Megan McVay posted on Tuesday, March 14, 2017 - 10:11 am
Help w/ this message
“THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES MAY NOT BE TRUSTWORTHY FOR SOME PARAMETERS DUE TO A NON-POSITIVE DEFINITE FIRST-ORDER DERIVATIVE PRODUCT MATRIX. THIS MAY BE DUE TO THE STARTING VALUES BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. THE CONDITION NUMBER IS 0.173D-18. PROBLEM INVOLVING PARAMETER 40.”
Our model: Autoregressive mediation model to see if treatment effects are mediated by psychosoc variables. 3 mediators @t2 predicting outcomes @t3,controlling for t0.
Model: outcomet3 on outcomet0 RamdomizationGroup mediatorAt2 mediatorBt2 mediatorCt2; MediatorAt2 on RamdomizationGroup MediatorAt0; MediatorBt2 on RamdomizationGroup MediatorBt0; MediatorCt2 on RamdomizationGroup MediatorCt0;
Randomizationgroup with outcomet0 mediatorAt0 mediatorBt0 mediatorCt0; Outcomet0 with mediatorAt0 mediatorBt0 mediatorCt0; mediatorAt0 with mediatorBt0 mediatorCt0; mediatorBt0 with mediatorCt0; mediatorBt2 with mediatorCt2;
Variables are continuous except RandomizationGroup. Parameter 40 is variance of outcomet0. Model fit=good.
We noticed big variance in outcome and big correlation of outcome @t0 & t3, so we tried w/ no luck: 1 Dropping outcomet0 from model 2 Using difference score(t3-t0) as outcome 3 Log transforming outcome(error message becomes covariance of outcomet0/mediatorCt0) 4 1 mediator only