Quantcast

Difference in p-values between Fixed Effects tables in LME output when adding interaction term

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Difference in p-values between Fixed Effects tables in LME output when adding interaction term

Joana Vasconcelos
Hi All,

Can anyone help me understand why a 2-category variable gives me a different p-value in the Type III Tests of Fixed Effects table and the Estimates of Fixed Effects table using an LME model when I add in the model an interaction term of this variable with another 2-category variable?

This difference in p-value only happens if I consider both variables as factors (not covariates) in the LME model...

Code Used:
MIXED VisualAcuity BY   id   time   BP   
/FIXED  time   BP   BP*time   | SSTYPE(3)
/RANDOM id
/PRINT SOLUTION TESTCOV.

Example:

Factors:
Time (1/2) 
BloodPressure (BP) (1/2)
Time x BP

In the Type III Tests of Fixed Effects table I get:
p-value for Time p=0.02
p-value for BP p=0.29
p-value for Time x BP p=0.84

In the Estimates of Fixed Effects table I get:
p-value for Time p=0.17
p-value for BP p=0.34
p-value for Time x BP p=0.84

Also, I've noticed that if I only put one of the variables as factor, say BP, and time as a covariate, the p-value of BP is the same in both tables (as desired, p=0.37)) but the p-value for the variable I specified as covariate (time) is not (p=0.02 / p=0.17) (despite the p-value for the interaction term still being the same in both tables).

Thank you very much.

PS: This problem doesn't happen if I compute the interaction term manually/as a pre-specified variable

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
Loading...