What are the basic methods of data analysis?

1. data trend analysis

Generally speaking, trend analysis is suitable for long-term tracking of product core indicators, such as click-through rate, GMV, active users, etc. Making a simple data trend chart is not trend analysis. Trend analysis needs to make clear the changes of data and analyze the reasons for the changes.

Trend analysis, the best output is the ratio. In trend analysis, several concepts need to be clarified: ring comparison, year-on-year comparison and fixed base ratio. Ring-on-ring refers to the comparison between the statistics of the current period and the statistics of the previous period, such as the comparison between February 20 19 and June 20 19. The ring-on-ring shows the recent changing trend, but there are some seasonal differences. In order to eliminate seasonal differences, the concept of year-on-year is put forward, such as comparing February 20 19 with February 20 18. The fixed base ratio is easy to understand, that is, it is compared with a certain base point, such as 20 18 1 month, and the fixed base ratio is compared with 20 19 February and 20 18 1 month.

For example, from 2065438 to February 2009, the number of monthly active users of an APP was 20 million, up 2% from the previous month and 20% from February last year. Another core purpose of trend analysis is to explain trends. For the obvious inflection point in the trend line, it is necessary to give a reasonable explanation for what happened, whether it is external or internal reasons.

2. Comparative analysis of data

The trend of data changes independently, in fact, in many cases, it does not explain the problem. For example, a company's profit growth 10%, we can't judge whether it is good or bad. If other companies in this company's industry generally have negative growth, it will be 5%. If other companies in the industry grow by an average of 50%, this is a very poor data.

Comparative analysis is to give isolated data a reasonable frame of reference, otherwise isolated data is meaningless. Here I recommend a big data technology exchange circle: 658558542 to break through the technical bottleneck and improve thinking ability.

Generally speaking, the comparative data is the fundamentals of the data, such as the situation of the industry and the situation of the whole station. Sometimes, in the iterative testing of products, in order to increase persuasiveness, the comparison benchmark will be set artificially. That is, A/B test.

The key to the comparative experiment is that only a single variable is kept in the A/B group, and other conditions remain unchanged. For example, to test the effect of homepage revision, it is necessary to keep the quality of A/B users unchanged, the online time unchanged, and the source channels unchanged. Only in this way can we get more convincing data.

3. Data decomposition analysis

When we get some preliminary conclusions, we need to further decompose them, because in the process of using some comprehensive indicators, some key data details will be erased, and the changes of indicators themselves need to analyze the reasons for the changes. The subdivision here must be a multi-dimensional subdivision. Common splitting methods include:

Time-sharing: whether there are changes in short data at different times.

Sub-channel: whether the traffic or products from different sources have changed.

Sub-users: whether there are differences between newly registered users and old users, and whether there are differences between advanced users and low-level users.

Sub-region: whether the data in different regions have changed.

Composition splitting: for example, search consists of search words, which can split different search words; Store traffic is generated by unused stores, and different stores can be split.

Subdivision analysis is a very important means. Asking more why is the key to drawing a conclusion, and step by step splitting is the process of constantly asking why.