Skip to contents

Compute Design Effect Ratios from a Sandwich Variance Object

Source: R/sandwich.R
compute_der.Rd

Returns the Design Effect Ratio (DER) for each fixed-effect parameter: $$\mathrm{DER}_p = V_{\mathrm{sand}}[p,p] / H_{\mathrm{obs}}^{-1}[p,p]$$

Usage

compute_der(sandwich)

Arguments

sandwich

An object of class "hbb_sandwich" from sandwich_variance().

Value

Named numeric vector of length \(D\).

Details

DER > 1 means the survey design inflates variance beyond what a naive (non-survey-corrected) analysis would produce. Expected values are typically 1–5 for stratified cluster designs (Kish DEFF ~ 3–4).

The following warnings are issued:

  • DER < 0.5: survey design appears to reduce variance substantially, which is unusual and may signal a problem.

  • DER > 20: very large design effect, possibly indicating model misspecification or extreme weight variability.

See also

sandwich_variance()

Other sandwich: compute_H_obs(), compute_J_cluster(), compute_score_matrix(), print.hbb_sandwich(), sandwich_variance()