In simplest terms, Statistical Power is the probability of ‘correctly’ rejecting your Null Hypothesis. Here ‘correctly’ signifies that you are not making a Type 1 or Type 2 error.

Consider 2 groups of kids - group A with no medicine given and B given with a medicine to increase heights. If your medicine indeed had a significant effect on the heights, the distribution curves of heights of both the groups will be clearly separated when plotted on the same axis.

In such a case, when you repeatedly perform hypothesis tests, you’ll probably get a low p-value most of the times, indicating that you can safely reject your null. But what about when you make a Type 2 error? That is, you get a high p-value and incorrectly fail to reject the null.

Such cases will be less depending on how well your distributions are separated. Or, how high is your Statistical Power. Therefore, higher the Statistical Power, more is the probability that you are correctly rejecting the null.

Similarly, when the medicine didn’t have a significant effect, the 2 distributions won’t be well separated. Now you’ll get a high p-value a lot of times and hence have a low probability to safely reject the null. Signifying a low Statistical Power.

#statistics #datascience #machinelearning