Feature request: sets of control variables

[masurp]

Feature request: sets of control variables

[masurp]

Dear Karl,

in the current version, you can add groups of control variables (and their combination) as follows:

library(specr)

results <- run_specs(
  df = dplyr::mutate(example_data, c3 = runif(dplyr::n())), 
  y = c("y1", "y2"), 
  x = c("x1", "x2"), 
  model = c("lm"), 
  controls = c("c1 + c2", "c1 + c3")  # note the different notation here
)
dplyr::distinct(results, controls)
#> # A tibble: 4 x 1
#>   controls         
#>   <chr>            
#> 1 c1 + c2 + c1 + c3
#> 2 c1 + c2          
#> 3 c1 + c3          
#> 4 no covariates  

As you can see, the function produces also the combination of the "combinations". You would need to remove this specification afterwards, if it doesn't make sense to you.

We have considered including all combinations of controls, but as you said that increases the amount of specifications considerably (and for a standard routine potentially unnecessarily). I'll have to think about this more and might add some more standardized procedures in the next version.

Hope this helps!

[karldw]

Hi @masurp,

Ah, that's very helpful, thank you!

Another tool that might be useful is purrr::cross_df and its .filter argument, which could be used in place of expand.grid in setup_specs to drop specifications before they're run.
(If that change sounds useful, I would be happy to submit a PR in a few weeks.)

[maswiebe]

results <- run_specs(
df = dplyr::mutate(example_data, c3 = runif(dplyr::n())),
y = c("y1", "y2"),
x = c("x1", "x2"),
model = c("lm"),
controls = c("c1 + c2", "c1 + c3") # note the different notation here
)
plot_specs(results)

One problem with this approach is that the graph doesn't recognize "c1+c2" or "c1+c3" as separate groups in the 'controls' section. Instead, they get lumped together as 'all covariates':

[masurp]

Hi,

run_specs() no longer puts them together in the newest development version. Please install the latest version, and it should print them all separately. Will push to CRAN soon.

Best,
Philipp

[maswiebe]

Great, that works!

[akarlinsky]

I just found out on my own that I can include "groups" using this notation and it's very useful. Can this be added to the guides on the website?