Does school pay off? How much?
Organisation: El Financiero (Costa Rica)
Publication Date: 04/08/2015
DescriptionCategory 3: Best News Data App of the Year (Small Newsroom) Objective: This web app aims to inform citizens about returns to education (the relationship between the number of years of schooling and income earned in the job market) based on scientific methods and the most reliable data in Costa Rica, in order to incentivize their investment in human capital to prevent young people from dropping out of school. Web app description: The web app lets citizens find out the average monthly salary someone with their profile can receive in Costa Rica’s job market. The estimate is based on seven personal characteristics (years of schooling, age, labor market, labor branch, gender, location and weekly worked hours). Once the web app calculates the user’s salary, it lets her interact with each of the seven variables to understand how each one has a decreasing or increasing effect on her salary (i.e. years of schooling), holding all the other variables constant. Finally, the web app lets the user share her results with her friends in social media to invite others to use our tool. Main findings: Using our app, each citizen could reach to her own and specific conclusions, but our model supports the following trends: 1. Obtaining a secondary school degree can increase a person’s average salary by 45%, compared to just having a primary school degree, holding all other variables constant. The effect of this difference in school degrees even influences the consumption patterns of individuals. For instance, families with individuals with higher school degrees spend more on health and cultural activities than those lower school degrees. 2. In general, obtaining a graduate school degree could double your salary compared to the average salary perceived from having a college degree,holding all other variables constant. 3. Years of schooling explains around 20% of the differences in monthly salaries. It is the second most influential factor according to our research. The first one is the increase of worked hours. 4. Graduate degrees are 62% more profitable in public sector jobs than on private sector jobs holding all other variables constant. 5.Women earn salaries 28% lower than men, holding all other variables constant. 6. The average peak earnings age in Costa Rica is 46. When comparing the peak salary conditional on the educational degree, the biggest increase in income comes from moving from a secondary education degree to college (62% increase). Scientific support and social importance of our app: Scientific evidence shows that lack of information and downward biased data regarding returns to education can make individuals lose hope in the future and avoid investing in education in the present, a conduct known in scientific literature as excessive discounting of the future. For instance, using data from the Dominican Republic, scientists found out that, while the measured returns to schooling are high, the returns perceived by students are extremely low, so their families do not invest in education. Also, students underestimate the returns to education in part because they rely heavily on information on the returns within their own community, which are downwards biased due to residential segregation by income (Jensen, 2008, http://goo.gl/Siqwto). Hopefully, there is a solution: students provided with information on the higher measured returns reported increased perceived returns several months later (ibid). Also, research regarding retirement savings behavior concludes that, next to removing the lure of immediate rewards or elaborating the future of rewards, allowing people to interact with age-progressed renderings of themselves will cause them to allocate more resources to the future—i.e. save more money and invest in education (Hershfield et al. 2011, http://goo.gl/gpaQ0l ). Based on these findings, our web app tries to offer citizens a useful tool to 1) get scientific-unbiased information about returns to schooling, and 2) estimate their average salary based on their present self and interact with variables such as age and years of schooling to find out how their future-self could be in economic terms (income). Therefore, by giving citizens access to this information in our app we can contribute, to some degree, to avoid young students from dropping out of school and motivate grown-ups to invest in graduate school. Methods: We based our series of articles and web app, mainly, on a linear regression model frequently used and tested to study the returns to education by economists and statisticians.By analyzing the data based on this model, we can know which factors, and by what amount, explain the differences (variation) in the average gross salary of Costa Rican workers between 15 and 65 years old (excluding wages in kind). We wanted our readers to explore directly the potential of our analysis, so this intuition is reflected in the functional design of the web app: it lets our readers interact with seven variables, using their smartphones or desktops, to know how each can modify the average salary a person like the user receives in the labor market. The variables are the following: age (as a proxy for experience), residence (urban, rural), labor market (seven main markets), labor branch (17 branches), education level, weekly hours worked, and gender. Together, they can explain around 60% of the variation in salaries, which is a fairly good percentage according to scientific literature. Our database comes from the National Institute of Statistics and Census of Costa Rica.We used the best income and expenses survey available in our country (Encuesta Nacional de Ingresos y Gastos de los Hogares, 2013). It uses a sample of 5,968 subjects, which represents around 1.5 million workers. This survey uses a complex sample design (area design, two-stages, stratified and replicated). The questionnaires were applied each month between October 2012 and December 2013. Two important caveats: 1- Our model does not pretend to predict the future income of our readers, it only shows the behavior of salaries in the labor market in Costa Rica between October 2012 and September 2013. 2- Our model does not pretend to estimate the exact salary our readers perceive, it aims to calculate the average salary the labor market pays to workers that share similar characteristics with our readers. Some very technical specifications about our model: -The model is based on Gaussian regression model with an identity link, which, according to information criteria of Akaike and Bayes, it adjusts better to the survey design we used than a Gamma regression. -Since the income variable (dependent variable) is very asymmetric, we used the Box Cox transformation to correct the heteroskedasticity problem with a natural logarithm function to make the distribution more normal. -We applied a extreme values analysis (externally studentized residuals, hat values, Cook distances and others) to eliminate observations that atypically influence our estimations. -The final estimates of our model do not present multicollinearity problems according to the inflation factor of the variance. -All the variables used in our model were statistically significant according to the Wald and likelihood-ratio tests. -The model shows an adjusted sum of squares of the regression of 58.5%. -The standard error of the average salary estimate is low enough, so we do not show the confidence intervals. Suggestion: We believe this project should be assigned to a small newsroom category, specifically "News data app of the year" (small newsroom). Also in "Best entry from a small newsroom, with fewer than 25 journalists". El Financiero has around 15 journalists, including editors and reporters.
Technologies used for this project:Techniques, methods or processes used in the production of this project: -For data analysis: we used a linear regression model using R software. -For the web app: we developed a responsive website using HTML5 y CSS3 for the tool to adjust to the resolution, size, design functionality and other characteristics of the browser used by our readers. This way the app could be visualized and used properly and easily in any device. Regarding animation, transition and interaction, we used jQuery and jQuery UI; and Toggles and Highlight libraries. For calculations and data processing we used Java Script, which interprets a JSON object with the statistical data.
You have to be connected to contribute
You have to be connected to follow
Leave this project and no longer be informed about this project
By joining this project, you will be informed by email when an update or a new contribution is posted on the website.
Thank you for your active participation !
The GEN Community Team