Interacting with Your Data

At Select we regularly analyse data to help our clients gain key insights into their business so that they can make effective decisions.

In many cases, clients will often be interested in the results for different scenarios (for example, ‘what happens if we change this input…’ or ‘how much stock is required if…’) and will want to demonstrate these results to other decision-makers within their organisation.

Enhancing our Service

To enhance this type of scenario planning, over the last few months we’ve been developing the capability to enable our clients to interact with their data via online interactive tools or web applications which they can access anywhere and at any time.

There are lots of different tools available to create bespoke web applications, but the tool we’ve focussed on is Shiny, a package that allows you to create web applications using the statistical software, R.  We use R to carry out the majority of our statistical analyses (if you’d like to find out more about R, why not read our introductory blog) and Shiny provides the perfect platform for combining the statistical functionality and great graphics from R to easily produce a fantastic online resource for use directly by our clients.  Essentially, Shiny provides our clients with a direct interface to our software and models allowing them to easily interact with the statistical codes that we have written, to re-do  analyses and to explore what would happen as they change the various inputs going into the model.

Presenting results via an interactive web application is also a very effective way of communicating the results of an analysis to others.  Not only do our clients continue to receive a report including details of the methodology and results, but they can now also use the interactive graphics in their presentations to help stakeholders better understand the results and their implications.

Split-ab-testing-web-app

A/B split testing Shiny example

A Shiny Example

We’ve put together a simple Shiny example based on our split A/B testing case study.  In this example visitors to a website are assigned to one of two groups.  In the first group visitors are directed to one version of the webpage with a new design scheme and the second group are directed to the original design.  To assess the effectiveness of the new website design we can compare the revenue and conversion rates between the groups for different products sold on the website to see whether there is evidence of a difference.

Shiny provides the perfect platform for combining the statisticial functionality and great graphics from R to produce a fantastic resourse for use directly by our clients.

Our A/B split testing example has three inputs that allow the user to select which product and metric they are interested in and at what significance level they would like to test at.  The first tab provides the results of the split A/B test and we also plot the distributions so that you can see the extent to which they overlap (the smaller the overlap, the greater the evidence that there is a difference between the two groups). The user can examine how different significance levels affect the result of the test. For example, there is evidence of a difference in the mean revenue of product B at the 95% level, but not at the 99% level.

The second tab provides an alternative way to visualise the data by plotting the daily time series of the revenue or conversion rate for both groups.  These types of plots are useful for spotting interesting patterns in the data such as seasonality or outliers.

Finally, you can view the underlying data in the Data tab and search the data for particular products or groups.  You can select the arrow at the top of each column and order it in either ascending or descending order.  This means that a user can easily identify any data points that are outliers.

This example is relatively simple, but effectively communicates the results of an A/B split test when trialling a new webpage.  The user can easily choose which product or metric they are interested in and experiment to see how the significance level affects the final result.  It’s also easy to drill down into more detail as the underlying data are available in a table that can be ordered depending on the variable of interest.  We’ve only used a few of Shiny’s capabilities when putting together this example, but there are many more features available that gives us the flexibility to create an online resource to suit each individual client’s needs, allowing them a whole new level of access to their data.