Methods & Meta-science

Analyzing visual world data using cluster randomization in R

Cluster randomization is a non-parametrical statistical method popular in neuroimaging for detecting where and when experimentally induced effects occur. Using this method to detect effects in visual world studies presents a number of advantages among which allowing the researcher to consider the entire dataset without having to break it into smaller time windows, dealing with within-trial dependencies and so on. What’s more, the p-values are generated through randomization, i.e. by generating new datasets by randomly redistributing condition labels within participants; the probability of observing the initial dataset given these randomly generated datasets. This avoids having to resort to assumptions about the population. The presentation will include an illustration of this method through a reanalysis of a subset of the data from Kronmüller and Barr (2007), performed using R.