datasets with millions or billions of observations — carries an almost magical aura. More data sounds like it should always be better. More data means smaller standard errors, tighter confidence intervals, and more power to detect effects. All of which is true — *when the data is representative*.