parameterise the model explicitly for the USA

[cwhittaker1000]

Currently, our data sources are quite UK focussed. Try and source data (for e.g. distribution of school sizes) for the USA so that all datasources being used to parameterise the model (esp. wrt locations) are derived from the US.

[athowes]

Specifically need to find US replacements for:

• Households (and therefore age distributions)
• Schools
• Workplaces: maybe*
• Leisure: maybe, as this currently I don't think is based on any data

*R.e. workplaces, in the report we have written that Fergusson et al. 2005 is based on US data, but is this the form of the distribution, and then the parameters are fit to Thai data? Ferusson does cite back to "Axtell, R. L. Zipf Distribution of U.S. Firm Sizes. Science 293, 1818-1820 (2001)." It's unclear to me exactly what was done based on the SI.

[athowes]

(In some sense I have doubts about how important this is to the results of the analysis, though I can imagine if someone were looking to question results them saying "oh but you didn't use US specific data" would be a very easy complaint to make and that it's worth preempting that.)

[cwhittaker1000]

Schools: https://nces.ed.gov/programs/digest/d21/tables/dt21_216.40.asp Workplaces: https://www.census.gov/data/tables/2019/econ/susb/2019-susb-annual.html see download "U.S. & states, NAICS, detailed employment sizes (U.S., 6-digit and states, NAICS sectors)" House/Age: I think we can use RTI SynthPop (https://www.rti.org/focus-area/rti-synthpoptm) - see here for more detail but basically they created a Synthetic US pop specifically for ABM stuff: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2875687/. I've emailed them to ask for access, but we can access what we need here I think: https://fred.publichealth.pitt.edu/syn_pops - I downloaded data for all of California and then picked a "people.txt" file at random and opening it up it had age and household ID which I think is all we need!

[athowes]

• Put raw data in data-raw
• Add code to make any cleaning modifcations in data-raw/DATASET.R
• Add documentation for data sets in R/data.R
• For schools, update generate_initial_schools_bootstrap to use new data
  • May want to generalise to take data as input as standardise possible data sources
• For workplaces, write new generate_initial_workplaces_bootstrap function (similar to the schools one)
  • Update it to take US workplaces input, follow same pattern as above
• For households, update generate_initial_households_bootstrap to use new data
• Stretch: find data to inform leisure settings

Based on chat with @cwhittaker1000:

• Add generate_initial_schools with arguments for type as either "empirical" or "synthetic" and country as either "UK" or "US"

[cwhittaker1000]

Household and age distributions now in via https://github.com/mrc-ide/helios/pull/97 - comments welcome!

[cwhittaker1000]

Note that I've started the USA workplaces update. A couple of things:

• The data I linked above ended up not being suitable as it has Enterprise Size but not Establishment size. Establishment (of which many can constitute an Enterprise) are what we need for the modelling.
• I've checked Ferguson et al 2006 (https://www.nature.com/articles/nature04795#Sec2) and the data they use is from the USA and looks pretty similar to the same data for 2023 (see here: https://data.bls.gov/cew/apps/table_maker/v4/table_maker.htm#type=13&year=2023&ind=10&supp=0) and compare: to January employment here:

I think given all this, we're fine to just use the parameters and sampling scheme from Ferguson et al. I'm going to update the structure to have "synthetic" vs "empirical" (where "synthetic" will be empty for now).

[cwhittaker1000]

Partially addressed by https://github.com/mrc-ide/helios/pull/98 and https://github.com/mrc-ide/helios/pull/97

[cwhittaker1000]

Fully closed by https://github.com/mrc-ide/helios/pull/99