That's not quite right. If the study is underpowered at n = 50 --- which is extremely likely --- statistically significant estimates are likely to be inflated.

That's not quite right. If the study is underpowered at n = 50 --- which is extremely likely --- statistically significant estimates are likely to be inflated. And as power declines, they also become more likely to have the wrong sign (e.g., the study will yield a positive estimate even though the true effect is negative). See https://doi.org/10.1177/1745691614551642 https://doi.org/10.1177/1745691614551642.