[FEATURE] Add Household Characteristic of Number of Workers

[Eric-Liu-SANDAG]

Is your feature request related to a problem? Please describe

As we seek to align Estimates Program output with the launch year of our Cohort Component Model and regional forecast (see SANDAG/Cohort-Component-Model#140), we require additional household characteristics to be added to the Estimates Program. Specifically, we need number of workers (0,1,2,3+) to be added as a household characteristic.

Describe the solution you'd like

Following the conventions of the household characteristics module, add number of workers (0,1,2,3+) ensuring that in addition to using census tract level distributions and possibly regional control totals, that the module does not generate unreliable results at the MGRA level. For example, the implied number of workers in the MGRA should not exceed the number of work-eligible persons nor should it disagree with the households by size characteristics for the MGRA.

[GregorSchroeder]

Follow a similar pattern to _create_hh_size() in the python/hh_characteristics.py module,

1. Set n_people_in_3_plus = 4. This hardcoded number is used for the average number of workers present in a household with 3+ workers. This value was arrived at following a similar methodology as [FEATURE] Average maximum size of households in the 7+ category #112 using the SQL query below

SELECT 
    [bucket] * 5 AS [percentile],
    (MAX([workers]) + MIN([workers])) / 2 AS [value]
FROM (
        -- Workers are defined as >=16 in the labor force
        -- see [ESR] and [COW] documentation
        SELECT [workers], NTILE(20) OVER (ORDER BY [workers]) AS [bucket]
        FROM (
            SELECT [hh].[SERIALNO], COUNT([ESR]) AS [workers]
            FROM [acs].[pums].[vi_5y_2020_2024_households_sd] AS [hh]
            INNER JOIN [acs].[pums].[vi_5y_2020_2024_persons_sd] AS [persons]
                ON [hh].[SERIALNO] = [persons].[SERIALNO]
            WHERE [ESR] IN (1,2,3,4,5)
            GROUP BY [hh].[SERIALNO]
            HAVING COUNT([ESR]) >= 3
        ) AS [hh]
    ) AS [bucket]
GROUP BY [bucket]
ORDER BY [bucket]

2. Grab the total number of households in each MGRA with the census tract distribution of households by number of workers (0,1,2,3+) from the 5-year ACS table B08202 and perform MGRA-level two-dimensional controlling.

3. For the next step of controlling, compute the maximum implied worker population and compare to the actual household population that is 16+ in each MGRA. The 16+ definition is derived from the ACS PUMS definitions of [COW] and [ESR]. While not every person 16+ is in the labor force it serves as a realistic maximum value that cannot be exceeded. Adjust the number of households in worker categories such that the maximum is never exceeded in each MGRA following a similar methodology to the adjust_mgra() function in _create_hh_size().

@Eric-Liu-SANDAG I think this is a strong methodology that avoids inconsistencies with the adult household population but it does not explicitly avoid potential inconsistencies with household by size distributions. For example, we could have no households of size 3 or greater in a MGRA but currently this methodology does nothing to adjust if it assigns 3+ workers in the same MGRA. Is this another control check we should add?

[Eric-Liu-SANDAG]

Agree with the methodology and with validating households by workers against households by size. UDM does this as well based on the order in which household characteristics are created, i.e. size --> children --> workers