The setting of user satisfaction problem

Little Story (a little bit long, a little round, some patience)

When you see apples and oranges at the fruit stand, are you worried about which one to buy? If you buy both, the problem will be solved. But there are many cases that need to be compared and cannot be handled in this way.

For example, I would like to participate in a postgraduate class, and finally selected the Mountain Palace training and new western schools, two institutions have math and English tutoring, and must be two subjects at the same time enrollment. Mountain Palace Training English teachers and the new western school math teacher's level has been widely recognized in the industry, so they are the other section of the guidance effect to do publicity focus, take out the http://www.aliyun.com/zixun/aggregation/8319.html " > Data speak, Mountain Palace Training said, "Our postgraduate training course, 460 students, the mathematics test highest score 145, the lowest score 95"; The new Western school said, "Our postgraduate class, 450 students, English test the highest points 140, the lowest points 100". I'm haunted, two institutions. How to measure the teaching level of different subjects and how to choose the students ' comprehensive quality?

Mountain Palace Training Mathematical average = 120, standard deviation = 15; New Western school English average = 125, standard deviation = 18, two institutions of students two subjects to graduate school scores have different mean and standard deviation. Statistically, the mean and standard deviation are not the same, and the value of one variable cannot be compared with the value of another variable.

One way to compare this is to change the scores of students in the middle of two institutions to the standard score (standard scores). First, the standard score is a relative distance value, remove the special effects of internal factors; Secondly, the median can reduce the effect of high and low, when there is no extreme value, the median will be similar to the mean, according to statistical experience, the two subjects to determine the middle position of the students, can represent the teaching level of the subject teacher.

In this case, the median number of training mathematics in the Mountain Palace is divided into 129, the standard score = (129-120)/15=0.60, and the median of the new Western school English is divided into 130, the standard score = (130-125)/18=0.28. Judging from this, I went up the Mountain Palace training in the postgraduate class will be better.

In fact, the direct use of discrete coefficients (standard deviation/mean) can be judged, or the use of functional transformation (linear, logarithmic, cotangent, etc.) can be judged. The application of different methods will have the prerequisite, which is the key to see what the problem to solve.

Point

In the daily operation of the website, in order to collect the user feedback of different products, we can hang out the feedback entrance in each product's page, which can help the product manager to listen to the user's voice indirectly, and understand the pain point of the product and obtain the improvement demand. At the same time, the feedback questionnaire can be added to the topic of satisfaction, to a certain extent, the user's feelings quantified, when the amount of feedback is too large to digest quickly, the quantification of the satisfaction of the topic can be faster to promote product improvement.

Set up the satisfaction problem in the user feedback questionnaire, pros and cons, the advantages are as before, can quickly grasp the feelings of some users, but also to long-term continuous access to some of the user's evaluation of the product; The main problem is that the data does not represent all users, the feedback entry copy, location, the user attributes of the product itself and other factors. Although the majority of users are silent, the study of satisfaction shows that users who are willing to make sounds are more likely to turn into loyal users, and it is these fans that have contributed to the development of the site. Therefore, the feedback questionnaire is still a more friendly way to obtain user experience.

Different product feedback questionnaire have satisfaction score, how to compare? Take Taobao as an example, landing page users are the vast number of buyers, sellers; The user of the rights page is mainly the buyer who encounters the transaction dispute; There are too numerous, these pages have user feedback portal, feedback questionnaires are satisfied with the topic, these product satisfaction scores for horizontal contrast, is it possible?

Everyone will have concerns, first of all products of different user groups, followed by the products to fill in the feedback questionnaire users with different emotions, and again the product to fill in the feedback of the will to fill in the feedback of the opportunity are not the same, simply to compare, certainly not appropriate.

The opening of the story provides a comparison of ideas, of course, this also needs to have a major premise is that each product satisfaction score in the middle of the user, can represent the average level of this product. With this premise in place, the analysis can be carried out under the existing conditions, otherwise there will be no consensus but stagnation. After standardization of product satisfaction data, the concerns mentioned in the previous paragraph can be weakened.

The setting of the satisfaction problem

Because is in the user feedback questionnaire to set up the satisfaction question, this kind of questionnaire is different from sends the invitation letter the formal research questionnaire, the user mainly is for the feedback opinion, therefore is not suitable to set up the multi-channel satisfaction question.

When comparing the satisfaction of each product, the index should be unified as far as possible in order to obtain the comparative data of multi-angle. The proposal to find the site's most critical 3-4 common components, such as typesetting, function, content, vision, and overall satisfaction, constitutes a simple satisfaction evaluation criteria.

Degree of satisfaction with the topic of five scale, or level seven or 10 scale can be developed according to research purposes, the indicators of each product to maintain unity. Or back to the background of this case, the satisfaction metric in the user feedback questionnaire is preferably level five, which is relatively easy to choose. Berdie (1994) According to the research experience, pointed out that in most cases, the five scale is the most reliable, the internal consistency is better, the option is more than five levels, the average person is difficult to have enough discernment.

The computational process of satisfaction score

1, cleaning the satisfaction of data (including the heavy, random answers, logic errors, etc.). This step can be properly interpolated based on the lack of data. Foreign scholars have studied that the missing value of level five scale is appropriately interpolated 3, and the negative effect on the result is less.

2. Standardize each index of a single commodity and find the median.

3, the median of the standard score into percentile. Available T=70+30*z (median) or t=60+40z (median), such conversion, similar to the effect coefficient method, each percentile fraction can be compared.

4, according to the sub-item and the overall satisfaction weight, calculates the final score. Weights can be calculated by statistical calculations, or by experience in advance.

This operation, as long as the calculation process is complete and reliable, the rules remain unchanged, can be multiple product satisfaction to do horizontal, vertical comparison.

Summary

The product satisfaction comparison mentioned in the paper is only a workaround under the existing conditions. If there is a condition, you can every six months or quarterly, or even monthly, product random user, put a complete satisfaction questionnaire, research data obtained through reasonable statistical analysis, horizontal, vertical contrast will be more convincing.

* Thank Wu Ling for the suggestions of the calculation methods involved in this paper.

Source Address: http://ued.taobao.com/blo......satisfaction/