Package refinement (variable inputs, helpfile, table output)

[mcaceresb]

• [x] Allow multe to take any multi-valued treatment T and automatically take the lowest value as the control
  • [x] NB: This should already be the case; we should code a test to make sure.
• [x] Allow multe to take any multi-valued control W.
  • [x] NB: This should already be the case; we should code a few tests to make sure.
• [x] Add code to drop strata without all the treatment levels
  • [x] This should include a message: "G control strata without sufficient overlap dropped (N obs)."
• [x] Conform helpfile to Stata standards.
  • [x] Add a section on stored results (r() results and optional mata results)
  • [x] Add options section to expand on what each option does and how to use it.
  • [x] Add references section with note this is based on ongoing work by GPHK.
  • [x] Typically Stata packages explain what they do; can we copy language from one of the write-ups?
  • [x] NB: The typical order in help files syntax, options table, options details, examples, results, references.
• [x] Output to the style of that of typical Stata packages.
  • [x] Make it easy to switch between ATE, 1-at-a-time, common weights
  • [x] Make it easy to switch between normal SE and oracle SE
  • [x] NB: Standard table output is achieved through _coef_table

[mcaceresb]

TODOS:

• We should come back to vcov computation after being done with the other tasks here.
• Add a panel for the decomposition results.

EDIT: See #4.

[jerrayc]

TODOs (me):

• [x] fill in short and long descriptions in helpfile
• [x] Populate option descriptions
• [x] Read through helpfile stored results and results lists to make sure it makes sense
• [x] Add references (working papers for now)

[jerrayc]

I populated the helpfile (sorry I forgot to reference this issue in my commit, not sure how to change that now). Pushed to issue3_refine.

Questions to answer: 1) (For Mauricio) in practice, how does one call the mata functions decomposition.tauhat() and decomposition.lambda() without the user having to manually creating Wm first? If there is no way currently, could we save the Wm matrix as an object to make tauhat() and lambda() easier to call? 2) (For Peter) Are the decomposition SEs heteroskedasticity-robust? 3) (For Peter) How do we want to format the decomposition output? 4) (For Peter) What references other than this paper should we include in the helpfile, if any?

[jerrayc]

Also, should we do a unit test on the dependent variable, so we can say for sure what type of variable you can use as the dependent variable?

[mcaceresb]

Continued in #5

[mcaceresb]

In this issue we completed or made drafts of all the TODOs here and here. Items not done 100% were marked for follow-up here.

[peterdhull]

Questions to answer:

2. (For Peter) Are the decomposition SEs heteroskedasticity-robust?
3. (For Peter) How do we want to format the decomposition output?
4. (For Peter) What references other than this paper should we include in the helpfile, if any?

2 Yes

3 & 4 @jerrayc let's chat about these in our next meeting / after I've taken a look through the file.