Can you describe a scenario where you would prefer using a non-parametric test over a parametric test?

In my statistics class, we conducted a survey to understand student preferences for online versus in-person learning. After collecting the data, we realized that the responses were on an ordinal scale (e.g., ranking preferences from 1 to 5). Since the data wasn't normally distributed, we chose to use the Mann-Whitney U test instead of a t-test. This allowed us to accurately analyze the differences between the two groups without making strict assumptions about the data distribution. The results helped us conclude that a significant number of students preferred online learning, which we presented in our final project.

While volunteering for a local non-profit, I assisted in gathering feedback from community members about a new program. The feedback was collected using a Likert scale, which provided ordinal data. Given the small sample size and the non-normal distribution of the responses, we opted for the Kruskal-Wallis test to analyze the differences between different demographic groups. This approach not only respected the data's nature but also allowed us to draw meaningful insights that helped improve the program based on community needs.

In my first job as a junior analyst, I worked with sales data that included several outliers and was not normally distributed. Instead of using a parametric approach, I applied the Wilcoxon signed-rank test for paired samples to compare sales performance before and after a marketing campaign. This non-parametric test was suitable as it didn’t rely on the assumption of normality, allowing us to uncover significant improvements in sales attributed to the campaign. This experience taught me the importance of selecting the correct statistical test based on data characteristics.