Do we understand risk? Much of classical statistical reasoning is structured around processes similar to a criminal trial. In a criminal trial, the accused must be found guilty beyond a reasonable doubt. This is very subjective. The judge or jury wants there to be a very small chance of sending an innocent person to jail. The trial is structured around assuming the defendant is innocent and then the prosecution presents evidence to prove guilt. The judge/jury must weigh whether it is possible to accept the evidence presented and still believe innocence. If it is possible, then they conclude not guilty but if they no longer believe that the evidence is possible with innocence, they conclude guilty. In statistical hypothesis testing, you might put the belief this marketing strategy is ineffective on trial. You then present evidence of using this marketing strategy on a test group, seeing sales increase, and then ask could this happen just by chance? Maybe sales went up just by chance and the strategy is ineffective. If the evidence is sufficiently compelling, you reject chance as the explanation and conclude that it is effective. We can set up procedures to calculate the probability of sales increasing just by chance. We conclude the marketing strategy is effective if this chance is very small below some threshold. This may sound confusing. It is! What should the threshold be? It is common to use either 1% or 5%. That is, the test procedure will lead you to adopting an ineffective marketing strategy 1% (or 5%) of the time, because sales increased just by chance, not because of the new strategy. Can you differentiate between a 1% risk and a 5% risk? Should the risk threshold be higher, say 10% chance of implementing an ineffective strategy? Should it be lower, say only a 0.1% (1 in a 1,000) chance of implementing an ineffective strategy. All decision involve a risk of making a wrong choice. But how do we decide how large a risk to take?