8.7 Conclusion

We started this chapter with the question Chapter 2 left open: what patterns of absenteeism would a manager want to know about? Five charts later, we have a much clearer picture.

  • When: absences cluster in March and July, with the patterns differing — March produces both more events and more total hours, while July’s hours come from fewer-but-longer absences. January is consistently low across the board.
  • Which days: Mondays produce the most events, Wednesdays the most hours. The midweek-Tuesday-through-Thursday band is hottest within the worst months.
  • How long: absence durations are heavily right-skewed. Most are short — a workday or less — but a long tail of multi-day or multi-week absences accounts for a disproportionate share of total hours lost.
  • Why: that tail is not random. It is concentrated in serious-illness, injury, and pregnancy/perinatal categories. Routine and administrative absences look entirely different — short, tightly clustered, and predictable.

These are observations, not conclusions. We have not tested whether the March spike is statistically distinguishable from natural variation, whether the Monday-vs.-Wednesday gap holds up after controlling for absence reason, or whether any of these patterns differ by employee characteristics like age, education, or distance from work. Visualization is for finding patterns; turning those patterns into reliable claims is the job of the modeling chapters that follow.

Beyond the specific findings, the chapter has also walked through a working method:

  1. Start with a business question, not a chart. Each section opened with a question — when, how often, how long, why — and only then chose a chart type.
  2. Match the chart to the relationship. Bar charts for category-level comparisons, histograms for distributions, box plots for comparing distributions across groups, heat maps for two-way grids. Chapter 7 introduced these in the abstract; this chapter applied them.
  3. Make the design choices explicit. Faceting vs. standardization, log vs. linear scales, calendar order vs. alphabetical, viridis vs. default colors. Defaults are convenient but rarely optimal. Knowing why you chose a setting is the difference between a chart that communicates and a chart that decorates.

In Chapter 9, we will move from describing patterns to formally modeling them. The visual exploration we did here is what tells us which models are worth fitting and which variables are worth including — a step that is too often skipped, and too often regretted.