2.4.1 Why Should You Expect Mistakes To Happen?

Often what we do at work is easier than what we learned at school. On exams we might average 90%, with most people probably closer to 80%, and perhaps even lower. Do you think that when you begin working everyone shifts to 100% correct? Of course not. Within data analysis the same problem holds true. Unfortunately time pressures can make even the simplest tasks difficult. Copy a page from a book. Simple enough. Now do it in 1 minute and with no mistakes. Do you even have time to check your work? Sometimes work can be like this. Within consulting, it can often be like this. Often you will be given a task that is not difficult, particularly if it was the only thing you had to do. You may give a statistician a simple task, but he may have 9 other simple things all at the same time, making things difficult. Making things difficult. If a statistician only had one thing to do and it was easy, it would not make good business sense to employ him or her. Everyone must perform a certain amount of work to justify being employed.

Assume that the average statistician can produce correct statistics 95% of the time. Not bad. People go from about 80% correct on exams to 95% correct in the real world for various reasons.

Let’s say that someone has 10 tasks a week, 10X.95=9.5 correct. In two weeks he has 20 simple tasks. Thus in two weeks he is expected to get 1 thing wrong. If you present one report a week to your boss containing the statisticians statistics, every other report (1/2) presented to him will contain a mistake. Why would your boss even want to read your work if he cant trust it? High level decisions will possibly be made on work that is incorrect.

We often say, anything is possible. We must decide and go with the probable. Certain things may be possible, but if they are highly unlikely you should have the work double checked. The possible but not probable is often a mistake. True we can not double check everything. Hypothetical scenario: A statistician has a probability of being correct 95% of the time. You are not confident in his work 10% of the time and have him check his work 10% of the time. Ideally, if you are very good you will catch all 5% of the mistakes. True, you will waste some of the statistician’s time. So what! You will now produce reports for your boss without mistakes, at least not statistical mistakes, which is what your boss expects of you. This would mean half the time you ask him to double to check he is wrong. You will earn respect if you catch his mistakes. If you always ask him to check and there is never anything wrong, that will be a problem. Remember: when he double checks, he may not find anything wrong, but this does not mean it is correct. This too happens and has happened to me more than once while managing statisticians. Fortunately we ultimately found the mistakes.