Outside and inside the regression “black box” from exploratory to interior data analysis

Good data analysis consists of three phases: (1) preliminary analysis, (2) confirmatory analysis (model testing), and (3) interior analysis (model checking). Social scientists doing quantitative research usually concentrate on only one of the three: confirmatory analysis. I argue that there is much to be learned from careful preliminary and interior analyses. I present an extensive example of data analysis for each of the three phases using the same data set in each phase. Rather than surveying all the possible tools available in each phase of data analysis, I concentrate on Exploratory Data Analysis techniques (stem-and-leaf plot, letter-value display, box plot, and power transformations) for the preliminary phase, on OLS for the confirmatory phase, and on residuals, leverage and single-case influence measures for interior analysis.