Simulating how voxel selection in fMRI caused regression to the mean in a recently-retracted paper

How selection bias perpetuates low power in science

How do you calculate a noise ceiling? What is the difference between the estimates calculated in RSA and those reported in the encoding model literature?

In which we study how statistical power and significance thresholding influences reproducibility by relating it to positive and negative predictive values.

We demonstrate how to apply pilab linear discriminant contrast methods to a real fMRI dataset, including searchlight mapping over the full volume.

We estimate multivariate discriminability with linear discriminant contrast (also known as cross-validated Mahalanobis distance, crossnobis).

We simulate a basic fMRI dataset, and estimate cross-validated prediction performance for a convolved design.

Syncing folders between computers can be surprisingly complicated if you work for an academic institution that takes a dim view of Dropbox and other cloud-based solutions. You are also sure to run out of space fast if you use it to sync e.g. neuroimaging data. Unison is a free …

It's a bit of a cliche that the best papers are the ones that raise more questions than they answer (in fact, many papers seem to answer hardly anything at all on close inspection and it doesn't mean they're great). But I think this might be one of those papers …

Getting a scientific python install up and running is still way too complicated. In this post I describe how I use a conda to keep a reproducible record of the packages I use.

In the past, I have usually hacked together my own developing environment through whatever tools were most …