Sumstats() Settings and Effect-Size Conversion

Are the latent variable “effects” (the est column in latent trait sumstats) always converted to continuous beta effects after GenomicSEM after using uswerGWAS () ?

Replying to “Are the latent variable “effects” (the est column …”:

Yes. This happens in the sumstats() step - it rescales effect sizes for all your traits to covariances.

Replying to “Are the latent variable “effects” (the est column …”:

Are they converted to continuous betas even if you have a binary trait? What equation or conversion is the sumstats doing?

Replying to “Are the latent variable “effects” (the est column …”:

Yes even when you have a binary trait. There are a number of characteristics of the input trait that will change how the effects size is transformed (e.g. ols or logistic; scale of standard errors, linprob model etc.) - check out 4. Common Factor GWAS · GenomicSEM/GenomicSEM Wiki · GitHub

Replying to “Are the latent variable “effects” (the est column …”:

In the methods section of the original Genomic SEM paper they describe the specific transformations done under the different scenarios

With the decision tree for sumstats, If you have a meta-analysis for PTSD that was measured as a continuous and discrete GWAS outcome and the sumstats have Z stats only, do you still pick the OLS = True.

Replying to “With the decision tree for sumstats, If you have a…”:

Yes, typically if the sumstats are a meta-analysis of both binary and continuous, then OLS = T, se.logit = F, linprob = F

Replying to “With the decision tree for sumstats, If you have a…”:

Thanks very much for confirming!