Multilevel Models

Question 1

What is aggregation?

Answer 1

Variables at lower level moved to higher level.

Ex:

Assigning schools to mean of student scores.

Question 2

What is disaggregation?

Answer 2

Variables at higher level moved to lower level.

Ex:

Asian American becomes:

Japanese American
Indian American
Bangladeshi American
Korean American
Lao American

Question 3

Why is aggregation naive?

Answer 3

Information is lost; leads to loss in statistical power (aggregation).

Sample size misleading, leading to larger number of false positives (disaggregation).

Ecological fallacy: inferences about individual drawn from group data.

Ignores correlation from 1st level clustered within 2nd level.

Question 4

What happens when data is correlated but not accounted for in analysis?

Answer 4

RLs may be erroneously described.

SE measures will be incorrect.

SE used in calculation of p-value and CI, so also incorrect and invalid.

Means statistical inferences will be incorrect and invalid.

Question 5

What are issues with relying on p-value?

Answer 5

Does not provide clinical importance

P-values are a function of sample size and show statistical significance. Do not show effect size or magnitude.

Question 6

How do multilevel models account for issues with naiive methods?

Answer 6

Account for individual AND group level predictors.

Dependency in data is accounted for.