tests/sim/20211028-sim/README.md

In jlivsey/Dvar: What the Package Does (One Line, Title Case)

Major change for this simulation

1. Objective function is weighted by inverse of the noise scale parameter.
2. \epsilon budget choose such that a. higher weight in more aggregated margins a. ratio of largest to smallest is 4 times.

This simulation has:

$$ \epsilon = 5 $$

and $\epsilon$ schedule,

| | State | County | Tract | [Total] | |--------------------------|-------|--------|-------|---------| | detailed | | | | .2 | | Own/rent | | | | .25 | | VoteAge x Hisp x CenRace | | | | .25 | | Age x Sex | | | | .3 | | [Total] | .4 | .3 | .3 | 1 |

For the following set of simulations we move outward in a nested structure from the no margins case. This follows Table 2 in the write up, starting in the top-right with only detailed-tract.

Nested Structure:

1. (0002 margins) non = just detailed at tract level
2. (0086 margins) tr1 = just own/rent margins
3. (0172 margins) tr2 = add (votingAge x hisp x cenRace) = (1, 1, 1)
4. (0256 margins) tr3 = add (votingAge x hisp x cenRace) = (2, 1, 1)
5. (0430 margins) tr4 = add (votingAge x hisp x cenRace) = (., 2, 1)
6. (0774 margins) tr5 = add (votingAge x hisp x cenRace) = (., ., 2)
7. (1290 margins) tr5 = add (votingAge x hisp x cenRace) = (., ., 3:4)
8. (1548 margins) tr6 = add all (votingAge x hisp x cenRace) AND all age x sex
9. (1800 margins) cnt = add all county
10. (1872 margins) all = all margins

jlivsey/Dvar documentation built on July 13, 2024, 6:10 a.m.