Week 1 SCM (loglinear models)

Question 1

briefly describe how a loglinear model works

Answer 1

• it begins by including all possible interactions and terms
• e.g in 3 conditions x,y,z it would include x*y, y*z,x*z, x,y,z, z*y*x
• it then removes a term and compares the new model with the previous one in which the term was present
• it starts with the highest order interactions
• it uses the likelyhood ratio to compare models

Question 2

what are assumptions of loglinear models

Answer 2

• data must be independent
• all cells must have expected frequencies greater than 1
• no more than 20% of cells can have expected frequencies smaller than 5

Question 3

if you wanted to use a loglinear model, but failed to have large enough expected frequencies, how would you remedy it?

Answer 3

• Collapse the data across one of the variables (the one you least expect to have an effect)
• Collapse levels of one of the variables
• Collect more data
• Accept loss of statistical power

Question 4

if your collapsing data across one of the variables to do a loglinear analysis what must be the case?

Answer 4

• the highest order interaction should be non-significant
• at least one of the lowest order interaction terms involving the variable should be non significant
• the categories should make theoretical sense to combine