Release 7-day average signals for Safegraph

[krivard]

• Link to issue
• Link to PR
• Proposed release version: 1.11

This work creates 7-day average signals for each of the following:

• median_home_dwell_time
• completely_home_prop
• full_time_work_prop
• part_time_work_prop

• write API documentation and/or changelog
• write API mailing list notification

• Statistical review (usually correlations)
• Signal / source name review (usually Roni)

• Visual review
• write Signal description pop-up text & review
• write Map release notes (I think we're not doing these by hand any more?)

[benjaminysmith]

Hi @sgsmob, take a look at #367 for an example of what we need for the initial steps.

[sgsmob]

Documentation update PR here: cmu-delphi/delphi-epidata#275

[sgsmob]

Draft mailing list notification:

Hi COVIDcast API users,

We are pleased to announce the availability of smoothed versions of existing signals
based on SafeGraph mobility data.

Each of the following signals previously available now has a corresponding additional 7-day trailing average signal, i.e., the reported value for May 10 is the average of values from May 4 - May 10:

1. Proportion of people at home for the whole day
2. Proportion of people working full time outside the home
3. Proportion of people working part time outside the home
4. Median number of hours spent at home

The signal names and technical details for each of these signals
are provided in our API documentation:
https://cmu-delphi.github.io/delphi-epidata/api/covidcast-signals/safegraph.html

[sgsmob]

Correlation plots for the smoothed signals look alright to me, although somewhat concerning that they are becoming less indicative recently.

[sgsmob]

Compared to the existing raw signals, the averaged signals look more useful/interpretable:

[krivard]

@sgsmob the additional unsmoothed correlations plot is helpful here, since it shows the same degradation in correlations over the past four weeks -- thanks for including it.

The decline is worrisome, and it's possible we might want to pause or even retire these signals (both raw and 7dav). I've put it on the agenda for Thursday's team leads.

[capnrefsmmat]

If I had to make a wild guess, I'd say the decline in correlation is because cases are now exponentially growing everywhere, and slightly lower mobility is no longer enough to prevent spread. But that's a wild guess.

I think the signals still are useful for research even if they don't correlate with case rates; for example, researchers studying the effects of social distancing orders, compliance with them, the economy, or mental health could all find these signals useful for their work.

[benjaminysmith]

Result of team leads was that we are going to keep the signals around, so this is unblocked.

[benjaminysmith]

@krivard for visual review it looks like the averages match the raw data. Seems reasonable to me. What do you think?

@sgsmob are the signal names in https://docs.google.com/document/d/10hGd4Evce4lJ4VkWaQEKFQxvmw2P4xyYGtIAWF52Sf8/edit#heading=h.vu3vsc11yl09 correct? If so, lets ask for Roni approval.