3: Reliability Flashcards
Reliable measure
Produces consistent results when repeated measurements of the same quantity are made under identical conditions
- increases confidence of individual measurements
- well-defined and objective operational definition
Test-retest
Consistent values every time you measure
Evaluating:
-r > or = .50 good
-composite score for extraversion
•remove items with poor internal reliability
• add responses to remaining internally reliable items on questionnaire, with reverse scored
-re-administer to same respondent at a later time
2 problems:
• respondent remember earlier answers(inflates r)
• respondents changing between administratations (deflates r)
When to use:
-appropriate for all but problematic for unstable variables
Inter-rarter
Consistent values no matter who is observing
Evaluating:
Quantitative
-r > or = .7 good
Categorical
- % agreement = total agreements/ total observations
- 70-80%
When:
-observational measures that aren’t automated
Internal
Consistent values no matter how you ask
-extent to respondents answer multiple questionnaire items designed to measure same construct consistently
Evaluating:
-cronbachs alpha
• from mean of all inter-item correlation
• alpha> or = .70 is good
When:
-self-report measures with multiple items
Evaluating reliability
Empirical question
- paired measurements of same cases
- diff times, raters, or asked in diff ways
Quantitative:
- scatterplot
- correlation coefficient(Pearson r) (+)=high reliability
Categorical:
-Percent agreement
Random measurement error (e)
Measured X= true value T+ bias B+ random error e
- measurements of same quantity Rarley be exactly the same due to random M. E.
- small random M. E. = more reliable
Reducing random M. E. :
- increases precision of measurement (reduces margin or error)
- increases stat power
Systematic measurement error(bias) (B)
-average measured value systematically differs from the true value
(Apparatus; due to miscalibrarion, researcher, and subject)
Smaller= more accurate
-problematicfor frequency claims:
• can correct if know
• less prob. For association and causal, since stat power may be unaffected
Accuracy
-produces results that agree with a known standard (closeness of measurement to true value)
-assess by measuring a known standard with the instrument
-less relevant when not using standard units
• can standardize a measure to have a specified mean and variability across a population
Reliability
Closeness of repeated measurements to each other
- essential
- if unreliable a single measurement will vary unpredictably from tribe value even if accurate