Reading Notes in simple data analysis

All models are incorrect, but some of them are useful.

The root cause of data analysis is to properly resolve the problem, set appropriate mental and statistical models for the data, and make correct judgments. However, it is not guaranteed that the correct answer is obtained for the next time.

1. Data Analysis Mode: confirm the problem -- break down the data -- evaluate the data -- make decisions; it is critical to ask questions and obtain the required information to confirm the problem.
2. test your theory: A good AB test; no statistical data is absolutely accurate; comparison is a magic weapon for cracking observation data, and data is only meaningful for comparison;
3. Find the maximum value: Use the plan to solve the problem. Set the formula and all constraints to find the maximum solution in the feasible region.
4. Data graphics: the root of Data graphics lies in correct comparison.

5. hypothesis test: the pseudo-positive method: removes basic impossible assumptions from each piece of information, establishes positive and negative correlation between various factors, and assigns positive and negative values to each hypothesis for each piece of information, the assumption with the maximum value is the most likely;

6. bayesian statistics: Bayes is amazing: If a cold is detected by default, the probability of positive test is 90%, or if no cold is detected, the probability of positive test is 9%; if your test is positive, the probability of a cold is (1% of people have a cold ). For example, everyone thinks that A is very likely to happen, And suddenly receives A message saying that A is likely not to happen. This is not to allow everyone to re-evaluate the probability of A happening, instead, A evaluates the probability that A receives the message and A does not receive the message, and then computes the message using Bayesian.

7. subjective probability: the intuition is solidified into subjective probability data, and the differences are determined by standard deviation;

8. Inspiration: Do you decide on impulse or on several carefully selected key data? Or is it best to build a model that contains all the variables and get the best answer?