A housing-authority waitlist analytics team must understand missing fields before joining open-data tables. Which EDA focus is most appropriate?

Before joining housing-authority waitlist tables to other open-data sources, EDA should profile missingness by department or source system to reveal patterned gaps. Missing phone numbers, income fields, or preference codes often cluster in specific bureaus, legacy migrations, or intake channels rather than occurring completely at random. Mapping null rates by source exposes whether a join will silently drop entire cohorts, bias downstream equity metrics, or require source-specific imputation rules. Pattern-aware profiling is the appropriate EDA focus when incompleteness is structural.

Assuming missingness is random and joining immediately without profiling fails when MNAR/MAR patterns exist; the join then looks “successful” while under-representing the departments with worse data entry. Filling every null with zero erases the missingness signal and invents false numeric meaning (zero income is not “unknown income”). Deleting all columns that contain any null is overly destructive for civic tables where sparse but important fields still carry operational value and where column deletion can remove join keys or protected attributes needed for fairness checks.

Exam caveat: EDA on missing data emphasizes profiling patterns by source before joins/imputation—not zero-fill-everything or drop-any-null-column extremes. Operational check: tabulate null rates by department for each waitlist field, visualize missingness heatmaps, decide per-field strategies (retain with indicator, impute, or exclude), and only then execute joins with documented row-loss counts for the housing analytics lead.