g₁ - Π)–IV estimation

Sector	All firms in manufacturing				Small firms in manufacturing
Regression	AR-IV	CH-IV	CR-IV	UY-IV	AR-IV	CH-IV	CR-IV	UY-IV
Constant	−6.887***	−2.02	−12.160**	1.402**	−5.763**	−2.13	−7.571	0.27
	(1.764)	(3.00)	(5.170)	(0.66)	(2.589)	(4.701)	(6.088)	(0.91)
Process only innovator (d)	−2.947	0.33	18.413*	−2.716**	−1.480	−3.38	15.415	−2.60
	(1.874)	(2.572)	(10.076)	(1.10)	(2.409)	(2.921)	(12.655)	(1.77)
Sales growth due to new products (g2)	0.631***	1.751***	1.015***	0.961***	0.780***	2.141*	1.051***	0.998***
	(0.031)	(0.653)	(0.050)	(0.040)	(0.064)	(1.205)	(0.068)	(0.060)
Located in the capital	NA	−0.36	1.361	NA	NA	5.21	7.194	NA
		(1.449)	(5.503)			(4.699)	(11.113)
Foreign owned (10% or more)	2.285	0.05	6.680*	1.37	2.389	0.87	−0.319	−3.16
	(1.865)	(1.04)	(3.843)	(1.19)	(1.963)	(3.394)	(6.049)	(2.50)
Two-digit industry dummies	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Time dummies	No	Yes	No	Yes	No	Yes	No	Yes
F test, g2 equation	151.9	35.91	78.16	170.80	19.74	6.94	51.12	89.62
P value	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00
Davidson–MacKinnon test of exogeneity for g2	3.551	2.71	13.79	10.39	0.112	2.23	7.15	10.66
P value	0.0596	0.10	0.00	0.00	0.738	0.14	0.01	0.00
R-squared	0170	0.25	0.65	0.42	0.294	0.14	0.58	0.34
Number of firms	3433	2049	208	2532	725	652	119	1353

Source: Authors' elaboration based on country studies for Chile, Costa Rica, and Uruguay.

Notes: Product innovators are firms that have introduced product innovations. Process innovators are firms that have introduced process innovations or organizational change innovations. Product-only innovators are firms that are product innovators but not process innovators. Process-only innovators are firms that are process innovators but not product innovators. Product or process innovators are firms that are product innovators or process innovators. Product and process innovators are firms that are both product innovators and process innovators. Robust standard errors in parentheses. Significance level: *** 1%, ** 5%, and * 10%. In the case of both Uruguay and Argentina was not possible to differentiate whether the firm was located not in the capital region. Endogenous variable g₂. Instruments used: AR: knowledge of public support for innovation activities; CH: obstacles for innovation averaged across firms in the same region; CR: increased range of goods and increase in productive capacity, UY: Increased range of goods and services and new markets.

There are two plausible interpretations for this result. First, a process innovation may not generate important productivity gains; hence, there is no displacement effect on employment. Second, a process innovation may generate productivity gains (displacement effect) which induce a demand enlargement through market competition (creation effect). In the end, the creation effect on employment compensates the displacement effect on employment.¹⁹

The IV results of the basic model are almost identical for SMEs (right panel) and for high- and low-tech samples (Table 4). The results are qualitatively similar. Perhaps the most interesting result is that in the case of Uruguay the displacement effect of process innovation is stronger in large firms and in the high-tech sector.

Table 4.

Employment growth. Low- and high-tech sectors dependent variable: l - (g₁ - Π) - IV estimation

Sector	AR	AR	CH	CH	UY	UY
Regression	IV-low-tech	IV-high-tech	IV-low-tech	IV-high-tech	IV-low-tech	IV-high-tech
Constant	−2.590	−7.879***	1.697	−2.733	1.115	1.670*
	(3.874)	(1.971)	(4.206)	−3.368	(0.944)	(0.929)
Process only innovator (d)	−1.401	−2.811	−0.517	−0.076	−2.524	−2.897**
	(4.199)	(2.094)	(3.787)	−2.835	(1.813)	(1.407)
Sales growth due to new products (g2)	0.614***	0.662***	1.403*	1.695**	0.956***	0.958***
	(0.051)	(0.039)	(0.846)	(0.728)	(0.056)	(0.056)
Located in the capital	No	No	Yes	Yes	No	No
Foreign owned (10% or more)	Yes	Yes	Yes	Yes	Yes	Yes
Two-digit industry dummies	Yes	Yes	Yes	Yes	Yes	Yes
Time dummies	No	No	Yes	Yes	Yes	Yes
F test, g2 equation	490.1	906.5	13.14	23.57	94.37	81.97
P value	0.00	0.00	0.00	0.00	0.00	0.00
Davidson–MacKinnon test of exogeneity for g2	1.083	0.533	0.4296	1.982	4.3178	5.496
P–value	0.298	0	0.5125	0.1594	0.038	0.0192
Sargan–Hansen test of overidentification	19.28	80.56	4.442	0.49	3.349	3.58
P–value (degrees of freedom)	0.0001	0, 466	0.1085	0.7823	0.646	0.611
R-squared	0.212	0, 151	0.2568	0.2464	0.423	0.419
Number of firms	751	2682	632	1417	1068	1464

Source: Authors' elaboration based on country studies.

Notes: Product innovators are firms that have introduced product innovations. Process innovators are firms that have introduced process innovations or organizational change innovations. Product-only innovators are firms that are product innovators but not process innovators. Process-only innovators are firms that are process innovators but not product innovators. Product or process innovators are firms that are product innovators or process innovators. Product and process innovators are firms that are both product innovators and process innovators. Low-tech and high-tech industries are categories constructed based on their innovation expenditure-to-sales ratios. Industries (considered at the two-digit level) with ratios below the country median are considered low-tech industries, and industries with ratios above country median are considered high-tech industries. Robust standard errors in parentheses. Significance level: *** 1%, ** 5%, and * 10%. In the case of Uruguay was not possible to differentiate whether the firm was located in the capital region. Endogenous variable g₂. Instruments used: AR: knowledge of public support for innovation activities, and new markets as a consequence of product innovation; CH: obstacles for innovation averaged across firms in the same region; UY: increased range of goods and services and new markets.

4.2 Robustness checks on the relationship between innovation and employment

Several robustness checks to evaluate the sensitivity of the results were performed in each of the country studies.²⁰ First, we included additional instruments using a Sargan–Hansen overidentification test. In the case of Argentina, the additional instruments used are a binary variable which indicates if the firms reach new markets as a consequence of the new product (new market dummy). Firms’ access to new markets is likely to be correlated with time-invariant firm attributes rather than productivity shocks. Thus, this additional variable seems like a suitable instrument. In the case of Costa Rica, the previous increase in productive capacity was used as an additional instrument. Based on the accelerator theory, it could be argued that before an increase in the demand of the goods produced, the firm has the option to increase its production by increasing its productive capacity. Thus, the production of new goods would be related to the increase in the productive capacity, but the increase in productive capacity would not necessarily be correlated to productivity shocks. In the case of Uruguay, the development of new markets was used as an additional instrument. In the questionnaire, firms are asked if the innovation helped maintain or increased market share. The opening up of new markets is expected to be related to the development of new products and an increase in demand for reasons other than changes in product prices and quality, complementing the increased range of goods that could be related to a change in price. Finally, in Chile the same instruments as before were used. In every case, the test of excluded instruments in the first stage suggests that the model is identified and weak instruments are not a concern. In addition, the Sargan–Hansen test does not reject exogeneity of the instruments. These results provide additional evidence of the validity of the originally chosen instrument(s).

Second, a model in which both g₂ and process innovation are endogenous was estimated. In every case, when including the estimation assuming d as an endogenous variable, the Davidson–MacKinnon test does not reject the null hypothesis if exogeneity of process innovation. Additionally, the results of the Sargan test indicate no problems with respect to the validity of the instruments.

Third, an interaction between g₂ and a dummy that is equal to 1 if product innovation occurs together with process innovation was added as an additional covariate. In the case of Argentina, the interaction between knowledge of support for innovation activity and the product and process innovator dummy as an additional instrument was used. For Uruguay, the instruments used are “increased range of good,” “development of new market,” and their interactions with the products and process innovation dummy. Again, in Chile the same three instruments were used. Unfortunately, in the case of Costa Rica it was not possible to check for changes in the slope, since there were no firms in the sample that meet such a condition. Even though the estimated coefficient on g₂ increases, the interaction is not significant, concluding that there is no compelling evidence to treat product and process innovators separately from product innovators. In the case of Uruguay, there is only weak evidence that the positive impact on labor growth of the introduction of new products is weaker when this innovation is introduced together with a process innovation. Process innovation only continues to have in general a negative impact on labor growth, but in some cases the coefficient is not significantly different from 0.²¹

4.3 Decomposition of employment growth

An interesting way to summarize the evidence obtained with our estimates is to use them to decompose the employment growth observed in each country (and type of firm) over three different components. Using our preferred specification [i.e., equation (6) with IV], we can write employment growth for each firm as follows:

$$l=\left[{\sum }_j\left({\widehat{\alpha }}_0+{\widehat{\alpha }}_{0j}\right){ind}_j+{\widehat{\alpha }}_{1}d\right]+\left[1-1\left(g_2>0\right)\right]\left(g_1-{\pi }_1\right)+1\left(g_2>0\right)\left(g_1-{\pi }_1+{\widehat{\beta }g}_2\right)+\widehat{u}$$

with the same notations as before and with ${ind}_j$ denoting the industry dummies and ${\widehat{\alpha }}_{0j}$ their estimated coefficients. For a given firm, the first component $\left[{\sum }_j\left({\widehat{\alpha }}_0+{\widehat{\alpha }}_{0j}\right){ind}_j+\widehat{{\alpha }_{1}d}\right]$ measures the change in its employment attributable to the (industry-specific) productivity trend in production of old products plus the firm-specific productivity growth due to process innovations in the production of old products; the second component (⁠$\left[1\hspace{0.17em}-\hspace{0.17em}1\left(g_2>0\right)\right]\left(g_1-{\pi }_1\right)$⁠) corresponds to the employment change associated with output growth of old products for firms that do not introduce new products; and finally, the fourth one $1\left(g_2>0\right)\left(g_1-{\pi }_1+{\widehat{\beta }g}_2\right)$ gives the net contribution of product innovation (i.e., contribution after allowing for any substitution of new products for old products). The last term (⁠$\widehat{u}$⁠) is a zero-mean residual component.

Table 5 reports the results of applying this decomposition to the four country samples using the proportion of firms and averages presented in Table 1 with the coefficients obtained in Table 3. First, in the case of all manufacturing firms (top panel), product innovations are an important source of firm-level employment growth. This is true even in situations of aggregate employment destruction, as in the case Uruguay. In this case, the destruction of employment is associated with a contraction in the production of old products. Finally, we observe that productivity growth associated with the production of old products normally leads to employment destruction, with the exception of a modest positive impact on employment in the case of Uruguay.

Table 5.

Decomposition of employment growth (from IV estimates). All manufacturing and small firms manufacturing

Manufacturing firms	AR	CH	CR	UY
Firms employment growth	2.2	−0.2	3.3	−0.7
Productivity trend and process innovation	−7.0	−7.1	−7.2	1.0
Output growth of old products	4.9	−0.3	2.9	−1.9
Product innovation	4.4	7.2	7.6	0.2
Small manufacturing firms	AR	CH	CR	UY
Firms employment growth	2.5	1.4	3.6	−3.7
Productivity trend and process innovation	−8.2	−5.0	−5.6	0.4
Output growth of old products	5.5	1.2	2.8	−3.8
Product innovation	5.2	5.3	6.4	−0.3

Source: Authors' elaboration based on country studies.

Notes: Argentina (AR)-Innovation Survey 2010–2012; Chile (CH): pooled regressions for the innovation surveys 1995, 1998, 2001, and 2007; Costa Rica (CR): innovation survey 2006–2007. Uruguay (UY): pooled regressions for the surveys 1998–2000, 2001–2003, and 2004–2006.

For small firms (bottom panel of Table 5), with respect to product innovation we observe that, with the sole exception of Uruguay, this type of innovation is an important source of employment growth at the firm level. The results for the output of old products and the productivity trend present a very similar picture to that exhibited by the whole sample of manufacturing firms.

The decomposition for the samples of low- and high-tech sector firms shows, first, that the drop in output of old product for non-innovator firms is mostly responsible for the almost ubiquitous drop in employment (with the sole exception of Argentina). Table 6 shows a consistent positive effect of product innovation for both low- and high-tech sectors. In the three countries analyzed, the positive effects on employment growth due to product innovation are present in both the high- and the low-tech sectors, although the impacts are always larger in the former ones.

Table 6.

Decomposition of employment growth (from IV estimates)–high and low tech

	AR	AR	CH	CH	UY	UY
	High-tech	Low-tech	High-tech	Low-tech	High-tech	Low-tech
Firms employment growth	1.8	3.8	0.1	−0.1	0.3	−2.1
Productivity trend and process innovation	−8.2	−0.2	−7.7	−4.4	1.2	0.8
Output growth of old products	6.0	0.6	−0.1	−0.5	−1.2	−2.9
Product innovation	4.0	3.4	7.9	4.8	0.3	0.1

Source: Authors' elaboration based on country studies.

Notes: Argentina (AR)-Innovation Survey 2010–2012; Chile (CH): pooled regressions for the innovation surveys 1995, 1998, 2001, and 2007. Uruguay (UY): pooled regressions for the surveys 1998–2000, 2001–2003, and 2004–2006.

5. Innovation effects on skill composition

In this section, we estimate equations (7) and (8), controlling for fixed effects at the industry level. Non-observable characteristics can be correlated with innovation variables; hence, as in the previous section, our preferred strategy relies on the use of an IVs approach. Given the validity of the instruments used so far, we will use the same set used in the previous section. Specifically, the share of skilled labor force in a given firm is measured as the percentage of professionals and technicians working for that firm in a certain period. Unfortunately, the analysis in this section is restricted to Argentina, Costa Rica, and Uruguay. We could not consider Chile in the analysis because innovation surveys in these countries do not report the classification of employment by skills.

5.1 Descriptive statistics

Table 7 shows descriptive statistics for the available data on the share of skilled workers for Argentina and Uruguay, distinguishing by type of innovative firm. Argentina presents a mean share of skilled labor in the manufacturing sector is slightly above 7.5%, while Uruguay is 9.5%. While product innovators have the highest share, the lowest is observed in non-innovators. That table also offers statistics for real growth rates of employment for each type of labor by type of firm. In both countries considered, we observe that skilled labor grows at a higher pace than unskilled labor (e.g., 1.4% vs. −0.7% in the case of Argentina, and 10.2% vs. 5.1% in the case of Uruguay). Growth rates for both types of labor are normally higher for innovators (whether process or product innovators). These general trends are consistent with a generalized process of skills upgrading in manufacturing in the three countries considered and also with some sort of skill-biased technical change driven by both process and product innovations.

Table 7.

Descriptive statistics, employment quality–all manufacturing firms

Manufacturing firms	Argentina	Uruguay
Share of skilled labor
All firms	7.5	9.5
Non-innovators (no process or product innovations)	5.7	7.4
Process only innovators (non-product innovators)	8.2	10.4
Product innovators	8.6	12.5
Employment (total) growth (%)
All firms	2.1	9.5
Non-innovators (no process or product innovations)	−2.4	3.3
Process only innovators (non-product innovators)	1.7	6.2
Product innovators	5.3	7.6
Skilled labor growth (%)
All firms	1.4	10.2
Non-innovators (no process or product innovations)	−9.5	6.3
Process only innovators (non-product innovators)	−5.2	13.4
Product innovators	9.9	14.1
Unskilled labor growth (%)
All firms	−0.7	5.1
Non-innovators (no process or product innovations)	−13.1	4.1
Process only innovators (non-product innovators)	−9.0	5.2
Product innovators	9.2	6.8

Source: Authors' elaboration based on country studies.

Notes: Argentina (AR)-Innovation Survey 2010–2012; Uruguay: pooled regressions for the surveys 2001–2003, 2004–2006, and 2007–2009. Product innovators are firms that have introduced product innovations. Process-only innovators are firms that have introduced process innovations or organizational change innovations excluding product innovators. Non-innovators are firms not classified as product or process innovators. Sample: Firms with less than 50 employees and with information in all the relevant variables for the empirical analysis. (i) Sales growth for each type of firm is the average of the variable g and averages for old and new products are the averages of variables g₁ and g₂, respectively, and (ii) prices computed for a set of industries and assigned to firms according to their activity.

Taken at face value, this might suggest a skill bias due to the introduction of innovations. Nevertheless, to fully assess the existence of skill bias due to the introduction of innovation, a model such as the one suggested in equation (6) is needed. Additionally, whether the coefficients found for each type of labor are statistically different must be assessed.

5.2 Econometric results for skill composition

In this section, we present the results for the model in equations (7) and (8). The dependent variables are the employment growth rate of type q_j labor minus sales growth rate (lqj−(g1 − π)) for each type of labor (i.e., skilled and unskilled labor), described below. The specifications include the process innovation dummy, d, sales growth rate of new products, g₂, a dummy controlling for the foreign ownership of the firm, whether the firm is located in the capital region of the country, and a constant capturing the productivity trend. The estimations also include industry fixed effects (at two-digit level). Given the problem with the potential endogeneity of the innovation variables, we only report the IV estimates.

The results summarized in Table 8 suggest some interesting patterns regarding the impacts of innovation on skills composition of the workforce. First, product innovation is always significant. While in the case of Uruguay the coefficient is close to 1, in Argentina it is smaller than 1 suggesting that new products are produced more efficiently than old products. We also find that the coefficient associated with the sales growth of new products is larger for skilled labor than unskilled labor in Uruguay and the opposite for Argentina. With regard to the process innovation dummies, the same ones are not statistically significant for skilled labor but are negatively significant for unskilled labor in the case of both Uruguay and Argentina. In summary, there seems to be skill-biased product and process innovation both in Argentina and Uruguay. These results are also very similar to those for small firms (tables available upon request).

Table 8.

Employment growth by type of labor (skilled and unskilled) manufacturing firms. IV estimation

Country	Argentina		Uruguay
Regression	Skilled	Unskilled	Skilled	Unskilled
	IV	IV	IV	IV
Constant	−3.561*	−6.975***	2, 934*	0.225
	(1.919)	(1.871)	(1.748)	(1.10)
Process only innovator (d)	−1.916	−3.406*	2.379	−3.373*
	(2.038)	(1.987)	(2.822)	(1.78)
Sales growth due to new products (g2)	0.524***	0.631***	1.087***	0.929***
	(0.034)	(0,033)	(0.12)	(0.075)
Located in the capital	No	No	No	No
Foreign owned (10% or more)	Yes	Yes	Yes	Yes
Two-digit industry dummies	Yes	Yes	Yes	Yes
Time dummies	No	No	Yes	Yes
F test, g2 equation	151.9	151.9	64.87	64.87
P value	0.00	0.00	0.00	0.00
Davidson–MacKinnon test of exogeneity for g2	0.206	3.960	5.37	1.16
P value	0.650	0.0467	0.02	0.28
Number of firms	3433	3433	1037	1037

Source: Authors' elaboration based on country studies.

Notes: Product innovators are firms that have introduced product innovations. Process innovators are firms that have introduced process innovations or organizational change innovations. Product-only innovators are firms that are product innovators but not process innovators. Process-only innovators are firms that are process innovators but not product innovators. Product or process innovators are firms that are product innovators or process innovators. Product and process innovators are firms that are both product innovators and process innovators. Robust standard errors in parentheses. Significance level: *** 1%, ** 5%, and * 10%. In the case of both Uruguay and Argentina was not possible to differentiate whether the firm was located in the capital region. For Uruguay, estimates also included whether the firms was fully foreign owned. Endogenous variable g₂. Instruments used: AR: knowledge of public support for innovation activities; UY: increased range of goods and services and new markets.

5.3 Employment growth decomposition

As in the previous section, we decompose the growth of the employment for each type of labor into three components: the productivity trend in the production of old products (including the effect of process innovation in the production of old products); the change in employment associated with output growth of old products for firms that do not introduce new products; and finally, the net contribution of product innovation (i.e., contribution after allowing for any substitution of new products for old products).

In the case of the whole sample for manufacturing firms, product innovation seems to contribute to the creation of both unskilled and skilled employment in Argentina but only to skilled employment in Uruguay (Table 9, top and lower panels, respectively). While in Uruguay the effect of product innovation on skilled labor is higher than on unskilled labor, the opposite is true for Argentina where an employment expansion occurred. Indeed, while product innovation contributed to 4.6% of unskilled employment growth in Argentina, its effect on skilled labor was 4.1%. In the case of Uruguay, product innovation displaced unskilled labor by −0.1%, while at the same time it added skilled labor at a rate of 1.5% per year.

Table 9.

Decomposition of unskilled and skilled employment growth manufacturing firms. IV estimation

Unskilled labor in manufacturing	AR	UY
Firms employment growth	2.2	5.1
Productivity trend and process innovation	−7.2	7.1
Output growth of old products	4.9	−1.9
Product Innovation	4.6	−0.1
Skilled labor in manufacturing	AR	UY
Firms employment growth	4.1	10.2
Productivity trend and process innovation	−4.8	10.7
Output growth of old products	4.9	−1.9
Product Innovation	4.1	1.5

Source: Authors' elaboration based on country studies.

Notes: Product innovators are firms that have introduced product innovations. Process innovators are firms that have introduced process innovations or organizational change innovations. Product-only innovators are firms that are product innovators but not process innovators. Process-only innovators are firms that are process innovators but not product innovators. Product or process innovators are firms that are product innovators or process innovators. Product and process innovators are firms that are both product innovators and process innovators. Robust standard errors in parentheses. Significance level: *** 1%, ** 5%, and * 10%. In the case of both Uruguay and Argentina was not possible to differentiate whether the firm was located in the capital region. For Uruguay, estimates also included whether the firms was fully foreign owned. Endogenous variable g₂. Instruments used: AR: knowledge of public support for innovation activities; UY: Increased range of goods and services and new markets.

With respect to productivity trends in the production of old products, the findings across are consistent with skill-biased productivity growth. In Argentina, productivity growth in old products destroys employment, but the displacement is always larger in the case of unskilled labor. On the other hand, in Uruguay, productivity growth in the production of old products actually creates labor (suggesting that compensating effects are larger than displacement effects). However, even for Uruguay, productivity growth is associated with higher recruitment of skilled labor.

6. Conclusions

Despite recent high economic growth, Latin America still faces the challenge to reduce poverty and inequality. Considering the key role played by employment generation in achieving these objectives, it is of particular interest to understand the effects of innovation on employment generation.

In this article, we have estimated a model based on Harrison et al. (2014) by using a source of comparable and representative data on innovation in manufacturing (by firm size) across four Latin American countries. Our results provide findings on a key yet barely explored topic in the region. They shed new light on the relative roles of displacement and compensation effects of product and process innovation on employment growth in manufacturing.

Our results highlight the fact that individual process innovation accounts for a small share of the changes observed in employment, inducing small displacement effects. More importantly, and fundamental for the search for more inclusive growth patterns in the region, we found that product innovations are an important source of firm-level employment growth. This is true even in situations of aggregate employment destruction.

We went beyond the received literature by using the same conceptual apparatus to assess the differential effects of innovation on skill composition of employment. Here, we found that although the regression coefficients suggest no clear evidence of a skill bias, the employment growth decompositions generate results that are more consistent with skill-biased product and process innovation. In other words, innovation, in particular product innovation, is good for employment; however, reaping its benefits requires the presence of a workforce with the requisite skills. Results are similar for small and large firms. However, when we looked at different sectors, we found that the skill bias of product innovation is more evident in the case of high-tech sectors.

Our results suggest the importance of maintaining and augmenting the public support for firm innovation in Latin America. Our findings provide evidence that public support has a rationale that goes beyond the productivity improvements. In fact, innovative firms capable of introducing new products in the market are an important driver to employment creation, hence enabling reducing inequality. At the same time, Latin American countries should foster training and talent creation programs that provides employees with novel capabilities and skills that complement the acquisition of technology and the implementation of new production processes and organizational changes.

Footnotes

1

Labor informality is a pervasive characteristic of the labor markets in Latin America. Evidence shows that labor informality exhibited a discernible downward pattern since the 2000s in most countries See Tornarolli et al. (2014) and Messina and Silva (2018) for an account of this reducing trend.

2

This applies to both process and product innovation. Although process innovations might displace labor in the short term, in the extent that productivity gains are passed through to prices and consumers react to price reductions; it might increase labor in the long term. The opposite applies to product innovations, in the extent that short-term demand shifts might be compensated by imitators later on.

3

Before the Inter-American Development Bank (IDB) project, there were only two papers on innovation and employment in Latin American Countries (LAC): Benavente and Lauterbach (2008) on Chile, and Fajnzylber and Fernandes (2009) on Brazil. While the first paper deals only with employment quantity issues in a cross-section setting, the second one only discusses composition effects and uses trade proxies as controls for technology, so it does not properly control for innovation. Benavente and Lauterbach (2008), found, for manufacturing in Chile, that product innovation is positively and significantly linked with firm-level employment. They do not find any significant impact of process innovation on employment dynamics, after controlling for investment and sectorial patterns. Fajnzylber and Fernandes (2009) also found that increased levels of international integration linked to an increased demand for skilled labor in a cross-section of Brazilian firms.

4

A companion paper (Crespi and Zuñiga, 2013) presents the results for the relationship between innovation strategies of the firm and employment growth and composition for the same set of countries and with the same disaggregates related to size and sector.

5

Unfortunately, this approach has some shortcomings. Vivarelli (2011: 14) notes that the microeconomic approach cannot fully take into account the indirect compensation effects, while the analysis might present a “positive bias” in which “microeconomic analyses generally show a positive link between technology and employment, since they do not consider the important effect on the rivals, which are crowded out by the innovative firms.”

6

See Vivarelli (2011) for an account of the evidence on the effects of process and product innovation in developed countries.

7

The evolutionary tradition has dealt extensively with the link between technology and employment. Saviotti and Pyka (2008), for example, set a simulation model discussing how new products and diversification raise employment, while process innovations are detrimental to employment. An adequate succession of new products, stemming from product innovations, leads to a higher rate of growth of variety and to a higher rate of growth of employment. However, growth in variety may not compensate for the loss of employment entailed by the evolution of pre-existing and mature sectors if the new sectors created are very small.

8

Readers interested in the theoretical analysis of the relationship between technical change and employment could refer to recent surveys such as Sabadash (2013), Vivarelli, (2014), and Calvino and Virgillito (2016).

9

This will be the case if relative prices do not change much over time or across new and old products.

10

For simplicity, we assume that w11∼w22.

11

Real growth in sales of old products, y₁ is the result of three different effects: the autonomous increase in firm demand for the old products, the compensation effect induced by any price variation following a process innovation, and the demand substitution effect resulting from the introduction of new products. As these components cannot be disentangled without additional data, in practice y₁ will be simply subtracted from l, so an alternative specification for equation (4) is to use the inverse labor productivity growth as the dependent variable.

12

Unfortunately, the sample size for the innovation survey in Costa Rica does not permit the exercise to be performed for high- and low-technology sectors. This analysis is presented only for the three other countries.

13

This is the sort of timing for investment decisions underlying Olley and Pakes (1996).

14

There are good reasons to think that process innovation can in fact be exogenous. As Harrison et al. (2014) noted, it seems realistic to assume that firms cannot predict future labor problems, supply chain disruptions, or organization shocks when deciding their process innovation investments. Accordingly, the hypothesis on exogeneity of process innovations is maintained in this article. However, we carry out some robustness checks to review whether this assumption is attainable.

15

See Aboal et al. (2011), Alvarez et al. (2011), and Monge-González et al. (2011). de Elejalde et al. (2015) was part of the project, although they exploited an earlier innovation survey covering the 1998–2001 period in Argentina. In the remainder of the artile, we present the results of Pereira and Tacsir (2019).

16

Actually, all of the surveys have a question asking for the share a sales at the end of the period that are the result of product innovations introduced over the past 2 or 3 years, depending on the survey. We call this share s. The surveys also include information on the nominal growth rates of total sales (g). Thus, it can be shown that given that the sales of new products at the beginning of the period is 0 by definition, the nominal growth rates of new products can be obtained as g₂ = s (1 + g).

17

These results are remarkably stable across firms of different size and from sectors with different degrees of technological intensity. Tables showing this can be provided upon request.

18

The analysis dividing manufacturing firms according to the innovation intensity of their respective sectors presents similar results to those described so far (results available upon request).

19

As in Harrison et al (2014), we tested different (extreme) specifications for process innovation, in which we assume that either the process innovations correspond entirely to old products or to new products. In the first case, while the coefficient on the process innovation only indicator remains unchanged, the coefficient on sales growth due to new products is slightly reduced suggesting lower employment growth associated with new products. In the second case, the interaction terms generate estimates suggesting smaller productivity due to new products. Give these results, and the available data, we maintain the estimates for equation (6) presented in Table 5 as our preferred specification.

20

To simplify the exposition, we focus on the impact of innovation on employment.

21

In the cases of Costa Rica and Argentina, a more restrictive definition for sales of new products was also implemented by excluding new goods sold by the firm that were already sold in the market by other firms. The reported results show that the new estimates were consistent with the previous results. Finally, in Argentina the estimation of innovation-employment model under non-constant returns to scale was performed as part of the robustness checks. The main result is that joint identification of efficiency and scale parameters is quite complicated, and assuming constant returns to scale is a sensible working assumption given those constraints.