Skip to content
Prev 5466 / 5632 Next

[R-meta] testing for and visualizing correlation between dependent traits in bivariate meta-analysis

Hi Sigurd,
To add to Wolfgang's note on outliers, could you share more about how these
traits are measured and what the units of the effect sizes are? Just based
on descriptive plots of the effect estimates, they appear to be strongly
right-skewed:

new_data_transf <-
  new_data %>%
  mutate(
    vi = if_else(dep.trait == "lambda", v1i, v2i),
    sei = sqrt(vi),
    lnyi = log(yi),
    sei_ln = sei / yi
  )

ggplot( new_data_transf, aes(yi, fill = dep.trait)) + geom_density(alpha =
0.4) + facet_wrap(~ dep.trait)
ggplot(new_data_transf, aes(yi, sei)) + geom_point() + facet_wrap(~
dep.trait)

Log transformation leads to a distribution much closer to symmetric:

ggplot(new_data_transf, aes(yi, fill = dep.trait)) + geom_density(alpha =
0.4) + facet_wrap(~ dep.trait) + scale_x_continuous(transform = "log")
ggplot(new_data_transf, aes(lnyi, sei_ln)) + geom_point() + facet_wrap(~
dep.trait) + scale_x_continuous()

Best,
James

On Thu, Oct 31, 2024 at 7:46?AM Viechtbauer, Wolfgang (NP) via
R-sig-meta-analysis <r-sig-meta-analysis at r-project.org> wrote:

            

  
  
Message-ID: <CAFUVuJwfpy9Ox57yYvAV=-XqAhk+EHEztmkXb+pTWTAtFpV8Lg@mail.gmail.com>
In-Reply-To: <AS8PR08MB9193E9A68CB9BB7D0018A4BF8B552@AS8PR08MB9193.eurprd08.prod.outlook.com>