Add high-income households

[nikhilwoodruff]

It's likely that a restraint on the success of the reweighting routine is the lack of observations of high-income households. For example, our targets include income tax revenue from individuals with £1m-2m and £2m+ market income, but the FRS has no such individuals. One simple approach to fix this might be to:

• Duplicate the top-n percent (by market income) of FRS individuals (duplicating their households in the process)
• Replace income variables (common in both the FRS and SPI) with those from individuals at the same market income quantile in the SPI
• Re-run the reweighting procedure

This is similar to the official SPI adjustment for the FRS.

@MaxGhenis how does this sound?

[MaxGhenis]

Here's what we just discussed live:

• Prior to reweighting, select all records that would be in the SPI (pay tax)
• Duplicate these records, strip all income variables that would occur in the SPI, and halve the weights
• Apply synthimpute to impute income variables from SPI for the stripped records
• Reweight

This will help not only with top incomes but also other variables that are more accurate in the SPI like dividends.

cc @rickecon per our recent chat on this