"all" should not include group levels

[strengejacke]

@DominiqueMakowski What would you say how "all" should behave? Return fixed effects, random variances and group level estimates? (as it is now)

Or only return fixed effects and random variances, as for non-bayesian models, and return group level estimates only when requested?

At the moment, we have no option for brms for return fixed effects and random variances only. In parameters, we return both random effects variances and group level effects, and simply filter out group level estimates when group_level = FALSE. This is inefficient for large models.

Thus, I suggest that "all" returns fixed and ran vars only.

[strengejacke]

Suggested change in this PR:

library(insight)
model <- insight::download_model("brms_chocomini_1")

find_parameters(model, effects = "fixed")
#> Loading required namespace: rstan
#> $conditional
#> [1] "b_Intercept"
#> 
#> $delta
#> [1] "b_delta_Intercept"
#> 
#> $k
#> [1] "b_k_Intercept"
#> 
#> $phi
#> [1] "b_phi_Intercept"

find_parameters(model, effects = "all")
#> $conditional
#> [1] "b_Intercept"
#> 
#> $random
#> [1] "sd_Participant__Intercept"
#> 
#> $delta
#> [1] "b_delta_Intercept"
#> 
#> $k
#> [1] "b_k_Intercept"
#> 
#> $phi
#> [1] "b_phi_Intercept"
#> 
#> $delta_random
#> [1] "sd_Participant__delta_Intercept"
#> 
#> $k_random
#> [1] "sd_Participant__k_Intercept"

find_parameters(model, effects = "random_variance")
#> $random
#> [1] "sd_Participant__Intercept"
#> 
#> $delta_random
#> [1] "sd_Participant__delta_Intercept"
#> 
#> $k_random
#> [1] "sd_Participant__k_Intercept"

find_parameters(model, effects = "grouplevel")
#> $random
#>  [1] "r_Participant[S001,Intercept]" "r_Participant[S002,Intercept]"
#>  [3] "r_Participant[S003,Intercept]" "r_Participant[S004,Intercept]"
#>  [5] "r_Participant[S005,Intercept]" "r_Participant[S006,Intercept]"
#>  [7] "r_Participant[S007,Intercept]" "r_Participant[S008,Intercept]"
#>  [9] "r_Participant[S009,Intercept]" "r_Participant[S010,Intercept]"
#> [11] "r_Participant[S011,Intercept]" "r_Participant[S012,Intercept]"
#> [13] "r_Participant[S013,Intercept]" "r_Participant[S014,Intercept]"
#> [15] "r_Participant[S015,Intercept]" "r_Participant[S016,Intercept]"
#> [17] "r_Participant[S017,Intercept]" "r_Participant[S018,Intercept]"
#> [19] "r_Participant[S019,Intercept]" "r_Participant[S020,Intercept]"
#> 
#> $delta_random
#>  [1] "r_Participant__delta[S001,Intercept]"
#>  [2] "r_Participant__delta[S002,Intercept]"
#>  [3] "r_Participant__delta[S003,Intercept]"
#>  [4] "r_Participant__delta[S004,Intercept]"
#>  [5] "r_Participant__delta[S005,Intercept]"
#>  [6] "r_Participant__delta[S006,Intercept]"
#>  [7] "r_Participant__delta[S007,Intercept]"
#>  [8] "r_Participant__delta[S008,Intercept]"
#>  [9] "r_Participant__delta[S009,Intercept]"
#> [10] "r_Participant__delta[S010,Intercept]"
#> [11] "r_Participant__delta[S011,Intercept]"
#> [12] "r_Participant__delta[S012,Intercept]"
#> [13] "r_Participant__delta[S013,Intercept]"
#> [14] "r_Participant__delta[S014,Intercept]"
#> [15] "r_Participant__delta[S015,Intercept]"
#> [16] "r_Participant__delta[S016,Intercept]"
#> [17] "r_Participant__delta[S017,Intercept]"
#> [18] "r_Participant__delta[S018,Intercept]"
#> [19] "r_Participant__delta[S019,Intercept]"
#> [20] "r_Participant__delta[S020,Intercept]"
#> 
#> $k_random
#>  [1] "r_Participant__k[S001,Intercept]" "r_Participant__k[S002,Intercept]"
#>  [3] "r_Participant__k[S003,Intercept]" "r_Participant__k[S004,Intercept]"
#>  [5] "r_Participant__k[S005,Intercept]" "r_Participant__k[S006,Intercept]"
#>  [7] "r_Participant__k[S007,Intercept]" "r_Participant__k[S008,Intercept]"
#>  [9] "r_Participant__k[S009,Intercept]" "r_Participant__k[S010,Intercept]"
#> [11] "r_Participant__k[S011,Intercept]" "r_Participant__k[S012,Intercept]"
#> [13] "r_Participant__k[S013,Intercept]" "r_Participant__k[S014,Intercept]"
#> [15] "r_Participant__k[S015,Intercept]" "r_Participant__k[S016,Intercept]"
#> [17] "r_Participant__k[S017,Intercept]" "r_Participant__k[S018,Intercept]"
#> [19] "r_Participant__k[S019,Intercept]" "r_Participant__k[S020,Intercept]"

find_parameters(model, effects = "random")
#> $random
#>  [1] "r_Participant[S001,Intercept]" "r_Participant[S002,Intercept]"
#>  [3] "r_Participant[S003,Intercept]" "r_Participant[S004,Intercept]"
#>  [5] "r_Participant[S005,Intercept]" "r_Participant[S006,Intercept]"
#>  [7] "r_Participant[S007,Intercept]" "r_Participant[S008,Intercept]"
#>  [9] "r_Participant[S009,Intercept]" "r_Participant[S010,Intercept]"
#> [11] "r_Participant[S011,Intercept]" "r_Participant[S012,Intercept]"
#> [13] "r_Participant[S013,Intercept]" "r_Participant[S014,Intercept]"
#> [15] "r_Participant[S015,Intercept]" "r_Participant[S016,Intercept]"
#> [17] "r_Participant[S017,Intercept]" "r_Participant[S018,Intercept]"
#> [19] "r_Participant[S019,Intercept]" "r_Participant[S020,Intercept]"
#> [21] "sd_Participant__Intercept"    
#> 
#> $delta_random
#>  [1] "r_Participant__delta[S001,Intercept]"
#>  [2] "r_Participant__delta[S002,Intercept]"
#>  [3] "r_Participant__delta[S003,Intercept]"
#>  [4] "r_Participant__delta[S004,Intercept]"
#>  [5] "r_Participant__delta[S005,Intercept]"
#>  [6] "r_Participant__delta[S006,Intercept]"
#>  [7] "r_Participant__delta[S007,Intercept]"
#>  [8] "r_Participant__delta[S008,Intercept]"
#>  [9] "r_Participant__delta[S009,Intercept]"
#> [10] "r_Participant__delta[S010,Intercept]"
#> [11] "r_Participant__delta[S011,Intercept]"
#> [12] "r_Participant__delta[S012,Intercept]"
#> [13] "r_Participant__delta[S013,Intercept]"
#> [14] "r_Participant__delta[S014,Intercept]"
#> [15] "r_Participant__delta[S015,Intercept]"
#> [16] "r_Participant__delta[S016,Intercept]"
#> [17] "r_Participant__delta[S017,Intercept]"
#> [18] "r_Participant__delta[S018,Intercept]"
#> [19] "r_Participant__delta[S019,Intercept]"
#> [20] "r_Participant__delta[S020,Intercept]"
#> [21] "sd_Participant__delta_Intercept"     
#> 
#> $k_random
#>  [1] "r_Participant__k[S001,Intercept]" "r_Participant__k[S002,Intercept]"
#>  [3] "r_Participant__k[S003,Intercept]" "r_Participant__k[S004,Intercept]"
#>  [5] "r_Participant__k[S005,Intercept]" "r_Participant__k[S006,Intercept]"
#>  [7] "r_Participant__k[S007,Intercept]" "r_Participant__k[S008,Intercept]"
#>  [9] "r_Participant__k[S009,Intercept]" "r_Participant__k[S010,Intercept]"
#> [11] "r_Participant__k[S011,Intercept]" "r_Participant__k[S012,Intercept]"
#> [13] "r_Participant__k[S013,Intercept]" "r_Participant__k[S014,Intercept]"
#> [15] "r_Participant__k[S015,Intercept]" "r_Participant__k[S016,Intercept]"
#> [17] "r_Participant__k[S017,Intercept]" "r_Participant__k[S018,Intercept]"
#> [19] "r_Participant__k[S019,Intercept]" "r_Participant__k[S020,Intercept]"
#> [21] "sd_Participant__k_Intercept"

^{Created on 2025-04-27 with reprex v2.1.1}

[strengejacke]

See also new docs:

For models of class brmsfit, there are additional options:

• "fixed" returns fixed effects.
• "random_variance" return random effects parameters (variance and
  correlation components, e.g. those parameters that start with sd_ or
  cor_).
• "grouplevel" returns random effects group level estimates (BLUPs), i.e.
  those parameters that start with r_.
• "random" returns both "random_variance" and "grouplevel".
• "all" returns fixed effects and random effects variances.
• "full" returns all parameters.

[DominiqueMakowski]

Thus, I suggest that "all" returns fixed and ran vars only.

Fine by me yes

[strengejacke]

I think we can do the same optimization for stanreg/rstanarm, too. The as.data.frame() method also has a pars argument, to filter parameters.