Clustered data Flashcards
When a sample is drawn using clusters, will traditional standard error formulas assuming iid draws be correct?
No
What is the design effect?
Ratio of the variance of xbar (sampling variation) under two sampling designs (cluster and SRS)
Why is there an increase in sampling variability under cluster sampling?
The greater the share of variability that is between cluster, the more you lose by a cluster sample design
What is the ICC?
A measure of the amount of clustering
If there is perfect homogeneity within groups, what will the iCC equal?
1
What does an ICC of 1 indicate?
perfect homogeneity within groups
Which sampling has larger standard errors: cluster or simple random sampling?
cluster
If ICC is 0, what are the implications for the design effect?
No clustering, standard errors same as SRS
If ICC is 1, what are the implications for the design effect?
Extreme clustering, standard errors are larger
If we have observations clustered within groups i and xi doesn’t vary across units within the same group, what do we need in order to get new information?
new clusters
If the deft is 3.3, what does this mean?
Standard errors are approximately 3.3 times higher than those reported by the traditional standard error formula.
True or false: cluster sampling provides less variation than what one would obtain under a simple random sample. Standard error formulas that assume a SRS (i.i.d. observations) will be incorrect.
true
What does power of hypothesis test depend on?
effect size of interest
alpha
standard error of sample mean–smaller standard error=greater power
n because n decreases standard errors
If you have clustering, will increasing the number of students per school help to increase power?
No
