[Experience sharing] Chi-square test practice-test the independence of the retention rate for the next day and User Classification

Problem background

In a recent internal game test, a group of internal test users were given and each user was markedRPGGame users,Casual game usersAndCard game users. It is expected that the retention rate of these types of users will be calculated on the next day, and whether there are significant differences to guide subsequent game operation activities.

The Chi-square test technology is used to verify the independence of user types and next-day behaviors as a memo.

Statistical results

After data cleaning and statistics, the following results are obtained:

User Type	Registration Date	Number of registrants	Remaining users the next day	Retention rate for the next day	Average retention rate for the next day
RPG	Day1	2,504	752	69.97%	74.50%
RPG	Day2	2,718	701	74.21%
RPG	Day3	2,538	692	72.73%
RPG	Day4	2,207	540	75.53%
RPG	Day5	2,004	464	76.85%
RPG	Day6	1,899	458	75.88%
RPG	Day7	1,243	294	76.35%
Leisure	Day1	2,983	925	68.99%	73.45%
Leisure	Day2	3,219	864	73.16%
Leisure	Day3	3,067	863	71.86%
Leisure	Day4	2,715	686	74.73%
Leisure	Day5	2,442	607	75.14%
Leisure	Day6	2,295	584	74.55%
Leisure	Day7	1,523	370	75.71%
Card	Day1	2,102	675	67.89%	73.04%
Card	Day2	2,198	584	73.43%
Card	Day3	1,908	542	71.59%
Card	Day4	1,643	422	74.32%
Card	Day5	1,454	372	74.42%
Card	Day6	1,382	370	73.23%
Card	Day7	883	208	76.44%

'

P.s .:The above data is processed without affecting the effect, and does not represent real business data.

By observing the above data, we can basically determine that there is no significant difference in the next day retention rate between different types of games. But is that true?

 

Chi-square test

The Chi-square [1] test can be used to verify whether the loss and retention of game users in different categories are independent of each other.

Hypothesis

Original hypothetical H-0: The next-day user behavior is independent of the user category.

Alternative Hypothesis ha: the next day's user behavior is related to the user category.

PS:The next-day behavior refers to the loss or retention of the next day. The user category is RPG., Casual or card

Select significance level Alpha = 1%

Statistics

In R, the following statistical results can be obtained through the chi-square test:

Chi-square: 9.36, P value [4]: 0.9%

Conclusion

Because the P value is 0.9%, less than the significance level α = 1%, You can reject the original hypothesis H0 and accept the alternative hypothesis H-A, that is, the user category does affect the user's next day retention, they are not independent of each other.

Download hereStatistical Code and data.

 

Reference

[1] Chi-square test Wiki

[2] Chi-square test practice

[3] Chapter 14th of deep dive into statistics/head first statistics

[4] P value Wiki

[Experience sharing] Chi-square test practice-test the independence of the retention rate for the next day and User Classification