Data from a statistical point of view (1)

We often see a certain industry, a certain company's average wage is 200,000 or something every year, and then if they are in this industry, look at their own wages, some people will not feel a bit confused and dissatisfied? In fact, these are deceptive statistical methods caused.

If a company has 200 people, ordinary staff 180 people, a monthly salary of 3500 yuan, management staff 19 people, the average salary is 5000 yuan, the boss 1 people monthly salary is 200,000, then the average monthly salary of the company is 13625 Yuan, employees a year's average wage a hundred thousand of, But the reality is not the same as the data show.

or our website revision, visual changes, or interactive function changes, daily clicks or visits than before the revision of 200,000 times, then whether the data can be based on this increase to show that our revision is successful? Obviously through the above example we cannot deal with the data so simply and draw a conclusion.

Today, we will discuss some simple and practical statistical methods to help us better understand the meaning of data in our work. From Z-score, T-Test, X2 test, variance analysis to regression equation, there are many kinds of statistical methods, which should be used well. I personally feel that the traditional statistical textbook is not interesting because the concept of the book is too much, divorced from the reality of the statistics, it is difficult to understand, or learn to forget, or encounter problems will not be used. If you can combine the various examples, it should become more clear. So, here we rely on some examples to introduce some common statistical methods and the scope of application, welcome to criticize the guidance.

Also take our example above, a website revision, the new version of the page did not change the original interactive operation, just changed the visual style, user access and click Volume changes, these changes are good or bad?

First, let's take a look at the analysis:

1 What we know is the number of clicks and the amount of data that the user visits before and after the revision

2 We want to know if this change is good or bad

What do you want to do? Calculate the number of users before and after the revision of the percentage and clicks, if the user after the revision of the amount of decline, the number of clicks is not the revision is not successful? Obviously we can't look at the problem so simply. To compare these two samples, we can use T-Test.

The T-Test (Student ' t test) is a test method for the degree of two average differences in small samples (the sample size is less than 30, the overall standard deviation is an unknown normal distribution).

But T-Test needs variance homogeneity to determine the result, but don't worry, the statistics software will help us to check.

OK, so we're going to input the data (this is not for me to say, TXT file on the line) to the statistical software, and then paired sample T-Test (equivalent to a process before and after testing, so using paired sample T test), get the results of the following table (with SPSS, the data are made by me):

We only focus on the yellow part, where the first is mean, STD is standard deviation, t value, DF represents freedom, SIG is P value, in this case, my confidence interval is 95%, so if the sig ". 05" represents a significant difference. From the table, the revision before and after the number of clicks and users two difference is not significant, so we can think that the revision at least did not cause any adverse effects.

Some people may find it pointless to draw such a trivial conclusion, but please think about it, with the amount of data on the increase or decrease in the happy to take credit or dejected plan to modify the program, perhaps the real statistics more can explain the problem, can let us calm down, think about, How we should improve our work.

Of course, the real problem is often more complex, only in the case of revision, we need to consider a number of issues, such as:

1 What has changed? Appearance or interaction? or appearance + interaction mode? How does the layout change? How does the change in interaction change the steps or clicks required to complete a task for a user?

2 Data acquisition before the revision how many days? After the revision of the data collection how many days?

3 the time before and after the revision of the corresponding whoever rescues in each year, the user's access to the amount of significant changes? What is the trend?

...

Here I just give a simple example to share with you the idea of statistics.

Statistics are like a bikini. What they reveal is interesting. But What they hide is vital.

The author of this article: went

Article Source: Ctrip ued