date math / off-by-one: previous day admissions versus current day admissions

[stephenturner]

From the flusight guidance:

Confirmed influenza hospital admission data are reported as “previous day” admissions in the raw data. Therefore, before aggregating to the weekly values, the gold standard or “truth” data will shift the values in the date column one day earlier so that the date aligns with the date of admission. As an example, if 17 confirmed influenza hospital admissions were reported in the previous_day_admission_influenza_confirmed field in a row where the date field was 2021-10-30, then the “truth” dataset would assign those 17 hospital admissions to a date of 2021-10-29. These cases would then be counted towards the weekly total computed for EW43, which runs from 2021-10-24 through 2021-10-30.

This should be considered when working on #9 , #10 , #12

[stephenturner]

There are lots of places throughout our code where we're doing something like mutate(date=date-1) %>% after pulling the hospitalization data from hdgov. Rather than try to handle this directly in #9 , #37 , etc., should we handle it directly in the get_hdgov_hosp function?

[vpnagraj]

good call. BUT lets be 101% sure that we fix all of those other places where we call date = date-1. may be good motivation to kill off orphan scratch scripts.

[stephenturner]

fixed in #47 - I'd go ahead and merge main into whatever you're working on. This fixes the function to shift days back one day, but also comments out all instances of mutate(date=date-1) %>% in scratch scripts.