Struct SmoothTermLrInference 

pub struct SmoothTermLrInference {
    pub name: String,
    pub term_idx: usize,
    pub statistic_lr: f64,
    pub ref_df: f64,
    pub bartlett_factor: f64,
    pub bartlett_factor_conditional: Option<f64>,
    pub rho_variation_shift: Option<f64>,
    pub statistic_corrected: f64,
    pub p_value_uncorrected: f64,
    pub p_value_corrected: f64,
    pub material: bool,
    pub correction: SmoothLrCorrection,
}

The Bartlett-corrected per-term significance report for one penalized smooth term (#1063). Unlike the summary table’s Wood rank-truncated Wald statistic, this is a genuine likelihood-ratio statistic from a constrained refit (the smooth dropped), so the exact Lawley LR Bartlett factor corrects the right quantity.

Fields

name: String

Smooth-term name (matches the summary row).

term_idx: usize

Smooth-term index within resolvedspec.smooth_terms.

statistic_lr: f64

The uncorrected likelihood-ratio statistic W = 2(ℓ_full − ℓ_null), floored at zero (a non-negative LR by construction).

ref_df: f64

Reference degrees of freedom d (the Wood truncation tr(F)²/tr(F²) on the term’s influence block, falling back to the term EDF).

bartlett_factor: f64

Lawley LR Bartlett factor c = E[W]/d = 1 + Δε/d when computable, else 1.0 (no correction).

bartlett_factor_conditional: Option<f64>

Fixed-λ conditional factor c_cond = 1 + Δε(ρ̂)/d when the estimated-λ correction was applied. None means the applied factor was either the fixed-λ factor itself or no Lawley correction was available.

rho_variation_shift: Option<f64>

Increment in Lawley’s LR mean shift due solely to ρ̂ sampling variation, 0.5 * tr(H_Δε Cov(ρ̂)), when estimated-λ correction was applied.

statistic_corrected: f64

Bartlett-corrected statistic W* = W / c.

p_value_uncorrected: f64

Uncorrected p-value P(χ²_d > W).

p_value_corrected: f64

Corrected p-value P(χ²_d > W*); equals the uncorrected value when no correction was applied.

material: bool

Whether the second-order correction is material (#939 deliverable 4): the per-test diagnostic “is n too small for first-order inference here?”. true when a correction was applied and it moves the result by more than SMOOTH_LR_MATERIAL_THRESHOLD — measured as the larger of the relative Bartlett-factor distance from one |c − 1| and the relative p-value change |p* − p| / max(p, p*, ε). false when correction is SmoothLrCorrection::None (no correction was applied).

correction: SmoothLrCorrection

Which statistic the corrected p-value is built from.

Trait Implementations

impl Clone for SmoothTermLrInference

fn clone(&self) -> SmoothTermLrInference

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for SmoothTermLrInference

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.