I am working on a piece of analysis which examines the characteristics of property buyers over time. All the information is drawn from administrative data. I am computing some very basic descriptive statistics - counts (number of properties bought by cohorts) along with median values (median price, age, income).
In theory, statistics derived from registers are supposed to represent the entire population and therefore there shouldn't be a need to compute confidence intervals in comparison to say a survey.
However, I am aware that there can be many issues in admin data (linkage error, missing data, incorrect values etc.). For missing data I attempted to impute but I found it made little difference and the results weren't particularity good.
When presenting my figures, which will be at different levels of granularity I would like to be able to present a barometer around the variability in the estimate given issues around the data itself (sample size etc.).
I was thinking of a confidence interval might be a simple metric to help people understand.
I have yet to find a paper which provides information as to how to construct confidence intervals for entire populations? Maybe a different metric would be better?
Any suggestions to make question are welcome.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling