How to add covariates(pretreatments) to setup_profile()

[jiyanxu]

Hi Kris,

I continued to follow the tutorial after solving issue 2 and tested my data with

profile_vvo_3 <- setup_profile(model_lasso_3, treatments(exper_vvo_3), treatments(exper_vvo_3))
samples_vvo_3 <- list(
  real = outcomes(exper_vvo_3),
  fitted = outcomes(sample(model_lasso_3, profile = profile_vvo_3)),
  altered = outcomes(sample(altered_lasso_3, profile = profile_vvo_3))
) |>
  bind_rows(.id = "source") |>
  bind_cols(treatment = rep(treatments(exper_vvo_3)$Diet_Type, 3)) |>
  pivot_longer(direct$outcome) |>
  mutate(name = factor(name, levels = unique(direct$outcome)))

then I got an error:

 Error in h(simpleError(msg, call)) : 
 error in evaluating the argument 'object' in selecting a method for function 'outcomes': object 'Female' not found

I checked the cran doc It seemed I could not add pretreatment to the profile directly, also :

> profile_vvo_3.1 <- setup_profile(
   model_lasso_3, 
   treatments(exper_vvo_3), 
   treatments(exper_vvo_3), 
   pretreatments = pretreatments(exper_vvo_3)  # Explicitly include covariates like "Female"
 )
Error in setup_profile(model_lasso_3, treatments(exper_vvo_3), treatments(exper_vvo_3),  : 
  unused argument (pretreatments = pretreatments(exper_vvo_3))

Here I attached the traceback() info for the first error:

> traceback()
20: h(simpleError(msg, call))
19: .handleSimpleError(function (cond) 
    .Internal(C_tryCatchHelper(addr, 1L, cond)), "object 'Female' not found", 
        base::quote(eval(predvars, data, env)))
18: eval(predvars, data, env)
17: eval(predvars, data, env)
16: model.frame.default(Terms, newdata, na.action = na.action, xlev = object$xlevels)
15: model.frame(Terms, newdata, na.action = na.action, xlev = object$xlevels)
14: predict.lm(fits[[i]], newdata = newdata, ...)
13: predict(fits[[i]], newdata = newdata, ...)
12: predict(fits[[i]], newdata = newdata, ...)
11: x@mediation@sampler(x@mediation@estimates, newdata = bind_cols(pretreatment, 
        profile@t_mediator[[i]]), indices = i, ...)
10: .local(x, size, ...)
9: sample(model_lasso_3, profile = profile_vvo_3)
8: sample(model_lasso_3, profile = profile_vvo_3)
7: outcomes(sample(model_lasso_3, profile = profile_vvo_3))
6: list2(...)
5: bind_rows(list(real = outcomes(exper_vvo_3), fitted = outcomes(sample(model_lasso_3, 
       profile = profile_vvo_3)), altered = outcomes(sample(altered_lasso_3, 
       profile = profile_vvo_3))), .id = "source")
4: list2(...)
3: bind_cols(bind_rows(list(real = outcomes(exper_vvo_3), fitted = outcomes(sample(model_lasso_3, 
       profile = profile_vvo_3)), altered = outcomes(sample(altered_lasso_3, 
       profile = profile_vvo_3))), .id = "source"), treatment = rep(treatments(exper_vvo_3)$Diet_Type, 
       3))
2: pivot_longer(bind_cols(bind_rows(list(real = outcomes(exper_vvo_3), 
       fitted = outcomes(sample(model_lasso_3, profile = profile_vvo_3)), 
       altered = outcomes(sample(altered_lasso_3, profile = profile_vvo_3))), 
       .id = "source"), treatment = rep(treatments(exper_vvo_3)$Diet_Type, 
       3)), direct$outcome)
1: mutate(pivot_longer(bind_cols(bind_rows(list(real = outcomes(exper_vvo_3), 
       fitted = outcomes(sample(model_lasso_3, profile = profile_vvo_3)), 
       altered = outcomes(sample(altered_lasso_3, profile = profile_vvo_3))), 
       .id = "source"), treatment = rep(treatments(exper_vvo_3)$Diet_Type, 
       3)), direct$outcome), name = factor(name, levels = unique(direct$outcome)))

Looking forward to hearing from you.

Yours sincerely,
Jiyan

[krisrs1128]

Thank you for sharing the informative issue and traceback, and I'm sorry for the error. Could you try passing a pretreatment argument to sample function? e.g.,

samples_vvo_3 <- list(
  real = outcomes(exper_vvo_3),
  fitted = outcomes(sample(model_lasso_3, profile = profile_vvo_3, pretreatment = pretreatments(exper_vvo_3))),
  altered = outcomes(sample(altered_lasso_3, profile = profile_vvo_3, pretreatment = pretreatments(exper_vvo_3)))
) |>
  bind_rows(.id = "source") |>
  bind_cols(treatment = rep(treatments(exper_vvo_3)$Diet_Type, 3)) |>
  pivot_longer(direct$outcome) |>
  mutate(name = factor(name, levels = unique(direct$outcome)))

I realize that this isn't documented anywhere well, but when we sample an object with pretreatment variables, it needs to be passed in a separate argument. The profile only has information about the counterfactual treatment statuses, so if the model expects pretreatment information, it needs to be passed in separately.

Please let me know if you continue to see issues, and thank you again for trying out the package.

[jiyanxu]

Dear Kris,

thank you very much for the hint, it works! That's wonderful.

Btw, I made a slight modification to my script (direct should be the df ). It might be obvious for many people, but just for someone's reference:

samples_vvo_3 <- list(
  real = outcomes(exper_vvo_3),
  fitted = outcomes(sample(model_lasso_3, profile = profile_vvo_3, pretreatment = pretreatments(exper_vvo_3))),
  altered = outcomes(sample(altered_lasso_3, profile = profile_vvo_3, pretreatment = pretreatments(exper_vvo_3)))
) |>
  bind_rows(.id = "source") |>
  bind_cols(treatment = rep(treatments(exper_vvo_3)$Diet_Type, 3)) |>
  pivot_longer(direct_3$outcome) |>
  mutate(name = factor(name, levels = unique(direct_3$outcome)))

All the best.

Yours sincerely,
Jiyan