[R-sig-ME] GAMM4: In mer_finalize(ans) : false convergence (8)

Sun Aug 14 12:55:12 CEST 2011

Hi all,

I know there are already some topics on the warning "In mer_finalize(ans) : false convergence (8)" but none of the advice given there could help me solving the problem.

Here's what I'm trying to do: I have a dataset on 400 (700 for the other group, I had to split the dataset) individuals. There are repeated measurements per individual (for some only one, for others more than 10) which is why I introduced a random effect for the individuals. The dependent variable is binary (0/1), and I fit a spline for the time trend, because it's especially interesting to see the trend in time for the dependent variable. The independent variables are age, binary variables and categorial variables with up to 4 categories.

For the analysis I used the gamm4 function:

model <- gamm4(aim ~ age+ s(time, bs="cr") +  kat1 + bin1 + kat2 + kat3 + bin2 +
                   kat4+ bin3 + bin4 + bin5,
                   random=~ (1|ID) , family=binomial(link = "logit"),
                   data=data)

Whenever I'm running this I get the warning "In mer_finalize(ans) : false convergence (8)".

I already tried verbose=TRUE which made no difference and I also tried it with standardized or logarithmic age which both did not help (age is not normally distributed). If I only take one or two of the categorial/binary variables there is no problem but if there are more independent variables in the model, I always get the warning.

Can anyone help me what else I could try in order to get it work properly? I'm scared this is all not reliable as long as I get the warning message.

The output is the following:

Generalized linear mixed model fit by the Laplace approximation
  AIC  BIC logLik deviance
 3886 4019  -1923     3846
Random effects:
 Groups Name        Variance   Std.Dev.  
 ID     (Intercept) 7.5614e+00 2.7498e+00
 Xr.1   s(time)     8.1108e-25 9.0060e-13
Number of obs: 5656, groups: ID, 621; Xr.1, 8

Fixed effects:
                Estimate Std. Error z value Pr(>|z|)    
X(Intercept)   -10.53860    1.56384  -6.739 1.60e-11 ***
Xage          0.29204    0.03091   9.449  < 2e-16 ***
Xkat12       -0.12581    0.30125  -0.418   0.6762    
Xkat13       0.26185    0.38071   0.688   0.4916    
Xbin11       0.22787    0.19604   1.162   0.2451    
Xkat22       0.09542    0.32195   0.296   0.7669    
Xkat23       1.19255    0.49638   2.402   0.0163 *  
Xkat24       0.79042    0.66865   1.182   0.2372    
Xkat32       -0.36311    0.27997  -1.297   0.1946    
Xkat33       -1.55390    0.67783  -2.292   0.0219 *  
Xkat34      12.31324  975.77796   0.013   0.9899    
Xbin21       -0.25221    0.26531  -0.951   0.3418    
Xkat42       -0.27021    0.73140  -0.369   0.7118    
Xkat43        0.50828    0.73869   0.688   0.4914    
Xbin32       -0.04130    0.30818  -0.134   0.8934    
Xbin41       0.43754    0.92027   0.475   0.6345    
Xbin52        -1.38775    0.33668  -4.122 3.76e-05 ***
Xs(time)Fx1      1.52712    0.23391   6.529 6.63e-11 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Correlation of Fixed Effects:
            X(Int) Xage   Xkat12 Xkat13 Xbin11  Xkat22 Xkat23 Xkat24 Xkat32   Xkat33   Xkat34   Xbin22 Xkat42 Xkat43 Xbin32 Xbin41 Xbin52
Xage        -0.847                                                                                                                
Xkat12     0.082 -0.179                                                                                                         
Xkat13     0.089 -0.131  0.354                                                                                                  
Xbin11     -0.105  0.023  0.005 -0.003                                                                                           
Xkat22     -0.132 -0.042  0.002  0.006 -0.006                                                                                    
Xkat23     -0.080 -0.044 -0.016 -0.030  0.005  0.361                                                                             
Xkat24     -0.036 -0.050 -0.032 -0.013  0.033  0.247  0.162                                                                      
Xkat32     -0.001 -0.080 -0.071 -0.182  0.001  0.012 -0.037 -0.078                                                               
Xkat33      0.053 -0.068 -0.029 -0.053 -0.019 -0.053 -0.049 -0.174  0.247                                                        
Xkat34      0.000  0.000  0.000  0.000  0.000  0.000  0.000  0.000  0.000  0.000                                                 
Xbin22      0.008 -0.065  0.053  0.020 -0.004  0.007  0.016  0.013 -0.037  0.012  0.000                                          
Xkat42     -0.440  0.003  0.002 -0.022  0.010  0.022  0.069  0.042 -0.015 -0.013  0.000 -0.089                                   
Xkat43     -0.474  0.031 -0.005 -0.044  0.025  0.106  0.099  0.057  0.017 -0.022  0.000 -0.086  0.915                            
Xbin32     -0.123 -0.061 -0.032 -0.045 -0.034  0.071  0.032  0.100 -0.078 -0.093  0.000  0.000  0.083  0.095                     
Xbin41     -0.266  0.258 -0.005 -0.094  0.005 -0.036  0.019 -0.066  0.032  0.034  0.000  0.028  0.087  0.063 -0.023              
Xbin52    -0.277  0.203  0.003 -0.018  0.355  0.292  0.178  0.084  0.014 -0.014  0.000 -0.019 -0.014 -0.105 -0.067  0.035       
Xs(time)Fx1  0.007  0.021  0.003  0.009 -0.261  0.006  0.008 -0.007 -0.012 -0.002  0.000 -0.012  0.001  0.005  0.000  0.007 -0.119

$gam

Family: binomial
Link function: logit

Formula:
aim ~ age+ s(time, bs="cr") +  kat1 + bin1 + kat2 + kat3 + bin2 +
                   kat4+ bin3 + bin4 + bin5

Estimated degrees of freedom:
1  total = 18

glmer.ML score: 3846.487

I'd be really thankful for any help,
Caro

___________________________________________________________
Schon gehört? WEB.DE hat einen genialen Phishing-Filter in die