What is Data?
What this domain tests
This domain tests whether you can describe, classify, and judge the quality of the data an insurer holds before you analyze it. Expect items on measurement scales, structured versus unstructured sources, and data-quality dimensions.
Core rules
- Data are recorded observations; a dataset is a collection of records, each record a set of fields and values.
- Structure: structured (claims tables), semi-structured (XML, JSON, EDI feeds), unstructured (adjuster notes, PDFs, images, recorded statements).
- Measurement scales: nominal (cause-of-loss code), ordinal (severity rank), interval (temperature), ratio (paid amount, which has a true zero).
- Discrete versus continuous; quantitative versus qualitative.
- Data quality dimensions: completeness, accuracy, consistency, timeliness, validity, uniqueness.
- Provenance and lineage matter: know which system produced a field, when, and under what business rules.
- Personal and health information must be handled on a need-to-know basis.
Worked example
A cause-of-loss field is coded 1–9. Reasoning: the code is nominal, so the correct summary is a frequency count or mode - averaging it produces a meaningless 4.6. Claimant age is ratio and can be averaged, though median is safer if the distribution is skewed. Paid amount is ratio with a true zero, but a blank paid field means "no payment recorded," not $0, so blanks must be inspected rather than filled with zero. Free-text loss addresses need standardization before any count of losses per address is trustworthy.
Traps that cost marks
- Averaging nominal codes or ordinal ranks.
- Treating NULL or blank as zero in payment fields.
- Assuming date fields are clean; spreadsheet serial dates and mixed formats are common.
- Counting parties by name string, which splits or merges entities incorrectly.