In what situations would you prefer using a non-parametric test over a parametric test, and why?

During my final year in college, I worked on a project analyzing survey data collected from classmates regarding their study habits. The data was ordinal, as it included responses like 'poor', 'average', and 'excellent'. Since the data didn't meet the assumptions required for parametric tests, like normal distribution, I opted for a non-parametric test, specifically the Mann-Whitney U test. This choice allowed me to accurately analyze the differences in study habits without violating the underlying assumptions, leading to meaningful insights that contributed to my project’s success.

While volunteering at a local NGO, I helped analyze feedback from community members about a new program. The responses were mostly categorical, indicating satisfaction levels from 'very dissatisfied' to 'very satisfied'. Given the non-numeric nature of the data, I chose to use the Chi-square test, a non-parametric test, to understand the association between different demographic groups and their satisfaction levels. This approach was effective in uncovering trends without the need for complex assumptions, ultimately helping the NGO improve their services based on community needs.

During my internship at a retail company, I noticed that the sales data for certain products often skewed heavily due to outliers. Recognizing that using a parametric test like ANOVA could misrepresent the true differences in sales performance, I suggested using the Kruskal-Wallis test instead. This non-parametric alternative allowed us to analyze the sales across different product categories without being influenced by outlier sales figures. The results provided a clearer picture of product performance and informed our marketing strategies effectively.