There's a sweet spot to be found, and I might even argue that their sample is too large if it isn't stratified in some meaningful sense. (It seems it isn't str

There's a sweet spot to be found, and I might even argue that their sample is too large if it isn't stratified in some meaningful sense. (It seems it isn't stratified: "Scores were not broken down beyond the state level.") We're talking an average of 2400 students from each state. That's plenty for certain tests, such as the question of whether or not all states perform equally well. Since it's not stratified, we're basically talking about a rank order test concerning 50 means (depending on how you formulate it). A sample size of 120,000 is so enormously huge (for a question like that) that it might even be detrimental. The quote about small sample sizes is just wrong. If anything, this sample is too large, such that even minuscule differences would appear to be statistically significant, when in fact that might not be "practically different."