Univariable GLM / LMM Screens Over Inputs and Outputs

Fits [glm_single()] or [lmm_single()] for every combination of an input (predictor) and an output (outcome). With `normalize = TRUE` the main estimate is the one on the 0 to 1 scaled predictor (comparable across inputs: the change in the output over the full observed range of the input); with `include_dy_dx = TRUE` the estimate on the original scale (the marginal effect `dy/dx`, change per unit of the input) is added with the suffix `_dy_dx`.

Usage

batch_glm_single(
  data,
  inputs,
  outputs,
  normalize = TRUE,
  include_dy_dx = TRUE,
  family = stats::gaussian(),
  conf_level = 0.95,
  p_breaks = c(0.001, 0.01, 0.05),
  p_labels = c("***", "**", "*", "ns")
)

batch_lmm_single(
  data,
  inputs,
  outputs,
  id_col = "subject_id",
  normalize = TRUE,
  include_dy_dx = TRUE,
  conf_level = 0.95,
  p_breaks = c(0.001, 0.01, 0.05),
  p_labels = c("***", "**", "*", "ns")
)

Arguments

data: A data frame (long, with `id_col`, for `batch_lmm_single()`).
inputs, outputs: Character vectors of predictor and response columns.
normalize: Fit on the 0 to 1 scaled predictor as the main result. If `FALSE`, only the original scale is fitted and `include_dy_dx` is ignored.
include_dy_dx: Add the original-scale columns (see above).
family: Family for `batch_glm_single()`, passed to [glm_single()].
conf_level: Confidence level.
p_breaks, p_labels: Significance scheme; `p_labels` has one more entry than `p_breaks`.
id_col: Subject column of `batch_lmm_single()` (random intercept).

Value

A tibble with `output`, `input`, `term`, `estimate`, `std.error`, `statistic`, `p.value`, `conf.low`, `conf.high`, `n`, `status`, `warning`, `p.adj`, `rank`, `significance` and, with `include_dy_dx`, the same estimate columns suffixed `_dy_dx`. A combination that could not be fitted keeps a row with its `status`.

Ranking and significance

`rank` orders the inputs within each output by decreasing absolute main estimate (1 = largest; ties share the lowest rank). `significance` is the label of `p_breaks` that the main p-value falls below (the first break it is strictly below, as in [plot_coef_forest()]), or the last label if none. No multiple-testing adjustment is applied to `p.value`; `p.adj` is the Benjamini-Hochberg adjustment over all combinations of the main fits.

Examples

batch_glm_single(mtcars, c("wt", "hp", "disp"), c("mpg", "qsec"))
#> # A tibble: 6 × 21
#>   output input term  estimate std.error statistic  p.value conf.low conf.high
#>   <chr>  <chr> <chr>    <dbl>     <dbl>     <dbl>    <dbl>    <dbl>     <dbl>
#> 1 mpg    wt    wt      -20.9      2.19     -9.56  1.29e-10   -25.4    -16.4  
#> 2 mpg    hp    hp      -19.3      2.86     -6.74  1.79e- 7   -25.2    -13.5  
#> 3 mpg    disp  disp    -16.5      1.89     -8.75  9.38e-10   -20.4    -12.7  
#> 4 qsec   hp    hp       -5.22     0.951    -5.49  5.77e- 6    -7.17    -3.28 
#> 5 qsec   disp  disp     -2.51     0.951    -2.64  1.31e- 2    -4.45    -0.565
#> 6 qsec   wt    wt       -1.25     1.28     -0.972 3.39e- 1    -3.87     1.37 
#> # ℹ 12 more variables: n <int>, status <chr>, warning <chr>,
#> #   estimate_dy_dx <dbl>, std.error_dy_dx <dbl>, statistic_dy_dx <dbl>,
#> #   p.value_dy_dx <dbl>, conf.low_dy_dx <dbl>, conf.high_dy_dx <dbl>,
#> #   p.adj <dbl>, rank <int>, significance <chr>