**Myth 4: Electoral survey projections were wrong. That is proof that statistics is useless.**

**Truth 4**

It is ironic that the type of reasoning used in this statement is intrinsically statistical (because a general characteristic of a whole is inferred from the empirical observation of a single result).

Under the premise that failure is gross discrepancy between a statistical prediction and reality, it is essential to note that this may be due to two main reasons:

- The first reason, "failure due to randomness", is due to the fact that data obtained at random can legitimately turn out to be atypical, and thus lead to an erroneous conclusion. A correctly specified statistical method cannot completely avoid failure due to randomness, but it can limit its occurrence.

- The second reason, "failure due to implementation", is ascribed to statistical malpractice. One cause could be that a statistical method was wrongly specified and that working assumptions did not apply. Another cause - even assuming that the method was correct - is that its implementation was wrong, for example, during the sampling and data recording phases. It is also possible that the wrong question was addressed, or that an error in the interpretation of results was committed.

Let us consider the following analogy: suppose a sick person has had a drug prescribed by a medical doctor, but that his health does not improve. It won't generally hold that medicine (the discipline) is put to blame, but rather that the doctor (practitioner of medicine) was wrong in the diagnosis and/or treatment. Others might blame the drug itself (the means) for the lack of improvement.

In this case, the concept of "failure due to randomness" is quite clear, because there is always a probability that a symptom does not yield to treatment, even if the treatment is undeniably correct. It is also possible that the illness is unheard-of. Therefore, when failures arise in studies in which statistics played a role, it is reasonable to challenge statisticians and the methods they have employed, but not the discipline of statistics. It is imperative to overcome the popular skepticism that arises against the field of statistics as a whole.

Under the premise that failure is gross discrepancy between a statistical prediction and reality, it is essential to note that this may be due to two main reasons:

- The first reason, "failure due to randomness", is due to the fact that data obtained at random can legitimately turn out to be atypical, and thus lead to an erroneous conclusion. A correctly specified statistical method cannot completely avoid failure due to randomness, but it can limit its occurrence.

- The second reason, "failure due to implementation", is ascribed to statistical malpractice. One cause could be that a statistical method was wrongly specified and that working assumptions did not apply. Another cause - even assuming that the method was correct - is that its implementation was wrong, for example, during the sampling and data recording phases. It is also possible that the wrong question was addressed, or that an error in the interpretation of results was committed.

Let us consider the following analogy: suppose a sick person has had a drug prescribed by a medical doctor, but that his health does not improve. It won't generally hold that medicine (the discipline) is put to blame, but rather that the doctor (practitioner of medicine) was wrong in the diagnosis and/or treatment. Others might blame the drug itself (the means) for the lack of improvement.

In this case, the concept of "failure due to randomness" is quite clear, because there is always a probability that a symptom does not yield to treatment, even if the treatment is undeniably correct. It is also possible that the illness is unheard-of. Therefore, when failures arise in studies in which statistics played a role, it is reasonable to challenge statisticians and the methods they have employed, but not the discipline of statistics. It is imperative to overcome the popular skepticism that arises against the field of statistics as a whole.

Return