Diverse Perspectives on Survey Research
- Survey vs poll
- Describers and modelers use survey data differently
o Describers: discover some fixed property of a set of people (ie unemployment rate)
§ Does sample reflect population
o Modellers – motivated by hypotheses about cause of social phenomena
§ Does survey measures fully capture concepts involved in their theory
- Two types of error
o One
§ Sampling error
· Error that arises because only subset of population was measured.
· Increase sample size
· Assure representation of different groups in sample through stratification
· Adjust probabilities of selection for different sub groups
· Use clustered selections to reduce cost.
o Two
§ Non sampling error
· Reduce non response and measurement errors
· Better worded questions
· Interviewer training
o Interviewer effect
· Reduce / measure this error
The themes of integration: errors and costs
o Costs of survey activities act as limiting influence on efforts to reduce survey error
The languages of error
o Statistics
o Psychology
o Economics
Classification of error within survey statistics
- Bias
o Constant error
- Variable errors
o Values differ over units (ie sampled persons, interviewers used)
- Errors of non-observation
o Coverage – eg, telephone surveys do not include people without a phone
o Non response – respondents in frame cannot be located / refuse
o Sampling
- Errors of observation
o Interviewer
o Respondent
§ Ie, not able to correctly remember
o Questionnaire / instrument error
o Mode of data collection
- A more restrictive view, sampling statistics
- Other error terms in survey statistics
o Accuracy
o Precision – converse of variance – stable over replications
o Reliability – not same meaning when used in psychometrics – here means same as precision
Terminology of errors in psychological measurement
- Notion of unobservable characteristic that is measured with survey indicator
- Problem is is not impossibility of measuring the characteristic, but weakness of measure
- Classical score theory
- Validity and reliability
- Estimating validity
- Other notions of validity
Language of Errors in Econometrics
- Language of estimation of GLM
- Selection bias
Debates About Inferential Errors
- How does researcher conceptualise impact of errors of non-observation
o Modeler: concentrate on correct specification of form of model and less so on errors of non-observation.
Important Features of Language Differences
- Sampler is committed to randomization process that on average produces samples with desirable properties / analyst is more concerned with ability of this single sample to describe population
Summary: The Tyranny of the Measurable
No comments:
Post a Comment