What is causal analysis?
Causal Analysis is to analyze the causal relationship between each other.
Causal inference is the inference of results based on causes, which is a part of causal analysis.
Causal analysis is an important method in data analysis and data science, and is widely used in A/B experiments, anomaly analysis, user growth and other fields.
This article attempts to use logic to deduce the basis, principles, methods, and applied knowledge system of causal analysis from a macro perspective.
Firstly, basic knowledge such as the concept, elements and classification of causal analysis are introduced;
Then, the principles of common causal analysis methods such as A/B experiment and fishbone diagram analysis of causal analysis are introduced;
Then, it introduces the application of causal analysis methods such as A/B experiment (strategy adjustment), abnormal analysis (DAU drop), and introduces R and Python causal inference tools;
Finally, the causal analysis is summarized, the essence of causal analysis is pointed out, the difference between causal relationship and correlation relationship is introduced, and the development trend is introduced.
The catalog of this article is as follows:
1. Basics of causal analysis
1.1. Concept of causal analysis
1.2. Elements of causal analysis
1.3. Classification of causal analysis
2. Principle of causal analysis
2.1. Principle of A/B experiment
2.2. Principle of fishbone diagram analysis
3. Application of causal analysis
3.1 . A/B experiment: strategy adjustment
3.2. Abnormal analysis: DAU decline
3.3. Causal inference tool
4. Causal analysis summary
4.1. The essence of causal analysis
4.2. Causal relationship VS correlation
4.3. The development trend of causal analysis
Next, let us walk into the world of causal analysis and try to discuss the foundation, principle, methodology and application of causal analysis.
1. Basics of causal analysis
1.1. The concept of causal analysis
Causation is the relationship between cause and effect, and causation is the relationship between cause and effect.
What is causal analysis?
Here is an excellent article from Zhihu: a clearer description of the concept of causal analysis: causality - Zhihu (zhihu.com) 
https://zhuanlan.zhihu.com/p/555170435
Causal analysis is the analysis of the causal relationship between each other.
1.2. Elements of causal analysis
There may be many reasons, and there may be many consequences.
Here we focus on the problem and only discuss the abstract causal relationship, so the causal analysis can be abstracted as follows:
The three elements of causal analysis are: cause, effect, and relationship.
1.3. Classification of causal analysis
- According to the three elements of causal analysis, causal analysis is divided into three categories:
- The first category: deduce the result from the cause, also known as causal inference (Causal Inference);
- The second category: finding the cause from the result;
- The third category: mutual derivation of cause and effect;
What are the methods of causal analysis?
The three typical representative methods of causal analysis are shown in the table below:
Here is a special explanation: Causal inference is the result of reasoning, also known as causal inference and causal inference. It is a type of causal analysis. It is widely used in data science and is more and more closely integrated with machine learning.
2. Principles of causal analysis
In the Internet field, A/B experiments and fishbone diagram analysis are commonly used causal analysis methods, so the principles of these two methods will be explained next.
2.1. The principle of A/B experiment
A/B experiment, also known as A/B test, A/B experiment, is a randomized controlled experiment used to experimentally verify cause and effect.
A/B experiment is a representative of causal experiment and the main means of causal attribution and data attribution.
A/B experiment is a type of single-factor attribution, suitable for verifying the causality of a single factor.
2.2. The principle of fishbone diagram analysis
Fishbone diagram analysis (Cause and Effect Analysis Chart, also known as causal analysis) is a typical method of finding the cause from the result.
The fishbone diagram analysis method is to classify and exhaustively list all the influencing factors on a problem for further analysis.
Among them, the fish head is the result (problem), the big fish bone is the category of the cause, and the small fish bone is the specific cause.
Fishbone diagram analysis is useful for brainstorming for multiple possible causes.
3. Application of causal analysis
Next, some typical application scenarios of causal analysis are introduced.
The application of A/B experiments that deduce the results from the causes; the application of abnormal analysis that finds the causes from the results.
3.1. A/B experiment: strategy adjustment
Microsoft's A/B experiment to adjust the color of the page and improve the click-through rate.
3.2. Abnormal analysis: DAU drops
In anomaly analysis, abnormality is the result, find out the cause of the abnormality, and give optimization suggestions.
3.3. Causal inference tools
There are many causal inference tools for R and Python, such as Dowhy, Causal ML, EconML, causalToolbox, etc. For more causal inference tools, causal inference methods, and causal inference models, please refer to the paper " A Survey on Causal Inference " and Guanhe causal products:
4. Summary of causal analysis
4.1. The nature of causal analysis
The essence of causal analysis is to demonstrate the sufficiency and necessity of causality.
4.2. Causation vs Correlation
A causal relationship is necessarily a correlation, but a correlation is not necessarily a causal relationship.
4.3. The development trend of causal analysis
Personally think:
The direction of reason and result, such as meta-analysis, causal inference machine learning;
The result is to find the direction of the reason, such as breaking boundaries and innovation.
For industrial application cases of boundary-breaking innovation, you can refer to "Boundary-breaking Innovation in the Application of Causal Analysis: From Effect to Cause, and then from Cause to Effect".
references:
Summarize:
The author of "Lean Data Analysis" believes: "Finding correlation can help you predict the future, and discovering causality means you can change the future", so pay attention to causality and causal analysis.
Causal analysis can locate problems, mine business value, and gain insight into opportunities. It has a wide range of applications in the Internet field, such as causal inference, A/B experiments, user growth, anomaly analysis, and churn analysis.
But business is a complex ecology, causal analysis is a good analysis tool, and specific users, businesses, and data must be combined for targeted analysis.
Conclusion:
Due to the limited personal experience, ability and level, mine may be one-sided or wrong, here is an example.
Due to the limited personal experience, ability and level, mine may be one-sided or wrong
Theory itself is unrealistic and needs practice, practice, and practice. You can refer to the practice of Ali, Kuaishou, and Guan He in causal inference.
Your feedback, both positive and negative, is valuable and helps me accelerate iterative upgrades—deeper and more comprehensive. Leave your wonderful comments!