Vignette of covariance adjustment with auxiliary data in a pair-matched study

[jwasserman2]

No description provided.

[jwasserman2]

@josherrickson thoughts on having added paireddata for this vignette as a dataset for the package?
@benthestatistician thoughts on what the vignette covers?

**Additionally, the site build keeps failing because we aren't storing the data necessary to run the vignette in the package. Should this go in /inst?

[benthestatistician]

Re vignette coverage, I much like what I'm seeing. I have some ideas about how the vignette could state its purposes to highlight what's special about propertee, but perhaps it would be more efficient for me to add and edit text myself sometime in the coming weeks than to start wordsmithing now. So for now I'll keep those ideas to myself.

I figured this vignette would offer a natural opportunity to show off the weights=ate(des)*measure_of_cluster_size construction, so was surprised not to see it here. I'd think we should; without measures of cluster size, we estimate the effect of the treatment on school level averages. But if we bring in cluster size as above, then we get to estimate student level effects (despite our being restricted to aggregated data).

I'll let Josh E chime in about the paireddata addition, but I'll comment that:

• I like having it but might suggest a name like "michigan_school_pairs"
• Would be good to say that the schools we're looking at here are the closest matches we found to the schools that had participated in an RCT in a neighboring county
• it would be worth clarifying that there was no known intervention was happening at the school pairs and triples, so the expected effects in our example are 0 (although the actual RCT it's modeled on did seem to have some effect).

[josherrickson]

No objection to adding more data, as long as it's small enough to not trigger CRAN's size warning. I believe it warns if any directory is greater than 1Mb, based upon this note on rlemon's checks:

 Result: NOTE
    installed size is  5.0Mb
    sub-directories of 1Mb or more:
      lemon   2.9Mb
      libs    1.9Mb

It wouldn't be a total stopper, but avoiding any notes should ease CRAN submissions.

[jwasserman2]

I like the name "michigan_school_pairs"; I'll re-name it. No issues about size--it's only half a kilabyte. In the vignette, I'll clarify we expect to find no effect given there was no intervention to our knowledge. I agree that you should make comments directly to the vignette, Ben.