estimate_expectation() for an interaction: predictions not provided for the full range of observed data

[sainsonjk]

Hi all,

I have used estimate_expectations to visualize the interaction between a factor (category) and a continuous variable (sr).

pred <- estimate_expectation(m, by = c("sr", "category"))
pred

Model-based Predictions

sr    | category | Predicted |   SE |           CI
--------------------------------------------------
-3.19 | 10-15    |      0.43 | 0.18 | [0.15, 0.76]
-2.57 | 10-15    |      0.49 | 0.16 | [0.22, 0.76]
-1.96 | 10-15    |      0.55 | 0.13 | [0.30, 0.77]
-1.34 | 10-15    |      0.60 | 0.10 | [0.40, 0.78]
-0.73 | 10-15    |      0.66 | 0.07 | [0.50, 0.79]
-0.12 | 10-15    |      0.71 | 0.05 | [0.60, 0.80]
0.50  | 10-15    |      0.76 | 0.04 | [0.67, 0.83]
1.11  | 10-15    |      0.80 | 0.04 | [0.71, 0.86]
1.73  | 10-15    |      0.83 | 0.04 | [0.73, 0.90]
-1.96 | 16-40    |      0.72 | 0.05 | [0.60, 0.81]
-1.34 | 16-40    |      0.73 | 0.04 | [0.64, 0.80]
-0.73 | 16-40    |      0.74 | 0.03 | [0.67, 0.80]
-0.12 | 16-40    |      0.75 | 0.03 | [0.69, 0.80]
0.50  | 16-40    |      0.76 | 0.03 | [0.70, 0.81]
1.11  | 16-40    |      0.77 | 0.03 | [0.70, 0.83]
1.73  | 16-40    |      0.78 | 0.04 | [0.69, 0.85]
-1.96 | 41-80    |      0.70 | 0.07 | [0.55, 0.82]
-1.34 | 41-80    |      0.68 | 0.06 | [0.56, 0.78]
-0.73 | 41-80    |      0.66 | 0.05 | [0.56, 0.75]
-0.12 | 41-80    |      0.64 | 0.05 | [0.54, 0.73]
0.50  | 41-80    |      0.62 | 0.06 | [0.50, 0.72]
1.11  | 41-80    |      0.60 | 0.07 | [0.45, 0.73]
1.73  | 41-80    |      0.57 | 0.09 | [0.39, 0.74]
-1.96 | 80+      |      0.75 | 0.06 | [0.62, 0.84]
-1.34 | 80+      |      0.73 | 0.05 | [0.63, 0.81]
-0.73 | 80+      |      0.72 | 0.04 | [0.63, 0.79]
-0.12 | 80+      |      0.70 | 0.04 | [0.62, 0.77]
0.50  | 80+      |      0.68 | 0.05 | [0.59, 0.76]
1.11  | 80+      |      0.66 | 0.06 | [0.54, 0.77]
1.73  | 80+      |      0.64 | 0.08 | [0.49, 0.78]

Variable predicted: cbind(count, tot - count)
Predictors modulated: sr, category
Predictors controlled: lat (-0.028)
Predictions are on the response-scale.

pred %>% plot()

I am confused why predicted values are not provided across the full range of observed values for sr within each factor level?

The categories 16-40, 41-80, and 80+ have observed sr values outside the predicted range (smaller than predicted), while for category 10-15 the range is extended:

data %>% group_by(category) %>% summarize(min = min(sr))
# A tibble: 5 × 2
  category   min
  <fct>       <dbl>
1 10-15       -2.21
2 16-40       -2.21
3 41-80       -2.21
4 80+         -3.19

With pred <- estimate_expectation(m, by = c("sr", "category"), preserve_range = FALSE), the response is predicted across the full global range of sr for all the category levels:

I would highly appreciate it if someone could explain why this happens! Thanks!