Increase resolution of household size

[virgesmith]

Request from Maria:

We had meeting this morning and we reviewed the job done. We have some thoughts about capturing the large families. By now, when there is more than 4 people - we don't know how many people there are.

Unless "4" is an analytical constrain for some reason, it would clever to refine the allocation algorithm - in order to include multi-generational or "joint" families (i.e. 5-6 people). This would mean to move the threshold from ">4" to ">6" (creating more categories), or to supply the actual number of people.

Would this be a problem for you?

[virgesmith]

My initial response:
The census tables we are using LC4404/LC4405 only contain the following data about the occupants:

"C_SIZHUK11": {
  "0": "All categories: Household size",
  "1": "1 person in household",
  "2": "2 people in household",
  "3": "3 people in household",
  "4": "4 or more people in household"

So the “cap” comes from the census (or at least our choice of table). The one other (2011) household size table available at OA resolution – QS406EW – which goes up to 8+. The problem here is this table’s only other category is “rural-urban” which means we have no “link” variables (e.g. to assign to the correct tenure)

However, what we could do is postprocess the data, taking all the 4+ occupants entries and and reassigning to 4,5,6,7,8+ according to the other table totals for the OA (if the totals match between the tables).

If this works we may even be able to apply this to rooms and bedrooms, which would give us a better estimate of persons-per-bedroom.