1
$\begingroup$

Does someone could tell me why these three models suggest such a great difference on the significance of factors? And which one is much reasonable?

M1<-glmer(BL~Gender+cYear+TR+T4+R5+Sun+(1|Group), weights = n, family = Gamma(link="log"),data=Aan_St
M2<-glmer(BL~Gender+Year+TR+T4+R5+Sun+(1|Group)+(1|fYear),  family = Gamma(link="log"),data=AT_St
M3<-gam(BL~Gender+TR+T4+R5+Sun+s(Group,bs="re",k=6)+s(Year,bs="tp",k=20)+ s(fYear,bs="re"), method="REML", family = Gamma(link="log"),data=AT_St)

My data has multiple observations per group in each year. Here for M1, I used aggregate data and weights=n to account for the different amount of sampling. M2 and M3 are using raw data and Year as random effect. The results suggest factors are great significant in M1 but in M2 and M3 are not. The summaries are below:

M1:

Generalized linear mixed model fit by maximum likelihood (Laplace Approximation) ['glmerMod']
 Family: Gamma  ( log )
Formula: BL ~ Gender + cYear + TR + T4 + R5 + Sun + (1 | Group)
   Data: Aan_St
Weights: n
Control: glmerControl(check.conv.singular = .makeCC(action = "ignore",  
    tol = 1e-04), optimizer = "bobyqa", optCtrl = list(maxfun = 2e+05))

     AIC      BIC   logLik deviance df.resid 
 12939.8  12964.9  -6460.9  12921.8      111 

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.70196 -0.70689  0.05356  0.59470  2.16005 

Random effects:
 Groups   Name        Variance Std.Dev.
 Group    (Intercept) 0.002477 0.04977 
 Residual             0.012303 0.11092 
Number of obs: 120, groups:  Group, 6

Fixed effects:
              Estimate Std. Error t value Pr(>|z|)    
(Intercept)  4.524e+00  5.137e-02  88.078  < 2e-16 ***
GenderMale   2.180e-02  7.681e-02   0.284    0.777    
cYear        2.227e-03  9.154e-05  24.330  < 2e-16 ***
TR           5.337e-03  5.149e-04  10.367  < 2e-16 ***
T4           1.249e-02  5.443e-04  22.947  < 2e-16 ***
R5          -2.269e-03  4.227e-04  -5.368 7.96e-08 ***
Sun         -4.089e-03  5.267e-04  -7.763 8.29e-15 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Correlation of Fixed Effects:
           (Intr) GndrMl cYear  TR     T4     R5    
GenderMale -0.670                                   
cYear       0.003  0.000                            
TR         -0.003  0.000 -0.226                     
T4         -0.001  0.001 -0.570 -0.327              
R5          0.000  0.000  0.131 -0.144 -0.168       
Sun         0.000  0.000  0.247  0.290 -0.464  0.482
convergence code: 0
Model failed to converge with max|grad| = 0.216916 (tol = 0.001, component 1)
Model is nearly unidentifiable: very large eigenvalue
 - Rescale variables?
Model is nearly unidentifiable: large eigenvalue ratio
 - Rescale variables?



M2: 

Generalized linear mixed model fit by maximum likelihood (Laplace Approximation) ['glmerMod']
 Family: Gamma  ( log )
Formula: BL ~ Gender + Year + TR + T4 + R5 + Sun + (1 | Group) + (1 |      fYear)
   Data: AT_St
Control: glmerControl(check.conv.singular = .makeCC(action = "ignore",  
    tol = 1e-04), optimizer = "bobyqa", optCtrl = list(maxfun = 2e+05))

     AIC      BIC   logLik deviance df.resid 
 19114.1  19174.4  -9547.0  19094.1     3080 

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.82972 -0.70170 -0.01325  0.66048  3.13209 

Random effects:
 Groups   Name        Variance  Std.Dev.
 fYear    (Intercept) 9.385e-05 0.009688
 Group    (Intercept) 2.480e-04 0.015749
 Residual             3.349e-03 0.057872
Number of obs: 3090, groups:  fYear, 20; Group, 6

Fixed effects:
               Estimate Std. Error t value Pr(>|z|)    
(Intercept)   0.1029023  0.3869789   0.266    0.790    
GenderFemale -0.0222147  0.0747871  -0.297    0.766    
Year          0.0022150  0.0001919  11.540   <2e-16 ***
TR            0.0074943  0.0059709   1.255    0.209    
T4            0.0102481  0.0066042   1.552    0.121    
R5           -0.0006590  0.0065327  -0.101    0.920    
Sun          -0.0013737  0.0055412  -0.248    0.804    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Correlation of Fixed Effects:
            (Intr) GndrFm Year   TR     T4     R5    
GenderFemal -0.093                                   
Year        -0.988 -0.014                            
TR           0.081 -0.048 -0.043                     
T4           0.082  0.022 -0.081 -0.323              
R5          -0.048  0.015  0.045 -0.216 -0.026       
Sun         -0.047 -0.004  0.047 -0.077 -0.244  0.350
convergence code: 0
Model failed to converge with max|grad| = 0.690565 (tol = 0.001, component 1)
Model is nearly unidentifiable: very large eigenvalue
 - Rescale variables?
Model is nearly unidentifiable: large eigenvalue ratio
 - Rescale variables?

M3:

Family: Gamma 
Link function: log 

Formula:
BL ~ Gender + TR + T4 + R5 + Sun + s(Group, bs = "re", k = 6) + 
    s(Year, bs = "tp", k = 20) + s(fYear, bs = "re")

Parametric coefficients:
               Estimate Std. Error t value Pr(>|t|)    
(Intercept)   4.5507217  0.0482135  94.387   <2e-16 ***
GenderFemale -0.0219563  0.0680126  -0.323    0.747    
TR            0.0051556  0.0046525   1.108    0.268    
T4            0.0030441  0.0053974   0.564    0.573    
R5           -0.0010902  0.0044934  -0.243    0.808    
Sun          -0.0003641  0.0041028  -0.089    0.929    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Approximate significance of smooth terms:
            edf Ref.df        F  p-value    
s(Group)  3.996  4.000 1215.456  < 2e-16 ***
s(Year)   2.984  3.059    5.301 0.000991 ***
s(fYear) 10.672 14.000   10.240 2.82e-11 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

R-sq.(adj) =  0.661   Deviance explained = 66.2%
-REML = 9594.6  Scale est. = 0.0032066  n = 3090
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.