`parameters()` does not extract needed columns for `mipo` objects

[vincentarelbundock]

The parameters() function does not return all relevant columns for mipo multiple imputation objects. For example, when estimating multinomial models, we should get a response column, but we don’t:

library(mice)
library(nnet)
library(parameters)

data = ampute(mtcars) ## Ampute mtcars and impute two data sets
imp = mice(data$amp, m = 2)
imp.l = complete(imp, action = "long")
model = list() ## Fit and pool models
for (i in 1:2) model[[i]] = multinom(carb ~ disp + hp, data = imp.l, subset = .imp == i)
pooled = pool(model)

parameters(pooled)
# # Fixed Effects
# 
# Parameter   | Coefficient |   SE |        95% CI | Statistic |     p
# --------------------------------------------------------------------
# (Intercept) |       -1.85 | 2.03 | [-6.18, 2.48] |     -0.91 | 0.377
# disp        |    6.51e-03 | 0.01 | [-0.02, 0.03] |      0.55 | 0.588
# hp          |        0.01 | 0.03 | [-0.06, 0.08] |      0.35 | 0.729
# (Intercept) |      -11.88 | 2.03 | [-6.18, 2.48] |     -0.91 | 0.377
# disp        |       -0.03 | 0.01 | [-0.02, 0.03] |      0.55 | 0.588
# hp          |        0.13 | 0.03 | [-0.06, 0.08] |      0.35 | 0.729
# (Intercept) |      -10.95 | 2.03 | [-6.18, 2.48] |     -0.91 | 0.377
# disp        |       -0.02 | 0.01 | [-0.02, 0.03] |      0.55 | 0.588
# hp          |        0.13 | 0.03 | [-0.06, 0.08] |      0.35 | 0.729
# (Intercept) |      -32.82 | 2.03 | [-6.18, 2.48] |     -0.91 | 0.377
# disp        |       -6.88 | 0.01 | [-0.02, 0.03] |      0.55 | 0.588
# hp          |        5.99 | 0.03 | [-0.06, 0.08] |      0.35 | 0.729
# (Intercept) |      -37.68 | 2.03 | [-6.18, 2.48] |     -0.91 | 0.377
# disp        |       -0.57 | 0.01 | [-0.02, 0.03] |      0.55 | 0.588
# hp          |        0.73 | 0.03 | [-0.06, 0.08] |      0.35 | 0.729
# 
# Uncertainty intervals (equal-tailed) and p-values (two-tailed) computed using a Wald distribution approximation.

The best way to deal with this would probably be to use the mice:::tidy.mipo() method, which ships by default with mice, and should thus always be available whenever the user has a mipo object. Note the y.level column at the very end, which is the one we need.

mice:::tidy.mipo(pooled)
#           term      estimate  std.error   statistic    p.value            b           df dfcom       fmi      lambda m          riv         ubar y.level
# 1  (Intercept)  -1.852990716 2.03453974  -0.9107665 0.37660359 1.102874e-02 1.523514e+01    17 0.1132365 0.003996547 2 4.012583e-03 4.122809e+00       2
# 2         disp   0.006513995 0.01176502   0.5536747 0.58830509 3.589520e-06 1.438477e+01    17 0.1494675 0.038899332 2 4.047373e-02 1.330315e-04       2
# 3           hp   0.011120602 0.03154816   0.3524960 0.72945522 1.755040e-05 1.474169e+01    17 0.1361974 0.026450282 2 2.716891e-02 9.689608e-04       2
# 4  (Intercept) -11.877096615 4.67299293  -2.5416466 0.02235546 4.321350e-02 1.525253e+01    17 0.1122170 0.002968386 2 2.977224e-03 2.177204e+01       3
# 5         disp  -0.029333751 0.02030589  -1.4445936 0.16883125 1.185566e-06 1.522970e+01    17 0.1135509 0.004312936 2 4.331618e-03 4.105506e-04       3
# 6           hp   0.134618347 0.06203947   2.1698824 0.04625323 1.430839e-05 1.520749e+01    17 0.1148087 0.005576297 2 5.607567e-03 3.827433e-03       3
# 7  (Intercept) -10.954181418 4.36789413  -2.5078862 0.02391829 4.721884e-02 1.524000e+01    17 0.1129545 0.003712465 2 3.726299e-03 1.900767e+01       4
# 8         disp  -0.024111291 0.01854744  -1.2999798 0.21293953 9.900675e-07 1.522963e+01    17 0.1135550 0.004317062 2 4.335780e-03 3.425223e-04       4
# 9           hp   0.127845630 0.05976495   2.1391407 0.04904225 1.357680e-05 1.520525e+01    17 0.1149337 0.005701585 2 5.734279e-03 3.551484e-03       4
# 10 (Intercept) -32.820498916 2.15430280 -15.2348588 0.99315334 3.093918e+00 4.710637e-04    17 0.9999897 0.999969197 2 3.246338e+04 1.429573e-04       6
# 11        disp  -6.880753231 4.54316572  -1.5145283 0.90370842 1.372584e+01 3.684304e-02    17 0.9991466 0.997500322 2 3.990516e+02 5.159423e-02       6
# 12          hp   5.993956276 3.79909400   1.5777331 0.83682645 9.572496e+00 7.313213e-02    17 0.9982006 0.994847157 2 1.930676e+02 7.437157e-02       6
# 13 (Intercept) -37.679041139 0.51623460 -72.9882139 0.97967691 1.776354e-01 2.578597e-03    17 0.9999436 0.999831029 2 5.917164e+03 4.503055e-05       8
# 14        disp  -0.566960948 0.30634672  -1.8507166 0.17811331 3.198792e-02 2.530805e+00    17 0.6880006 0.511270642 2 1.046122e+00 4.586643e-02       8
# 15          hp   0.726340764 0.30756516   2.3615834 0.08095960 2.485393e-02 3.799515e+00    17 0.5723222 0.394105145 2 6.504514e-01 5.731543e-02       8

Unfortunately, I don’t have the bandwidth right now to work on this, but I thought I’d open an issue anyway

[strengejacke]

Thanks, should be fixed after merging the two referenced PRs.

[vincentarelbundock]

This is awesome, thanks a ton!