Estimating reliability

Question 1

Alternate forms

Answer 1

Reliability is estimated based on consistancy of scores across two versions of the test.

Only in parallel tests

problems
- difficult to use in psychological constructs. Making another test might mean measuring slightly different attributes
- Carryover effects, can cause correlated error scores

Question 2

Test-retest

Answer 2

Estimated based on consistency of scores across two times of testing

Problems
- carryover effects
- Changes in true scores between testing (pscych constructs are unstable over time)
- difficulty having respondents participate twice

Question 3

Internal consistency

Answer 3

Estimated based on consistency of scores across parts of the test

Three tests;
split half
Raw alpha
Standardized alpha
omega

Problems
- carryover effects

Question 4

split half estimates

Answer 4

Split test in two halves, with use of split half correlation you can calculate the estimated reliability. rests on assumption of parallel tests. otherwise you obtain different correlation based on the split. in speeded tests it’s better to use alternate forms, bc the correlation would be close to 1 in split-half

Question 5

Raw coefficient Alpha (aka Cronbach’s alpha)

Answer 5

Item-level approach, seeing each item as a subtest.
- in a matrix you can check the covarience, which also gives you info about how well the items measure the construct (you want positive covar)
- The sum of covar reflects general consistency (larger the sum the more cons)