Fix cohen d for within-subjects (paired) pretest-posttest measures about effsize HOT 3 CLOSED

Hi,

I am not 100% certain of what I say below so correct me if you think I am wrong.

According to Howell (link below), it is not possible to calculate 95% confidence intervals around Cohen's d by directly subtracting and adding a constant to d (as it is the case in the effsize cohen.d function, the constant in the source code is called Z = -qt((1 - conf.level)/2, df)).
It makes sense when taking account of the non-central and non-symmetrical distributions of the lower and upper distributions of the non-centrality parameter (ncp) around an observed t-value.

Knowing that, we programmed a little function that is seeking for the ncp that reduces probability at lower and upper boundaries of confidence for Cohen's d (based on a single differential variable, for instance mean after treatment - mean at baseline) using a "fsolve" function.

For independant measures, the idea is the same but the computation of the boundaries is defined by a slight different calculation (see Howell 2010).

Let me know if you have any questions on this method.

Ladislas

https://www.uvm.edu/~dhowell/methods7/Supplements/Confidence%20Intervals%20on%20Effect%20Size.pdf

from effsize.

I read the document, and added the computation in the cohen.d function.

from effsize.

Available in version 0.6.o

from effsize.