Exploratory Data Analysis and Data Quality Questions

When would you reach for the mean versus the median versus the mode to summarize a variable during EDA, and why does the choice matter? Give an example where reporting the mean instead of the median would mislead a business audience.

You're exploring a labeled dataset with two or more annotators. Explain inter-annotator agreement and how Cohen's kappa measures it beyond chance agreement. What would a low kappa alongside a high raw percent-agreement tell you about the labels?

How do you structure an EDA project so it's reproducible and shareable with the rest of your team: notebook vs script organization, where raw vs processed data and charts live, versioning of datasets and code, and what you'd automate so someone else can regenerate your findings?

Explain the three missing-data mechanisms: missing completely at random (MCAR), missing at random (MAR), and missing not at random (MNAR). Give a realistic example of each from a business dataset, and name one diagnostic you'd run during EDA to help tell them apart.

Given a small sample table with a few missing cells across two columns, decide whether the missingness in each column looks most consistent with MCAR, MAR, or MNAR based on what else you can see in the rows, and name two diagnostics you'd run to confirm your read on a larger dataset.