smooth_estimates() returns random smooths for levels on in the by smooth

[StefanoMezzini]

Hi Gavin,

I have a dataset with control and treatment animals (C_* and T_*, respectively). Plotting the partial effect of s(doy, animal_year, by = group, bs = 'fs') returns the random smooths for all levels for both groups rather than only the levels within each group. This results in unnecessary, flat smooth estimates with the sample-level CIs. (alpha is 0.1 for all geom_ribbons.)

> unique(filter(partials, term == 's(doy,animal_year):groupControl')$animal_year)
 [1] C_100 2023 C_100 2024 C_102 2023 C_102 2024 C_106 2023 C_107 2023 C_108 2023
 [8] C_161 2023 C_161 2024 C_200 2024 T_151 2023 T_151 2024 T_152 2023 T_153 2023
[15] T_153 2024 T_155 2023 T_155 2024 T_158 2023 T_158 2024 T_159 2023 T_160 2023
[22] T_163 2023 T_163 2024 T_166 2023 T_166 2024 T_169 2023
26 Levels: C_100 2023 C_100 2024 C_102 2023 C_102 2024 C_106 2023 ... T_169 2023

Cheers,
Stefano

[gavinsimpson]

Thanks @StefanoMezzini; I'll take a look at what's going wrong here, most likely is the special internal method use for fs smooths isn't doing the required subsetting of the levels for the by smooth.