This repository contains an A/B test analysis comparing two bidding strategies — Maximum Bidding and Average Bidding — based on user conversion performance. The analysis was conducted using a dataset simulating user behavior on a fictional e-commerce platform.
The project walks through data preparation, statistical assumption checks, and hypothesis testing to evaluate whether the new bidding strategy leads to significantly more purchases.
To determine if the new bidding method (Average Bidding) results in a higher average number of purchases compared to the existing method (Maximum Bidding).
The dataset includes two groups:
- Control Group: Exposed to Maximum Bidding
- Test Group: Exposed to Average Bidding
Each group includes the following metrics:
Impression: Number of ad impressionsClick: Number of clicks on the adsPurchase: Number of purchases after clickingEarning: Revenue generated from purchases
- Data loading and exploration
- Combining control and test groups
- Defining null and alternative hypotheses
- Running normality and variance homogeneity tests
- Performing an independent samples t-test
- Interpreting test results
- Statistical tests (Shapiro-Wilk, Levene, and t-test) were conducted.
- p-value = 0.3493 → No statistically significant difference in average purchases between groups.
- Recommendation: Since no meaningful uplift was observed with the new strategy, sticking with the current method may be more cost-effective unless other business metrics suggest otherwise.
- Python
- Pandas
- NumPy
- SciPy
- Matplotlib