What are the three types of validity?
Construct, internal, external
What is construct validity?
A measure or test has construct validity to the extent that it measures what it claims, or purports, to measure
Are the operations valid - do manipulations consistently change the construct’s state? Do the measures closely and consistently reflect the constructs current state?
What is internal validity?
Is the causal relationship valid - construct validity and causal relationship has been established
What is external validity?
Do the results generalise to other populations, stimuli, etc.
What are some of the possible failures in construct validity?
Construct defined too broad or too narrow
Construct not captured at all
Overlooking obvious aspects of a construct
True or False: reliability is often required for construct validity (the measure should be providing a replicable measure of a construct)
True
How can reliability be assessed
Interclass correlation coefficient (between 2 instances of administering the test)
What is the range of the intraclass correlation coefficient?
0 -> 1
What does a intraclass correlation coefficient of .70 mean?
Good correlation - 70% signal, 30% noise
What are the 3 types of reliability?
Test-retest reliability, split-half reliability, parity reliability
Out of test-retest reliability, split-half reliability, and parity reliability, which one should show the highest reliability coefficients ?
Parity reliability - scores between odd and even trials would show the least noise and be free of confounds - highest reliability
What is test-retest reliability?
Correlation between scores at 2 time points
What is split-half reliability?
Correlation between scores in 1st and 2nd half of experiment
What is parity reliability?
Correlation between scores on odd v even trials
What is convergent validity?
Showing that measurements of constructs that should be theoretically related to each other are indeed related
What is discriminant validity?
Showing that measurements that are theoretically not supposed to be related are in fact unrelated
What is internal validity?
The degree to which a study can establish that a factor indeed caused the behaviour
What is Simpsons Paradox?
A statistical phenomenon where a trend that appears in different groups of data reverses or disappears completely when the groups are combined. This happens because the aggregation of data masks the influence of a lurking variable, or confounding variable, which is systematically distributed across the subgroups.
