Skip to contents

Block-stratified sensitivity fit

Source: R/statistics-models.R
cr_blocked_effect.Rd

Re-estimates each contrast within a blocking factor, typically the plate. Units on one plate share a preparation run, an imaging session and a single control denominator, so they are not independent and a pooled estimate carries the between-plate variance inside its standard deviation. Fitting value ~ group + block on unit-level data removes the additive block effect and shows whether the pooled headline estimate is conservative or inflated by pooling.

Usage

cr_blocked_effect(
  data,
  value,
  group_var,
  reference_level,
  comparison_levels = NULL,
  block_var,
  by = NULL,
  unit = NULL,
  conf_level = 0.95,
  min_blocks = 2
)

Arguments

data

A data frame with one row per unit (or per cell when unit is given), or a cr_experiment.

value

Name of the numeric response column.

group_var

Name of the column holding the compared groups.

reference_level

Reference level of group_var.

comparison_levels

Levels contrasted against the reference. NULL (default) uses every other level present.

block_var

Name of the blocking column, for example the plate.

by

Optional character vector of stratifying columns, for example the compound.

unit

Optional column identifying the unit of replication. When given, rows are averaged per unit before the model is fitted.

conf_level

Confidence level for the coefficient interval.

min_blocks

Minimum number of blocks that must carry data for a contrast to be fitted (default 2; with one block the model is the unblocked one).

Value

A tibble with the by columns, contrast, n_units, n_blocks, shift_within_block (the group coefficient), ci_low, ci_high, resid_sd, d_within_block and p_value. Contrasts that could not be fitted are returned in the skipped attribute.

Details

The reported d_within_block is the group coefficient divided by the residual standard deviation of the blocked fit, so it is on the same scale as Cohen's d from cr_effect_grid() but conditions on the block.

See also

cr_effect_grid(), cr_unit_variability().

Other screen statistics: cr_unit_variability()

Examples

set.seed(1)
units <- data.frame(
  compound = rep(c("CompoundA", "CompoundB"), each = 16),
  arm = rep(rep(c("reference", "interval_short"), each = 8), 2),
  plate = rep(rep(c("P1", "P2"), each = 4), 8),
  log2_fc = stats::rnorm(32)
)
units$log2_fc <- units$log2_fc +
  ifelse(units$compound == "CompoundA" & units$arm == "interval_short",
         -1.2, 0) +
  ifelse(units$plate == "P2", 0.5, 0)
cr_blocked_effect(units, value = "log2_fc", group_var = "arm",
                  reference_level = "reference",
                  block_var = "plate", by = "compound")
#> # A tibble: 2 × 10
#>   compound  contrast n_units n_blocks shift_within_block ci_low ci_high resid_sd
#>   <chr>     <chr>      <int>    <int>              <dbl>  <dbl>   <dbl>    <dbl>
#> 1 CompoundA referen…      32        2             -1.28  -1.98   -0.576    0.973
#> 2 CompoundB referen…      32        2             -0.250 -0.896   0.397    0.894
#> # ℹ 2 more variables: d_within_block <dbl>, p_value <dbl>