Slicecast

Found an interesting slideshow from Realeyes in the UK via Usableworld that presents the importance of the correct sample size when conducting Eyetracking studies.

| View | Upload your own

Eyetracking I have found to be an interesting topic; largely because it has provided some very interesting things but also because I have found that many studies have failed in the eyes of a client - to coin a phrase "this study points out the bleedin' obvious, tell me something I don't know". True. The value of Eyetracking, alongside the newer mouse based heatmaps need to be one more weapon (albeit at a sizable investment) in the UX / Usability / Optimisation toolkit, is that it provides that insight into the decision making process of site visitors.

The presentation makes it quite clear that selection of the correct sample size can have a big impact on the outcome of the test; and therefore on the conclusions that you ould draw from the test. It illustrates one of the principle laws of statistics - "The Central Limit Theorum" - that essentially states you can take any dataset and if you take enough random samples from it, you can plot out a near perfect bell curve. In other words, under normal circumstances, everything regresses to the mean.

The example to the left shows the theorum in a simple way, the higher the sample size the higher the probability the distribution will be normal. The same basic rules apply in the methodologies used to determine correct sample sizes for A/B and Multivariate tests.

Interesting stuff, and the lesson is - make sure your sample sizes for your Eyetracking study are large enough to get you close to the mean result; too little (if for example your budget does not permit the study of more than 4 or 5 people) and you will end up with semi-directional heuristic analysis at best.