updates for glmmTMB changes

[bbolker]

There are two changes here:

1. account for the fact that Gaussian dispersion is parameterized as log-SD rather than log-variance now
2. change starting value for theta to avoid a singular fit (caused by slight changes in numerical stability due to Gaussian reparameterization)

also some minor spelling corrections (sorry)

[strengejacke]

Thanks!

This test is also failing:

══ Failed tests ════════════════════════════════════════════════════════════════
── Failure ('test-glmmTMB.R:899:3'): get_parameters ────────────────────────────
get_parameters(m0)$Estimate (`actual`) not equal to c(200.03431, -99.71491, 1.6014, 0.69323) (`expected`).

  `actual`: 200.034 -99.715 3.203 1.386
`expected`: 200.034 -99.715 1.601 0.693

[ FAIL 1 | WARN 0 | SKIP 10 | PASS 3987 ]

Am I right that glmmTMB 1.1.9 is introducing these "errors"? If so, we should conditionally run the affected tests on that version only, i.e. adding skip_if_not_installed("glmmTMB", minimum_version = "1.1.9"), and probably require that version in the DESCRIPTION file as well?

[strengejacke]

This may also break code in performance (but I'm planning a new release the next days anyway). I think functions like icc() or r2_nakagawa() should be affected, if variance components change. On the other hand, if the breaking change is:

account for the fact that Gaussian dispersion is parameterized as log-SD rather than log-variance now

then tests shouldn't probably be changed, but rather we want to fix insight::get_variance(), so that indeed the variance, not SD, is returned again? I think that makes more sense, in order to get correct results for r2/icc again.

[bbolker]

I think I fixed that one too?

Yes, these tests should be conditional.

I don't see any need to require version 1.1.9 in DESCRIPTION.

[bbolker]

Have to run, will respond later, but hopefully get_variance would be using accessor methods such as sigma() or VarCorr that would be robust to these internal changes?

[strengejacke]

I think I fixed that one too?

Yeah, the PR still used glmmTMB 1.1.8, that's why tests are failing now.

[strengejacke]

Have to run, will respond later, but hopefully get_variance would be using accessor methods such as sigma() or VarCorr that would be robust to these internal changes?

Ok, I'll also take a look.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 55.79%. Comparing base (ad16942) to head (a864be5).

❗ Current head a864be5 differs from pull request most recent head 0c36061. Consider uploading reports for the commit 0c36061 to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #861      +/-   ##
==========================================
+ Coverage   55.75%   55.79%   +0.03%     
==========================================
  Files         125      125              
  Lines       15514    15514              
==========================================
+ Hits         8650     8656       +6     
+ Misses       6864     6858       -6     

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

[strengejacke]

r2/icc are identical between glmmTMB version 1.1.8 and 1.1.9, so no need for action here.

[strengejacke]

Thanks, looks good!