Data and Analysis
Domain 1: Data and Analysis
CB2 tests whether you can source, assess, prepare and summarise data before any model is fitted. Questions often ask you to choose an appropriate summary or chart, or to explain how poor data quality affects later conclusions.
Data types and sources
- Structured data sit in rows and columns; unstructured data include text and images.
- Primary data are collected for the purpose; secondary data already exist and may use different definitions.
- Cross-sectional data give a snapshot; time-series data are ordered in time and may show trend, seasonality and autocorrelation.
- Quantitative variables are discrete or continuous; qualitative variables are nominal or ordinal.
Data quality
Assess accuracy, completeness, consistency, timeliness, granularity and relevance. Cleaning includes removing duplicates, reconciling definitions, treating missing values and investigating outliers. Do not delete an outlier automatically: it may be a genuine large claim.
Summarising data
Location: mean, median, mode. Spread: range, interquartile range, standard deviation, variance. Skewness and percentiles matter for claim-size data. Use a histogram for continuous data, a bar chart for categories, a boxplot to compare groups and a scatterplot for association.
Worked example
Nine claims (£000): 2, 3, 4, 5, 6, 7, 8, 9, 100. The mean is 144/9 = 16; the median is 6. The mean is pulled up by the single 100 outlier, so the median better represents a typical claim. A 10% trimmed mean removes the smallest and largest values: (3+4+5+6+7+8+9)/7 = 6. This illustrates why skewed insurance data are often summarised by medians and percentiles rather than the mean alone.
Traps
- Reporting a mean for heavily skewed data without noting the distortion.
- Using a pie chart for continuous claim amounts.
- Treating correlation as causation.
- Ignoring changes in definition when combining data sources.
- Failing to state units or the date of the data.