test_predictions() for zero inflated model

[YyLu5]

Thank you so much for developing this useful package for various analyses. I'm working on zero inflated negative binomial model with glmmTMB, and I used predict_response() for both the zero inflated part and the count part. Now I need to perform pairwise compare between levels of my predictor, so I used test_predictions(), but when I use the prediction from either the zero part or count part, the function results are the same. I was expecting different results for zero part and count part. My example codes as below:

preds_count=predict_response(model,term)
preds_zero=predict_response(model,term, type = "zero_inflated")
test_predictions(preds_count)
test_predictions(preds_zero)

for the predict_response I can set a "type" parameter to indicate which part of the component I want to condition on, is there anything similar I can do for the test_predictions? If no extra parameter used, which part of the model used for pairwise comparison?

[strengejacke]

Thanks for reporting this issue. This was indeed unclear. I revised test_predictions(), and you should now be able to compute contrasts / comparisons for the different model components of zero-inflation-models.

This is how it looks like:

library(ggeffects)
data(Salamanders, package = "glmmTMB")
m <- glmmTMB::glmmTMB(
  count ~ mined + (1 | site),
  ziformula = ~mined,
  family = poisson(),
  data = Salamanders
)
# count-model
pr <- predict_response(m, "mined")
pr
#> # Predicted counts of count
#> 
#> mined | Predicted |     95% CI
#> ------------------------------
#> yes   |      1.09 | 0.69, 1.72
#> no    |      3.42 | 2.86, 4.09
#> 
#> Adjusted for:
#> * site = NA (population-level)

test_predictions(pr)
#> # Pairwise comparisons
#> 
#> mined  | Contrast |       95% CI |      p
#> -----------------------------------------
#> yes-no |    -2.39 | -3.18, -1.60 | < .001
#> 
#> Contrasts are presented as conditional means.

# full model (count and zero-inflation)
pr <- predict_response(m, "mined", type = "zero_inflated")
pr
#> # Predicted counts of count
#> 
#> mined | Predicted |     95% CI
#> ------------------------------
#> yes   |      0.26 | 0.12, 0.41
#> no    |      2.21 | 1.77, 2.64
#> 
#> Adjusted for:
#> * site = NA (population-level)

test_predictions(pr)
#> # Pairwise comparisons
#> 
#> mined  | Contrast |       95% CI |      p
#> -----------------------------------------
#> yes-no |    -1.99 | -2.41, -1.58 | < .001
#> 
#> Contrasts are presented as counts.

# zero-inflation-probabilities
pr <- predict_response(m, "mined", type = "zi_prob")
pr
#> # Predicted zero-inflation probabilities of count
#> 
#> mined | Predicted |     95% CI
#> ------------------------------
#> yes   |      0.76 | 0.66, 0.83
#> no    |      0.36 | 0.30, 0.41
#> 
#> Adjusted for:
#> * site = NA (population-level)

# no CIs for default engine
test_predictions(pr)
#> # Pairwise comparisons
#> 
#> mined  | Contrast | 95% CI | p
#> ------------------------------
#> yes-no |     0.40 |        |
#> 
#> Contrasts are presented as probabilities.

# use "ggeffects" for CI
test_predictions(pr, engine = "ggeffects")
#> # Pairwise comparisons
#> 
#> mined  | Contrast |     95% CI |      p
#> ---------------------------------------
#> yes-no |     0.40 | 0.30, 0.50 | < .001
#> 
#> Contrasts are presented as probabilities.

^{Created on 2024-06-04 with reprex v2.1.0}