In data analysis, data analysis thinking is a framework guide, which can be used quickly in some common analysis scenarios, and it is also helpful to build data analysis models in the future.

Here are 20 common data analysis methods:

First, index thinking

The content we usually express can be basically divided into facts and opinions. Facts can not be denied, while opinions can not be recognized, but most people are easy to confuse facts and opinions

The same is true for data analysis. The purpose of data analysis is to improve the decision-making level of the company. If objective facts cannot be described, information is easily distorted in the process of communication.

If you say in a meeting that sales are down sharply this month, different people will interpret it differently. Some people think a drop of more than 20% is a big drop, some people think a drop of 10% is a big drop.

What if you take away “substantial”? The sentence “sales fell this month” is a fact, but it may be a different fact to different people because it is not clear what the comparison is.

Second, structured thinking

In fact, induction is the process of decomposing a complex problem into a variety of single factors, and summarizing and arranging these factors to make them orderly and programmatic. The process is like pulling the silk out of the cocoon, making a mess of things smooth.

Three, structured thinking

When faced with such a problem, the first thing a structured approach does is not immediately start cleaning the data.

Instead, it draws a mind map for data analysis based on the understanding of the business, which is equivalent to taking out baidu map to search the road map of taking transportation to the hotel you stay in when you come to a strange city.

In fact, structured thinking is the famous “pyramid thinking” proposed by McKinsey, as shown in the figure below:

Funnel thinking

Funnel thinking is to determine the key links, and then complete a set of flow analysis ideas, in all walks of life has the corresponding application, such as registration conversion rate analysis, user browsing path analysis, traffic monitoring, etc..

Taking the analysis of user conversion rate as an example, there are actually five key steps in a web page from display to order by using structured thinking analysis: exposure, click, browse, consultation and order

Quadrant method

Through the division of the two dimensions, the desired value is expressed in the form of coordinates, and the value is directly transformed into strategy, so as to promote some landing. Quadrant method is a strategy driven thinking, widely used in strategic analysis, product analysis, market analysis, customer management, user management, commodity management and so on.

The figure below is the RFM model, which divides customers into eight quadrants according to the three dimensions of Recency, Frequency and Monetary.

Six, multidimensional method

Multidimensional method refers to objects from multiple dimensions to analysis, there are generally three dimensions, each dimension has a different data classification, represents the total number according to the big cube is divided into small squares, fall in the same small block of data has the same properties, so that we can through the contrast analysis of data within a small square. Here is a multidimensional table of takeout orders from a fast food restaurant:

7. Comparative thinking

The comparison is mainly divided into the following types:

• Crosscutting Comparison: Comparisons are made according to the crosscutting dimensions of subdivisions, such as cities and categories
• Swatch comparison: To compare swatch maintenance in subdivision, such as conversion rate at different stages of funnel
• Comparison of objectives: common in management by objectives, such as completion rate
• Time comparison: month-on-month, week-on-month; Comparison of 7-day moving average and extreme value within 7 days

8. Dimensional thinking

** Drill down: ** Changes between different levels of dimension, from the top down to the next level, or breaking down aggregate data into more detailed data, such as drill down huawei’s total sales data for 2018 to see sales data for each phone model.

** Roll-up: ** The reverse operation of drill-down, that is, aggregation from fine-grained data to higher-level data. For example, the sales data of Jiangsu province, Shanghai and Zhejiang province are summarized to check the sales data of Jiangsu, Zhejiang and Shanghai.

** Slice: ** select specific values in the dimension for analysis, such as iPhone sales data only, or 2017 phone sales data.

** Slice: ** Select the data of a specific interval in the dimension for analysis, such as the sales data of 2016 and 2017.

** Rotation: ** is the swap of the position of dimensions, just like the row and column transformation of a two-dimensional table. In the figure, the swap of product and regional dimensions is realized through rotation.

Ninth, traceability thinking

After repeated subdivision and comparison, we can basically confirm the problem. At this point, you need to confirm with the business side whether the data is abnormal due to some business action, such as a new release coming online, or optimization of activity policy, etc.

If you still don’t have a clue, you can only start from the smallest granularity, such as user log analysis, user interviews, external environment understanding, such as external activities, changes in policy and economic conditions, and so on

Ten, twenty-eight thinking

The 80/20 rule is also known as the Pareto rule, from the classic 80/20 rule. For example, in terms of personal wealth, 20% of the world’s people hold 80% of the wealth.

In data analysis, it can be understood that 20% of the data produces 80% of the effect and needs to be mined around this 20% of the data.

It is often associated with ranking when using the 80/20 rule, with the top 20% being considered valid. The 80-20 method is to grasp the key analysis, applicable to any industry.

Find the focus, find the characteristics, and then think about how the other 80 percent can be converted to the 20 percent to improve performance.

Eleven, false try

In some cases, such as the sales volume entering a new market or the change of the sales volume after the price increase, there may not be detailed data to analyze, so fake methods are needed. Pseudomanipulation is assuming that a variable or ratio is true and then working backwards from part of the data. This is a thought-provoking technique. The general process is to assume, then verify, and then judge the results of the analysis.

12. Goal thinking

The same as the first step in the process of data analysis, analysis the purpose must be clear, you need to think about: what method is used to achieve the purpose of analysis, chart was able to fully show the end what you want to express intentions, these a few analysis dimension is comprehensive, whether can support analysis, etc., is a natural problem analysis accordingly.