How do I handle missing data in a Business Analysis project without biasing the results?

All answers to this question.

Deleting rows (listwise deletion) is usually a bad idea because you lose valuable info in other columns. Using the "Mean" is a quick fix, but it reduces variance and can lead to "Boring" models that miss outliers. A better approach for Business Analysis is "Median Imputation" if the data is skewed, or "K-Nearest Neighbors (KNN) Imputation," which looks at similar customers to guess the missing age/income. If the data is "Missing Not At Random" (e.g., wealthy people refusing to disclose income), then even the best math won't save you—you have to acknowledge that limitation in your executive summary.

   Answered 2024-11-17 by Patricia Robinson

Have you checked if the "Missingness" itself is a signal? Sometimes customers who don't provide an age are more likely to churn because they are less engaged with the brand.

   Answered 2024-11-19 by Charles Lewis

• Charles, that is a brilliant point! In many churn models, a "NULL" value is actually a predictor of low brand loyalty. Instead of filling it in, Steven could create a "Missing_Flag" column (0 or 1). This allows the model to see that the absence of data is a feature in itself. I’ve seen this improve model accuracy by 5% in telecom datasets.

     Commented 2024-11-20 by Thomas Walker

I always recommend Multiple Imputation by Chained Equations (MICE). It sounds complex, but most modern analytics software has a checkbox for it now.

   Answered 2024-11-22 by Susan Hall

• MICE is definitely the gold standard for statistical rigor. It’s a bit overkill for a basic slide deck, but great if the project is high-stakes.

     Commented 2024-11-23 by Patricia Robinson