This relates to the discussion in #381. There @mightyphil2000 suggests to correct for underdispersion (underlined by the given literature sources).
However, checking the implementations of ivw:
|
se_ivw_fe <- stats::coefficients(mod_ivw)[1,2] / max(mod_ivw$sigma, 1) |
|
se_ivw_re <- stats::coefficients(mod_ivw)[1,2] |
and MR egger:
|
se1_egger_fe <- stats::coefficients(mod_egger)[2,2] / max(mod_egger$sigma, 1) |
|
se1_egger_re <- stats::coefficients(mod_egger)[2,2] |
in the mr_rucker() function I could not find any underdispersion correction.
For the fixed effect IVW results this can lead to much smaller standard errors and pvalues then reported by the ivw_fe() implementation for the same IVs/ summary statistics (in some extreme cases 0 p-values).
What's the reason for the different implementation in mr_rucker() & mr()?
This relates to the discussion in #381. There @mightyphil2000 suggests to correct for underdispersion (underlined by the given literature sources).
However, checking the implementations of ivw:
TwoSampleMR/R/rucker.R
Line 141 in 6268c2c
TwoSampleMR/R/rucker.R
Line 153 in 6268c2c
and MR egger:
TwoSampleMR/R/rucker.R
Line 179 in 6268c2c
TwoSampleMR/R/rucker.R
Line 191 in 6268c2c
in the mr_rucker() function I could not find any underdispersion correction.
For the fixed effect IVW results this can lead to much smaller standard errors and pvalues then reported by the ivw_fe() implementation for the same IVs/ summary statistics (in some extreme cases 0 p-values).
What's the reason for the different implementation in mr_rucker() & mr()?