U-shaped wage curve and the Internet: The Colombian case

Martin, Diego A. Article U-shaped wage curve and the Internet: The Colombian case Estudios de Economía Provided in Cooperation with: Department of Economics, University of Chile Suggested Citation: Martin, Diego A. (2018) : U-shaped wage curve and the Internet: The Colombian case, Estudios de Economía, ISSN 0718-5286, Universidad de Chile, Departamento de Economía, Santiago de Chile, Vol. 45, Iss. 2, pp. 173-202 This Version is available at: https://hdl.handle.net/10419/194264 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-sa/4.0/ U-shaped wage curve and the Internet: The Colombian case / Diego A. Martin 173Estudios de Economía. Vol.45 - Nº2, Diciembre 2018. Págs. 173-202 U-shaped wage curve and the Internet: The Colombian case*1 Internet y curva salarial en forma de U: El caso colombiano Diego A. Martin** Abstract While there is a broad consensus in the literature that there is a positive correlation between Internet usage and labor income in the richest countries, this link has not been proven in the developing world. This paper uses propensity score matching techniques and household survey data to estimate the effect of the Internet on wages in Colombia, a country that has experienced a relatively rapid diffusion of information and communications technology in recent years. The empirical results confirm that there is a positive and statistically significant relationship between Internet use and income in this country. Consistent with evidence gathered on developed countries in previous studies, the empirical results also suggest that workers in the middle of the skill distribution receive the lowest wage premium for using the Internet. However, contrary to most evidence from developed countries, low-skilled workers in Colombia enjoy the highest wage premium from Internet use, which illustrates the potential for new technologies to address inequality gaps between occupations. Key words: Propensity score matching, skills, wage differentials, internet, income distribution. JEL Classification: C14, J24, J31, L86, O15. Resumen Si bien existe un amplio consenso en la literatura acerca de la existencia de una correlación positiva entre el uso de Internet y el ingreso laboral en los países * I thank the seminary of Universidad del Rosario and the anonymous peer reviewers for all the useful comments. I am also very grateful to Juan Vargas for his guidance in this document, which is my master’s thesis for the Universidad del Rosario. ** CAF-Development Bank of Latin America, Research Department. E-mail: dmartin@caf. com Received: 24 de septiembre de 2017. Accepted: 7 de mayo de 2018. Estudios de Economía, Vol.45 - Nº2174 más ricos, este vínculo no se ha demostrado en el mundo en desarrollo. Este documento utiliza propensity score matching y datos de encuestas de hogares para estimar el efecto de Internet sobre los salarios en Colombia, un país que ha experimentado una difusión relativamente rápida de la tecnología de la información y la comunicación en los últimos años. Los resultados empíricos confirman que existe una relación positiva y estadísticamente significativa entre el uso de Internet y los ingresos en este país. De acuerdo con la evidencia reunida en países desarrollados en estudios previos, los resultados empíricos también sugieren que los trabajadores en el medio de la distribución de habilidades reciben la prima salarial más baja por usar Internet. Sin embargo, contrariamente a la evidencia de los países desarrollados, los trabajadores poco calificados en Colombia disfrutan de la prima salarial más alta del uso de Internet, lo que ilustra el potencial de las nuevas tecnologías para abordar las brechas de desigualdad entre las ocupaciones. Palabras clave: Propensity score matching, habilidades, diferencia salarial, internet, distribución salarial. Clasificación JEL: C14, J24, J31, L86, O1. 1. Introduction Do information and communications technologies (ICTs) impact productivity? This question has been extensively studied in the developed world since the introduction of the first computing technology in the 1950s. In the United States, for example, a large body of evidence demonstrates the positive effect of ICTs on economic dimensions such as wages, supply and demand of factors (Katz & Murphy, 1992; Krueger, 1993; Acemoglu, 1999). These patterns are also found in other developed countries, where ICTs have been found to contribute 0.2-0.9 percentage points per year to economic growth (Colecchia & Schreyer, 2002). The canonical model presents a framework to explain why ICTs impact economic variables. This model incorporates supply and demand for different skills, which are imperfect substitutes but produce similar goods. Technology can complement either highor low-skilled workers, which in turn produces an increase in the demand for one group or the other (Katz & Murphy, 1992). The idea of a direct link between a specific type of worker and ICTs implies that technology has a heterogeneous impact on the labor market. According to this model, technology has simplified routine tasks and solved complicated problems, creating a specific wage distribution curve in which highly skilled workers are at the top and middle-skilled workers are at the bottom in the developed world (Autor etal., 2003; Michaels etal., 2014). Although this model is very useful for calculations and explanations, Acemoglu and Autor (2011) developed an improved version to demonstrate that in rich countries technology has depressed the real wages of low-skilled workers, created non-monotone U-shaped wage curve and the Internet: The Colombian case / Diego A. Martin 175 changes in the earnings distribution, and reduced the need to hire middle-skilled workers, among other effects. The heterogeneous effect found in the wage distribution can also be observed in social relationships, including personal and political behavior, in which ICTs follow a model of “the rich become richer”, which means that technology exacerbates pre-existing inequalities, and the exact effects depend mainly on individual personalities. For example, extroverts using the Internet have better results in social involvement than introverts (Kraut etal., 2002). However, other studies suggest the Internet is a tool that can help close the gap between the least and most advantaged populations (Steinfield etal., 2008). This technology can even build a bridge of communication between politicians and their electorates to reduce, at least to some degree, the asymmetry of information between government actions and what people need to improve their living conditions (Garcia-Murillo, 2013). The public sector can use ICTs to develop policies to reduce gaps along different dimensions, particularly in rural areas. Do ICTs have a similar effect in developing countries, especially in Latin America? This relationship has been less frequently explored in this region for three main reasons. First, there is a lag in the absorption of ICTs in the developing world. Europe and North America, for example, have around 84 Internet users per 100 people, while South America has only 51 (World Bank, 2015). Second, there are no specialized databases that contain common information between countries, which is needed in order to produce standardized indicators to compare within the region. Third, estimating the impact of ICTs on different dimensions requires smart approaches because non-random access complicates the process of conducting an experiment. Nonetheless, the effect of ICTs in Latin America is a highly relevant question since they can be a powerful tool to improve living conditions and overcome poverty traps. Colombia offers a unique opportunity to analyze how ICTs affect different dimensions of the labor market of a developing country that has experienced a recent growth in the number of Internet users because it has a database, which has never been used before, that makes it possible to present heterogeneous effects depending on workers’ skill levels. It is also possible to test whether this country follows the prediction of the canonical model, and its expansion as carried out by Acemoglu and Autor (2011), or whether it is necessary to develop another model for developing countries. The following three questions focus on the effect of the Internet in Colombia’s wage distribution curve. First, is the increased efficiency in some work tasks generated by ICTs valued and compensated for in the labor market? A sophisticated improvement using ICTs, such as computers programming by themselves, is not necessary; whenever workers can either communicate easier or access information faster, they may receive a salary increase. Indeed, it is also important to analyze how this effect changes over time, since there is evidence that the wage premium has decreased in the United States from 2000 to 2001 (DiMaggio & Bonikowski, 2008). Second, given that people are more likely to use the Internet at work in productive activities than at home, is there a higher Estudios de Economía, Vol.45 - Nº2176 wage premium for using ICTs in the workplace? Third, the Internet offers different kinds of applications for education, communication and entertainment, but can the labor market differentiate workers’ efforts, at least imperfectly, in productive activities? For example, do employees using the Internet for education receive a higher reward than those who use it for entertainment purposes? Finally, there is considerable evidence from the developed world that technology has polarized the labor market into low-skilled workers with low salaries and high-skilled workers with high salaries (Acemoglu & Autor, 2011; Autor, 2015). To explore whether a developing country such as Colombia follows this pattern, I analyze the heterogeneous effects of ICTs on workers’ income levels along three dimensions: places of access, activities on the web and workers’ skills. This empirical support presents important ideas about how developing countries, especially those that are the most socioeconomically disadvantaged, can obtain the greatest benefits from using the Internet. This paper uses Propensity score matching (PSM) techniques and nonparametric models to estimate the effects of the Internet on labor income. Although this model has some drawbacks, including the requirement to have individuals with the same probability of using the Internet in the treated and control groups (a technique known as a common support), the results must not change depending on the exclusion or inclusion of controls, and they must be robust to changing the matching method used (Abadie & Imbens, 2006). The paper presents several tests to demonstrate that these issues are overcome, and that it is possible to reliably estimate average treatment effects on the treated population. The results address the hypotheses individually and demonstrate a positive and significant correlation between using the Internet and labor income. On average, Internet users earn 6.5% of one standard deviation (SD) more than non-users. Using this technology at work increases wages 11% more than using it at home, and users in productive activities earn 7.1% of one SD more than those who use the Internet for other applications. The wage premium depends on the tasks performed in a specific occupation, which proxies for skill level. Workers in the middle of the skill distribution in Colombia have benefited the least from using the Internet, which parallels trends in the developed world. Nonetheless, and contrary to the evidence gathered from the richest countries, the highest wage premium is found in the lowest tail of the distribution, where farmers and miners using the Internet gain 20.7% of one SD more than nonusers in the same occupation. I find that the canonical model is applicable to the Colombian case, while Acemoglu and Autor’s (2011) extension of the model partially applies in this situation given the fact that the salary distribution curve has a u-shaped form. In the developing world, the relationship between ICTs, income, education and other variables has been explored extensively. To the best of my knowledge, this study is the first to use household surveys, non-parametric models and fixed effects of firm size, occupation and economic sector to analyze the Internet’s impact on the wage distribution in Colombia. Therefore, it presents U-shaped wage curve and the Internet: The Colombian case / Diego A. Martin 177 new evidence and methodology related to Colombia, a benchmark for other developing countries, and provides corresponding technical support to justify investments in ICTs. The remainder of the paper is structured as follows. Section 2 reviews the literature on the mechanism of the impact of new technologies and earnings. Section 3 describes the basic statistics of Internet use around the world as well as the descriptive statistics related to the hypotheses. It also presents the empirical strategy, which uses non-parametric models. Section 4 describes the results of tests of the matching model and addresses each of the hypotheses individually. Finally, Section 5 discusses the results and how the Colombian case is related to the international literature. 2. New Technologies and Earnings Many studies have explored the relationship between technology and wages, mainly in the developed world. Since the start of the computing era, there has been a rise in the demand for college graduates who can use this new technology, which creates a wage premium that favors technology users over the long term. This is also known as the canonical model, which has empirically explained the evolution of the skill premium generated by computers (Acemoglu & Autor, 2011). Computers can replace workers in some repetitive manual tasks, and help them carry out non-routine tasks such as communication and problem solving (Card & DiNardo, 2002; Autor etal., 2003). The variations in wages are not uniform across skill levels, because the lower premium is at the middle of the distribution rather than the bottom, and the relationship between wages and skills has a u-shaped distribution. Indeed, the automation generated by technology reduces the demand for middle-skilled workers more than for lower-skilled workers (Autor & Dorn, 2013). This pattern is also found in 16 countries in Europe (Goos etal., 2009) and the United Kingdom (Acemoglu, 1999; Goos & Manning, 2007). The current labor market mainly requires the use of ICTs to access quality jobs because online skills help people perform their work more efficiently. Workers of relatively high ability are more likely to move to non-routine cognitive occupations, while people with low abilities tend to stay in routine jobs with a lower potential for salary growth (Cortes etal., 2016). The Internet also makes it easier for workers to move to better jobs, because the demand for labor is concentrated in urban areas (McDonald & Crew Jr, 2006). In the United States, workers who use the the Internet earn 13.5% more than non-users. This wage premium is even higher in industries that are less intensive in technology, in which the only people who use Internet are those who can perform the most difficult tasks (Goss & Phillips, 2002). Using panel data, DiMaggio and Bonikowski (2008) also found a 20% increase in the hourly wages of workers for using Internet, not just computers. Bartel etal. (2007) found that companies encourage workers to undertake training in computers when they invest in ICTs. Estudios de Economía, Vol.45 - Nº2178 Gust and Marquez (2004), for example, show that the more flexible regulation in information technology in the United States partly explains its higher productivity than some countries in the OECD, such as Germany, France, Italy and the United Kingdom. Paunov and Rollo (2016) emphasize that using the Internet also increases productivity in developing countries, and find that firms with the most sophisticated productivity experience the greatest gains from this technology. The massive shift within the media to online channels has made the web a center of culture, education, politics and personal relationships, which has made the Internet as important in the labor market as in daily life. This technology can help reduce individuals’ sense of isolation, social exclusion and decrease the impact of physical disabilities, which also affect productivity (Foley, 2004; Chigona etal., 2009; Dobransky & Hargittai, 2016). Big changes in technology are not required in order to produce significant effects on the quality of life. Blanco and Vargas (2014), for instance, show that sending messages to vulnerable populations explaining their rights and where they can ask for them increases access to public aid and their welfare. Thus, this study explores the uses of technology in the developing world, which can inform policies and projects designed to increase the use of the Internet. 3. Data and Empirical Strategy There is generally a positive correlation between Internet users and income around the world (Katz & Murphy, 1992; Colecchia & Schreyer, 2002; S.-Y. T. Lee etal., 2005). Figure A-1 shows that an increase in the number of Internet users is positively correlated with the logarithm of per capita GDP. However, there is heterogeneity between regions. The developed world is at the top right of the figure with high levels of income and Internet users, while Latin America is in the middle of the figure with around 45 Internet users per 100 people and lower income levels. Whereas in 1994 only 0.10% of Colombians used the Internet, by 2016 this had increased to 58%, representing an average growth of 22% per year over the last two decades (World Bank, 2015). The United States, a benchmark in the developed world, surpassed Colombia’s current Internet penetration rate in 2002, and in the last year it reached 76% (World Bank, 2015). However, the average growth in the United States over the last 20 years has only been 8% per year, a much lower rate than in Colombia (World Bank, 2015)1. To analyze the hypotheses in Colombia, this study uses the Great Integrated Household Survey (GEIH, acronym in Spanish) conducted yearly from 2009 to 2011. This is a nationally representative cross-section survey and it contains demographic and industry data at the industry level. The dependent variable is standardized monthly labor income, and the sample is restricted to workers aged 1 The Colombian government invests 0.8% of its total budget in ICT projects; this sector ranks 15 out of 30 with the highest investment (MinHacienda, 2014). U-shaped wage curve and the Internet: The Colombian case / Diego A. Martin 179 18 to 65. This sub-sample is used in order to be able to compare the Colombian case with other countries with different levels of income, presented by DiMaggio and Bonikowski (2008) and Benavente etal., (2011). The results are robust to using the full population. Only this wave included questions on whether respondents had used the Internet in the last 12 months and where they accessed the Internet2. The survey also asked respondents whether they use the Internet for education or finance, which I used to analyze productive activities3. Finally, I used responses related to the frequency of Internet use in my robustness checks 4 . Controlling for economic activities, occupation and firm size is one of the most important aspects of this study because this information proxies for productivity, which is an important improvement compared with other studies. These variables reduce possible omitted variable bias because they create a cell, as small as possible, in which two workers have the same ability, but one uses the Internet and the other does not. Economic activity is divided into 13 sectors using the Classification of All Economic Activities (ISIC, 2008): agriculture, mining, manufacturing, electricity, construction, transport, financial service, housing sector, public administration, education, health and domestic service. Occupations are classified into 10 categories using the Standard Occupational Classification System (SOCPC, 2010): professional specialty, executive and managerial, service, sales, machine operator, cleaners and laborers, professional products, transportation, farming and mining. Firm size comprises five categories depending on the number of workers. Each category of the variable is included as a fixed effect in all estimations. Therefore, these covariates indicate the productivity of each worker and compare Internet users and non-users in the same firm’s economic activity (sector), in the activities that workers perform (occupation) and in the places they work (firm size). Using this database, it is possible to build a general profile of Internet users. Table1 presents the respondents’ individual characteristics by location of access and productive activities on the web. It is clear that these aspects are not mutually exclusive, because an individual can use the Internet anywhere. On average, Internet users are women, younger, better educated, well paid, and have more experience with technology than non-users. Around 40% of people in both groups are homeowners, and thus enjoy a similar standard of living. Both groups have similar locations of access and activities in the Internet. These characteristics 2 The options were accessing the Internet at home, at work, in educational institutions, in free public access centers, in paid access centers, in the house of another person. 3 The original question was worded “For which of the following services or activities, do you use the Internet?” The options are obtaining information, communication, electronic banking and other financial services, education and learning, transactions with government agencies, entertainment. 4 The original question was worded “How often do you use Internet?” The options are at least once a day; at least once a week, but not every day; at least once a month, but not every week and less than once a month. Estudios de Economía, Vol.45 - Nº2180 show that it is possible to find similar individuals who use and do not use the Internet, which is an advantage of using a household survey. Table2 shows Internet users distributed by firm size, economic sector, occupation and type of worker. Whereas big companies have the highest number of Internet users (78% of workers), the number of productive users is smaller than non-productive users in all firm sizes. Small companies seem to have the poorest control over employees’ activities on the web, because only 22% of workers use the Internet for education and financial activities in these companies. Access to and activities on the web by occupation does not present a ushaped distribution in Colombia. The financial services sector, which has the highest percentage of Internet users and average wage, has more than twice as many people using the Internet for education and financial services as the lower-income domestic service sector. Indeed, there are almost three white-collar workers for every two blue-collar workers using this technology. This does not represent evidence against a possible u-shape in the wage curve given the fact that the low supply of Internet users in a certain occupation can result in a wage premium depending on whether there is a large demand for workers with this skill in a particular sector. The literature is divided over which way of estimating the relationship between new technologies and earnings is more effective: parametric models using TABLE 1 SOCIO-DEMOGRAPHIC STATISTICS BY PLACE OF INTERNET USE Nonusers Everywhere At work At home Productive non-users users Logarithm of income 13.276 13.632 13.839 13.836 13.588 13.699 Standardized income –0.326 0.211 0.590 0.519 –0.152 0.301 Panel A: Population Schooling 11.812 13.797 14.545 14.513 13.614 14.468 Age 35.757 33.073 34.871 35.178 31.004 35.074 Women 0.479 0.525 0.562 0.524 0.481 0.603 Panel B: Technology experience Telephone at home 0.419 0.618 0.650 0.779 0.630 0.618 Satellite TV at home 0.653 0.808 0.837 0.907 0.814 0.799 PC at home 0.263 0.615 0.669 0.925 0.640 0.614 Panel C: Proxy for wealth Property owner 0.415 0.423 0.440 0.478 0.407 0.463 Notes: The average of each variable is presented in the table. The sample is restricted to employed workers aged 18 to 65. The standard deviation of the labor income is $262 and the mean is $295 for this sample. Internet users are workers accessing from any of the following places: at home, at work, in educational institutions, in free public access centers, in paid access centers, in the house of another person (relative, friend, neighbor). Source: Colombian Great Integrated Household Survey (GEIH), 2009 to 2011. U-shaped wage curve and the Internet: The Colombian case / Diego A. Martin 187 Second, the matching method has to close most of the gaps in observable and unobservable characteristics between Internet users and non-users. Table5 presents the differences in averages between treated and controlled individuals before and after the matching process, using two different methods: the nearestneighbor distance with five neighbors and the Mahalanobis distance. In the unmatched sample, all the differences are statistically different from zero. After using the nearest-neighbor method, there are still many differences between Internet users and non-users; even using 1 neighbor or 10 neighbors the gaps do not close, as shown by TableA-2. Using the Mahalanobis distance, however, closes the gaps for all the variables, showing that the matrix of variances and covariances, in this case, is much more informative for matching individuals. Closing the gaps reduces the possibility of reverse causality, because the small cells compare a clone in socio-demographic characteristics, in the same firm, performing the same occupation, and it is unlikely that workers were already rich before accessing the Internet. Finally, estimated coefficients have to be robust to changes including controls and the matching method used. Table6 presents a statistically significant positive correlation between the Internet and labor income in two matching methods and different specifications. Column 5 shows that Internet users earn 6.5% of one standard deviation more than non-users, using the Mahalanobis distance method and all the covariates. Is this coefficient large or small in magnitude? Considering that 54% of Colombia’s population lives with one minimum wage income, which is defined as the minimum amount of money to have a basic standard of living Dep. Var: Internet use (1) Internet at work (2) Internet at home (3) Productive users (4) Administrative 0.175*** (0.000) 0.368*** (0.000) 0.301*** (0.001) 0.153*** (0.001) Support Service 0.072*** (0.000) 0.202*** (0.001) 0.133*** (0.001) 0.008*** (0.001) Sales 0.101*** (0.000) 0.256*** (0.000) 0.174*** (0.001) 0.039*** (0.001) Machine operator 0.042*** (0.001) 0.103*** (0.001) 0.074*** (0.001) –0.030*** (0.001) Equip. cleaners and laborers 0.036*** (0.000) 0.106*** (0.001) 0.058*** (0.001) 0.012*** (0.001) Precision production 0.026*** (0.000) 0.049*** (0.001) 0.033*** (0.001) –0.027*** (0.001) Transportation 0.015*** (0.000) 0.019*** (0.001) 0.054*** (0.001) –0.045*** (0.001) Municipality and time FE üüüü Notes: Robust standard errors are shown in parentheses. *** is significant at the 1% level, ** is significant at the 5% level, * is significant at the 10% level. FE denotes municipality and time fixed effects. Estudios de Economía, Vol.45 - Nº2188 (MinTrabajo, 2014). An increase of 6.5% of one standard deviation is equivalent to 11% above one minimum salary. This means that using ICTs can definitely help workers pass this threshold. How does this coefficient change over time? Using the available information, TableA-3 estimates the matching model for each year separately. Whereas Internet users earned 7.8% more than non-users in 2009, the coefficient is 22% less in 2011. The literature identifies a wage premium depending on where people use the Internet. The matching method presented in Table7 shows that differences in wages amount to 28.4% and 17.4% of one standard deviation from using the Internet at work and at home, respectively. This means that the labor market TABLE 5 DIFFERENCE IN MEANS BETWEEN TREATED AND CONTROLS BEFORE AND AFTER MATCHING SELECTED VARIABLES Variable Unmatched (U) Matched (M) Mean Difference Treated Control Schooling more than 12 U 0.603 0.190 0.412*** M-NN 0.503 0.504 –0.001 M-MD 0.503 0.503 0.000 Age U 32.989 35.543 –2.554*** M-NN 33.856 33.565 0.291*** M-MD 33.847 33.871 –0.024 Age squared U 1194.20 1374.30 –180.1*** M-NN 1259.70 1242.90 16.80*** M-MD 1259.10 1258.90 0.200 Woman U 0.493 0.525 –6.800*** M-NN 0.492 0.509 –0.017** M-ND 0.492 0.492 0.000 Telephone at home U 0.603 0.399 0.205*** M-NN 0.567 0.563 0.003*** M-MD 0.566 0.566 0.000 Satellite TV at home U 0.812 0.651 0.161*** M-NN 0.792 0.787 0.006** M-MD 0.792 0.792 0.000 PC at home U 0.608 0.239 0.369*** M-NN 0.528 0.517 0.011*** M-MD 0.529 0.528 0.001 Property owner U 0.406 0.400 0.006** M-NN 0.409 0.409 –0.000 M-MD 0.409 0.409 0.000 White collar U 0.430 0.139 0.291*** M-NN 0.337 0.347 –0.010*** M-MD 0.338 0.337 0.000 Notes: Robust standard errors are shown in parentheses. *** is significant at the 1% level, ** is significant at the 5% level, * is significant at the 10% level. The row M-NN shows the result using the Nearest-Neighborhood matching method with 5 neighbors. The row M-MD presents the results using the Mahalanobis distance method. U-shaped wage curve and the Internet: The Colombian case / Diego A. Martin 189 TABLE 6 NON-PARAMETRIC MODEL: PROPENSITY SCORE MATCHING FOR INTERNET USE Dep. Var: Standardized labor income (1) (2) (3) (4) (5) Nearest Neighbor 0.166*** 0.123*** 0.115*** 0.094*** 0.068*** (0.005) (0.005) (0.004) (0.004) (0.005) Mahalanobis Distance 0.165*** 0.122*** 0.114*** 0.092*** 0.065*** (0.005) (0.005) (0.004) (0.004) (0.004) Controls Socio-demographic üüüüü Firm size üüüü Economic sector üüü Occupation ü ü Municipality and time FE ü R-squared 0.369 0.383 0.394 0.419 0.429 No. of observations 421,349 421,349 418,798 415,813 415,813 Notes: Robust standard errors are shown in parentheses. *** is significant at the 1% level, ** is significant at the 5% level, * is significant at the 10% level. FE denotes municipality and time fixed effects. The nearest-neighbor method uses 5 neighbors. TABLE 7 PLACES AND ACTIVITIES Dep. Var: Standardized labor income Treatment: Internet Use (1) At work (2) At home (3) Productive users (4) OLS 0.069*** (0.005) 0.291*** (0.006) 0.183*** (0.006) 0.076*** (0.005) Matching Mahalanobis Distance 0.065*** (0.005) 0.284*** (0.005) 0.174*** (0.005) 0.071*** (0.004) All controls ü ü ü ü R-squared 0.429 0.439 0.432 0.430 No. of observations 451,813 451,813 451,813 451,813 Notes: Robust standard errors are shown in parentheses. *** is significant at the 1% level, ** is significant at the 5% level, * is significant at the 10% level. All the controls are sociodemographic, firm size, economic sector, occupation, municipality and time fixed effects. The control group for columns 1, 2 and 3 is Internet non-users, and for column 4 it is Internet users who use the Internet for non-productive activities. Internet users are workers accessing from any of the following places: at home, at work, in educational institutions, in free public access centers, in paid access centers, in the house of another person (relative, friend, neighbor). The number of observations shows the individuals in the common support for the PSM using Mahalanobis Distance. The OLS number of observations is 462,014. Estudios de Economía, Vol.45 - Nº2190 (imperfectly) identifies what people do on the web, and higher incomes relate to using the Internet for productive activities. What about actions using this technology? The last column in Table7 shows that workers who use the Internet for education and financial services earn 7.1% of one standard deviation more than those who use the Internet for less productive tasks, such as searching for music and videos. An additional dimension analyzed in this paper is the intensity of using this technology. TableA-4 estimates the four treatments only for people who claim to use the Internet every day. Internet users and productive users who access it daily earn 18% and 13%, respectively, more than those who do not necessarily use it daily6. Is there a heterogeneous effect depending on the workers’ skill levels? Table8 estimates the results of using the Internet divided into occupations. Those who benefited the most from using the Internet are the lowest-skilled workers in farming or mining, who earn 20% of one standard deviation more than non-users in the same occupation, followed by the highest-skilled workers in professional specialties, who earn 0.2% less than farmers or miners who use the Internet. Seven sectors out of ten have a positive and statistically significant coefficient between 2.7% and 10% of one standard deviation. Figure A-3 summarizes the results of this table, showing that the effect of using the Internet in the Colombian labor market has a u-shaped distribution depending on occupation. 5. Conclusion Although a large international literature demonstrates that ICTs have a positive impact on salaries, the evidence in the developing world is scarcer. This paper uses the Colombian case to analyze the heterogeneous effects of Internet use in the wage distribution curve, using matching methods. It finds a positive and significant effect of using the Internet. Whereas the lowest wage premium is for Internet users in the middle of the skill distribution, the highest increase in income is for Internet users who work in the lowest-skilled occupations, farming or mining, followed by highly skilled professionals. In relation to other international experiences, the effect of using ICTs has decreased over time due to an increase in the supply of workers who know how to use these tools. Colombia presents the same pattern as developed countries: even though the United States has more Internet users than Colombia, both present a decline in the effect of Internet use over time. In terms of heterogeneous effects across abilities, the evidence for the developed world partially holds in a developing country such as Colombia. On the one hand, there is a complementarity between 6 Guataqui, Martin, and Porto (2016) find that self-employed workers and salaried workers are very different in several characteristics, including who pays for the Internet. Whereas the former probably pay directly for this service, the latter do not. Thus, Table A-5 shows that self-employed workers earn around twice as much as salaried workers who use the Internet. U-shaped wage curve and the Internet: The Colombian case / Diego A. Martin 191 TABLE 8 INTERNET USE BY SKILL LEVEL Dep. Var: Standardized labor income Farming and mining (1) Transportation (2) Precision production (3) Treatment: Internet use Cleaners and laborers (4) Machine operator (5) Sales (6) Service (7) Administrative support (8) Executive administrative (9) Professional specialty (10) OLS 0.207*** 0.049*** 0.060*** 0.027*** 0.074*** 0.087*** 0.101*** 0.096*** 0.156*** 0.184*** (0.058) (0.012) (0.011) (0.007) (0.037) (0.011) (0.009) (0.010) (0.031) (0.020) Matching Mahalanobis Distance 0.196*** 0.048*** 0.058*** 0.023*** 0.071*** 0.084*** 0.098*** 0.091*** 0.152*** 0.178*** (0.061) (0.013) (0.014) (0.006) (0.039) (0.012) (0.009) (0.011) (0.033) (0.021) All controls ü ü ü ü ü ü ü ü ü ü R-squared 0.472 0.288 0.285 0.247 0.348 0.241 0.397 0.351 0.336 0.361 No. of observations 3,888 33,010 27,618 63,671 3,029 72,478 45,160 33,937 64,225 68,796 Notes: Robust standard errors are shown in parentheses. *** is significant at the 1% level, ** is significant at the 5% level, * is significant at the 10% level. All the controls are socio-demographic, firm size, economic sector, occupation, municipality and time fixed effects. Internet users are workers accessing from any of the following places: at home, at work, in educational institutions, in free public access centers, in paid access centers, in the house of another person (relative, friend, neighbor). The number of observations shows the individuals in the common support for the PSM, using Mahalanobis Distance. The OLS number of observations is around 10% more of the mentioned. Occupation is built using the Standard Occupational Classification System (SOCPC, 2010). Estudios de Economía, Vol.45 - Nº2192 higher-skilled workers and the Internet. On the other hand, this relationship is not unique and lower-skilled workers also gain from using this technology. As shown above, there is a low supply of Internet users in this occupation, compared with other sectors, which makes Internet users very valuable in the labor market. One of the key findings of the paper is that the Internet seems to have the smallest impact on the middle of the skill distribution. One possible reason for this is the lack of complementarity of ICTs with these occupations. Transportation workers, cleaners and laborers are at the bottom of the wage distribution presumably because the Internet is not frequently used in these occupations. There are some companies looking for ingenious solutions to reduce costs in the communication between demand and supply in these sectors, which may suggest that perfecting these applications could increase the magnitude of the effect. Another possible reason for this finding is that these occupations are at a high risk of disappearing in the future. Workplace automation could decrease the demand for workers in these tasks, at the expense of an increase in the demand for these new technologies. It is therefore necessary to delve more deeply into the possible impacts of these applications in the developing world, which despite going a step backwards in terms of absorbing ICTs can lead to undesired effects. These results show that the Internet could reduce the income inequality gap between occupations, which is already large in Latin America. Colombia, for example, has one of the highest Gini coefficients in the world, around 51. Some previous studies show that ICTs can reinforce pre-existing socio-economic inequalities, although the u-shape curve in the Colombian labor market wage distribution shows that the Internet serves as an equalizing factor in the wage distribution, at least between the highestand lowest-skilled workers. The developing world needs public policies on ICTS to help overcome poverty. Since the worldwide trend is to increase access to ICTs, future studies should examine the heterogeneous effects of such access depending on the place of access and particularly the effects of activities on the web. It is very important to understand which access-use patterns are more directly associated with an increase in productivity, and which sectors benefit the most from these policies. References Abadie, A., & Imbens, G. W. (2006). “Large sample properties of matching estimators for average treatment effects”. Econometrica, 74 (1), 235-267. Abadie, A., & Imbens, G. W. (2016). “Matching on the estimated propensity score”. Econometrica, 84 (2), 781-807. Acemoglu, D. (1999). “Changes in unemployment and wage inequality: An alternative theory and some evidence”. The American Economic Review, 89 (5), 1259-1278. Retrieved from http://www.jstor.org/stable/117057 Acemoglu, D., & Autor, D. (2011). “Skills, tasks and technologies: Implications for employment and earnings”. Handbook of Labor Economics, 4, 1043-1171. U-shaped wage curve and the Internet: The Colombian case / Diego A. Martin 193 Angrist, J. D., & Pischke, J.-S. (2009). Mostly harmless econometrics: an empiricist’s companion (Vol. 1). Princeton, NJ: Princeton University Press. Arabsheibani, G. R., Emami, J., & Marín, A. (2004). “The impact of computer use on earnings in the UK”. Scottish Journal of Political Economy, 51 (1), 82-94. Arceneaux, K., Gerber, A. S., & Green, D. P. (2006). “Comparing experimental and matching methods using a large-scale voter mobilization experiment”. Political Analysis, 14 (1), 37-62. Autor, D. H. (2015). “Why are there still so many jobs? The history and future of workplace automation”. The Journal of Economic Perspectives, 29 (3), 3-30. Retrieved from http://www.jstor.org/stable/43550118 Autor, D. H., & Dorn, D. (2013). “The growth of low-skill service jobs and the polarization of the US labor market”. The American Economic Review, 103 (5), 1553-1597. Retrieved from http://www.jstor.org/stable/42920623 Autor, D. H., Levy, F., & Murnane, R. J. (2003). “The skill content of recent technological change: An empirical exploration”. The Quarterly Journal of Economics, 118 (4), 1279-1333. Retrieved from http://www.jstor.org/ stable/25053940 Barrera-Osorio, F., & Linden, L. L. (2009). The use and misuse of computers in education: evidence from a randomized experiment in Colombia. Bartel, A., Ichniowski, C., & Shaw, K. (2007). “How does information technology affect productivity? Plant-level comparisons of product innovation, process improvement, and worker skills”. The Quarterly Journal of Economics, 122 (4), 1721-1758. Re-trieved from http://www.jstor.org/ stable/25098887 Benavente, J. M., Bravo, D., & Montero, R. (2011). “Wages and workplace computer use in Chile”. The Developing Economies, 49 (4), 382-403. Blanco, M., & Vargas, J. F. (2014). “Can SMS technology improve low take-up of social benefits?” Peace Economics, Peace Science and Public Policy, 20 (1), 61-81. Borghans, L., & Ter Weel, B. (2004). “Are computer skills the new basic skills? The returns to computer, writing and math skills in Britain”. Labour Economics, 11 (1), 85-98. Card, D., & DiNardo, J. (2002). “Skill-biased technological change and rising wage inequality: Some problems and puzzles”. Journal of Labor Economics, 20 (4), 733-783. Retrieved from http://www.jstor.org/ stable/10.1086/342055 Chigona, W., Beukes, D., Vally, J., & Tanner, M. (2009). “Can mobile internet help alleviate social exclusion in developing countries?” The Electronic Journal of Information Systems in Developing Countries, 36. Chiswick, B. R., & Miller, P. W. (2007). “Computer usage, destination language proficiency and the earnings of natives and immigrants”. Review of Economics of the Household, 5 (2), 129-157. Colecchia, A., & Schreyer, P. (2002). “ICT investment and economic growth in the 1990s: is the United States a unique case? A comparative study of nine OECD countries”. Review of Economic Dynamics, 5 (2), 408-442. Estudios de Economía, Vol.45 - Nº2194 Cortés, G. M., etal. (2016). “Where have the middle-wage workers gone? A study of polarization using panel data”. Journal of Labor Economics, 34 (1), 63-105. Dehejia, R. H., & Wahba, S. (2002). “Propensity score-matching methods for nonexperimental causal studies”. The Review of Economics and Statistics, 84 (1), 151-161. DiMaggio, P., & Bonikowski, B. (2008). “Make money surfing the web? The impact of internet use on the earnings of us workers”. American Sociological Review, 73 (2), 227-250. DiNardo, J. E., & Pischke, J.-S. (1997). “The returns to computer use revisited: Have pencils changed the wage structure too?” The Quarterly Journal of Economics, 112 (1), 291-303. Retrieved from http://www.jstor.org/ stable/2951283 Dobransky, K., & Hargittai, E. (2016). “Unrealized potential: Exploring the digital disability divide”. Poetics, 58, 18-28. Entorf, H., Gollac, M., & Kramarz, F. (1999). “New technologies, wages, and worker selection”. Journal of Labor Economics, 17 (3), 464-491. Foley, P. (2004). “Does the internet help to overcome social exclusion?”. Electronic Journal of e-government, 2 (2), 139-146. García-Murillo, M. (2013). “Does a government web presence reduce perceptions of corruption?” Information Technology for Development, 19 (2), 151-175. Goos, M., & Manning, A. (2007). “Lousy and lovely jobs: The rising polarization of work in Britain”. The Review of Economics and Statistics, 89 (1), 118-133. Goos, M., Manning, A., & Salomons, A. (2009). “Job polarization in Europe”. The American Economic Review, 99 (2), 58-63. Goss, E. P., & Phillips, J. M. (2002). “How information technology affects wages: Evidence using internet usage as a proxy for it skills”. Journal of labor Research, 23 (3), 463-474. Guataqui, C., Martin, D., & Porto, I. (2016). El raro caso de los trabajadores cuenta propia en Colombia. Unpublished. Gust, C., & Márquez, J. (2004). “International comparisons of productivity growth: the role of information technology and regulatory practices”. Labour Economics, 11 (1), 33-58. Heckman, J. J., Ichimura, H., & Todd, P. (1998). “Matching as an econometric evaluation estimator”. The Review of Economic Studies, 65 (2), 261-294. Heckman, J. J., LaLonde, R. J., & Smith, J. A. (1999). “The economics and econometrics of active labor market programs”. Handbook of Labor Economics, 3, 1865-2097. ISIC (2008). International standard industrial classification of all economic activities. Revision 4. United Nations Publications. Retrieved from http:// unstats.un.org/unsd/publication/seriesM/seriesm 4rev4e.pdf Katz, L. F., & Murphy, K. M. (1992). “Changes in relative wages, 1963-1987: supply and demand factors”. The Quarterly Journal of Economics, 107 (1), 35-78. U-shaped wage curve and the Internet: The Colombian case / Diego A. Martin 195 Kraut, R., Kiesler, S., Boneva, B., Cummings, J., Helgeson, V., & Crawford, A. (2002). “Internet paradox revisited”. Journal of Social Issues, 58 (1), 49-74. Krueger, A. B. (1993). “How computers have changed the wage structure: Evidence from microdata”. The Quarterly Journal of Economics, 108 (1), 33-60. LaLonde, R. J. (1986). “Evaluating the econometric evaluations of training programs with experimental data”. The American Economic Review, 604-620. Lee, S.-H., & Kim, J. (2004). “Has the internet changed the wage structure too?”. Labour Economics, 11 (1), 119-127. Lee, S.-Y. T., Gholami, R., & Tong, T. Y. (2005). “Time series analysis in the assessment of ICT impact at the aggregate level-lessons and implications for the new economy”. Information & Management, 42 (7), 1009-1022. McDonald, S., & Crew Jr, R. E. (2006). “Welfare to web to work: Internet job searching among former welfare clients in Florida”. J. Soc. & Soc. Welfare, 33, 239. Michaels, G., Natraj, A., & Van Reenen, J. (2014). “Has ict polarized skill demand? Evidence from eleven countries over twenty-five years”. Review of Economics and Statistics, 96 (1), 60-77. MinHacienda (2014). Presupuesto de 2015. Ministerio de hacienda y crédito público. Ministerio de Hacienda. Colombia. MinTrabajo (2014). Definición de salario mínimo. Ministerio de Trabajo. Colombia. Navarro, L. (2010). The impact of internet use on individual earnings in Latin America (Tech. Rep.). Institute for Advanced Development Studies. Oosterbeek, H., & Ponce, J. (2011). “The impact of computer use on earnings in a developing country: Evidence from Ecuador”. Labour Economics, 18 (4), 434-440. Paunov, C., & Rollo, V. (2016). “Has the internet fostered inclusive innovation in the developing world?”. World Development, 78, 587-609. Porto, I. (2016). Impacto del programa de alimentación escolar en el trabajo infantil: Una aproximación desde la toma de decisiones familiares. Rosenbaum, P. R., & Rubin, D. B. (1983). “The central role of the propensity score in observational studies for causal effects”. Biometrika, 70 (1), 41-55. Rubin, D. B. (1978). “Bias reduction using Mahalanobis metric matching”. ETS Research Report Series, 1978(2). Smith, J. A., & Todd, P. E. (2005). “Does matching overcome Lalonde’s critique of nonexperimental estimators?” Journal of Econometrics, 125 (1), 305-353. SOCPC (2010). Standard occupational classification. Bureau of Labor Statistics On behalf of the Standard Occupational Classification Policy Committee. Retrieved from https://www.bls.gov/soc/soc structure 2010.pdf Steinfield, C., Ellison, N. B., & Lampe, C. (2008). “Social capital, self-esteem, and use of online social network sites: A longitudinal analysis”. Journal of Applied Developmental Psychology, 29 (6), 434-445. Retrieved from http://www.sciencedirect.com/ science/article/pii/S0193397308000701 (Social Networking on the Internet) doi: https://doi.org/10.1016/j. appdev.2008.07.002 World-Bank. (2015). World Bank indicators. The World Bank Group. Estudios de Economía, Vol.45 - Nº2196 Appendix TABLE A-1 DIFFERENCES IN MEAN BY TYPE OF WORKER Variables Salaried worker Self-employed worker Difference Logarithm of labor income 13.526 13.111 0.416*** (0.0013) (0.0024) (0.005) Age 33.385 37.010 -3.625*** (0.0188) (0.0243) (0.054) Age squared 1,217.343 1,488.661 -271.318*** (1.392) (1.900) (4.112) Schooling 13.035 12.594 0.442*** (0.0047) (0.0053) (0.013) 12 or more years of education 0.512 0.412 0.100*** (0.0009) (0.0010) (0.003) Telephone at home 0.622 0.556 0.066*** (0.0009) (0.0011) (0.003) PC at home 0.770 0.717 0.053*** (0.0007) (0.0010) (0.002) Satellite TV at home 0.519 0.473 0.047*** (0.0009) (0.0011) (0.003) House owner 0.399 0.427 -0.028*** (0.0009) (0.0011) (0.003) 2-5 employees 0.192 0.823 -0.631*** (0.0007) (0.0008) (0.002) 6-10 employees 0.074 0.033 0.040*** (0.0004) (0.0003) (0.001) 11-50 employees 0.159 0.040 0.119*** (0.0006) (0.0004) (0.001) 51-100 employees 0.056 0.009 0.047*** (0.0004) (0.0002) (0.001) 101 and over employees 0.520 0.095 0.425*** (0.0009) (0.0006) (0.002) Agriculture 0.016 0.021 -0.006*** (0.0002) (0.0003) (0.001) Mining 0.001 0.001 0.000 (0.0000) (0.0000) (0.000) Manufacturing 0.171 0.120 0.051*** (0.0007) (0.0007) (0.002) Electricity 0.013 0.002 0.010*** (0.0002) (0.0001) (0.000) Construction 0.033 0.048 -0.015*** (0.0003) (0.0004) (0.001) Commerce 0.235 0.331 -0.095*** (0.0007) (0.0010) (0.002) Transport 0.077 0.130 -0.053*** (0.0004) (0.0007) (0.002) Financial service 0.038 0.007 0.032*** (0.0003) (0.0001) (0.001) Housing sector 0.084 0.111 -0.028*** (0.0005) (0.0007) (0.002)