4 EFA and Reliability Flashcards
What is the formula for classical test theory?
Observed Score = True Score + Error Score
Conceptually, Cronbach’s alpha coefficient is a summary metric of the __________ of ______.
Conceptually, Cronbach’s alpha coefficient is a summary metric of the equivalence of items.
In SPSS Reliability Analysis, the squared multiple correlation for an item is….?
The R2 for each item predicted by the rest of the items. Like if you regressed all the other items on the one item in question.
What is one problem with high inter-item correlation?
It can constrain validity by limiting the assessment of complex constructs by a narrow focus on redundant variance (i.e. redundant items).
How does reliability change with number of items in a scale?
More items means higher reliability, AOTBE.
How high should Cronbach’s be?
For clinical diagnosis, should be more than .90. For research purposes should be more
than .70.
Exploratory Factor Analysis is used for what? Two things
- Aid with the Internal (Construct) Validity by demonstrating underlying factors.
- Summarise the relationships among items, with a set of fewer constructs (factors/traits)
How many items do you need to create a reliable, robust factor, such as Gf?
3 should be the bare minimum – at least 3 items.
A ______ is much more stable than an _______.
A factor is much more stable than an item.
When are variables argued to represent the same latent construct?
When they correlate highly enough. What is ‘enough’ is determined by the statistical methods.
_______ construction is facilitated when the inter-item correlations indicate that items belong to certain ______ (_____).
Scale construction is facilitated when the inter- items correlations indicate that items belong to certain clusters (factors)
What is the difference between Principal Components Analysis and Exploratory Factor Analysis?
In Principal Components Analysis, the full variance of each item/variable is assumed to load on a factor. No unique variance for each item is considered.
In Exploratory Factor Analysis, the correlations between items are measured to group them into a non-specified number of clusters. In PCA, all of the observed variance is analyzed, while in EFA it is only the shared variances that is analyzed.
How much of the original variance in responses should be captured by the factors?
If less than 50%, then you’re not capturing enough. If 80%, maybe some redundancy in scale.
What is communality (h2)?
h2 is the proportion of each variable’s variance that can be explained by the factors (e.g., the underlying latent continua).
How high should h2 be?
Somewhere between .20 (item has little in common with the factors) and .80 (items are redundant)
