By Nadine Amersi-Belton
As a data scientist working for Northwind Traders, I have been tasked with drawing insights from a given database. In particular, I will seek to answer the following questions, which are of interest to stakeholders:
- Do discounts have a statistically significant effect on the number of products customers order? And if so, at what level(s) of discount?
- Do discounts have a statistically significant effect on the amount a customer spends on their order? And if so, at what level(s) of discount?
- Does the average order amount vary significally between customers in different regions?
The Jupyter Notebook is our key deliverable and contains the answers to the above questions.
The data was provided in the form of a SQLite database and can be found in this repository. It was originally provided by Flatiron School in the following repository.
- The relevant data was queried from the table and stored as a Pandas DataFrame.
- Data manipulation was undertaken as required (e.g. creating feature columns).
- Null and alternative hypothesis were formulated.
- EDA and visualisations were created.
- Statistical (t-tests, ANOVA) were used to formally assess the impact of discounts.
- Hypothesis testing revealed that discounts have a statistically significant effect on the number of products customers order. Orders with discounts contain 5 more items on average, which is an increase of 25%. However there are no statistically significant differences between the different levels of discount.
- Hypothesis testing revealed that discounts have a statistically significant effect on the amount a customer spends on their order. On average customers spend USD143 on orders with discounts, which is a 10% increase. However there are no statistically significant differences between the different levels of discount.
- Hypothesis testing revealed that there are statistically significant differences in the average order revenue depending on the customer location.
Photo by Artem Beliaikin on Unsplash