You need 16 times the sample size to estimate an interaction than a main effect

Re:You need 16 times the sample size to estimate an interact... by rripken on Friday March 16, 2018 @03:50PM (#16603534)

That 16x number is specific to a set of assumptions in the article. Is this just the Curse of Dimensionality in action? https://en.wikipedia.org/wiki/Curse_of_dimensionality

[ Parent | on HN ]

Re:You need 16 times the sample size to estimate an interact... by RA_Fisher on Friday March 16, 2018 @05:51PM (#16604272)

Andrew calls it the blessing of dimensionality: http://andrewgelman.com/2004/10/27/the_blessing_of/

Multi-level modeling's concept of partial pooling is really a killer feature.

[ Parent | on HN ]

Re:You need 16 times the sample size to estimate an interact... by xitrium on Friday March 16, 2018 @11:28PM (#16605466)

Bob Carpenter did a great case study on the this topic, for anyone new to it: http://mc-stan.org/users/documentation/case-studies/curse-di...

Check out the graphs, especially of the euclidean distance from the mode of a random point as dimensions increase.

[ Parent | on HN ]

Re:You need 16 times the sample size to estimate an interact... by evandijk70 on Saturday March 17, 2018 @03:55AM (#16605996)

It's part of it, but the described effect is amplified by the fact that if there are n variables, there are (n^2-n)/2 potential interactions. This increases the risk of finding spurious correlations due to multiple testing.

[ Parent | on HN ]