Nature of Sampling Error
- Sampling error is an error of non observation
- Why does sampling error occur à heterogeneity of frame population
Measuring Sampling Error
Non Probability Samples
- Prev called quota sampling
- Now called model based samples
Comparing Probability and non probability samples
Four Effects on Design on sampling error
- Stratification
o Sorting of frame into separate groups
o Separate samples drawn from each group
- Probabilities of selection
o Equal probabilities of selection to each element in frame population
§ Practical reason to depart
· Cost
· Desire to study sub group of population intensively
o Disproportionate sampling
- Clustering
o Clusters of people are selected and individuals within the clusters are interviewed
§ Eg, samples of classrooms of students
§ Loss of precision – not inherent in design – sociological fact that most natural groupings of people contain persons who are similar in many variables
- Sample size
- Combinations of sample design features
The Effects of Non Sampling Errors on sampling Error Estimates
- Response = true value + interviewer effect + error
- Under very restrictive cuircumstances, sampling error estimates will reflect some non-sampling errors
o Only for variable non-sampling errors
o Only when mechanisms producing non sampling error coincide with sampling unit definitions
o Only for non sampling errors that are independent of true values of observation unit
o One design under which traditional sampling error calculations fully incorporate measurement error and other non sampling errors associated with interviewers
§ Cluster sample
§ One interviewer assigned at random to each sample cluster
§ That interviewer completes all interviews in cluster
Measuring sampling Errors on Sample Means and Proportions from Complex Samples
- Error properties of statistics require both statistical and social science knowledge for understanding.
Debate on Reflecting the Sample Design in Estimating Analytic Statistics
Reflecting the sample design when estimating complex statistics
Summary
- Sampling error is erroneously interpreted as a summary measure of all the errors in the estimates
- Other difficulties in data collection process can reduce quality of survey statistics
- Probability sampling – attractions disappear ehen errors of non-response are considered
o Two areas
§ First, 50% response
§ Second – 90% response
§ Will be different from survey with 100% response in both areas.
"Finally, because sampling errors are measurable and reductible under a wide variety of designs,
they have yielded themselves to explicit cost and error modelling. Optimal sampling designs under
fixed research resources can be identified. These minimise sampling error relative to all other designs.
This use of cost and error modelling is the foundation of attempts in this book to acknowledge cost constraints as limits on efforts
to reduce nonsampling errors. The raeder will note these attempts are only partially successful and are critically affected by the weak
measures available for some of the non sampling errors".
-
No comments:
Post a Comment