Let's start with a scenario: Suppose the platform sells two mobile phones A and B. 800 people like mobile phone A and 200 people dislike it; mobile phone B has 9 people like it and 2 people dislike it. So, which phone do users prefer? I believe that this scene has been encountered by all my friends in daily life and work. How do you usually make judgments? I hope that through today's article, I can give you a new perspective and a more scientific solution. consumer email list A common measurement method I think everyone's first reaction should be to measure according to the ratio? therefore, A mobile phone preference rate=800÷(800+200)=80% B mobile phone preference rate=9÷(9+2)=82%80%<82%
Therefore, users prefer B mobile consumer email list phones. is this correct? It looks fine. After all, the higher the like rate, the more users like it! However, I believe my friends can also see the clues of this example: the total sample size of mobile phone B is only 11. Although the like rate is high, the sample size is so low that any data change will have a huge impact on the results. Therefore, according to this ratio method, is the calculated like rate "reliable"? In statistical language, confidence? 2. Wilson's score Above we feel that it is a bit difficult to measure according to the simple like rate. But if you don't compare by like rate, how else can you calculate it? That's our topic today: Wilson's score. 1.
Formula Definition Let’s first look at the specific Wilson score calculation formula: Wilson Score: The sample size is too small, how to measure the degree of preference scientifically? A common problem in data analysis u is the number of positive cases (likes), v is the number of negative cases (dislikes), n is the total number of instances (total samples), p is the like rate, z is the quantile of the normal distribution (parameter), and S is the final Wilson's score. The higher the score, the more like the degree, the greater the probability of like.