IASIU CIFA: Data Analysis Steps

What this domain tests

This domain tests the disciplined sequence of an analysis, from question definition to reproducible delivery. Items often ask what step is missing or what should happen next.

Core rules

1. Define the question and the decision it supports.
2. Identify sources, grain, and definitions of every measure.
3. Extract with the query and date of extraction logged.
4. Profile and clean; document every change and keep raw data immutable.
5. Explore distributions, outliers, timelines, and missingness.
6. Analyze against a stated plan, not against whatever looks interesting.
7. Validate: reconcile totals to finance or claim-system reports, test sensitivity to assumptions, seek a peer review.
8. Communicate findings with limitations.
9. Preserve the extract and steps so the work can be reproduced.

Worked example

Question: are windshield claims rising? Reasoning: convert to a rate - glass claims per 1,000 auto policy-years by month - because counts rise with exposure. Extract payment and policy tables with the run date logged. Profile for duplicate claim numbers and for test claims, and reconcile total paid to the finance ledger. Explore seasonality using two prior years. Compare the current year against the same months, then test whether a single glass vendor or region drives the change. If the rate is flat and one vendor explains the count rise, the conclusion is vendor concentration, not an industry trend.

Traps that cost marks

• Changing the question midway to fit the data (data dredging).
• Cleaning without documenting, or editing the source file.
• Skipping reconciliation so totals are wrong from the start.
• Comparing a partial year to a full year.