Fix 17 test failures: probability vectors, object init, vestigial dataset, boundary SE detection

[Copilot]

17 test failures caused by mismatched probability vector lengths, a forward-reference in data.frame(), a test referencing a removed dataset, and incomplete large-SE boundary detection logic.

Changes

tests/testthat/test-calc-risk-diff.R

• Age probability vector (create_cachar_inspired_data): 18:70 is 53 values; original prob argument supplied 73 weights. Replaced with explicit length-matched array normalized to sum to 1:

  ages <- 18:70
  age_probs <- numeric(length(ages))
  age_probs[ages >= 18 & ages <= 39] <- 0.8
  age_probs[ages >= 40 & ages <= 59] <- 1.2
  age_probs[ages >= 60 & ages <= 70] <- 0.6
  age_probs <- age_probs / sum(age_probs)
  sample(ages, n, replace = TRUE, prob = age_probs)

• Object initialization (create_convergence_challenge_data): exposure was referenced inside data.frame() before it existed. Refactored to build the frame first, then assign outcome_prob and outcome as follow-on steps.

• Removed birthweight test: birthweight dataset was replaced by cachar_sample in a prior iteration; test referencing it was dead code causing a load failure.

R/boundary_detection.R

• Large SE detection: Previous logic only flagged large SEs via relative ratio (SE / |coef| > 2), which misses cases where the coefficient itself is near zero. Detection now triggers on either absolute SE > 2.0 or relative ratio > 2.0 (guarded against all-small-coefficient false positives).

Original prompt

Fix Test Failures: Correct Probability Vectors, Object Initialization, and Remove Vestigial Data

This PR fixes 17 test failures in the riskdiff test suite:

Critical Issues Fixed:

1. Test Data Generation Errors (9 failures)

• Root Cause: Mismatched probability vector lengths in sample() calls
• Example Problem: sample(18:70, prob = c(rep(0.8, 22), rep(1.2, 30), rep(0.6, 21))) attempts to sample from 53 values (18:70) but provides 73 probability weights
• Fix Applied: Create explicit probability distribution array of correct length (53) and normalize

In create_cachar_inspired_data(), replace:

age = sample(18:70, n, replace = TRUE,
             prob = c(rep(0.8, 22), rep(1.2, 30), rep(0.6, 21)))

With:

ages <- 18:70
age_probs <- numeric(length(ages))
age_probs[ages >= 18 & ages <= 39] <- 0.8   # Young
age_probs[ages >= 40 & ages <= 59] <- 1.2   # Middle
age_probs[ages >= 60 & ages <= 70] <- 0.6   # Older
age_probs <- age_probs / sum(age_probs)
age = sample(ages, n, replace = TRUE, prob = age_probs)

2. Object Initialization Error (1 failure)

• Root Cause: create_convergence_challenge_data() uses exposure variable in dplyr::mutate() before it's fully defined
• Error Message: object 'exposure' not found
• Fix Applied: Create complete data frame first, then pipe to mutate

3. Remove Vestigial "birthweight" Dataset (1 failure + cleanup)

• Context: The "birthweight" dataset was completely replaced with "cachar_sample" in a previous package iteration
• Issue: Test references non-existent birthweight dataset
• Fix Applied: Remove entire test section referencing birthweight data

4. Boundary Detection Large SE Test (1 failure)

• Root Cause: Large standard error detection logic only checks relative ratio, missing edge cases
• Fix Applied: Enhanced detection to trigger on EITHER:
  • Large absolute SE (> 2.0), OR
  • Large relative SE (ratio > 2.0)
  • Properly handles small coefficient edge cases

Files Modified:

1. tests/testthat/test-calc-risk-diff.R (complete rewrite of test data generation)

   • Fixed create_cachar_inspired_data()
   • Fixed create_convergence_challenge_data()
   • Removed birthweight test section

2. R/boundary_detection.R (lines ~120-140)

   • Improved large standard error detection logic

Expected Results:

• Before: 545 passing, 17 failing
• After: 562 passing, 0 failing

Testing:

All changes maintain backward compatibility with existing passing tests. Only failing tests are fixed.

This pull request was created from Copilot chat.

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.