What is the reason why you shouldn't believe too much in 'mean'?
When determining the specifications of web services and software, it is often the case that the 'average value' is calculated from the user behavior data to derive the 'average user image'. However, investor
How the US Air Force Ditched the 'Average' and Saved Lives
https://mannhowie.com/average-pitfall
Mann cites the problems associated with the development of aircraft in the United States Air Force in the 1940s as an example of where it was unsuitable to use the average. In the 1940s, the United States Air Force destroyed 23.6 fighters for every 100,000 hours of flight. Analyzing the aircraft in which the accident occurred rarely found any failures that could lead to the accident, so the Air Force attributed the accident to 'the cockpit does not fit the pilot's physique.' I tried to order a cockpit for a fighter aircraft design company according to the 'average value of the pilot's physique'.
However, Lieutenant Gilbert Daniels, who was working on 'study on student hand size' at Harvard University when he was a student, realized that there are not always pilots who match the 'average physique'. He disagreed with the Air Force's decision. Lieutenant Daniels actually measured more than 140 items such as 'finger length' and 'distance from eyes to ears' for more than 4000 pilots and calculated the average value of each item. Furthermore, from 140 items, items that are considered to be closely related to accidents such as height and chest circumference are extracted, and all the extracted items are 'pilots of average physique' that satisfy the condition of '30% before and after the average value'. I asked for the number of people.
As a result of the above analysis, it was found that there is no 'pilot of average physique'. In other words, even if the cockpit is designed according to the 'pilot of average physique' based on the body measurement data of all pilots, there is no pilot whose cockpit fits perfectly. Based on the analysis results, Lieutenant Daniels advised that 'the cockpit of a fighter should be tailored to all pilots, not the average pilot.' As a result, the cockpit was designed with position adjustment functions similar to the driver's seat of a modern car, and the number of fighter jets damaged was drastically reduced to 4.3 per 100,000 hours of flight.
What is the reason for the frequent occurrence of 'crash accidents due to operational mistakes' that was found by measuring the dimensions of 4000 pilots? --GIGAZINE
In addition to the 1940s US Air Force case above, Mann also presents examples of modern software development. For example, suppose that when performing a task called 'improving the performance of an analysis service for users' on a service, the user's behavior record first reveals that the 'average number of analysis executions per month' is 32 times. ..
However, if the number of analysis executions is shown in a graph (horizontal axis is the number of executions, vertical axis is the ratio of users), 60% of users execute the analysis only 0 to 9 times, and execute the analysis 70 to 89 times. You can see that there are more than 20% of users in total. On the other hand, we can also see that few users perform the average number of '32 times'.
When developing an analysis function for the above user group, if you keep in mind that 'it is used 32 times on average per month', it is possible to develop a function that is far from the user's request. There is sex. Based on these examples, Mr. Mann advises, 'If you refer to the average value, make sure to check how many examples correspond to the average value.'
Related Posts:
in Note, Posted by log1o_hf