How to calculate the effect size when conducting a discriminant analysis?

While classification accuracy seems to provide some insights about the effect size, as classification involves many parameters (i.e. classifier, cross-validation, averaging of samples, etc.), which can modulate its magnitude, it does not directly predict the effect size. I was wondering if there has been any studies which systematically compared the two?

Using signal detection theory, we have calculated d prime for our research looking at false memories for faces.

We have found a significant correlation between facial recognition ability and d' for faces. Does that mean that the higher the d' prime score, the better you are at discriminating between hits and false alarms? Thus, your false memories for faces are lower?

I use nonparametric tests due to small groups and the absence of normal distribution. For Mann-Whitney U test I calculate the effect size by dividing U with the product of the two group sizes (as suggested by Ronán M. Conroy as well as others).

How do I most appropriately calculate effect size for Wilcoxon signed rank test (related samples)? 

I believe Cohen's d is not appropriate since it requires normal distribution and is based on means, hence the alternative effect size above for M-W U. Agree?

abstract

introduction

main body

conclusion

Hi,

Need suggestion on the SEM testing. Is measurement model testing is must while wring paper in SEM testing ? or we can directly move to SEM?

Regards,

Uday Bhale

In SmartPLS 4's Construct reliability and validity output have Composite reliability (rho_a) and Composite reliability (rho_c).

What is the difference between Composite reliability (rho_a) and (rho_c)?

Which one should use for my study?

I will like to modify a matlab .p file, the computer where thé initial code was write is damage, but a copy of the .p is available in another computer. I need a help

Dear all,

I've been conducting ANOVA followed by post-hoc (Tukey) tests and using partial eta squared as my effect size estimate. However, partial eta squared is simply estimating the fraction of the total variance that can be explained by the treatments and does not say anything about how different a particular group mean is from another, thus providing an "incomplete" answer. Therefore, to quantify the contrasts between the different groups, I feel Cohen's d would be more revealing in terms of the actual differences between the multiple pairwise comparisons. Would you agree and what are the pitfalls of such an approach? Am open to trying out any alternate methods to "completely" answer the question in my mind.

Thanks a lot!