The α level depends largely on how comfortable researchers are with the balance between wrongly concluding that there is an effect and not detecting a real effect (discussed in the “Statistical Power, β, Type II Error, and CIs” section). For instance, in our example, the researchers might have selected a lower α level (eg, 0.01) so that they could further reduce the risk of wrongly concluding that the chemotherapy is effective, particularly in light of its potential side effects, costs, and limited availability. However, even if the P value is as low as 0.01, chance might still be responsible for the observed results (in this case, it is a low probability of one chance in 100). As such, the probability of chance being responsible for observed results never goes to zero because it is always possible for chance to be responsible for the observed results. However, the key question is: How probable is it that chance is responsible for the observed results? Wrongly concluding that there is a difference (or effect) when one truly does not exist is called a “Type I” error—with the P value indicating the probability of this error occurring (ie, that the observed results were not the result of a real effect but actually occurred by chance).