How does it increase sensitivity to data operations?

Enhance the data sensitivity, mainly by accumulating experience, looking at the data, the more the quantity, the richer the kind, the stronger the sensitivity.

But this does not mean that casual look, random look, but targeted to increase their ability to refer to the following three directions:

Predictive power

Ability to discover data exceptions

The ability to translate data into knowledge

Predictive power

Before seeing the actual value, we can estimate the value according to other data and subjective feeling, and control the estimation error.

This ability is easier to exercise. Before looking at the data, make a subjective guess, then compare the guesses with the actual value and adjust your feeling according to the actual value. Lots of practice.

For a series of data, such as sales, site visits, active users, and so on, need to understand that in the absence of anomalies, the series of data is composed of regular + random number. For example, Monday site traffic around 100W, this is the law, up and down fluctuations of 95% of the possible within 5W, which is random number.

Ability to discover data exceptions

Discovering data anomalies is an extension of the predictive power. The stronger the predictive power, the better the speed and alignment of discovering anomalies.

There are two kinds of exceptions:

The correct exception, that is, the actual occurrence of special events, reflected in the data is abnormal.

The error is normal, but the statistic error causes the data to be abnormal.

There are also some tips for finding errors, and share them as follows:

Order of magnitude

Observe the final data to ensure that there is no large error in order of magnitude. For example, by asking the demand side, the number of charges per day is about 30, and if the results are 80 or 8, the result is likely to be problematic.

Summary of dimensions

For different latitude analysis, the total value of each latitude is consistent

Interactive report, the rollup after the drill is consistent with the drill before.

Distribution

The distribution of statistical results often has its characteristics.

For example, the number of pay per day, there will be continuity and periodicity of the two characteristics, continuity is not appear suddenly high and low large changes, cyclical is the weekend will have a certain degree of improvement or reduction, 7 days a repeat.

For example, statistical staff workload, the total amount of work per week by staff summary, generally from high to low a uniform decline.

Sampling

Sampling inspection has been the most effective way. Take the details of the statistics and check the numbers directly.

When selecting the checklist, try to ensure that the sampling can cover each layer, as well as randomness.

The ability to translate data into knowledge

The evolution from data and information to knowledge is as follows:

The difference between data, information and knowledge:

This part of the ability is the hardest to develop. By looking at the analysis reports from all walks of life, learn from the data into the common ideas of knowledge, such as: induction, subdivision, contrast and so on.

In addition, for your daily work, set the fundamental purpose to solve the problem, rather than complete the data requirements. Only in-depth understanding of the operation, in-depth understanding of the first line of data use scenarios, to complete the transformation of data to knowledge.

Wen/Sun Wenliang