Calculate Wellbeing Statistics by subgroups for UC analysis outputs

[andrewbaxter439]

Hi @amorison - building on the improvements in #503 there's a few other key things that the health stats should show which should be simple to incorporate.

Alongside gender subgroups similar subgrouping for age groups, household structures, education. Had set the age groups and household structures previously as:

    List<AgeRange> ageGroups = Arrays.asList(
            new AgeRange(16, 17),
            new AgeRange(18, 24),
            new AgeRange(25, 34),
            new AgeRange(35, 49),
            new AgeRange(50, 64),
            new AgeRange(65, 130)
    );

    List<HouseholdStructure> householdStructures = Arrays.asList(
            new HouseholdStructure(true, true, Gender.Male),
            new HouseholdStructure(true, false, Gender.Male),
            new HouseholdStructure(false, true, Gender.Male),
            new HouseholdStructure(false, false, Gender.Male),
            new HouseholdStructure(true, true, Gender.Female),
            new HouseholdStructure(true, false, Gender.Female),
            new HouseholdStructure(false, true, Gender.Female),
            new HouseholdStructure(false, false, Gender.Female)
    );

Does this now need a new file and new class for WellbeingByHousehold etc.? In previous experimental code I had opted for one Statistics class with an update(filter) method override to get the different splits, e.g.:

 public void update(SimPathsModel model, String gender_s, SimPathsCollector.AgeRange ageRange) {

Would best approach now be to create new files or to change WellbeingByGender to allow other splits?

[amorison]

Hi @andrewbaxter439 I assume this is about your feature/uc-effects-mental-health branch.

I'm not sure what you mean with WellbeingByHousehold or WellbeingByGender. The way I see it, those AgeRange and HouseholdStructure are Predicates in disguise that can be built from the Filters brought in with #503.

As it is, HealthStatistics has two concerns (which is one too many). It is responsible for computing a set of interesting statistics on a set of persons, but also how to get that set of persons. As a result, it needs to expose a state that you mutate and examine (hoping its internal state has been updated properly). That's not great for composability (and for correctness). You could have a HealthStatistics object that takes a List<Person> on construction (possibly hidden behind a Supplier depending on caching considerations) and gives you back a set of interesting Stats for that list (again, possibly hidden behind a Supplier). HealthStatistics shouldn't need to care about gender, agegroup or HouseholdStructure, all it needs to do its job is a list of agents. How that list of agents is built is an orthogonal concern (and how to label it is yet another one).

I'm happy to setup a call to discuss this if this is more convenient.

[andrewbaxter439]

Thanks for this Adrien, can chat either later today or Monday if we miss each other.

The crucial thing I need out of this is one Wellbeing statistics sheet per subgroup breakdown, so estimating mean/median etc. for health statistics for male/female/total, for agegroups, for household structures, for education values. For example, here's stats we can extract from the previous branch:

Ideally this needs to be working for next week as we'll need to do the runs Thursday/Friday and get the stats ready for presentation by Monday 7th. Is there a manageable fix for this?