Generate several arbitrary treatment priorities

[JamesPHoughton]

study 1

• [ ] dyads
• [ ] one republican, one democrat
• [ ] uniformly sample over age and gender, balanced so that we get an even number of republicans at each age/gender combo, etc.
• [ ] assign everyone who comes, with no leftovers if possible
• [ ] a group of 3 as a backup to account for odd numbers is preferable to having leftovers

study 2

• [ ] groups of 3
• [ ] equal number of times when women are in the minority as there are times when men are in the minority

study 3

• [ ] group sizes to vary between 2 and 5 so that I have a uniform distribution of groups across the size dimension

study 4

• [ ] dyads
• [ ] (evenly sample the space for m/f group with male age as one axis and f age as the other) + (evenly sample over the space of f/f groups where younger persons age is one axis and older persons age is the other) + (same for m/m groups)

[JamesPHoughton]

Deciding that having mixed continuous and discrete dimensions is too difficult for this thesis - simplify to only continuous dimensions, so that we can use gaussian kde and a prioritization function that is continuous.

Study 1:

• [ ] dyads
• [ ] sample so that we get a uniform distribution over space that has {dim1: age of younger partipant, dim2: age of older participant} in the triangle where that is valid.

We'll come back and do more of this next week.

[Alan-Qiao]

We noticed that since our goal is to suggest interventions for all scenarios in the scenario space, we value rare cases equally as common cases. As a result, this seems to suggest that we would need to place higher sampling priority on rarer cases as its less likely to get data on those cases if we omit any such potential grouping.

We want to thus consider two scoring functions:

1. Integrate the linear difference between the two surfaces.
2. Integrate the quadratic difference between the two surfaces.