[R-sig-ME] Lmer- Convergence issues

arun smartpink111 at yahoo.com
Wed May 9 19:21:04 CEST 2012


Dear R mixed group,

Last week I posted a convergence issue (Warning message:In mer_finalize(ans) : false convergence (8)) while I was running one of my models.

(fmNC<-lmer(Response~Wavelength*Start_Resp*time+(1|resid)+(1+time|Subject)+(1|RepNest), verbose=TRUE,family=binomial,data=BehavdatOrig2)).  


Then I came across David Hugh's blog (http://davidhughjones.blogspot.com/2009/11/lme-false-convergence.html) about the lmer false convergence. 


I divided the variable time by 10 (as it's estimate for beta was the smallest) and called it a new variable time1.  When I run the new model,

(fm<-lmer(Response~Wavelength*Start_Resp*time1+(1|resid)+(1+time1|Subject)+(1|RepNest), verbose=TRUE,family=binomial,data=BehavdatOrig2)), the convergence issue was solved and I am getting estimates of coefficients closer to what I get in other models.  Only difference in the coefficients were in the estimates and standard errors involving time.  It is now one decimal to the left (estimates for time I was getting previously in other models were 0.08, 0.086 with same z value and p value.  Same is true for the interactions).  


I guess I need to convert to one decimal right for the time variable when I report.  


I tried several ways to converge (from other posts), but none of them worked except the division,  


Any comments will be appreciated.    


Thanking you,
A.K.    



Generalized linear mixed model fit by the Laplace approximation 
Formula: Response ~ Wavelength * Start_Resp * time1 + (1 | resid) + (1 + time1 | Subject) + (1 | RepNest) 
   Data: BehavdatOrig2 
  AIC  BIC logLik deviance
 1325 1420 -645.7     1291
Random effects:
 Groups  Name        Variance   Std.Dev.   Corr   
 resid   (Intercept) 6.9806e-11 8.3550e-06        
 RepNest (Intercept) 9.0018e+00 3.0003e+00        
 Subject (Intercept) 7.0247e+00 2.6504e+00        
         time1       1.2760e+01 3.5722e+00 -0.833 
Number of obs: 1960, groups: resid, 1960; RepNest, 98; Subject, 98

Fixed effects:
                                  Estimate Std. Error z value Pr(>|z|)    
(Intercept)                        -4.0456     0.9875  -4.097 4.19e-05 ***
WavelengthGreen                     0.5634     1.4388   0.392   0.6954    
WavelengthRed                      -5.1001     3.2495  -1.570   0.1165    
Start_RespL                        10.9094     1.9002   5.741 9.41e-09 ***
time1                               0.8834     0.8693   1.016   0.3095    
WavelengthGreen:Start_RespL        -2.7382     2.4821  -1.103   0.2700    
WavelengthRed:Start_RespL           1.9070     3.9939   0.477   0.6330    
WavelengthGreen:time1              -0.9994     1.2802  -0.781   0.4350    
WavelengthRed:time1                 1.0487     2.4436   0.429   0.6678    
Start_RespL:time1                  -3.8931     1.5605  -2.495   0.0126 *  
WavelengthGreen:Start_RespL:time1  -0.1811     2.1123  -0.086   0.9317    
WavelengthRed:Start_RespL:time1     3.5655     3.2699   1.090   0.2755   




More information about the R-sig-mixed-models mailing list