"Apparent" missingness in GHT

[huisaddison]

Actual Behavior:

"Apparent" missingness: for most days, what appears to be missing data is actually a 0 in the GHT indicator. (There is the meta-issue regarding whether a 0 should be interpreted as a signal that indicates a low, but nonzero volume of searchers, or whether it should be interpreted as actual missing data). The map renders an exact zero as a solid gray (rather than a diagonally hatched gray, which indicates true missingness as a matter of fact about what's in our API). To me, the pipeline is working properly, and the discussion should be whether anosmia and i' can't smell or taste now has low recall and therefore we should adjust the terms in our bag.

Expected behavior

Should we change our bag of terms to something with higher recall now that it's no longer March, and people's search habits have changed?

Context

https://delphi-org.slack.com/archives/C0130CSQRN3/p1601495049023500?thread_ts=1601480482.022300&cid=C0130CSQRN3

cc @ryantibs who pointed this out originally
cc @nmdefries who is working on Google Symptoms, which may replace GHT
cc @jingjtang who is the current pipeline "maintainer" to my knowledge

[jingjtang]

The GHT pipeline has been automated starting from Aug 11.

[huisaddison]

Said something relevant to a different issue - sorry!

[huisaddison]

@jingjtang thanks! I only tagged you here to keep you in the loop (not sure if there are any official maintainers after a pipeline is automated). Your insight would be useful for #293, though

[nmdefries]

the discussion should be whether anosmia and i can't smell or taste now has low recall and therefore we should adjust the terms in our bag.

The strength of the GHT signal (compared to baseline) does appear to be decaying relative to Google Symptoms, more info here.

So if we intend to continue using the GHT indicator, we should adjust the search terms used, keeping in mind that this is something that would probably need to be repeated periodically. How were the initial set of search terms chosen?

Via Katie: we've already decided not to alter bag of search terms, and incoming GS indicator should be more robust to changes in user search term selection.

[capnrefsmmat]

Some useful context is in #138.

[nmdefries]

GHT has been deprecated (on the Google side) and replaced by Google Symptoms, which we expect to be more robust to shifts in specific search terms used.