rstudent.deconv: Regression Deletion Diagnostics

In cellGeometry: Geometric Single Cell Deconvolution

Regression Deletion Diagnostics

Description

Functions for computing regression diagnostics including standardised or Studentized residuals as well as Cook's distance.

Usage

## S3 method for class 'deconv'
rstudent(model, ...)

## S3 method for class 'deconv'
rstandard(model, ...)

## S3 method for class 'deconv'
cooks.distance(model, ...)

Arguments

model	'deconv' class object
...	retained for class compatibility

Details

Residuals are first adjusted for gene weights (if used). rstandard and rstudent give standardized and Studentized residuals respectively. Standardised residuals are calculated based on the hat matrix:

H = X (X^T X)^{-1} X^T

Leverage h_{ii} = diag(H) is used to standardise the residuals:

t_i = \cfrac{\hat{\varepsilon_i}}{\hat{\sigma} \sqrt{1 - h_{ii}}}

Studentized residuals are calculated based on excluding the i th case. Note this corresponds to refitting the regression, but without recomputing the non-negative compensation matrix. Cook's distance is calculated as:

D_i = \cfrac{e_i^2}{ps^2} \left[\cfrac{h_{ii}}{(1 - h_{ii})^2} \right]

where p is the number of predictors (cell subclasses) and s^2 is the mean squared error. In this model the intercept is not included.

Value

Matrix of adjusted residuals or Cook's distance.

See Also

stats::influence.measures()

cellGeometry documentation built on April 20, 2026, 1:06 a.m.