Observation-level orthogonal PLQR scores at a supplied coefficient (experimental)

Computes, for each observation \(i\), the orthogonal PLQR score $$ \psi_{i} \;=\; (\tau - \mathbf{1}\{\,Y_{i} \le D_{i}^{\top}\theta + g_{0}(U_{i})\,\})\, (D_{i} - m_{D}(U_{i})) \, , $$ from supplied nuisance columns. No weights are applied and no localization is performed; plqr_dnn() aggregates these scores with DNN rank weights.

Usage

plqr_score_matrix(data, y_col, D_cols, g_col, mD_cols, theta, tau)

Arguments

data

A data.frame containing the outcome, regressor, and nuisance columns.

y_col

Name or index of the outcome column in data.

D_cols

Names or indices of partially linear regressor columns in data.

g_col

Name or index of the shape-nuisance column in data holding the values of \(g_0(U)\).

mD_cols

Names or indices of nuisance columns for m_D(U); must have the same length and order as D_cols.

theta

Numeric coefficient vector with length length(D_cols).

tau

Quantile level in (0,1).

Value

A numeric matrix of size \(n \times p_d\) of observation-level scores.

Lifecycle

Experimental. PLQR functionality is outside the CATE and regression workflow focus of the initial public release. Its API and inference interfaces may change in later releases.

See also

plqr_dnn(), plqr_demo_data().

Examples

demo <- plqr_demo_data(n = 100, p = 2, p_d = 2, tau = 0.5, seed = 1)
scores <- plqr_score_matrix(
  data = demo$data,
  y_col = "Y",
  D_cols = c("D1", "D2"),
  g_col = "g_oracle",
  mD_cols = c("mD1", "mD2"),
  theta = demo$truth$theta_true,
  tau = 0.5
)
head(scores)
#>               D1         D2
#> [1,]  0.12282055  0.2681021
#> [2,] -0.50666199  0.3141894
#> [3,]  0.47597653  0.5914012
#> [4,] -0.09927234 -0.1150896
#> [5,]  0.68557066 -0.4962436
#> [6,]  0.74929848  0.4536638