mismatch between `{emmeans}`\s and `model_parameters()`'s adjusted p-values

[mattansb]

library(emmeans)
library(parameters)

warp.lm <- lm(breaks ~ wool * tension, data = warpbreaks)

myc <- emmeans (warp.lm,  ~ wool + tension) |> 
  contrast("pairwise")

myc
#>  contrast  estimate   SE df t.ratio p.value
#>  A L - B L   16.333 5.16 48   3.167  0.0302
#>  A L - A M   20.556 5.16 48   3.986  0.0030
#>  A L - B M   15.778 5.16 48   3.059  0.0398
#>  A L - A H   20.000 5.16 48   3.878  0.0041
#>  A L - B H   25.778 5.16 48   4.998  0.0001
#>  B L - A M    4.222 5.16 48   0.819  0.9627
#>  B L - B M   -0.556 5.16 48  -0.108  1.0000
#>  B L - A H    3.667 5.16 48   0.711  0.9797
#>  B L - B H    9.444 5.16 48   1.831  0.4561
#>  A M - B M   -4.778 5.16 48  -0.926  0.9377
#>  A M - A H   -0.556 5.16 48  -0.108  1.0000
#>  A M - B H    5.222 5.16 48   1.013  0.9115
#>  B M - A H    4.222 5.16 48   0.819  0.9627
#>  B M - B H   10.000 5.16 48   1.939  0.3919
#>  A H - B H    5.778 5.16 48   1.120  0.8706
#> 
#> P value adjustment: tukey method for comparing a family of 6 estimates

model_parameters(myc)
#> contrast  | Coefficient |   SE |          95% CI | t(48) |      p
#> -----------------------------------------------------------------
#> A L - B L |       16.33 | 5.16 | [  5.96, 26.70] |  3.17 | 0.133 
#> A L - A M |       20.56 | 5.16 | [ 10.19, 30.93] |  3.99 | 0.016 
#> A L - B M |       15.78 | 5.16 | [  5.41, 26.15] |  3.06 | 0.169 
#> A L - A H |       20.00 | 5.16 | [  9.63, 30.37] |  3.88 | 0.022 
#> A L - B H |       25.78 | 5.16 | [ 15.41, 36.15] |  5.00 | < .001
#> B L - A M |        4.22 | 5.16 | [ -6.15, 14.59] |  0.82 | > .999
#> B L - B M |       -0.56 | 5.16 | [-10.93,  9.81] | -0.11 | > .999
#> B L - A H |        3.67 | 5.16 | [ -6.70, 14.04] |  0.71 | > .999
#> B L - B H |        9.44 | 5.16 | [ -0.93, 19.81] |  1.83 | 0.873 
#> A M - B M |       -4.78 | 5.16 | [-15.15,  5.59] | -0.93 | > .999
#> A M - A H |       -0.56 | 5.16 | [-10.93,  9.81] | -0.11 | > .999
#> A M - B H |        5.22 | 5.16 | [ -5.15, 15.59] |  1.01 | > .999
#> B M - A H |        4.22 | 5.16 | [ -6.15, 14.59] |  0.82 | > .999
#> B M - B H |       10.00 | 5.16 | [ -0.37, 20.37] |  1.94 | 0.822 
#> A H - B H |        5.78 | 5.16 | [ -4.59, 16.15] |  1.12 | 0.998 
#> 
#> p-value adjustment method: Tukey

^{Created on 2025-05-08 with reprex v2.1.1}

---

It looks like we have a huge internal .p_adjust() function that is trying to apply all these corrections, accounting for emmeans's by-variables, etc... why aren't we simply extracting the p-values given by emmeans? It seems (in model_parameters.emmGrid()) that we are going out of our way to unadjusted and re-adjust p-values...

Is there any objection to relying on the values given by emmeans?

[mattansb]

We also seem to be calling summary(<emmGrid>) more than once (more than once in {insight} and then again in {parameters}, which seems inefficient.

Is there any objection to simplifying this redundancy?

[strengejacke]

Is there any objection to simplifying this redundancy?

No :-)