Trend Analysis of Time series

Whether it is a Web analytics tool, BI Report or data, it is difficult to see the data isolated point of view, usually the data is in the form of sequence, grouping and so on, the reason is very simple, we can not find from a single data, the data used for analysis must contain context. The context of the data is like setting one or more reference lines for each metric, through these reference and comparison process to analyze the pros and cons of the data, like the physical example of high school, if we do not use the ground as a reference, we can not tell whether the train is still or moving, north or south.

In the actual view of the data, we may have inadvertently used the context of the data, trend analysis, scale analysis, segmentation and distribution are all we set the appropriate reference environment for the data. So this side through a topic--the context of the data, to summarize and organize our daily data analysis can be used in the data reference system, the previous several mainly based on the internal baseline (Internal Benchmark) of the formulation, followed by the external baseline (External Benchmark). Today this is the first, the main introduction of the trend based on time series analysis, repeat the next year and the chain, before the site of the new and old user analysis of this article, has been used and the chain of simple application examples.

The definition of year-on-year and chain

Define this thing here or a few more words, because do not understand the definition can not be applied, familiar friends can skip.

Year-on-year: In order to eliminate the impact of cyclical fluctuations in data, the data in this cycle and the previous cycle of the same time point of data comparison. Early application is the sales industry and other affected by the season more serious, in order to eliminate the seasonal impact of trend analysis, introduced the concept of year-on-year, so more is the year's quarterly data or monthly data compared to the same period of the previous year, calculate the year-on-year growth rate.

Chain: The response is the trend of continuous data changes, the current period of data and the previous cycle of data comparison. The most common is this month's data compared to last month's data, which calculates the quarter-on-quarter growth rate, because the data is compared to the previous one, so it is used to observe the continuous changes in the data.

Buy two to send one, and then give a concept-the base ratio (in fact, Baidu Encyclopedia included): all the data with a baseline data comparison. Usually this baseline is a milestone or important point in the development of a company or product, comparing the following data with the baseline, thus reflecting the company's development in the context of this important basis.

The application environment of the year and the chain

In fact, the chain does not have a strict scope of application or targeted applications, all need to analyze the time series on the changes in the data or indicators can be used year-on-year and the chain.

But my suggestion is to set up the data context of the year and the chain for the target index of the website, such as the site's revenue, the number of active users of the site, the number of key actions of the site, such indicators need to be clear long-term growth trend, and the chain can provide a strong reference for the overall development of the site operation.

Another suggestion is not to be the same year and the chain of the most primitive or the most common application of the shackles: Year-on-year is the annual data per month or quarter compared with the same period last year, the chain is this month's data compared with last month. For the application of the method needs to be based on the actual application of the environment, make reasonable modifications, choose the most appropriate way. So the year and the chain does not necessarily take years as the cycle, it is not necessarily the monthly, quarterly time granularity of the statistics, we can choose any suitable cycle according to the need, such as your company's product operations in weeks, months, or even a specific number of months of the year for cycle changes, which can be used as a cycle.

Especially for the fast-changing environment of the Internet, commonly used year to year, the quarter or month as the size of the statistics may no longer fit, in order to adapt to rapid changes in the monthly cycle, week cycle of the year, to the days of granularity, the size of the hour of the statistical data to be the chain will become a common way, Because to adapt to this rapid change, we need to make more rapid decisions and adjustments, of course, the data to adapt to the need for rapid decision-making.

Application examples

And the chain is widely used in various fields, in Google's pictures in the search and the chain will have a rich list of the year-on-year chain of the chart shown in front of you, so here is a simple example: because a lot of Internet product data changes will be "Week" For the cycle of fluctuations (the weekend will be a significant increase or downward trend), so here to a week of data as an example to see the Year-on-year and the chain show effect. or virtual data that is temporarily set up for display needs:

Monday Tuesday Wednesday Thursday Friday Saturday Sunday last week earnings 113 134 123 145 137 196 187 Week earnings 129 122 134 149 146 215 208 year-on-year Growth 12.4%-9.84% 8.21% 2.68% 6.16% 8.84% 10 .10% quarter-on-quarter Growth-44.96%-5.74% 8.96% 10.07%-2.05% 32.09%-3.37%

From the diagram you can see the trend of the data in the week, weeks and weekends there is a clear difference between the weekend's earnings will have a significant increase in the use of the time when the need to catch this kind of data cyclical changes in the law, so that the comparison of data can more effectively reflect the changes in data. Also, in Excel, you can draw a trendline directly for a set of data based on time series, as the dotted line shows, this week's trend in earnings is obvious, with an exponential fit, and the trend line in Excel provides linear, exponential, logarithmic, idempotent regression analysis, It also includes the methods of trend analysis such as polynomial and moving average.

Finally, we often use the web analytics tool inside there is no year-on-year and the chain function? Here to Google Analytics and Baidu statistics for example cut two pictures, first look at Baidu statistics log in after the website overview:

Baidu statistics by default for us to provide a comparative environment, the above table is today and yesterday's data contrast and changes, but also provides a predictive function; the line chart below shows changes in data per hour, The same day before or last week (Baidu may have realized that most sites will have a trend of week-changing cycles, so many places provide reference data in weeks) for each of the whole point of data control, and can choose different time intervals and various indicators. And look at Google Analytics's dashboard:

Google does not like Baidu as a go in to see the contrast data, we need to choose manually, in the time interval of the selection interface provides the "Compare to Past" check button, if the default is nearly one months of data, then the reference data is pushed forward one months of daily change data, Timeline's selection panel is very cool, can be customized to select any valid time range, of course, also provides different reference indicators, the mouse to move to the corresponding date in the diagram after the point will show the specific data and the size of the difference.

Year-on-year and chain is the most simple and intuitive trend analysis method based on time series, by observing the changes in key indicators to gain insight into the development and operation of the site, while measuring the degree of achievement of the goal. So the theme of this article is to use trend analysis method to set the context for the site's goals, the next one will focus on the KPI indicators for the selection and setting of the data context.
Source Address: http://webdataanalysis.net/personal-view/time-series-analysis/