altmeta (version 4.4)

pb.timelag: Flexible Nonparametric Assessment of Time-Lag Bias in Meta-Analysis

Description

Performs permutation-calibrated fixed-gamma tests and an adaptive hybrid test for assessing time-lag bias in meta-analysis introduced in Xing et al. (2026).

Usage

pb.timelag(y, v, time, B = 5000, tau2 = NULL)

Value

A list containing the following components:

p.value

the permutation p-values for T1 through T8, TInf, and the hybrid test.

statistic

the observed values of T1 through T8 and TInf.

tau2

the supplied or estimated between-study variance.

B

the number of permutation iterations.

Arguments

y

a numeric vector specifying the observed effect sizes in the collected studies.

v

a numeric vector specifying the corresponding within-study variances.

time

a numeric vector specifying the publication times of the studies.

B

a positive integer specifying the number of permutation iterations. The default is 5000.

tau2

an optional non-negative value specifying the between-study variance. If NULL, it is estimated using the restricted maximum-likelihood method.

Author

Zhiyuan Yu, Xing Xing, Lifeng Lin

Details

The studies are first ordered by publication time. For each position in the ordered sequence, the function constructs a standardized cumulative contrast between the corresponding study and all later studies. The resulting contrasts are combined using powers 1 through 8 and the maximum statistic.

The null distributions of the fixed-gamma statistics are obtained by jointly permuting each study's effect size and within-study variance with respect to publication order. The hybrid test uses the minimum fixed-gamma permutation p-value and an additional permutation calibration to account for the adaptive selection across candidate statistics.

When multiple studies have the same value of time, their ordering is determined by their original order in the supplied vectors.

References

Xing X, Yu Z, Xiao M, Tong J, Lin L (2026). "Flexible nonparametric assessment of time-lag bias in meta-analysis." Biostatistics, in press.

See Also

pb.bayesian.binary, pb.hybrid.binary, pb.hybrid.generic

Examples

data("dat.sherrington")

set.seed(100)
out <- pb.timelag(
  y = dat.sherrington$yi,
  v = dat.sherrington$vi,
  time = dat.sherrington$year,
  B = 100
)
out$p.value