Errors and Security

1
Errors and Security

• Validity
• Do we measure what we want to measure?
• Reliability
• Do we get the same result each time?
• Degree of details
• Accuracy
• Comparison

2
Validity

• Concept - Do we measure what we really want to
  measure?
• Statistical
• Can we draw conclusions?
• Internal
• Historical event
• Maturity - respondent changes attitudes
• Selection - wrong respondents
• Faulty - not at home, invalid responds
• External
• Can the results be generalised?

3
Reliability

• Concept - Can the same results be achieved in
  other circumstances or if another person is
  repeating the observations
• Equipment
• The ruler always measure the same distance, but
  the measurement may not be correct
• Data Collection
• Is the interviewer reliable?
• Data processing
• Coding error?

4
Random sampling error - chapters 15 16
Sample 1
Population
inference
sampling
Sample 2
5
Systematic Sampling Error

• Sampling bias
• consistent deviation from true value
• SSE - also called nonsampling errors
• Can be explained by
• Respondent error
• Administrative error

6
Non Response Error

• No information at all
• Nonrespondents - especially mail surveys
• Not-at-home - when do you call?
• Refusals - follow up procedures, incentives
• Incomplete information
• Refusals
• Self-selection bias - extreme responses
• Cultural bias - Copenhagen/Thy

7
Response Bias - p 171

• Deliberate falsification
• Unconscious misrepresentation
• acquiescence bias - always agrees
• extremity bias - Copenhagen rather than Thy
• interviewer bias - impress interviewer
• auspices bias - who is asking (which
  organisation?)
• social desirability bias - exceed level of salary

8
Administrative error

• data processing error
• sample selection error
• interviewer error
• interviewer cheating

9
Validity

• The ability to make sure that a measurement -
  with the help of a scale - actually measures what
  was meant to be be measured.
• How many times did you visit the hospital last
  year ?

10
Reliabilitet og validitet On Target
Old Rifle New Rifle New Rifle
Sunglare Low Reliability High
Reliability Reliable but Not Valid (Target
A) (Target B) (Target C)