Permute data while respecting grouping structure(s) of observations

Usage

permute_by_predictor(
  jlmer_spec,
  predictors,
  predictor_type = c("guess", "between_participant", "within_participant"),
  n = 1L
)

Arguments

jlmer_spec: Data prepped for jlmer from make_jlmer_spec()
predictors: A vector of terms from the model. If multiple, the must form the levels of one predictor.
predictor_type: Whether the predictor is "between_participant" or "within_participant". Defaults to "guess".
n: Number of permuted samples of the data to generate. Defaults to 1L.

Value

A long dataframe of permuted re-samples with .id column representing replication IDs.

Examples

# \donttest{

# Example data setup
chickweights_df <- ChickWeight
chickweights_df <- chickweights_df[chickweights_df$Time <= 20, ]
chickweights_df$DietInt <- as.integer(chickweights_df$Diet)
head(chickweights_df)
#> Grouped Data: weight ~ Time | Chick
#>   weight Time Chick Diet DietInt
#> 1     42    0     1    1       1
#> 2     51    2     1    1       1
#> 3     59    4     1    1       1
#> 4     64    6     1    1       1
#> 5     76    8     1    1       1
#> 6     93   10     1    1       1

# Example 1: Spec object using the continuous `DietInt` predictor
chickweights_spec1 <- make_jlmer_spec(
  formula = weight ~ 1 + DietInt,
  data = chickweights_df,
  subject = "Chick", time = "Time"
)
chickweights_spec1
#> ── jlmer specification ───────────────────────────────────────── <jlmer_spec> ──
#> Formula: weight ~ 1 + DietInt
#> Predictors:
#>   DietInt: DietInt
#> Groupings:
#>   Subject: Chick
#>   Trial:
#>   Time: Time
#> Data:
#> # A tibble: 533 × 4
#>   weight DietInt Chick  Time
#>    <dbl>   <dbl> <ord> <dbl>
#> 1     42       1 1         0
#> 2     51       1 1         2
#> 3     59       1 1         4
#> # ℹ 530 more rows
#> ────────────────────────────────────────────────────────────────────────────────

# Shuffling `DietInt` values guesses `predictor_type = "between_participant"`
reset_rng_state()
spec1_perm1 <- permute_by_predictor(chickweights_spec1, predictors = "DietInt")
#> ℹ Guessed `predictor_type` to be "between_participant"
# This calls the same shuffling algorithm for CPA in Julia, so counter is incremented
get_rng_state()
#> [1] 38

# Shuffling under shared RNG state reproduces the same permutation of the data
reset_rng_state()
spec1_perm2 <- permute_by_predictor(chickweights_spec1, predictors = "DietInt")
#> ℹ Guessed `predictor_type` to be "between_participant"
identical(spec1_perm1, spec1_perm2)
#> [1] TRUE

# Example 2: Spec object using the multilevel `Diet` predictor
chickweights_spec2 <- make_jlmer_spec(
  formula = weight ~ 1 + Diet,
  data = chickweights_df,
  subject = "Chick", time = "Time"
)
chickweights_spec2
#> ── jlmer specification ───────────────────────────────────────── <jlmer_spec> ──
#> Formula: weight ~ 1 + Diet2 + Diet3 + Diet4
#> Predictors:
#>   Diet: Diet2, Diet3, Diet4
#> Groupings:
#>   Subject: Chick
#>   Trial:
#>   Time: Time
#> Data:
#> # A tibble: 533 × 6
#>   weight Diet2 Diet3 Diet4 Chick  Time
#>    <dbl> <dbl> <dbl> <dbl> <ord> <dbl>
#> 1     42     0     0     0 1         0
#> 2     51     0     0     0 1         2
#> 3     59     0     0     0 1         4
#> # ℹ 530 more rows
#> ────────────────────────────────────────────────────────────────────────────────

# Levels of a category are automatically shuffled together
reset_rng_state()
spec2_perm1 <- permute_by_predictor(chickweights_spec2, predictors = "Diet2")
#> ℹ Guessed `predictor_type` to be "between_participant"
#> ℹ Shuffling all levels of the factor together ("Diet2", "Diet3", and "Diet4")
reset_rng_state()
spec2_perm2 <- permute_by_predictor(chickweights_spec2, predictors = c("Diet2", "Diet3", "Diet4"))
#> ℹ Guessed `predictor_type` to be "between_participant"
identical(spec2_perm1, spec2_perm2)
#> [1] TRUE

# }