Hello,

I was wondering if p-values for estimated genetic correlations can be obtained directly from the model output, in addition to confidence intervals?

I know that a genetic correlation is significant when zero is not included in the CI, but I need a p-value for multiple comparison correction as I will be running many models.

Any help would be much appreciated,

Marc

Hi Marc,

The likelihood based confidence interval is used because OpenMx does not assume a particular distribution for the estimated parameters. However, we do provide estimates and standard errors of those estimates. The estimate could be considered the mean of the distribution of that parameter value, and similarly the standard error can be thought of as the standard deviation of that distribution. The p-value can then be obtained by assuming a distribution for the parameter.

As an example, a parameter (est) could be normally distributed with standard error (sde).

If the estimate is a correlation, it will not be normally distributed, but Fisher-Z transforming it will make it pretty close to normally distributed. Of course, the standard error would need to be transformed as well.