We have 8 years of data that includes info on all crimes reported to MPs and all crimes the commanders submitted for prosecution. This is not a sample - it is the entire population of crimes committed by members of the Army. We are studying domestic violence and among other things we are looking at the relationship between offender rank and the actions taken by the commanders (who function as prosecutors). The data is primarily nominal and ordinal data and normally we would run significance tests but those are designed for samples. Are p-values even meaningful when you have the entire population and do not need to wonder if the sample is representative of the population? Any suggestions for methods of analysis?
Maybe you could break the population down into groups or factors, perhaps using a factor analysis or just identifying the themes you want to look at, and then do an intra-population analysis using the usual techniques on each group?
It maybe helpful to think that all crimes in your case are a sample of imagined space and time. For example, all crimes in the future of 8 yrs or in a country with the same context. In these situations, you can proceed by treating all crimes as a sample from a data generation process.
You may want to look at my lengthy responses here - in short I argue that even with a 'census' you have to make inferences under conditions of uncertainty
https://www.researchgate.net/post/Are_p-values_and_significance_tests_still_meaningful_in_census_data_analysis
- Badges
- Science topic