There's a great - and very common - p-hacking trick where you just run your study for as long as it takes to get the result you want, and then immediately stop. Why increase the size of your dataset if there's a risk it will wash out the anomaly that confirms your priors?