Model Comparison with bgmCompare

Introduction

The function bgmCompare() extends bgm() to independent-sample designs. It estimates whether edge weights and category thresholds differ across groups in an ordinal Markov random field (MRF).

Posterior inclusion probabilities indicate how plausible it is that a group difference exists in a given parameter. These can be converted to Bayes factors for hypothesis testing.

ADHD dataset

We illustrate with a subset from the ADHD dataset included in bgms.

library(bgms)

?ADHD
data_adhd = ADHD[ADHD$group == 1, -1]
data_adhd = data_adhd[, 1:5]
data_no_adhd = ADHD[ADHD$group == 0, -1]
data_no_adhd = data_no_adhd[, 1:5]

Fitting a model

fit = bgmCompare(x = data_adhd, y = data_no_adhd, seed = 1234)

Posterior summaries

The summary shows both baseline effects and group differences:

summary(fit)
#> Posterior summaries from Bayesian grouped MRF estimation (bgmCompare):
#> 
#> Category thresholds:
#>      parameter   mean  mcse    sd    n_eff  Rhat
#> 1    avoid (1) -2.676 0.010 0.389 1411.000 1.000
#> 2 closeatt (1) -2.249 0.011 0.376 1198.435 1.002
#> 3 distract (1) -0.486 0.013 0.337  637.870 1.001
#> 4   forget (1) -1.580 0.010 0.330 1070.957 1.000
#> 5 instruct (1) -2.416 0.014 0.401  797.064 1.008
#> 
#> Pairwise interactions:
#>           parameter   mean  mcse    sd    n_eff  Rhat
#> 1    avoid-closeatt  0.993 0.017 0.459  723.538 1.000
#> 2    avoid-distract  1.704 0.009 0.365 1598.070 1.003
#> 3      avoid-forget  0.486 0.014 0.381  709.989 1.009
#> 4    avoid-instruct  0.387 0.014 0.464 1029.911 1.001
#> 5 closeatt-distract -0.257 0.011 0.393 1238.404 1.001
#> 6   closeatt-forget  0.146 0.008 0.312 1513.112 1.003
#> ... (use `summary(fit)$pairwise` to see full output)
#> 
#> Inclusion probabilities:
#>                  parameter  mean    sd  mcse n0->0 n0->1 n1->0 n1->1    n_eff
#>               avoid (main) 1.000 0.000           0     0     0  1999         
#>  avoid-closeatt (pairwise) 0.821 0.383 0.015   209   149   148  1493  678.075
#>  avoid-distract (pairwise) 0.391 0.488 0.012   779   438   438   344 1703.716
#>    avoid-forget (pairwise) 0.819 0.385 0.017   244   118   118  1519  496.957
#>  avoid-instruct (pairwise) 1.000 0.022     0     0     1     1  1997 2002.003
#>            closeatt (main) 1.000 0.000           0     0     0  1999         
#>   Rhat
#>       
#>      1
#>      1
#>  1.031
#>  1.291
#>       
#> ... (use `summary(fit)$indicator` to see full output)
#> Note: NA values are suppressed in the print table. They occur when an indicator
#> was constant (all 0 or all 1) across all iterations, so sd/mcse/n_eff/Rhat
#> are undefined; `summary(fit)$indicator` still contains the NA values.
#> 
#> Group differences (main effects):
#>            parameter   mean    sd mcse n_eff  Rhat
#>     avoid (diff1; 1) -2.523 0.728            1.000
#>  closeatt (diff1; 1) -3.008 0.726            1.002
#>  distract (diff1; 1) -2.552 0.689            1.001
#>    forget (diff1; 1) -2.831 0.654            1.007
#>  instruct (diff1; 1) -2.374 0.884            1.003
#> Note: NA values are suppressed in the print table. They occur here when an
#> indicator was zero across all iterations, so mcse/n_eff/Rhat are undefined;
#> `summary(fit)$main_diff` still contains the NA values.
#> 
#> Group differences (pairwise effects):
#>                  parameter   mean    sd  mcse    n_eff  Rhat
#>     avoid-closeatt (diff1)  1.296 0.911 0.031  890.715 1.000
#>     avoid-distract (diff1)  0.219 0.362 0.011 1018.055 1.000
#>       avoid-forget (diff1)  1.235 0.831 0.032  676.485 1.006
#>     avoid-instruct (diff1) -2.805 0.970 0.036  740.334 1.004
#>  closeatt-distract (diff1) -0.174 0.348 0.012  795.120 1.000
#>    closeatt-forget (diff1)  0.169 0.333 0.012  715.853 1.000
#> ... (use `summary(fit)$pairwise_diff` to see full output)
#> Note: NA values are suppressed in the print table. They occur here when an
#> indicator was zero across all iterations, so mcse/n_eff/Rhat are undefined;
#> `summary(fit)$pairwise_diff` still contains the NA values.
#> 
#> Use `summary(fit)$<component>` to access full results.
#> See the `easybgm` package for other summary and plotting tools.

You can extract posterior means and inclusion probabilities:

coef(fit)
#> $main_effects_raw
#>                baseline     diff1
#> avoid(c1)    -2.6758947 -2.523160
#> closeatt(c1) -2.2489431 -3.007718
#> distract(c1) -0.4859767 -2.551852
#> forget(c1)   -1.5803152 -2.830772
#> instruct(c1) -2.4164664 -2.373887
#> 
#> $pairwise_effects_raw
#>                     baseline      diff1
#> avoid-closeatt     0.9927303  1.2964029
#> avoid-distract     1.7035207  0.2190339
#> avoid-forget       0.4858398  1.2349851
#> avoid-instruct     0.3870157 -2.8050136
#> closeatt-distract -0.2573117 -0.1735373
#> closeatt-forget    0.1456844  0.1691829
#> closeatt-instruct  1.5675969  0.6163591
#> distract-forget    0.4044984  0.2301470
#> distract-instruct  1.2586856  1.2586632
#> forget-instruct    1.1314787  0.8426717
#> 
#> $main_effects_groups
#>                  group1    group2
#> avoid(c1)    -1.4143149 -3.937475
#> closeatt(c1) -0.7450839 -3.752802
#> distract(c1)  0.7899495 -1.761903
#> forget(c1)   -0.1649295 -2.995701
#> instruct(c1) -1.2295228 -3.603410
#> 
#> $pairwise_effects_groups
#>                        group1     group2
#> avoid-closeatt     0.34452892  1.6409318
#> avoid-distract     1.59400382  1.8130377
#> avoid-forget      -0.13165272  1.1033324
#> avoid-instruct     1.78952252 -1.0154911
#> closeatt-distract -0.17054303 -0.3440803
#> closeatt-forget    0.06109296  0.2302758
#> closeatt-instruct  1.25941733  1.8757764
#> distract-forget    0.28942488  0.5195718
#> distract-instruct  0.62935398  1.8880172
#> forget-instruct    0.71014281  1.5528145
#> 
#> $indicators
#>           avoid closeatt distract forget instruct
#> avoid    1.0000   0.8210   0.3915 0.8190   0.9995
#> closeatt 0.8210   1.0000   0.3745 0.3915   0.6030
#> distract 0.3915   0.3745   1.0000 0.3970   0.8440
#> forget   0.8190   0.3915   0.3970 1.0000   0.7485
#> instruct 0.9995   0.6030   0.8440 0.7485   1.0000

Visualizing group networks

We can use the output to plot the network for the ADHD group:

library(qgraph)

adhd_network = matrix(0, 5, 5)
adhd_network[lower.tri(adhd_network)] = coef(fit)$pairwise_effects_groups[, 1]
adhd_network = adhd_network + t(adhd_network)
colnames(adhd_network) = colnames(data_adhd)
rownames(adhd_network) = colnames(data_adhd)

qgraph(adhd_network,
  theme = "TeamFortress",
  maximum = 1,
  fade = FALSE,
  color = c("#f0ae0e"), vsize = 10, repulsion = .9,
  label.cex = 1, label.scale = "FALSE",
  labels = colnames(data_adhd)
)

Next steps