Module 10 Lesson: Non-Parametric Thinking and Assumption Caution
Classroom Explanation
Most standard summaries assume the values are not wildly lopsided. When they are, methods based on ranks describe the data more honestly than methods based on magnitudes.
In this module, the goal is to recognize when common tests need assumption caution and simpler alternatives. The important part is not only the calculation. The important part is the thinking before and after the calculation.
Why This Matters
The common mistake is calculating an average rank and treating it as an exact measurement. This mistake can make a report sound more confident than the evidence deserves.
Key Ideas
1. Assumptions describe what must be reasonably true for a method to be trustworthy.
Assumptions are conditions behind a method. If they are badly broken, a precise answer can become precisely wrong.
2. Ordinal ratings have order, but the gaps between scores may not be equal.
Ordinal values have order, but the distance between levels may not be equal. Treating them like exact measurements can overstate the result.
3. Ranking can be safer when data is skewed, ordinal, or strongly affected by outliers.
Ordinal values have order, but the distance between levels may not be equal. Treating them like exact measurements can overstate the result.
Worked Example
Satisfaction ranks from 1 to 5 are ordered. A 4 is higher than a 3, but the distance from 3 to 4 may not equal the distance from 1 to 2.
Here is the dataset used in this module.
| response_id | group_name | satisfaction_rank | completion_time_minutes | unusual_flag |
|---|---|---|---|---|
| O01 | Original | 3 | 18 | no |
| O02 | Original | 2 | 22 | no |
| O03 | Original | 4 | 19 | no |
| O04 | Original | 3 | 21 | no |
| O05 | Original | 1 | 35 | very slow completion |
| O06 | Original | 3 | 20 | no |
Look at the shape before choosing a method. On strongly skewed data a mean can sit above almost every observation, and a rank-based summary describes a typical case — at the cost of telling you nothing about the total, which is sometimes what the decision needs.
How To Think Through It
- Look at the distribution of values.
- Identify anything extreme and decide how to treat it.
- Choose between magnitude-based and rank-based summaries.
- Report the summary with its spread.
- Say what your chosen summary does not tell you.
Common Mistake
The common mistake is calculating an average rank and treating it as an exact measurement.
How To Write The Result
The updated group appears to have higher satisfaction ranks, but because the data is ordinal, we should describe the direction cautiously and consider rank-based methods.
Practice Prompt
Identify why a standard mean comparison may be weak for satisfaction ranks.
Takeaway
Rank-based thinking trades detail for robustness; say which you chose and why.