That definitely changes the calculus :) As for payscale, I meant the locality pay tables: https://www.opm.gov/policy-data-oversight/pay-leave/salaries-wages/2

That definitely changes the calculus :) As for payscale, I meant the locality pay tables: https://www.opm.gov/policy-data-oversight/pay-leave/salaries-wages/2018/general-schedule/ https://www.opm.gov/policy-data-oversight/pay-leave/salaries... The federal government aims to have everyone at the same grade and step to make roughly the same (adjusted for cost of living). Sounds similar to your goal. To aid that they broke the USA into a group of metro areas. You could look at the locality tables, and if there's states that have multiple entries on the GS payscale, adjust accordingly in your algorithm. For example, Pennsylvania would have a low rate if you're in a small town like the capitol, a medium rate if you're in Pittsburgh (home to CMU) and a higher rate in the east near Philadelphia. Your model (IIRC) had one COL value for all of PA. Similarly, you could look to the US State Department's per diem table to find a list of international localities that should have their own calculation. Many organizations rely on it. When I traveled on business for a multinational, they used the State Department Per Diem rates to determine what is a reasonable amount of money per day to give: https://aoprals.state.gov/web920/per_diem_action.asp?MenuHide=1&CountryCode=0000 https://aoprals.state.gov/web920/per_diem_action.asp?MenuHid... While the per diem rate itself may or may not be useful, a list of cities that should have a unique value for COL is a useful piece of data in itself. To be honest, I'm surprised that nobody with input to the algorithm used either data source or a similar one - I don't know if say, Canada or the EU do their own calculations or leverage the US federal govt's. I know a few multinationals that use the federal guidelines for setting per diem costs rather than spend the effort calculating that in house. But the general idea - that a larger entity had to tackle this issue long ago and published their solution... I think it's a useful data source.