Guess the Correlation

http://guessthecorrelation.com/

The most common effect size you’ll encounter in this course is a correlation, whether it’s the phenotypic correlation between twins, or the correlation between a genetic variant and a phenotype, or the genetic correlation between two phenotypes.

Practice visualizing what different correlations imply in terms of how consistent or spread out the data might be.

It’s a simple website that for some reason is really strict about picking a unique username even though it won’t let you sign in to retrieve your results later. I usually just bang on my keyboard to generate a unique username to get in.

p-hacker

https://www.shinyapps.org/apps/p-hacker/

This web app illustrates how p-hacking (or running a bunch of different kinds of analyses - add or dropping participants until you get a significant result, looking at multiple outcomes without correcting the significant p-value threshold, running subgroup analyses without a theory or plan or correction for multiple testing) can get you a significant (low) p-value just by chance alone. It will simulate, analyze, and plot data according to your instructions.

The Manual drop-down menu at the top left will give some tips. To make it feel genetic-y, label the experimental group “minor allele present (AB, BB)” and the control group “minor allele absent (AA)” and “test” for effects of the “genotype” on multiple “phenotypes” (set the number of DVs to 10).

Instead of a correlation, the effect size used is cohen’s d (which describes the average difference between two groups). Larger values mean a bigger difference between the two groups. It’s a little different from a correlation in that it’s not bounded by +/- 1 (it can be infinitely large). But 0 still means “no effect”, or no difference between the two groups, or no relationship between the exposure and the outcome, and the interpretation of its p-values is still the same (it’s a combination of the size of the effect and the number of observations - larger effects and/or more observations lead to smaller, more significant p-values).

Next: 4.5. Activity: Estimate Heritability from Twin Data using Two Methods in R

Previous: 4.3. Additional Readings

Home: Table of Contents