[R-sig-ME] Mixed Multinomial Model built with brms::brm: diagnostic and goodness of fit

[Romaine TCHEDJI]

 Hi, all.
I'm new at Bayesian modelling and I just fitted a Mixed Multinomial model
with brms::brm function.
The outputs look like follow:

Family: categorical
  Links: mu5059Kg = logit; mu6069Kg = logit; mu70Kg = logit
Formula: dependent.variable ~ Predictor1 + Predictor2 + Predictor3
+Predictor4 + (1 | grouping.variable)
   Data: QEP_df (Number of observations: 305)
  Draws: 4 chains, each with iter = 1000; warmup = 500; thin = 1;
         total post-warmup draws = 2000

Multilevel Hyperparameters:
~Patient_id (Number of levels: 302)
                       Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS
Tail_ESS
sd(mu5059Kg_Intercept)  3870.72   2237.74  1008.75  9990.76 1.86        6
    15
sd(mu6069Kg_Intercept)  2918.78   2119.11   723.89  8392.54 2.16        5
    11
sd(mu70Kg_Intercept)      65.00    127.51     2.65   446.45 1.23       14
    31

Regression Coefficients:
                                      Estimate Est.Error   l-95% CI  u-95%
CI Rhat Bulk_ESS Tail_ESS
mu5059Kg_Intercept                      930.06   1237.78   -1855.51
2999.83 1.21       16       28
mu6069Kg_Intercept                    -3528.80   3203.72  -11462.34
21.29 1.82        6       25
mu70Kg_Intercept                      -7167.48   7525.71  -23059.82
-490.98 2.30        5       26
mu5059Kg_Predictor1 level1             3680.97   1448.03    1503.36
6862.25 1.46        8       28
mu5059Kg_Predictor1 level2            -4249.19   8310.04  -22142.60
9823.92 2.28        5       19
mu5059Kg_Predictor1 level3            -46342.99  25341.93 -104113.23
 -6878.36 2.26        5       14
mu5059Kg_Predictor2                      -26.60    268.69    -603.03
 412.51 1.88        6       14
mu5059Kg_Predictor3                        0.15      3.19      -5.36
 8.39 1.19       20       31
mu5059Kg_Predictor4 level2              -199.51   1143.60   -1960.80
2663.52 1.23       15       36
mu6069Kg_Predictor1 level1              5299.62   2827.60    1867.48
 12239.12 1.98        6       11
mu6069Kg_Predictor1 level2              11270.73   9092.68    2558.59
 30427.83 2.30        5       20
mu6069Kg_Predictor1 level3                873.77  17756.14  -25351.15
 44191.74 2.04        6       13
mu6069Kg_Predictor2                     -294.70    247.52    -887.61
42.60 2.14        5       11
mu6069Kg_Predictor3                       -0.28      2.65      -5.60
 5.40 1.45        8       14
mu6069Kg_Predictor4 level2              -160.26   1019.22   -1888.31
1804.02 1.11       26       53
mu70Kg_Predictor1 level1                8548.41   8203.24    1256.54
 25529.99 2.67        5       19
mu70Kg_Predictor1 level2               12166.31  11819.17    1562.80
 37022.18 2.54        5       23
mu70Kg_Predictor1 level3               52159.88  53607.67    4477.90
169288.72 2.67        5       12
mu70Kg_Predictor2                        -40.10     64.78    -247.26
31.76 1.39        9       19
mu70Kg_Predictor3                          1.23      1.45      -0.70
 4.78 1.20       17       47
mu70Kg_Predictor4 level2               -1307.78   1123.37   -3664.23
 331.77 1.57        7       18
Draws were sampled using sampling(NUTS). For each parameter, Bulk_ESS
and Tail_ESS are effective sample size measures, and Rhat is the potential

Please, help me answer the following questions:
1)
As you can see, most of the estimates have very large values: is this a
symptom of an issue?
If yes, how can I fix this issue?
In general, what tools can I use to make the diagnostic of such models
(built with brm or other bayesian modelling functions)
2)
Could you please suggest some tools to assess the goodness of fit for this
model (this type of models)?
3)
It was hard for me to find 02 functions to fit Mixed Multinomial
regression: one frequentist (mixcat::npmlt) and one Bayesian (brms::brm)
Any suggestion for other functions and/or documentation to build such
models will be much appreciated.

Thanks in advance for your answers.
Best regards.
Romaine

	[[alternative HTML version deleted]]