make_exposure_prob: Create exposure probabilities.

Description Usage Arguments Details Value References Examples

Description

Create matrices of exposure probabilities.

Usage

1
2
3
4
5
6
make_exposure_prob(
  potential_tr_vector,
  adj_matrix,
  exposure_map_fn,
  exposure_map_fn_add_args = NULL
)

Arguments

potential_tr_vector

an R * N matrix of 0, 1 entries such as that produced by make_tr_vec_permutation, or an R * N matrix containing R permuted treatment assignments for N units.

adj_matrix

an N * N numeric matrix of 0, 1 entries such as those retuned by make_adj_matrix, or a N * N matrix, where N is the number of units.

exposure_map_fn

function which returns the exposure mapping such as make_exposure_map_AS.

exposure_map_fn_add_args

list of additional arguments which are passed to exposure_map_fn. adj_matrix and tr_vector should not be specified here because they are already passed with values derived from the arguments to make_exposure_prob. If using make_exposure_map_AS, hop must be specified here.

Details

make_exposure_prob produces the units' probabilities of being subject to each of the possible exposure conditions which are used for estimating exposure-specific causal effects, and the joint exposure probabilities used for variance estimators.

Value

A list of 3 lists:

I_exposure:

A list of K N * R numeric matrices of indicators for whether units N are in exposure condition k over each of the possible R treatment assignment vectors. The number of numeric matrices K corresponds to the number of exposure conditions.

prob_exposure_k_k:

A list of K symmetric N * N numeric matrices each containing individual exposure probabilities to condition k on the diagonal, and joint exposure probabilities to condition k on the off-diagonals.

prob_exposure_k_l:

A list of permutation(K,2) nonsymmetric N* N numeric matrices each containing joint probabilities across exposure conditions k and l on the off-diagonal, and zeroes on the diagonal. When K = 4, the number of numeric matrices is 12; permutation(4,2).

References

Aronow, P.M. & Samii, C. (2017). Estimating average causal effects under general interference, with application to a social network experiment. The Annals of Applied Statistics, 11(4), 1912–1947.

Aronow, P.M. et al. (2020). Spillover effects in experimental data. arXiv preprint, arXiv:2001.05444.

Examples

1
2
3
potential_tr_vector <- make_tr_vec_permutation(N = 9, p = 0.2, R = 20, seed = 357)
adj_matrix <- make_adj_matrix(N = 9, model = 'sq_lattice')
make_exposure_prob(potential_tr_vector, adj_matrix, make_exposure_map_AS, list(hop=1))

szonszein/interference documentation built on Jan. 10, 2022, 6:35 p.m.