Real or big?
Two different questions. Real: is the estimate distinguishable from zero (does its 95% interval exclude zero)? Big: is it large enough to matter? Pick a study, vote, then reveal. Change the sample size and the tool draws more (or less) data, so the estimate moves and the interval changes, just as it would with real data.
More data is simulated: each doubling adds a new batch of people drawn from the true process, so the original sample's quirks wash out. Less data is a random subsample. The true effect is something you only get to see in a simulation.