1. Reduce the absolute value of data for calculation. For example, the value of each data item is very large, and the calculation of many such values may exceed the range of common data types. At this time, taking logarithm will reduce the value; In the empirical model, the correlation coefficient will be larger after reducing the numerical value (the original numerical value may need four to five significant figures). 2. After taking logarithm, multiplication calculation can be converted into addition calculation. 3. In some cases, the differences between different intervals in the whole data range will have different effects.

It can be seen from the image of log function that the smaller the value of independent variable X, the faster the change of function value Y, which is the same as the above example, but the difference is 300, but log500-log200 >: Log800-log500, because the former pair is smaller than the latter pair. That is to say, the sensitivity to the difference of the part with small value is higher than that of the part with large value. This is also in line with common sense of life. For example, for the price, if the price difference is several hundred yuan, it can greatly affect your decision-making, but when buying a car, you will ignore the price difference of several hundred yuan.