- Jose E. Galdon-Sanchez, Professor of Economics, Universidad Pública de Navarra (UPNA)
- Ricard Gil, Associate Professor, Smith School of Business, Queen’s University
This blog article is derived from the authors’ paper titled Big Data Adoption and Employment in Small and Medium Enterprises,, a project of the Economics of Digital Services (EODS) research initiative led by Penn’s Center for Technology, Innovation and Competition (CTIC) and The Warren Center for Network & Data Services. CTIC and The Warren Center are grateful to the John S. and James L. Knight Foundation for its generous support of the EODS initiative.
Technology and the diffusion of knowledge in their many forms have been responsible for large advances in societal well-being for as long as we can keep count. For example, we would easily agree that the quality, durability, and reliability of goods produced nowadays is an order of magnitude higher than those of goods produced a few decades ago, and potentially at more affordable prices. Yet it is also easy to agree that production of goods and services is more mechanized today than it was 100 years ago. This means that technological progress may have increased employment in certain occupations and created unemployment in others.
Technological progress can indeed have ambiguous effects on employment. On the one end, it can raise labor productivity, thus increasing firms’ demand for labor, increasing employment, and reducing unemployment. On the other end, it may also raise the returns of investment in capital, consequently driving firms to substitute capital for labor reducing and increasing unemployment. This ambiguity is the center of the debate around the adoption and diffusion of recent technologies such as robots in manufacturing and AI most generally. There is little debate, however, regarding the impact of the adoption of Big Data and data analytics on employment.
In this project, we aimed to contribute to this debate by estimating the impact of a Big Data information-sharing program deployed by a large European bank in Spain among its small and medium-sized enterprises (SMEs hereafter) customers on job creation and unemployment at the municipal level. We estimated both the overall effect and heterogeneous impacts on different types of contracts (indefinite employment versus temporary short-term contracts) and on different gender-age groups of workers (male vs. female; younger than 25, 25-44, 45, and older).
In previous research, we evaluated the impact of adoption of this program, and we found a causal increase on establishment revenue by 9%. Here we studied whether this increase in establishment revenue can cause an increase in job creation and decrease unemployment. We hypothesized that even if the adoption occurs at the establishment level, we may observe an increase at the municipal level because of two distinct mechanisms directly related to establishment-level adoption. First, those establishments adopting the technology may increase employment directly if their marginal productivity of labor directly increases upon program adoption. Second, non-adopters may also increase their number of employees to compete directly with adopters and potentially keep up with the increase in their rivals’ customer service.
An important contribution of this study is that we investigated whether adoption of Big Data and data analytics technology biases job creation and unemployment levels towards a particular gender and/or age groups in the population. This part of our analysis has important policy implications because gender bias in the marketplace can manifest on dimensions other than salaries and compensation such as number of jobs and contract precariousness (temporary jobs versus indefinite contracts). Our analysis also acknowledges the difficulties of young people (under 25 years old) to find stable jobs in southern European economies or of those later in their careers between jobs and looking for a new job (above 45 years old). Our project investigated if Big Data and data analytics adoption corrects or exacerbates existing market bias against these gender-age groups in the labor force.
In short, our analysis found that adoption of Big Data and data analytics was associated with more employment and lower unemployment at the city level. A closer look at these results showed that adoption of this technology improve job quality (more indefinite jobs instead of short-time contracts) and increase employment for women, youth and older workers. These are positive and desirable outcomes that ought to shape future policies facilitating the adoption of Big Data and data analytics technologies by SMEs.
Our Big Data and data analytics program adoption information come from a large European bank. Amidst its large market share in Spain, the bank launched a pilot program for its SMEs clients in one region of Spain in the fall of 2014 and went national in the spring of 2016. The program aimed to bring the benefits of Big Data and data analytics technology to the SMEs using the bank’s credit card POS (point of sales). The bank provided this program for free, and adoption was voluntary. While the bank did not compensate its employees for the diffusion of this program, bank employees would offer the adoption of the program to their client portfolio as a source of value added to an already existing business relationship.
After establishments adopted the program, the bank generated a monthly report for each adopter, which became available through the program’s online platform. The report contained summary statistics regarding the number and value of credit card transactions in the previous month. The report disaggregated this information on credit card transactions by client demographic groups such as age, gender, and zip code, as well as other classifications such as new vs. returning customers or the time and day of transactions. The report also contained the same set of aggregated information for business competitors in the same zip code. This set of information on each store’s direct competitors provided a reference point and allowed program participants to discover differences between their own performance and client portfolio and those of their closest competitors. In other words, each monthly report effectively provided precise market research information on the local market in which each program participant operated.
The data used in this project comes mainly from two sources. First, there was proprietary data on Big Data and data analytics adoption at the establishment level from the bank that rolled out the big data program previously described. Program adopters are spread across 1,206 municipalities representing all provinces in the country. The average population of an adopting municipality is 60,389 from a population average of 12,567. Second, we used data from the Labor Department of the Spanish Government (Ministerio de Trabajo y Economia Social) on monthly changes in municipal job creation and monthly levels of unemployment from November 2014 to October 2018. This data set is rich in that it disaggregates unemployment by age, gender and sector, and job creation by sector and type of contract (temporary or indefinite contracts as well as indefinite contracts converted from temporary contracts).
After merging the data from both sources, our final data set is composed by all cities in Spain out of which 957 are adopting cities distributed as follows: 247 adopting cities with population levels below 5,000; 196 adopting cities with population levels between 5,000 and 10,000; 395 adopting cities with population levels between 10,000 and 50,000; and 119 adopting cities with population levels above 50,000. The average city in our final sample creates 664 jobs in a month and has 1,453 unemployed people in any given month. Service sectors are responsible for more jobs created and more people unemployed, 463 and 970 respectively, than non-service sectors combined (agriculture, construction, industry), 212 and 483 respectively.
Big Data impact on employment
Before presenting our results, a brief note on identification and endogeneity. Program adoption at the establishment level is obviously not exogenous or random. This would be a real concern if we were estimating the relationship between technology adoption and labor demand at the establishment level, but our dependent variables of job creation and unemployment levels are aggregated at the municipality level which attenuates the concern that aggregated job creation and unemployment levels are determined by establishment-specific shocks. Yet we may be concerned if adoption at the municipal level is driven by city-specific shocks varying over time. First, let us note here that if this were the case it is hard to justify that we did not see more widespread program adoption. Second, our identifying assumption is that the error term in our regression specifications is orthogonal to the adoption decision variable conditional on our controls, that is, city fixed effects, city-sector fixed effects, and province-specific time trends.
Moreover, our specifications are also informative about differences on the impact of adoption on employment across age and gender groups, which may uncover patterns of substitution across such groups correlated with potentially unobserved endowments of skill sets. Differences in results across these variables are informative of bias correction driven by technology adoption and hardly justifiable for gender-specific or age-specific knowledge embedded in the new technology and city-specific shocks.
Our evidence exploits two different empirical strategies. First, we exploited within-city variation comparing adopters and non-adopters within a province and saw that adoption increases total employment by 1.4%. This increase in total job creation is the result of increases in indefinite and temporary contracts for both men and women. The most drastic increases come from temporary contracts that are converted into indefinite contracts for both men and women, with 14% and 16% increases respectively.
When measuring the impact of technology on unemployment, our findings showed no differences in total unemployment levels due to program adoption. Having said this, we found decreases in unemployment levels for men under 25, men over 45, and for women under 45. These findings are consistent with Big Data adoption correcting some of market biases against younger women and men, and old men looking for employment.
Our second empirical strategy exploited between-city variation by matching each adopting city with a non-adopting city with the closest population level in its province. Then for each match, we estimated difference-in-difference regressions, creating a dummy that takes value 1 for the adopting city in each match, a dummy that takes value 1 when the adopting city in the matched adopts the technology, and an interaction dummy by multiplying both dummies. The coefficient in the interaction variable is the coefficient of interest in our analysis. Our results using this empirical strategy showed no statistically significant change in total jobs created. Interestingly, we found statistically significant effects on the number of jobs created for men converted to indefinite contracts from temporary contracts (+12%) and on the number of indefinite contracts offered to women with a 9% increase in direct contracts and 14% increase in converted contracts from temporary to indefinite. We find no statistically significant change in unemployment levels, in total or by gender and age group.
Policy implications and conclusions
This project contributes to the current policy debate about the impact of technology on employment by examining the adoption and diffusion of a particular type of technology, namely, Big Data and data analytics, on city-level job creation and unemployment levels. In particular, we examined the diffusion of an information-sharing program that aimed to bring the benefits of Big Data and data analytics to SMEs without bearing any of the costs.
Our research has important implications for the future design of policies governing information sharing programs among firms and businesses as well as Big Data technologies and data analytics management. First, our findings are consistent with adoption driving up employment and job creation while, if anything, decreasing unemployment levels. Second, our findings show that the adoption of these technologies increases job creation for women under 45 years of age as well as young males under 25 and males 45 years of age and older. In a nutshell, adoption of this type of technology favors those gender-age groups more sensitive to unemployment. Therefore, our evidence would support actions where governments facilitate the creation, adoption, and diffusion of information-sharing programs among SMEs to increase employment, improve employment quality (indefinite contracts over temporary contracts), and reduce unemployment of women as well as both younger and older workers (male and female).
It is important to remember that in our setting (an average OECD economy), large firms (more than 50 employees) account for less than 1% of all firms in the country and 48% of employment whereas SMEs account for more than 50% of employment and almost 99% of firms. These patterns in the size distribution of firms and employment are representative for all industrialized and OECD countries. To the extent that our results provide estimates of the impact of Big Data and data analytics adoption of SMEs on local employment, intervention and government policy aiming that aim to correct for socially inefficient adoption is desirable if it is possible to increase employment and correct for market biases that drive the inefficient use of resources in local economies.