Source: China Statistical net
"Why-What-How" in explaining the concept and execution is a good mental model, the paper according to cases Click to split the data analysis framework. White may not have a lot of ideas for data analysis, for the time being to sort out here from a personal point of view, by reference. To help you better understand this article, first posted a mind map:
A, WHY: Why do data analysis
Personal understanding, data analysis is to be able to quantify a way to analyze business problems and concluded, of which there are two key words: quantification and business.
Quantification is to unify the cognitive and ensure that the path can be traced back, can be copied. In addition to "quantify" One more important word is "business." Only by addressing the problem analysis in order to create business value, including the value of personal values and corporate values.
So, how to stand in the perspective of the business side to think of it, four words sum up " concern they consider, to its desired ":
- Communicate fully
- Conclusion concise
- Provide information and recommendations may fall
- Seeks feedback
In the communication, the sender wants to determine what business analysis , put forward a more reasonable measure and analyze professional manner, while making node synchronization, should a road go black. For example in terms of the business side said long stay depends on the page, but he actually wants, the user may want to measure the quality, the retention rate, conversion rate target is more appropriate indicators.
Elaborate on the analysis results, remember conclusion first, layer by layer to explain, and then provide arguments. Because the business side or management time is limited, a large piece eloquent message, did not see the first faint, but nobody thought to see you in the end what is analyzed.
Providing information and recommendations can be landed on, first to understand what information : Provides information they do not know. Tomorrow the sun rises in the east is not the amount of information, from the West is rising.
Two, WHAT: What is Data Analysis
Nature of the data analysis is to seize changed and unchanged. "Change" is the basis for data analysis, if a single business orders 10,000 a day, or are based on steady growth rate of 10 percent every day, there is no need of the analysis. And if you want to seize the change, we must first form a "change" of consciousness.
Therefore, I suggest that a beginner should form a habit, the first time go to work every day to view the data: real-time & day week Monthly; recording key data (list & reporting)
On the basis of "no change" on the index will be able to gradually develop sensitivity, awareness of that is the ability to deviate from the index. This is primarily maintained by the chain, monitoring week month year on year, and a variety of daily curiosity day. We Questmobile from a list, simply look "deviation index" is how everyday applied to the analysis:
Here to begin share with you how look at this list:
- Look Overall Ranking: APP which the top side to see beyond your unexpected
- Sub-sectors Ranking: Ranking and change of watch industry
- Look growth: What APP grow faster
- Long and other indicators to see when using
Customized data analysis, as well as foreign a book of business analysis is defined as a footnote:
Three, HOW: how data analysis
Any data analysis is " broken, contrast, traceability " These three behaviors continue to cross. The most common breakdown contrast dimension is time, through our week month year period, found that data anomalies, subdivided on the dimension or the process, step by step dismantling find the problem.
1, subdivision
On the way segments, mainly in the following three ways
- Cross : indicators for segmentation and cross-analysis based on a dimension
- Slitting : time axis variation, the downstream segmentation indicator
- Endo : divide from within the object based on a model
Transection
On the cross, we are doing the classification and cross dimensions and metrics, when a certain type of problem indicators, we know from the analysis of what dimension. During cross analysis, often you need to use multiple dimensions crossed.
Slitting
The longitudinal, purposeful path, the analysis funnel. There is no object path, by trajectory analysis. No object without path, the log analysis.
Naisetsu
The endo mainly analyzed according to the conventional common market model, RFM, Cohort Segment, and the like. RFM and most recent purchase time, frequency and amount of three indicators to determine overall customer loyalty and stickiness.
2, compared to
Contrast is divided into the following categories:
- Cross comparison: comparison, such as city and subdivided according to category of cross-cutting dimension
- 纵切对比：与细分中的纵切维护进行对比，如漏斗不同阶段的转化率
- 目标对比：常见于目标管理，如完成率等
- 时间对比：日环比，周月同比；7天滑动平均值对比，7天内极值对比
3、溯源
经过反复的细分对比后，基本可以确认问题所在了。这时候就需要和业务方确认是否因为某些业务动作导致的数据异常，包括新版本上线，或者活动策略优化等等。
如果仍然没有头绪，那么只能从最细颗粒度查起了，如用户日志分析、用户访谈、外在环境了解，如外部活动，政策经济条件变化等等
4 、衍生模型
在「细分对比」的基础上，可以衍生出来很多模型。这些模型的意义是能够帮你快速判断一个事情的关键要素，并做到不重不漏。这里列举几个以供参考：
- Why-How-What
- 5W1H
- 5Why
- 4P模型（产品，价格，渠道，宣传）
- SWOT 模型（优势，劣势，机会，威胁）
- PEST 模型（政治，经济，社会，科技）
- 波士顿矩阵
四、How：数据分析如何落地
以上讲的都偏「道术技」中的「术」部分，下面则通过汇总以上内容，和实际工作进行结合，落地成「技」部分。
1、数据分析流程和场景
根据不同的流程和场景，会有些不同的注意点和「术」的结合
2、数据分析常见谬误
控制变量谬误：在做 A/B 测试时没有控制好变量，导致测试结果不能反映实验结果。或者在进行数据对比时，两个指标没有可比性。
样本谬误：在做抽样分析时，选取的样本不够随机或不够有代表性。举例来讲，互联网圈的人会发现身边的人几乎不用「今日头条」，为什么这 APP 还能有这么大浏览量？
定义谬误：在看某些报告或者公开数据时，经常会有人鱼目混珠。「网站访问量过亿」，是指的访问用户数还是访问页面数？
比率谬误：比率型或比例型的指标出现的谬误以至于可以单独拎出来将。一个是每次谈论此类型指标时，都需要明确分子和分母是什么。
Causally related fallacy: mistakenly when the relevant cause and effect, ignore the mediating variables. For example, it was found that ice cream was significantly related to sales and the number of children drowning rivers, ordered to cut ice cream sales. In fact, it may just be because both occur in hot summer weather.
Simpson's Paradox: Simply put, the difference between the two when packet data more added in the comparison groups are dominant party, will be in the general but the party is losing ground.
to sum up
Data accuracy is the first one, standing on the perspective of the business side of thinking: worry they consider, to its pace, and the definition of "change" and "unchanged", subdivision, contrast, traceability.