We compute a number that should be higher or lower than a criterion. If the effect size is larger than a critical value, it the t-value is larger, or a p-value is smaller, you can say there is something because the long run probability that you are fooling yourself is small enough.
