Fixing Messy Data When Aggregating

Earlier chapters of this book introduced techniques to generate summary statistics on a whole DataFrame. We used methods such as describe, mean, and quantile to do that. This chapter covers more complicated aggregation tasks: aggregating by categorical variables and using aggregation to change the structure of DataFrames.

After the initial stages of data cleaning, analysts spend a substantial amount of their time doing what Hadley Wickham has called splitting-applying-combining—that is, we subset data by groups, apply some operation to those subsets, and then draw conclusions about a dataset as a whole. In slightly more specific terms, this involves generating descriptive statistics by key categorical variables. For the nls97 dataset, this might be gender, marital status, and the highest degree received. For the COVID-19 data, we might segment the data by country or date.

Often, we need to aggregate data to prepare it for subsequent...