In your experience, how do you determine whether to use a parametric or non-parametric statistical test for your analysis?

In my statistics course, I worked on a project analyzing survey data to assess student satisfaction with online learning. We collected responses on a Likert scale, which wasn't normally distributed. I recognized that using a parametric test, like t-test, would be inappropriate due to the violation of normality. Instead, I opted for the Mann-Whitney U test to compare two groups. This choice allowed us to accurately interpret the results without making unfounded assumptions about the data, ultimately leading to a clear understanding of student sentiment.

As a volunteer for a local charity, I helped analyze the effectiveness of a community health program. We had gathered clinical data from participants, but the sample sizes were unequal and some data points were outliers. I decided to go with a non-parametric test, specifically the Kruskal-Wallis test, to compare the health outcomes across different demographics. This approach was crucial because it ensured that our findings were robust despite the data's irregularities, providing reliable insights that helped the charity secure further funding.

During my first job as a data analyst, I was tasked with reporting on customer feedback scores from various product lines. While most data were normally distributed, some product lines had skewed distributions due to a few extreme scores. I carefully assessed the data and decided to use both parametric and non-parametric tests. For normally distributed data, I utilized ANOVA, while for the skewed data, I used the Wilcoxon signed-rank test. This approach allowed me to provide a comprehensive analysis to my team, ensuring that all data was accurately represented and actionable.