An empirical analysis of the variables was performed using panel data from Argentina, Brazil, Colombia, Costa Rica, Mexico, Panama and Uruguay for the period 1996-2011. . These countries were selected based on the criteria of highest investment in science and technology (S&T) as share of their GDP.

In addition, two sets of countries were created using GDP per capita as a measure for differentiation in order to separate the effect their particular national economic structure may have within the model. Figure 1 shows Argentina, Brazil, Mexico and Uruguay maintaining an average GDP per capita of $9,000 (in constant USD), evidencing higher growth rates than the cases of Colombia, Costa Rica and Panama where GDP per capita averaged approximately $6,000 USD (in constant USD) during the observed period. Given the difference between the two groups, the analysis was done separately.

Figure 1

Due to data availability as Monroy et al (2015) mentioned in their publication, the approach conducted here aimed to analyze if there was a relationship between innovation (measured by spending on science and technology) and unemployment; in other words, if national investments in S&T could create technological unemployment in the selected countries. Furthermore, in order to give greater significance to the model, other economic and social variables were included as shown in Table 1.¹

Table 1

Source: Authors summary.

As previously stated, the relationship between innovation (measured as spending on science and technology) could have two effects on the unemployment rate: either it increases it due to the automation of tasks previously performed by humans, or it decreases it because of growth in production levels that may require additional skilled labor and non-automated processes.

With regard to the GDP per capita (GDPpc), it is believed that an increase in production would foster business activity and industrial growth, thus stimulating job creation, hence the inverse relationship between GDPpc and unemployment. Wage levels were also considered as inversely related to unemployment assuming that higher salaries would attract more talent to the work force. As for the last variable, the effects of educational attainment on labor markets and productivity have been widely researched, this is why education is considered to be inversely related to unemployment (Becker, 1962; Mincer, 1991; Autor & Handel, 2009).

The regression analysis proposed is set based on the characteristics of the dependent variable. The unemployment rate is restricted between values of 0 and 1, so that its growth will be mitigated when reaching its maximum value, as will its decline when reaching its minimum value.

Thus, unemployment by means of automation - Technological Unemployment (TU) - is represented by a sigmoidal growth (''S'' Form - See Figure 1) which is curbed as it reaches its peak, given it is calculated within a finite limit (0 ≤ TU ≤ 1).

Even though it may have an exponential growth at one stage of its evolution, it does not reach infinite rates but it is stabilized in the long term, either by implementing public policies that control it or by the natural limit of the population.

Figure 2

Given this specification, consistent estimations of β can be obtained by maximizing the logarithm of the likelihood function

(1)

The distribution of probability will come given by a probit model, which function of distribution originates by the integral:

(2)

Where the probability of success, Pit, were defined by the area under the X_itβ curve. In addition, random effects were assumed:

(3)

Since there is assumed that it exists strict exogenous influence from X_it subject to μi, and that the Y_it are independent in time given X_it and μ_i (Gujarati, 2004).²

Considering this, in order to have a statistically significant model, Y turns into a binary variable where: Y = 1 if the event is successful or Y = 0 for the opposite case. In this sense, the unemployment variable was transformed to ''1'' in those values in which the unemployment rate increased above the country average and ''0'' when the unemployment rate was below the country average for the years studied.

According to Inter-American Development Bank, IDB, (2010), the countries of Latin America must maintain policies that improve the quality of life of their populations, attending to priorities such as education, health, poverty, among others; nevertheless, it also affirms that the challenges of modern society force the countries to face the technological changes that affect their economies. On a separate note, Crespi and Tacsir (2012) argued that the productive structure of the region, dominated by commodity production, preponderance of Small and Medium Enterprises (SMEs) and low technological intensity of manufactured goods could generate different results in the mix of innovation and employment.

Rodrik (2015) singled out Latin America as one of the regions most vulnerable to ''premature deindustrialization'' suggesting that globalization, trade and labor-saving technologies have hindered the development of a strong manufacturing base (and jobs) in the area. Additionally, the author commented on the political and economic backlash of this phenomenon, including political instability and democratic failure.

In the literature, it was found that the region had made changes regarding the levels of expenditure dedicated to Science and Technology. According to the IDB (2010), Latin America showed substantial advances in some indicators of recorded patents and increased output of scientific publications. However, the progress was focused in only a few countries, therefore, the region continued to lag behind in regard to science and the technology investment and achievements compared to the rest of the world. Figure 3 shows some of the gaps in the levels of expenditure in S&T between different countries of the region.

Brazil stands out as the country that invested the most in S&T as a share of GDP compared to its neighbors: it even doubles the regional average. On the other hand, countries like Colombia were behind regional peers and below average.³

Figure 3

In Figure 4, we can see the increasing trend of the expense in S&T by the countries of the sample. The investment in this segment rose (on average) from 0.26% of the GDP in 1996 to 0.51 % in 2011; in the year 2009 it reached a peak at 0.53%, being the countries with high GDPpc those that invested the most in this category. The principal characteristic of the indicator is its volatility during the period of study, which demonstrates high rates of growth particularly in the year 2000 and 2006.

Figure 4

Regarding the unemployment rate, the regional trend was downwards from 2002 and the volatility of this indicator was minimal. Between 1996 and 2000 the data showed an increase in the unemployment levels, due to real exchange rate appreciation as argued by Ros (2005); after 2002 the fall in unemployment is supported by some hypotheses associated with the commodities boom experienced by some regional countries during most of the first decade of the 21^st century (World Bank, 2015).

Vivarelli (2012) suggested that one of the compensation mechanisms of automation is the readjustment in wages, taking into account the implementation of innovations that reduce the need to use workforce - an adjustment was assumed in the price of work. When comparing the behavior of wages and the level of expenditure in S&T, see Figure 5, a relative stagnation was observed in the wages of the region in spite of the increase in the S&T investment. This stagnation trend in wages was documented in advanced economies, as the average productivity increased the return of the workforce, (labor share) has declined (Karabarbounis & Neiman, 2013; Autor, 2014).

Figure 5

The IDB (2010) argued the existence of a synergy in which investments in innovation, productivity and per capita income were mutually reinforced and allowed the countries to achieve growth rates sustained over time. Therefore, the stagnation of wages experienced in regional economies was perceived as a barrier to promote growth and technological development.

The econometric estimations were calculated to assess the correlation between Science and Technology expenditure as share of GDP, GDPpc, median wages and the expenditure in education on the unemployment rate.

Table 2 summarizes the results: the estimations (1), (3) and (5), include the regression with four independent variables proposed in this research, whereas the estimations (2), (4) and (6) showed only the statistically significant regressions of the models.

Table 2

Note: (*) 10% significance. (**) 5% significance (***) 1% significance.

The first regressions of every group proved not to be significant to explain the unemployment rate during the period 1996-2011. Nevertheless, after some transformations in the line of regression it was found that results provided some explanation hypotheses about the variations of the unemployment in the region.

According to the statistical results, there was no evidence that expenditures in Science and Technology were related to the unemployment rate in the region; contrasting what was found in specialized literature on advanced economies, technological innovation in the observed period did not generate unemployment in Latin America.

The model showed that unemployment rates were only explained by the GDP and the wages, a 1% increase in the GDPpc increased the probability of reducing the unemployment by 10%, whereas a 1% increase in the salary raised the probability of reducing the unemployment by 2.4%. A higher GDPpc suggested an increase in production, assuming that this increase involved a higher employment of workforce, thus a decrease of the unemployment rate. Regarding salaries, it was inferred that an increase in wages motivated the non-employed to enter the formal labor market, thus influencing the unemployment rate.

For countries with higher levels of income measured in GDPpc, it was found that only the expenditure in education was not adequate for explaining the unemployment. As in the case above, both GDP and salaries had a negative relation with unemployment, in agreement with the results of the model, a 1% increase of the GDPpc increased the probability of reducing the unemployment by 9%, whereas a 1% increase in the salary raised the probability of reducing the unemployment by 2.5%.

The resemblance of the results between the Latin America sample and the high GDPpc sample helped to infer that the particular economic structure of the countries could lead to bias in the results. It was important to understand the limitation of the model in this regard, and for the analysis to separate the countries depending on their income measured as GDPpc.

In the model, S&T expenditure turned out to be statistically significant to estimate the unemployment levels, however the relation was inverse: a 1% increase of the expense in S&T increased the probability of reducing the unemployment rate by 0.57%. In those countries with higher income, where the investment in investment in S&T was higher, it was observed that increased productivity did not necessarily cause unemployment; on the contrary, productivity gains from innovation could probably turn into higher labor intake.

As for the countries in the sample with lower income, the findings of the model were that only GDPpc explained the unemployment rate, thus supporting the negative relation of these two variables, based on the estimation: a 1% increase of the GDPpc raised the probability of reducing the unemployment in 22%.

As expected, in countries whose investment in S&T was high, no significant impacts were seen on the unemployment rate. The probability was even less in countries with lower GDPpc. Based on these results it was deducted that public policies directed to promote investment in education or increasing wages would not have a significant impact on the levels of unemployment.

The results were further analyzed by exploring the individual relation between unemployment and the expense in Science and Technology shown in Figure 6. The scatter plot 5a shows the correlation that exists in the countries with higher GDPpc, highlighting two trends. First, a high concentration of values was observed towards the left side of the graph, and secondly, a minor concentration of values to the right.

In the majority of the observations, no direct relation was perceived between S&T and unemployment. As for the second phenomenon, it seemed to validate that if the levels of investment in S&T increase, there is a minor decrease in the unemployment rate, as suggested by the model.

This behavior particularly matched the case of Brazil where there was a higher investment in S&T, high levels of production as seen in the GDPpc indicator, higher median wages and larger amounts invested in education that other countries in the sample. In sum, the application of these diverse public policy instruments might reduce unemployment.

On the other hand, Figure 6b shows the relation of both variables for the lower GDPpc group, minor increases in S&T expenditure did not seem to affect the level of unemployment of the sample. In this case, it is worth emphasizing that the lowest levels of unemployment related to the highest levels of expense in S&T were in Costa Rica, where the average percentage amount invested was larger than the average of the region. But this situation was not reflected in changes on the unemployment rate.

Figure 6

In order to explain why some variables had to be excluded from the model, we estimated the Pearson correlation coefficients of independent variables in every group, to verify if the presence of linear correlation might explain these behaviors. In reality, as the Table 3 demonstrates, a high correlation exists between the expenditure in S&T and the expenditure in Education, mainly for the case of Latin America and the group of countries with lower GDPpc.

Table 3

Source: Summary of results. Calculations by the authors.

The interdependence seen in the expenditures in education and S&T demonstrated that the allocation of these expenditures seemed to be restrictive between these two activities (opportunity cost): in other words, if it was assigned to one, it would not be assigned to other. Therefore, public policies should be devised to promote the interrelation and synergy of funds allocation in these categories.

Accordingly, it is advisable for Latin American decision makers to formulate public policies and private sector initiatives to tackle the negative effects, minimize the risk and mitigate the political and economic consequences of potential massive unemployment (Marchant et al., 2014; Rodrik, 2015).

So far, three major ideas have been found in the literature. First, the creation of a system for Universal Basic Income in which eligible citizens are granted an amount of money unconditionally. Even though the policy idea was proposed some time ago, it is not until recently (after the 2008-2009 financial crisis) that it gained traction in public policy circles. Experiments on this matter are currently under evaluation in Canada and Finland (De Wispelaere and Stirton, 2004; Van Parijs, 2004).

Secondly, some suggest solutions based on the human capital theory, life-longlearning, retraining, upskilling and reskilling. These ideas are not particularly new but are definitely relevant in a contemporary debate about automation replacing jobs (Becker, 1962; Davis & Reeve, 2000; Cairo & Cajner, 2014).

Finally, the third option involves reducing the weekly working time: some European countries have taken the lead by reducing the weekly working hours and sharing work seems to be a fresh strategy to mitigate the unemployment wave predicted by the probabilistic models (Bosch & Lehndorff, 2001; Ohanian, Raffo& Rogerson, 2006; Askenazy, 2013).

Even though the model showed that investments in S&T did not affect the unemployment rate in the sample but contributed to labor productivity, it is important to devise, test and implement public policies and private sector plans that may contribute to mitigate the effects of innovation such as technological unemployment.