IZA DP No. 13459: Exploratory Data Analysis on Large Data Sets: The Example of Salary Variation in Spanish Social Security Data
published online in: BRQ Business Research Quarterly, First Published September 9, 2020
New challenges arise in data visualization when a sizable database is used in the analysis. With many data points, classical scatterplots are non-informative due to the cluttering of points. On the contrary, simple plots such as the boxplot that are of limited use in small samples, offer great potential to facilitate group comparison in the case of an extensive sample. This paper presents Exploratory Data Analysis (EDA) methods that are useful when a large dataset is involved. The EDA methods, (introduced by Tukey in his seminal book of 1977) encompass a set of statistical tools aimed to extract information from data using simple graphical tools. In this paper, some of the EDA methods like the Boxplot and Scatterplot are revisited and enhanced using modern graphical computational devices (as, e.g., the heat-map) and their use illustrated with Spanish Social Security data. We explore how earnings vary across several factors like age, gender, type of occupation and contract and in particular, the gender gap in salaries is visualized in various dimensions relating to the type of occupation. The EDA methods are also applied to assessing competing regressions with earnings as the dependent variable. The methods discussed should be useful to researchers to assess heterogeneity in data, across group-variation, and classical diagnostic plots of residuals from alternative models fits.