Sample decentralized processing SPSS

1. According to Hou Jietai: the so-called centralization refers to subtracting the mean value of a variable from its expected value. For sample data, each observation value of a variable is subtracted from the sample average value of the variable, and the transformed variable is centralized
for your question, subtract the mean from each measurement.

2. Centralization is to subtract the mean and Z-score is to divide it by the standard deviation. Both of them are centralization methods.

3. First, the goal of refined operation. For example, if your proct is just a tool, I'm afraid it can't be said that there are too many refined operations. Generally, it's enough to do a good job in routine user behavior analysis, and then cooperate with user qualitative research to guide proct design; If it is a content-based proct, or a proct with both function and content, it really needs to be considered. 2. Design statistical framework suppose that users will frequently interact and use functions on your app, and browse or generate content at the same time, so you need to design your statistical framework well while designing procts. 2、 Data collection first lists the data items you need, then evaluates which parts need to be reported by app and which parts can be counted in the background, and then adds them on the front and back platforms. Generally speaking, the collected data reported by app must be carefully checked and tested before release, because once the version is released and the data collection goes wrong, not only the previous efforts are wasted, but also a lot of dirty data will be brought. At the same time, the running efficiency of the client may be reced, and the gain is not worth the loss. 2. After data collation and data collection, all kinds of original data need to be processed into intuitive and visible data needed by proct managers. Here, we need to do some basic data logical association and display, so we won't repeat it. 3. Data analysis according to the statistical framework designed at the beginning, you can clearly see the data you need. Of course, the above is just a more basic analysis. If you get these data, you can analyze that users who use a function also like B function. The two are closely related. Can you consider more integration or interface adjustment in front-end design; For example, by analyzing the click stream, what are the paths for most users to visit or use the app, and do they hide the core functions too deeply? For another example, we can analyze different user attributes, such as male users and female users. Do they have significant differences in user behavior? wait. There is a big gap between the data analysis methods and models of different procts, which cannot be explained at once. So the above are more examples. 3、 Some principles need to be noted: 1. The data itself is objective, but the data interpreted must be subjective. The same data analyzed by different people is likely to draw completely opposite conclusions, so we must not analyze it with opinions in advance (for example, if we have hypotheses, we can use the data to demonstrate them); 2. Data collection by app must be of low priority. It can't affect proct performance and user experience because of data collection, and it can't collect user's privacy data (although many domestic apps don't do this); 3. Data is not omnipotent. You should trust your own judgment.

4. The more troublesome
needs to understand the theory

5.

If you can't sell it, you can only keep it by yourself. Let's wait for appreciation or take over. Of course, the probability of this happening is very small. If it happens, there may be the following possibilities:

1. China, the United States or the European Union suddenly announced the ban on bitcoin and its circulation

2. Bitcoin exposes fatal weaknesses and defects, which are difficult to overcome, especially security factors

3. Bitcoin has not been used as a killer for a long time, and its application scenarios are strictly limited, so people graally lose information about bitcoin

4. The emergence of a better alternative to bitcoin or the global joint issue of a virtual currency has won global recognition

6.

SPSS can eliminate multicollinearity by stepwise regression analysis

1. The importance of explanatory variables is ranked according to the size of resolvable coefficient

2. Based on the regression equation corresponding to the explanatory variable which contributed the most to the explained variable, the other explanatory variables were introced one by one according to the importance of the explanatory variable. There are three situations in this process

(1) if the introction of a new variable improves the R-square and the t-test of the regression parameter is statistically significant, the variable is retained in the model

(2) if the introction of a new variable does not improve the R-square and has no effect on the t-test of the estimated values of other regression parameters, it is considered that the variable is rendant and should be discarded

(3) if the introction of the new variable fails to improve the R-square, and significantly affects the sign and value of the estimated values of other regression parameters, at the same time, its own regression parameters can not pass the t-test, which indicates that there is a serious multicollinearity and the variable is abandoned

extended data:

other methods to eliminate multicollinearity:

1, &8194; Direct combination of explanatory variables

when multicollinearity exists in the model, the relevant explanatory variables can be directly combined without losing practical significance, so as to rece or eliminate multicollinearity

2 、 Combining explanatory variables with known information

Through the deep understanding of theory and practical problems, additional conditions are introced to explain the multicollinearity, so as to weaken or eliminate the multicollinearity

3, increasing the sample size or re sampling

This method is mainly suitable for multicollinearity caused by measurement error. When re sampling, the measurement error is overcome and multicollinearity is eliminated. In addition, increasing the sample size can weaken the degree of multicollinearity

7. The requirement is at least 20 samples and 10 variables
1. Principal component analysis (PCA) is a linear transformation of the original variables; Factor analysis lies in the analysis of the original variables. Attention is to analyze and decompose them into common factors and special factors
2. The new variables obtained by these two methods, namely components or factors, are not the remaining variables after screening or proposing the original variables
3. Factor analysis can only explain part of the variation (common factor), and principal component analysis can explain all the variation (if all the components are extracted)
4. Principal component analysis, there are at least a few components in a few variables, generally only extract components that can explain more than 80%; In factor analysis, there are several variables, not necessarily several common factors, because the factors here are common factors. The potential existence of each variable needs to be decomposed from each variable, and the unexplained part is the special factor
5. The impact of SPSS factor analysis on dimensions and units among variables is automatically standardized by default, so it is not necessary to carry out data standardization separately before the start, because the results of standardization are the same
6. Important result of SPSS factor analysis: kmo value. Whether this value is calculated is related to the number of variables and samples, and may not be displayed every time. If there is no such result, it can be achieved by adjusting the proportion of variables and samples.