3 reasons to learn Bayesian Statistics in the new year

What is Bayesian Statistics

Bayesian Statistics (or Probabilistic Programming) take a more effective and deep approach to perform analysis of any given data and situation. A/B testing is one of the hottest topics on the internet nowadays. In this testing, you simply consider two different groups, A and B, to analyze the performance of both groups. In the end, the results obtained from the analysis are used to determine the best performing group. When I did some interviewing on which companies were using PyMC3 I discovered that a lot where using it in A/B testing. This makes sense, since it’s a fundamentally powerful lever in improving your bottom line. Especially in e-commerce.

What is A/B Testing?

If you want to compare two different web pages to choose the best one, A/B testing provides you with the best outcomes. In the testing phase, the analyst has to create two different versions of the same webpage. Users are requested to visit both pages. The page that generates more revenue and better conversions, wins the race.

Every web page on the internet is created for a specific purpose. A/B testing supports the results and conversions by indicating the best one to meet the end goals. Just consider some typical types of web pages that we mostly visit during our normal internet surfing routine.

  • Online stores want to turn the visitors into their permanent buyers
  • Online service providers want to sell their services and packages to their readers
  • Blogs want their want to get more and more clicks on their ads and subscriptions
  • Educational sites want more enrolments in their offered courses

Visitors are not the main concern here. If you are not getting sales and conversions, alone visitors are somehow useless for the business. Businesses want to turn their ads campaigns into successful sales. The A/B testing helps the businesses to grow their market by suggesting the best possible landing page for their products.

Why Bayesian Statistics in A/B Testing

In a simple analysis, the sample data for the A/B testing is too small and these experiments end before coming to any significant statically proven results. So, the best approach is to go for more accurate results before applying these results to your business.

Bayesian Statistics helps to get rid of common mistakes and errors that we face in the traditional way of analysis and statistics. The results obtained with the help of Bayesian Statistics are easier to understand and better decision making is enabled with the help of this theorem.

3 Reasons to use Bayesian Statistics in A/B Testing

Bayesian Statistics approach is used in A/B testing to get better results with higher conversions. Here are some main reasons why analysts must use Bayesian Statistics for A/B testing. I’ve been heavily influenced by Sean J. Taylor

1.   The ability to borrow strength/ share information

A common feature of Bayesian analysis is leveraging multiple sources of data (from different groups, times, or geographies) to share related parameters through a prior. This can help enormously with precision.

For example in A/B testing you may be doing A/B test by UK, France and Germany (or other groups). This is trickier to do in traditional frequentist statistics, it is even more importantly difficult to do right in frequentist statistics.

2.   Bayesian statistics allows you to model check while building your model

Good Bayesian analyses consider a wide range of models that vary in assumptions and flexibility in order to see how they affect substantive results. There are principled, practical procedures for doing this. This isn’t necessarily the case in machine learning.

Screenshot 2018-11-17 14.43.17
In this model I’ve been able to check if the model is well specified using the posterior plots.

 

3.   Bayesian Statistics allows analysts to interpret posteriors

What a posterior means makes more intuitive sense to people than most statistical tests. Validity of posterior rests on underlying assumption about correctness of model, which is not hard to to reason about. Furthermore you can add loss-functions based on the posteriors, and that will allow you to better explain your results. In A/B testing you may have some sort of loss function that’s not symmetric.

Screenshot 2018-12-15 13.36.42.png
It’s easier to interpret a posterior than a t-test say

Want to learn more?

If you’d like to learn more. I’d recommend you check out my online course where I do an in depth discussion of Bayesian Statistics and Probabilistic Programming.

Leave a Reply

Your email address will not be published. Required fields are marked *