Two parameteric weight functions for selection models were introduced in Iyengar and Greenhouse (1988):
$$w_1(x; \beta, q) = |x|^\beta / t(q, \alpha)$$
$$w_2(x; \gamma, q) = e^{-\gamma}$$
if \(|x| \le t(q, \alpha)\) and \(w_1(x; \beta, q) = w_2(x; \gamma, q) = 1\) otherwise. Here, \(t(q, \alpha)\) is the \(\alpha\)-quantile of a \(t\) distribution
with \(q\) degrees of freedom. The functions \(w_1\) and \(w_2\) are used to model the selection process that may be present
in a meta analysis, in a model where effect sizes are assumed to follow a \(t\) distribution. We have implemented estimation of the parameters in
this model in IyenGreenMLE
and plotting in IyenGreenWeight
.
The functions normalizeT
and IyenGreenLoglikT
are used in computation of ML estimators and not intended to be called by the user.
For an example how to use IyenGreenMLE
and IyenGreenWeight
we refer to the help file for DearBegg
.
normalizeT(s, theta, b, q, N, type = 1, alpha = 0.05)
IyenGreenLoglikT(para, t, q, N, type = 1)
IyenGreenMLE(t, q, N, type = 1, alpha = 0.05)
IyenGreenWeight(x, b, q, type = 1, alpha = 0.05)
See example in DearBegg
for details.
Quantile where normalizing integrand should be computed.
Vector containing effect size estimates of the meta analysis.
Parameter that governs shape of the weight function. Equals \(\beta\) for \(w_1\) and \(\gamma\) for \(w_2\).
Degrees of freedom in the denominator of \(w_1, w_2\). Must be a real number.
Number of observations in each trial.
Type of weight function in Iyengar & Greenhouse (1988). Either 1 (for \(w_1\)) or 2 (for \(w_2\)).
Quantile to be used in the denominator of \(w_1, w_2\).
Vector in \(R^2\) over which log-likelihood function is maximized.
Vector of real numbers, \(t\) test statistics.
Vector of real numbers where weight function should be computed at.
Kaspar Rufibach (maintainer), kaspar.rufibach@gmail.com,
http://www.kasparrufibach.ch
Note that these weight functions operate on the scale of \(t\) statistics, not \(p\)-values.
Iyengar, S. and Greenhouse, J.B. (1988). Selection models and the file drawer problem (including rejoinder). Statist. Sci., 3, 109--135.
For nonparametric estimation of weight functions see DearBegg
.
# For an illustration see the help file for the function DearBegg().
Run the code above in your browser using DataLab