search

Mouse-Imaging-Centre / RMINC

Statistics for MINC volumes: A library to integrate voxel-based statistics for MINC volumes into the R environment. Supports getting and writing of MINC volumes, running voxel-wise linear models, correlations, etc.; correcting for multiple comparisons using the False Discovery Rate, and more. With contributions from Jason Lerch, Chris Hammill, Jim Nikelski and Matthijs van Eede. Some additional information can be found here:
https://mouse-imaging-centre.github.io/RMINC
Other
22 stars 17 forks source link

Extension for *effectsize, partial correlations for continuous predictors #236

Open gdevenyi opened 5 years ago

gdevenyi commented 5 years ago

Also available from Nakagawa, S., Cuthill, I.C., 2007. Effect size, confidence interval and statistical significance: a practical guide for biologists. Biol. Rev. Camb. Philos. Soc. 82, 591–605. https://doi.org/10.1111/j.1469-185X.2007.00027.x

Is methods to compute an effect size for continuous predictors via partial correlation.

cfhammill commented 5 years ago

Jason and I were talking about this the other day, we're still not quite sure what we're gaining here over simply standardizing the betas by their empirical sds. Or seemingly equivalently back transforming the t by multiplying by the param specific sqrt N. Do Nakagawa and Cuthill make a compelling argument against doing that?

gdevenyi commented 5 years ago

From the paper: The t values for a continuous predictor variable can be converted to r using a rather unintuitive equation below: r = t / sqrt(t^2 + df) Effect size calculated using in this way takes covariates into account. This form of r value is often referred to as a partial correlation coefficient. The partial correlation between y and x1, controlling for x2, is numerically equivalent to the correlation between the residuals from the regression of y on x2 and the residuals from the regression of x1 on x2. Thus the partial coefficient for a given predictor removes the variance explained by other predictor variables from both variables, and then quantifies the remaining correlation.

They cover a lot of useful things about sample size biases and non-heterogeneity correction as well which I haven'y copy-pasted

gdevenyi commented 5 years ago

I may be insufficiently informed to know if what Nakagawa et. al. recommend is any different than what you're asking about though.... especially if you and Jason are discussing it :)

cfhammill commented 5 years ago

No no, that equation is quite different than what we're thinking. I don't immediately see why we need to do extra compensation for variance explained by other predictors, but I guess I'll just have to read the paper.