How does one determine whether to use a parametric or non-parametric test when analyzing data?

During my final year of college, I worked on a project analyzing student performance data across different teaching methods. We collected test scores from students taught via traditional lectures and those taught through interactive sessions. Given the small sample size and the non-normal distribution of scores, I chose to use the Mann-Whitney U test, a non-parametric method, instead of a t-test. This decision allowed us to draw valid conclusions about the effectiveness of the teaching methods without violating the assumptions of parametric tests. The findings highlighted the advantages of interactive learning, contributing to our final presentation and receiving positive feedback from our professors.

As a volunteer for a local charity, I helped analyze feedback from a community event. We distributed a survey with ratings on a scale from 1 to 5. The data collected was ordinal, meaning it didn't meet the assumptions required for parametric tests like ANOVA. Instead, I applied the Kruskal-Wallis test, which suited our data's nature. This choice enabled us to compare satisfaction levels across different groups effectively and provided insights for future events, ensuring we focused on improving areas that received lower ratings.

In my first job as a data analyst, I often encountered situations where I had to decide between parametric and non-parametric tests. For instance, when analyzing customer satisfaction survey data, I noticed it was skewed and had outliers. Instead of applying a t-test, I opted for a Wilcoxon signed-rank test. This experience taught me the importance of thoroughly assessing data characteristics and understanding when to apply the appropriate statistical methods, ultimately leading to more reliable insights for our marketing strategies.