Sign in or start a free trial to avail of this feature.
6. Preparing AB Trial Results
In this lesson, we will consolidate the historic and trial data, performing any necessary formatting and aggregation to prepare for analysis.
Preparing Trial Data
- In order to run the AB analysis, we need to merge our trial data to match our historic data, however the trial data must be correctly formatted first
- Trial data needs to be separated into the same sample groups as historic data
- Trial data must be aggregated to the weekly level
So far,we've determined the 20 Ben's Beverages stores that are most appropriate for the initial six-week new store design trial.
This group of stores is known as the treatment group.
Each treatment group store has been matched with four stores which will not be changed, or a control group.
The six-week trial has now concluded, and the managing director of Ben's Beverages has sent us over the results.
Our goal on this lesson is to prepare the results for our final analysis.
We'll achieve this through three key steps.
First, we'll import and format our trial data.
Next, we'll separate the treatment stores into groups, just as we did with the historic data.
As a final step, we'll combine our trial and historic data so we can analyze the results.
We'll begin by connecting the trial receipt data set, then connecting both an auto field and select tool, and running the workflow.
We'll then change the data type for the store field to V string, as we've done previously.
We'll eventually want to aggregate this data to the weekly level, so we'll need to determine the week for each sales transaction, like we did for the historic data.
We'll copy the week date formula tool in the historic sales data container, paste it, and connect it to our trial receipt data work stream.
We're now ready to move on to step two, and separate the stores from the trial into the sample groups.
We'll do this by joining the trial data with the grouped high level store data.
We'll bring down a join tool, connect the trial receipt work flow to the right input node, and the tile tool from the store data container, to the left input node.
We'll specify to join on the field store, and run the workflow.
If we click on the J node, we can see that we have sales and receipt data for each of our stores over the trial period, together with the high level store information and tile numbers.
In order to compare this trial data with our original data, we must aggregate it to the weekly level.
We'll do this now by bringing a summarize tool onto the canvas, and connecting it to the J output node of the join tool.
We'll navigate to the configuration window, and group by store, tile number, format, range, and week date.
We'll also sum by open, in store promotion, receipts, and sales.
We'll then run the workflow again.
We're now ready to move on to our final step, and consolidate this new data with our old data set so that we have a complete copy of the weekly data for the entire period.
We'll bring a union tool onto the canvas, and connect it to both the summarize tool we just added, and the summarize tool from the historic sample groups container.
Now that we've consolidated our data, we'll stop the lesson.
In the next lesson, we'll analyze the results of the trial.