Run linear regression model on nested dataset after grouping (multiply imputed dataset) - r

Question

*I want to group nested (multiply imputed) dataset and then apply linear regression on each dataset. I have tried a number of approaches, including the map options (2) and the for loop (3). I have had no luck at all. I want the model results to look like results from summary(mod1). Does anyone know what I could be doing wrong?
# get dependencies
library(mice)
library(tidyverse)
# impute the boys dataset from mice package
boys_imp <- mice(boys)
# 1) I want to run a model like this on my multiply imputed dataset
mod <- boys %>%
group_by(reg) %>%
do(tidy(
lm(
data=.,
formula = wgt ~ bmi),
conf.int = T))
summary(mod1)
# A tibble: 12 × 8
# Groups: reg [6]
reg term estimate std.error statistic p.value conf.low conf.high
<fct> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 north (Intercept) -81.9 9.84 -8.32 2.48e-12 -101. -62.3
2 north bmi 6.84 0.500 13.7 2.53e-22 5.85 7.84
3 east (Intercept) -75.3 7.62 -9.89 3.21e-18 -90.4 -60.3
4 east bmi 6.29 0.420 15.0 4.53e-32 5.46 7.12
5 west (Intercept) -91.9 6.31 -14.6 2.48e-34 -104. -79.4
6 west bmi 7.17 0.347 20.7 3.49e-54 6.49 7.86
7 south (Intercept) -79.8 6.73 -11.9 1.83e-24 -93.1 -66.5
8 south bmi 6.47 0.373 17.3 1.63e-40 5.73 7.20
9 city (Intercept) -92.0 13.9 -6.61 6.75e- 9 -120. -64.2
10 city bmi 6.95 0.757 9.18 1.39e-13 5.44 8.46
11 NA (Intercept) -88.6 43.8 -2.02 2.92e- 1 -645. 468.
12 NA bmi 6.46 2.89 2.24 2.68e- 1 -30.2 43.1
# 2) the map way --------------------------------------------------------
mod_imp <- boys_imp %>%
mice::complete("all") %>%
map(group_by, reg) %>%
map(lm, formula = wgt ~ bmi) %>%
pool()
summary(mod_imp)
term estimate std.error statistic df p.value
1 (Intercept) -85.473428 3.5511961 -24.06891 715.1703 0
2 bmi 6.793622 0.1945322 34.92287 693.7835 0
# 3) for loop way-------------------------------------------------------
# nest the mids dataset
boys_imp2 <- boys_imp %>%
mice::complete("all")
dat1 <- replicate(length(boys_imp2), NULL) # preallocate same size
# run the for loop
for (i in seq_along(boys_imp2)) {
dat1[[i]] <- boys_imp2[[i]] %>%
group_by(reg) %>%
do(lm(wgt ~ bmi, data = boys_imp2[[i]]))
}
|==================================================================|100% ~0 s remaining Error in `do()`:
! Results 1, 2, 3, 4, 5, ... must be data frames, not lm.
Run `rlang::last_error()` to see where the error occurred.*

Answer 1

I have found a solution to the problem. This involve grouping the data by ID and variable of interest, subsequently I map lm on to the datasets. I then finish off with unnesting the data
boys_imp %>%
mice::complete("long", include = FALSE) %>%
group_by(.imp, reg) %>%
nest() %>%
mutate(lm_model = map(data, ~lm(bmi ~ phb, data = .))) %>%
group_by(reg) %>%
summarise(model = list(tidy(pool(lm_model),conf.int = T))) %>%
unnest_wider(model) %>%
unnest(cols = c(term, estimate, std.error,
statistic, p.value, conf.low, conf.high))
# A tibble: 30 × 16
reg term estimate std.error statistic p.value conf.low conf.high b df dfcom fmi lambda m riv ubar
<fct> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 north (Intercept) 19.3 0.332 57.9 0 18.6 19.9
2 north phb.L 5.10 0.678 7.53 1.81e-10 3.75 6.46
3 north phb.Q 1.25 0.800 1.56 1.24e- 1 -0.357 2.86
4 north phb.C -0.430 0.882 -0.487 6.30e- 1 -2.25 1.39
5 north phb^4 -1.10 0.948 -1.16 2.57e- 1 -3.07 0.862
6 north phb^5 -0.156 1.08 -0.144 8.87e- 1 -2.41 2.10
7 east (Intercept) 18.7 0.244 76.8 0 18.3 19.2
8 east phb.L 4.83 0.509 9.48 4.44e-15 3.82 5.84
9 east phb.Q 1.10 0.692 1.60 1.27e- 1 -0.343 2.55
10 east phb.C -0.518 0.671 -0.772 4.49e- 1 -1.91 0.878
# … with 20 more rows
# ℹ Use `print(n = ...)` to see more rows