CESD: mixed effects model

Outcome variable: social-cognitive perceptions of the depression items

This model is to examine how the raters’ background affects their perceptions of depression items.
This is a multidimensional model with five aspects of perceptions.

# multidimensional model (random intercept)
m2 <-lmer(s_cog_value ~ gender +age + education + (1|id_item) + (-1+s_cog|rater_id), data=dat2, REML = FALSE)
summary(m2)

## Linear mixed model fit by maximum likelihood  ['lmerMod']
## Formula: s_cog_value ~ gender + age + education + (1 | id_item) + (-1 +  
##     s_cog | rater_id)
##    Data: dat2
## 
##      AIC      BIC   logLik deviance df.resid 
##  12032.2  12158.9  -5995.1  11990.2     3051 
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -3.5729 -0.5616  0.0301  0.5581  3.3136 
## 
## Random effects:
##  Groups   Name             Variance Std.Dev. Corr                   
##  rater_id s_cogavailabil   3.19866  1.7885                          
##           s_cogcomfort     4.27171  2.0668   -0.88                  
##           s_cogdesirab     2.27899  1.5096   -0.49  0.69            
##           s_cogspecificity 2.07464  1.4404    0.12  0.05 -0.16      
##           s_cogvague       3.07196  1.7527   -0.83  0.84  0.42  0.15
##  id_item  (Intercept)      0.07442  0.2728                          
##  Residual                  2.53508  1.5922                          
## Number of obs: 3072, groups:  rater_id, 31; id_item, 20
## 
## Fixed effects:
##               Estimate Std. Error t value
## (Intercept)  3.062e-01  3.772e-01   0.812
## gender       1.215e-01  1.757e-01   0.692
## age          9.903e-06  3.931e-03   0.002
## education   -1.030e-01  6.526e-02  -1.578
## 
## Correlation of Fixed Effects:
##           (Intr) gender age   
## gender    -0.492              
## age       -0.469  0.122       
## education -0.818  0.234  0.045

Outcome variable: item responses of the depression scale (20 items)

This model is to understand how the raters’ background (i.e., age, education and sex) and their perceptions on the items affect their responses on the depression items.
This model used presence data in which item responses are coded to 0 and 1.

# random intercepts for both person and item 
m3 <- glmer(response ~ -1 +  gender +age + education + specificity + availabil + comfort + vague +
             desirab +  (1|id_item) + (1|rater_id), data=data.p,family = binomial)

## Warning in checkConv(attr(opt, "derivs"), opt$par, ctrl = control
## $checkConv, : Model failed to converge with max|grad| = 0.0102 (tol =
## 0.001, component 1)

summary(m3)

## Generalized linear mixed model fit by maximum likelihood (Laplace
##   Approximation) [glmerMod]
##  Family: binomial  ( logit )
## Formula: 
## response ~ -1 + gender + age + education + specificity + availabil +  
##     comfort + vague + desirab + (1 | id_item) + (1 | rater_id)
##    Data: data.p
## 
##      AIC      BIC   logLik deviance df.resid 
##    679.3    723.3   -329.6    659.3      595 
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -6.5496 -0.6069  0.2858  0.5786  3.7302 
## 
## Random effects:
##  Groups   Name        Variance Std.Dev.
##  rater_id (Intercept) 1.3401   1.1576  
##  id_item  (Intercept) 0.9243   0.9614  
## Number of obs: 605, groups:  rater_id, 31; id_item, 20
## 
## Fixed effects:
##               Estimate Std. Error z value Pr(>|z|)   
## gender      -0.4894556  0.4309964  -1.136  0.25611   
## age          0.0007896  0.0101857   0.078  0.93821   
## education    0.2639205  0.1086339   2.429  0.01512 * 
## specificity  0.0069497  0.0637375   0.109  0.91317   
## availabil    0.2760412  0.0846789   3.260  0.00111 **
## comfort     -0.2365754  0.0894554  -2.645  0.00818 **
## vague        0.1747309  0.0778301   2.245  0.02477 * 
## desirab      0.2104592  0.0716541   2.937  0.00331 **
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##             gender age    eductn spcfct avalbl comfrt vague 
## age         -0.106                                          
## education   -0.308 -0.599                                   
## specificity -0.036  0.233 -0.049                            
## availabil   -0.020  0.073  0.128  0.066                     
## comfort     -0.111 -0.089 -0.064 -0.106  0.239              
## vague       -0.005 -0.192  0.059 -0.413  0.203 -0.111       
## desirab     -0.043 -0.048  0.093  0.038  0.019 -0.247  0.041
## convergence code: 0
## Model failed to converge with max|grad| = 0.0102 (tol = 0.001, component 1)

## persistence responses (depression 1-4)
## Note. Need to find models for ordinal categorical outcome and NOT used for the conference

m4 <- lmer(response ~ -1 +  gender +age + education + specificity + availabil + comfort + vague +
             desirab +  (1|id_item) + (1|rater_id), data=dat1, REML = FALSE)
summary(m4)

## Linear mixed model fit by maximum likelihood  ['lmerMod']
## Formula: 
## response ~ -1 + gender + age + education + specificity + availabil +  
##     comfort + vague + desirab + (1 | id_item) + (1 | rater_id)
##    Data: dat1
## 
##      AIC      BIC   logLik deviance df.resid 
##   1719.5   1768.0   -848.7   1697.5      594 
## 
## Scaled residuals: 
##      Min       1Q   Median       3Q      Max 
## -2.71084 -0.75020 -0.09062  0.72008  2.83339 
## 
## Random effects:
##  Groups   Name        Variance Std.Dev.
##  rater_id (Intercept) 0.3879   0.6228  
##  id_item  (Intercept) 0.1166   0.3415  
##  Residual             0.8150   0.9028  
## Number of obs: 605, groups:  rater_id, 31; id_item, 20
## 
## Fixed effects:
##              Estimate Std. Error t value
## gender       0.302112   0.216568   1.395
## age          0.006066   0.005005   1.212
## education    0.332166   0.052806   6.290
## specificity  0.004951   0.023835   0.208
## availabil    0.010735   0.029013   0.370
## comfort     -0.052786   0.030890  -1.709
## vague        0.096671   0.027455   3.521
## desirab      0.089529   0.024567   3.644
## 
## Correlation of Fixed Effects:
##             gender age    eductn spcfct avalbl comfrt vague 
## age         -0.130                                          
## education   -0.333 -0.654                                   
## specificity -0.023  0.181 -0.040                            
## availabil   -0.014  0.058  0.072  0.066                     
## comfort     -0.097 -0.061 -0.036 -0.097  0.260              
## vague       -0.005 -0.144  0.034 -0.434  0.180 -0.106       
## desirab     -0.036 -0.045  0.039 -0.016  0.000 -0.182  0.072

next step

1. for model-2: (a) make binary outcome using presence data; (b) find mixed modeling for ordinal categorical outcome

2. for model-1: (a) check residual distributions; (b) check references for outcome distribution assumptions; (c) add positive and negntive key (predictor) e.g. (-1 + key|id_item)

CESD: mixed effects model

Yan

July 16, 2017

Outcome variable: item responses of the depression scale (20 items)

next step