8.7 Chapter Review
8.7.1 Chapter summary
Comprehensive EDA moves from understanding structure and quality to examining distributions, relationships, and relevant dimensions. Every important finding should be connected to a decision about definitions, cleaning, modelling, interpretation, or project scope, and the full workflow should be reproducible from source data to report.
8.7.2 Common mistakes
- producing many charts without connecting them to analytical questions;
- silently deleting missing values or outliers without investigating their causes;
- interpreting identifiers, codes, or duplicated records as ordinary measurements;
- using transformed data without preserving and documenting the original values; and
- reporting interesting patterns without explaining their consequence for later analysis.
8.7.4 Exercises
Exercise 1. NVRW has 2,400 missing acquisition-channel values. Almost all occur before a new registration system was introduced. Should the team impute a channel for these records?
Check Your Work
Not automatically. The missingness is strongly connected to the collection period, so imputation may create unsupported historical detail. Document the system change, compare periods with and without reliable channel data, and limit channel-based conclusions to comparable periods unless a defensible reconstruction source exists. Use sensitivity analysis if the older records materially affect a later model.
Exercise 2. Use data/nvrw/raw/visits.csv to create a data-quality table containing total rows, distinct visit identifiers, duplicate identifiers, missing durations, and durations above 360 minutes. Explain which issues affect visit counts and which affect duration analysis.
Check Your Work
Your table should separate duplicate visit identifiers from duration problems. Confirmed duplicate identifiers affect visit counts and should be resolved before aggregation. Missing or extreme durations affect duration summaries but do not automatically invalidate an otherwise legitimate visit record. Report both the number of affected rows and the decision applied to each issue.
Exercise 3. Compare monthly visit totals by facility. Then divide visits by active member-periods and explain why the second comparison answers a different question.
One Possible Answer
Monthly totals measure workload or overall demand at each facility. Visits per active member-period measure intensity of use relative to the size of the eligible membership base. A large facility can have the highest total visits but a lower visit rate, so the two displays support different capacity and engagement decisions.
Exercise 4. Investigate program fill rates. Identify offerings below 60 percent and at or above 95 percent, then examine whether the pattern is stable across year, facility, and program category.
Check Your Work
Calculate fill rate with a documented capacity denominator and registration-status rule. List low- and high-fill offerings, then summarize their frequency by year, facility, and category. A pattern should be described as stable only if it persists across multiple relevant periods and is not driven by a few small-capacity or incorrectly coded programs.