Exploratory Data Analysis and Data Quality Questions

Explain Simpson's paradox and give a practical example where an aggregate-level EDA finding would mislead a business conclusion. What EDA steps would catch it before you act on the aggregate number?

Beyond basic distribution plots, which visualizations are most useful early in exploration specifically for spotting anomalies, duplicates, and irregular time patterns -- think calendar heatmaps, run charts, and missingness heatmaps? What does each reveal that a summary statistic wouldn't?

Tell me about a time exploratory data analysis revealed an unexpected insight that changed your project's direction. What did you find, how did you confirm it wasn't noise, and how did you communicate the pivot to your team?

A stakeholder wants a 'customer satisfaction score' on a weekly dashboard but gives no definition. How would you run exploratory analysis to propose a reproducible one: which data sources you'd inspect, what distributions and segmentations you'd look at, and how you'd sanity-check the metric before it ships?

Distinguish random measurement error from systematic measurement error (like a sensor that slowly drifts, or a logging pixel that changes mid-series). How would you detect a systematic shift in your data during EDA, and how would you estimate its magnitude before deciding whether to correct historical records?