make parameters() show fixed effects restricted to 0

[strengejacke]

Fixes #715

[codecov]

Codecov Report

Merging #902 (3e16c50) into main (dd574ba) will decrease coverage by 0.25%.
The diff coverage is 12.16%.

❗ Current head 3e16c50 differs from pull request most recent head 22798aa. Consider uploading reports for the commit 22798aa to get more accurate results

@@            Coverage Diff             @@
##             main     #902      +/-   ##
==========================================
- Coverage   57.56%   57.31%   -0.25%     
==========================================
  Files         186      186              
  Lines       13785    13857      +72     
==========================================
+ Hits         7935     7942       +7     
- Misses       5850     5915      +65     

Files Changed	Coverage Δ
R/display.R	0.00% <ø> (ø)
R/print_html.R	0.00% <0.00%> (ø)
R/print_md.R	0.00% <0.00%> (ø)
R/utils.R	45.76% <0.00%> (-25.29%)	⬇️
R/utils_format.R	68.94% <4.54%> (-4.97%)	⬇️
R/1_model_parameters.R	85.83% <100.00%> (+0.11%)	⬆️
R/format.R	71.27% <100.00%> (+0.12%)	⬆️
R/print.parameters_model.R	60.36% <100.00%> (+0.48%)	⬆️

[vincentarelbundock]

ooooh, i'm excited about this!

[vincentarelbundock]

love it!

[strengejacke]

Not sure how to print Species [setosa (ref.)], tough. Better Species [setosa] (ref.) or something completely else? hm...

[vincentarelbundock]

Maybe the (ref.) is superfluous? The estimate is 0 and the std errors NULL, so it's quite evident what is happening, and removing "ref." would make things uniform in the term names.

If you think they're really necessary, I would put it outside the brackets to have consistent labels (and to make it easier for downstream packages like mine to remove "ref." if we don't want it using regexes.

Just a thought.

[strengejacke]

I have no better model at hand, here's an example for glm:

data(mtcars)
mtcars$cyl <- as.factor(mtcars$cyl)
mtcars$am <- as.factor(mtcars$am)
m <- glm(vs ~ cyl + gear + am, mtcars, family = binomial())
parameters::parameters(m)
#> Parameter   | Log-Odds |       SE |                 95% CI |         z |     p
#> ------------------------------------------------------------------------------
#> (Intercept) |   151.52 | 71190.45 | [-16813.63,  52679.23] |  2.13e-03 | 0.998
#> cyl [6]     |   -43.16 | 23084.31 | [-36096.45, -21647.68] | -1.87e-03 | 0.999
#> cyl [8]     |  -109.42 | 41823.10 | [         ,   2786.97] | -2.62e-03 | 0.998
#> gear        |   -21.78 | 13269.07 | [         ,   2712.57] | -1.64e-03 | 0.999
#> am [1]      |   -42.63 | 25814.34 | [         ,   3345.63] | -1.65e-03 | 0.999
#> 
#> Uncertainty intervals (profile-likelihood) and p-values (two-tailed)
#>   computed using a Wald z-distribution approximation.
#> 
#> The model has a log- or logit-link. Consider using `exponentiate =
#>   TRUE` to interpret coefficients as ratios.

print(parameters::parameters(m), add_reference = TRUE)
#> Parameter      | Log-Odds |       SE |                 95% CI |         z |     p
#> ---------------------------------------------------------------------------------
#> (Intercept)    |   151.52 | 71190.45 | [-16813.63,  52679.23] |  2.13e-03 | 0.998
#> cyl [4] (ref.) |     0.00 |          |                        |           |      
#> cyl [6]        |   -43.16 | 23084.31 | [-36096.45, -21647.68] | -1.87e-03 | 0.999
#> cyl [8]        |  -109.42 | 41823.10 | [         ,   2786.97] | -2.62e-03 | 0.998
#> gear           |   -21.78 | 13269.07 | [         ,   2712.57] | -1.64e-03 | 0.999
#> am [0] (ref.)  |     0.00 |          |                        |           |      
#> am [1]         |   -42.63 | 25814.34 | [         ,   3345.63] | -1.65e-03 | 0.999
#> 
#> Uncertainty intervals (profile-likelihood) and p-values (two-tailed)
#>   computed using a Wald z-distribution approximation.

print(parameters::parameters(m, exponentiate = TRUE), add_reference = TRUE)
#> Parameter      | Odds Ratio |       SE |           95% CI |         z |     p
#> -----------------------------------------------------------------------------
#> (Intercept)    |   6.37e+65 | 4.54e+70 | [0.00,      Inf] |  2.13e-03 | 0.998
#> cyl [4] (ref.) |       1.00 |          |                  |           |      
#> cyl [6]        |   1.80e-19 | 4.14e-15 | [0.00,     0.00] | -1.87e-03 | 0.999
#> cyl [8]        |   3.01e-48 | 1.26e-43 | [    ,      Inf] | -2.62e-03 | 0.998
#> gear           |   3.48e-10 | 4.62e-06 | [    ,      Inf] | -1.64e-03 | 0.999
#> am [0] (ref.)  |       1.00 |          |                  |           |      
#> am [1]         |   3.06e-19 | 7.91e-15 | [    ,      Inf] | -1.65e-03 | 0.999
#> 
#> Uncertainty intervals (profile-likelihood) and p-values (two-tailed)
#>   computed using a Wald z-distribution approximation.

^{Created on 2023-09-11 with reprex v2.0.2}

[strengejacke]

Maybe the (ref.) is superfluous?

Yes, it could be removed, indeed.

[strengejacke]

when dropping coefficients, or models with multiple components whould work now, too:

library(parameters)
data("fish")
m1 <- glmmTMB::glmmTMB(
  count ~ child + camper + zg + (1 | ID),
  ziformula = ~ child + camper + (1 | persons),
  data = fish,
  family = glmmTMB::truncated_poisson()
)
print(model_parameters(m1), add_reference = TRUE)
#> # Fixed Effects (Count Model)
#> 
#> Parameter         | Log-Mean |   SE |         95% CI |     z |      p
#> ---------------------------------------------------------------------
#> (Intercept)       |     1.41 | 0.18 | [ 1.06,  1.75] |  8.02 | < .001
#> child             |    -0.53 | 0.12 | [-0.77, -0.29] | -4.40 | < .001
#> camper [0] (ref.) |     0.00 |      |                |       |       
#> camper [1]        |     0.58 | 0.10 | [ 0.39,  0.78] |  5.93 | < .001
#> zg                |     0.13 | 0.04 | [ 0.05,  0.21] |  3.17 | 0.002 
#> 
#> # Fixed Effects (Zero-Inflation Component)
#> 
#> Parameter         | Log-Odds |   SE |         95% CI |     z |      p
#> ---------------------------------------------------------------------
#> (Intercept)       |    -0.39 | 0.65 | [-1.67,  0.89] | -0.60 | 0.551 
#> child             |     2.05 | 0.31 | [ 1.45,  2.66] |  6.63 | < .001
#> camper [0] (ref.) |     0.00 |      |                |       |       
#> camper [1]        |    -1.01 | 0.32 | [-1.64, -0.37] | -3.12 | 0.002 
#> 
#> # Random Effects Variances
#> 
#> Parameter          | Coefficient |       95% CI
#> -----------------------------------------------
#> SD (Intercept: ID) |        0.28 | [0.13, 0.60]
#> 
#> # Random Effects (Zero-Inflation Component)
#> 
#> Parameter               | Coefficient |       95% CI
#> ----------------------------------------------------
#> SD (Intercept: persons) |        1.17 | [0.54, 2.57]
#> 
#> Uncertainty intervals (equal-tailed) and p-values (two-tailed) computed
#>   using a Wald z-distribution approximation.
#> 
#> The model has a log- or logit-link. Consider using `exponentiate =
#>   TRUE` to interpret coefficients as ratios.

data(mtcars)
mtcars$gear <- as.factor(mtcars$gear)
m <- glm(vs ~ wt + gear, data = mtcars, family = "binomial")
print(model_parameters(m, exponentiate = TRUE, drop = "(Intercept)"), add_reference = TRUE)
#> Parameter       | Odds Ratio |   SE |        95% CI |     z |     p
#> -------------------------------------------------------------------
#> wt              |       0.07 | 0.09 | [0.00,  0.52] | -2.05 | 0.040
#> gear [3] (ref.) |       1.00 |      |               |       |      
#> gear [4]        |       3.21 | 3.98 | [0.27, 41.36] |  0.94 | 0.348
#> gear [5]        |       0.03 | 0.07 | [0.00,  1.47] | -1.41 | 0.159
#> 
#> Uncertainty intervals (profile-likelihood) and p-values (two-tailed)
#>   computed using a Wald z-distribution approximation.

^{Created on 2023-09-11 with reprex v2.0.2}

[strengejacke]

ok, I removed (ref.) from the output now.
I bet there are edge cases, and tests only seem to work interactively, so I'll merge and wouldn't promote that feature too much, but rather consider it as "experimental".

[strengejacke]

What's not yet properly working could be if the reference level is changed, e.g. to the last level (see screenshot in #715)