Small updates to PR #40

[svteichman]

This builds off @shirmath's PR #40. It adds tests of glm_test() with logistic and gaussian models (to see if this addresses the lower code coverage from that PR), and increments the version to denote a minor update.

[codecov]

Codecov Report

❌ Patch coverage is 89.36170% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 87.86%. Comparing base (aaf92ec) to head (a8b5731).
⚠️ Report is 29 commits behind head on main.

Files with missing lines	Patch %	Lines
R/V_matrix_contribution.R	83.33%	3 Missing ⚠️
R/D_matrix_contribution.R	85.71%	1 Missing ⚠️
R/gee_test.R	50.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #42      +/-   ##
==========================================
+ Coverage   87.43%   87.86%   +0.43%     
==========================================
  Files          23       26       +3     
  Lines         748      783      +35     
==========================================
+ Hits          654      688      +34     
- Misses         94       95       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[svteichman]

@shirmath this PR takes yours and adds tests to increase code coverage (also simplifies function for Si matrix and updates Di matrix function to not hit matrix multiplication error for clusters with size 1). I think this is all set, conditional on the question I noted on your PR #40, about n-p vs n-1 in the denominator for the Gaussian sigma2_tilde.

[svteichman]

@shirmath Do you mind reviewing this PR when you get the chance, to make sure that all of my updates make sense to you?

[shirmath]

@svteichman Why does the denominator cancel out in the test statistic in the Gaussian case? All the changes made sense to me outside of that

[adw96]

Hi Shirley — if you consider it a constant factor c in the definition of the matrix V (?) in the Guo and coauthors paper, you’ll see that with the repeated matrix inversions and multiplications that constant factor cancels out ultimately in the test statistic. (I think it turns to it to be c^-2 in the meat of the sandwich and c in the bread, so constant overall) Amy Willis, Ph.D. // Associate Professor // UW Biostatistics // sent from a device with tiny keys

…

On Mon, Sep 15, 2025 at 7:40 AM, shirmath ***@***.***> wrote: *shirmath* left a comment (statdivlab/raoBust#42) <https://urldefense.com/v3/__https://github.com/statdivlab/raoBust/pull/42*issuecomment-3292529253__;Iw!!K-Hz7m0Vt54!nffe0iwQRcJEfQEtT9Z-OnjkV2P6AOjKjYWiLK3VO8V7vD2soiSLf0sFncQYQeI8H5z6mRiSm-FsK2QnDxr67_0$> @svteichman <https://urldefense.com/v3/__https://github.com/svteichman__;!!K-Hz7m0Vt54!nffe0iwQRcJEfQEtT9Z-OnjkV2P6AOjKjYWiLK3VO8V7vD2soiSLf0sFncQYQeI8H5z6mRiSm-FsK2QnlEyeC-Q$> Why does the denominator cancel out in the test statistic in the Gaussian case? All the changes made sense to me outside of that — Reply to this email directly, view it on GitHub <https://urldefense.com/v3/__https://github.com/statdivlab/raoBust/pull/42*issuecomment-3292529253__;Iw!!K-Hz7m0Vt54!nffe0iwQRcJEfQEtT9Z-OnjkV2P6AOjKjYWiLK3VO8V7vD2soiSLf0sFncQYQeI8H5z6mRiSm-FsK2QnDxr67_0$>, or unsubscribe <https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/AEGHKUV6GSWFFTV3LUBHH4L3S3FVLAVCNFSM6AAAAACFLJYY62VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTEOJSGUZDSMRVGM__;!!K-Hz7m0Vt54!nffe0iwQRcJEfQEtT9Z-OnjkV2P6AOjKjYWiLK3VO8V7vD2soiSLf0sFncQYQeI8H5z6mRiSm-FsK2Qn7tdEWAE$> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[svteichman]

@adw96 this should be all set to review and merge when you get the chance!

[adw96]

This is excellent, @shirmath and @svteichman -- well done!! This is clear, well-designed, modular code. I think this will flourish as an R package 😻

One minor thing I noticed (that don't need to be addressed to merge) is that (family == "binomial" & link == "logit") isn't tested, reducing coverage a little. If it's easy to add a test, that'd be awesome.

Regardless, feel free to merge when ready!

[svteichman]

@adw96 I think that a test with binomial family & logit link is in "test-basic.R" lines 51-61. Is this the kind of thing you were hoping for?

test_that("test logistic example", {
  
  expect_type(glm_test((dist > 43) ~ speed, data = cars, family=binomial(link = "logit"))$coef_tab,
              "double")
  
  cars$id <- c(rep(1:5, 9), 6:10)
  expect_type(gee_test(formula = (dist > 43) ~ speed, data = cars, id = "id", 
                       family=gaussian(link = "identity"))$coef_tab,
              "list")
  
})

[adw96]

weird!! covr shows me this

so... please ignore my concern 😊