ppi_mmmm {scorematchingad}

A PPI Score-Matching Marginal Moment Matching Estimator (dimension=3 only)

Description

Computes a marginal moment matching estimator (Section 6.2, Scealy and Wood 2023), which assumes \beta is a known vector with the same value in each element, and b_L = 0. Only A_L is estimated.

Usage

ppi_mmmm(Y, ni, beta0, w = rep(1, nrow(Y)))

Arguments

Y

Count data, each row is a multivariate observation.

ni

The total for each sample (sum across rows)

beta0

\beta=\beta_0 is the same for each component.

w

Weights for each observation. Useful for weighted estimation in Windham().

Details

\beta=\beta_0 is fixed and not estimated. b_L is fixed at zero. See (Section 6.2 and A.8 of Scealy and Wood 2023). The boundary weight function in the score matching discrepancy is the unthresholded product weight function

h(z)^2 = \min\left(\prod_{j=1}^{p} z_j^2, a_c^2\right).

Value

A vector of estimates for A_L entries (diagonal and off diagonal).

References

Scealy JL, Wood ATA (2023). “Score matching for compositional distributions.” Journal of the American Statistical Association, 118(543), 1811–1823. doi:10.1080/01621459.2021.2016422.