Randomisation based approaches
Randomisation based approaches included the survey sampling and experiment approach.
- Have common concept of popluation (usually finite and existing).
- Use Laplacian concept of probability.
- Probability enters only through random selection or assignment.
Populations
Pure or strict view is that population must be finite and existing (not subject to arbitrary definition). All units may be individually identified and assigned discrete labels. Common problems:
- Non-static populations (restrict inference to time period of interest)
- Populations of unknown size
- Hypothetical and constructed populations
- Arbitrary defined population units (inference conditional on same
- definition, be careful that attributes still appropriate)
Attributes and responses
Sampling approach assumes that there is a fixed value of interest, an attribute, associated with each unit. Not realisation of some random mechanism but fixed, immutable characteristics.
Under experimental approach, call measured or observed values as responses. Essentially the same as for sampling, but are influenced by external factors within the time frame of the study.