A lot has been said lately about how to apply statistics and machine learning techniques for SEO projects, although the truth is that there are not many real cases where some of these disciplines are applied. At Internet República we have built a web app to measure if an event has had a statistically significant impact on the SEO ranking of a web. We explain you how:
How does it work?
First of all, we have relied on the library “CausalImpact” developed by Google, that works by estimating from a pile of data whether an intervention in a specific moment has a causal impact. To find this out, it generates a model to predict the trend that should be followed depending on the data previous to the event, and it compares it with the real data.
To obtain the data to analyse, we use the API of Search Console, with which we analyse the magnitudes of this tool: impressions, clicks, CTR and average position of a maximum of 90 days (the maximum period that Search Console offers).
The good thing about this tool is that you need not understand how it works. The tool itself calculates all the statistics, offering the results after the analysis.
How to use it?
To use it, first of all the tool has to be granted access to your Search Console account clicking on the blue button “Login via Google.”
After that, choose from the list the web you want to analyse, and very importantly, indicate the event and metric to analyse following this procedure:
The tool has three dates selectors. The first of them, “Select the dates for the analysis”, loads by default the last 90 days to obtain as many data from Search Console as possible (recommended). On the second selector, “Select the dates previous to the event to be analysed” we have to indicate the period previous to the event we want to try out. This will define the data to generate the predictive model to measure the impact. Finally, the third selector is for “Select the dates ulterior to the event to be analysed”, the period where we want to see if the event has generated changes that are statistically significant for the data.
Clicking on “Obtain data from Search Console” we would generate the report by default, which is made based on web searches and clicks, although we can also choose other values or apply filters to limit the study just as we can do with Search Console under the section “Data filter.”
The results are what we get from the library “Causal Impact” and which are divided in two different parts: graph and summary. The first part is divided into a first panel with the tag “original”, where the results obtained from Search Console and the prediction made by the library are displayed. The second part is Pointwise, that shows the difference between the real data and the predicted data, whereas the third includes the accumulative data of the difference displayed on the second panel.
Finally, the summary offers a detailed explanation of the graphs together with the conclusion, i.e. if the results obtained after the event are those which could be predicted before said event or, on the contrary, it has had an impact on them.
This has many practical uses, as it helps us check out easily if the SEO changes that have taken place in a web or URL have had a real impact on the ranking. We also have to bear in mind that there might be other variables affecting the data in real time that can render them false. We have to analyse in detail everything that may have happened during that period before reaching definitive conclusions.
We hope you will find this useful.
NOTICE: You can find the application on a free hosting, and it has a limit set to the hours it can be used monthly, so permanent availability is not guaranteed.