Return detected blocks plus info

Given the results of the splitting and testing algorithm, report on the blocks where the null of no effects could be rejected at level alpha. Currently calculates rejections using an FWER style criteria (p of a node = max of all previous nodes) if the final alphas are all the same as the scalar alpha OR if fwer=TRUE.

Usage

report_detections(
  orig_res,
  fwer = TRUE,
  alpha = 0.05,
  only_hits = FALSE,
  blockid = "blockF"
)

Arguments

orig_res

results data.table output from the find_blocks function.

fwer

(default is TRUE) means that a block is detected (or not) using the maximum p-value associated with the block (or the groups containing that block). fwer=FALSE to detect blocks (or groups of blocks) using FDR control.

alpha

Is the false positive rate used for detecting an effect if it is constant (i.e. not an FDR-style approach).

only_hits

(default FALSE) returns only the detected blocks instead of all of them

blockid

Name of block variable (the blocking variable is a factor)

Value

A data.table adding columns to the res data.table: hit (TRUE when the block is detected either by its own test or by an unattributed group rejection; never NA); hit_type ("single" = the block's own singleton test rejected; "group" = the block's family parent rejected while no child of that parent rejected, so the effect is localized to the parent but not attributed to specific blocks; "none" otherwise – including blocks under a rejected parent that a sibling's individual rejection already explains); group_p (for "group" blocks, the rejecting parent's p-value; NA otherwise); and fin_depth (the block's own final tested depth). pfinalb from find_blocks is the running maximum of p-values along the block's tested path, so for "group" blocks it shows the non-significant child-level p while group_p shows the parent rejection that constitutes the finding.

Examples

if (FALSE) { # \dontrun{
# Use example data and run find_blocks
data(example_dat, package = "manytestsr")
library(data.table)
library(dplyr)

# Create block-level dataset
example_bdat <- example_dat %>%
  group_by(blockF) %>%
  summarize(
    nb = n(),
    pb = mean(trt),
    hwt = (nb / nrow(example_dat)) * (pb * (1 - pb)),
    .groups = "drop"
  ) %>%
  as.data.table()

# Run find_blocks
results <- find_blocks(
  idat = example_dat,
  bdat = example_bdat,
  blockid = "blockF",
  splitfn = splitCluster,
  pfn = pOneway,
  fmla = Y1 ~ trtF | blockF,
  parallel = "no"
)

# Report detections using FWER control
detections_fwer <- report_detections(results$bdat, fwer = TRUE, alpha = 0.05)
head(detections_fwer[, .(blockF, hit, pfinalb)])

# Report only significant blocks
hits_only <- report_detections(results$bdat, fwer = TRUE, only_hits = TRUE)
print(hits_only)
} # }