

Exponential decay growth model with E... 

Message/Author 


Trying to fit exponential decay growth model with estimator = bayes, can't get around this problem: *** ERROR For analysis with ESTIMATOR=BAYES, NEW parameters must appear on the lefthand side in at least one MODEL CONSTRAINT equation. No equation found containing parameter on the lefthand side: ERATE Is there a workaround? MODEL: INIT BY ANX1@1; INIT BY ANX2ANX5* (L12L15); RATE BY ANX1@0; RATE BY ANX2ANX5* (L22L25); ASYMP BY ANX1@0; ASYMP BY ANX2ANX5* (L32L35); !Name means of latent vars. [ASYMP] (AS); [INIT] (IN); [RATE@0]; MODEL CONSTRAINT: !Create a new var equals for mean of latent rate of change. NEW(ERATE); L12=exp(1*Erate); L13=exp(2*Erate); L14=exp(3*Erate); L15=exp(4*Erate); L22=(AS  IN)*1*exp(1*Erate); L23=(AS  IN)*2*exp(2*Erate); L24=(AS  IN)*3*exp(3*Erate); L25=(AS  IN)*4*exp(4*Erate); L32=1exp(1*Erate); L33=1exp(2*Erate); L34=1exp(3*Erate); L35=1exp(4*Erate); 


Mplus Bayes does not currently accept restrictions on parameters like here, only new, free parameters (so LHS). Don't know about a workaround except to try ML (with BS). 


Ah... that's a shame. I'll need to stick with ML then. The issue that I'm getting with this approach (and hence the motivation for trying Bayesian estimation) is that I keep getting a negative variance estimate for RATE (and it's not very close to zero, so no quick 'fixing to close to 0' type solution seems correct!). I've tried fixing it to be > 0 but then I get a correlation > 1 popping up... 


Could you use exp(erate) instead of erate to keep it nonnegative? With ML, that is. 

Back to top 

