7.2.1: Exploring a Standard Session
A practical, 'hands-on', example
This lesson bridges theory and practice—helping Session Creators apply everything they’ve learned about LLMs and data preparation to their own assessment contexts, and giving a template for confident, insightful analysis moving forward
Introduction to the Session
The example is taken from a school setting, however like all things in RM Compare the lessons learnt are universal. In this lesson we will be using 'dummy-data' from a previously run session described in the short video below which you should take time to watch as an introduction.
Session Data Walkthrough
The data from the session is HERE. It can be copied, linked or downloaded as required. We are using the James Standard Session 1 data set for this lesson.
Remembering what you have learnt so far it is time to inspect. You will see that the spreadsheet contains a number of tabs that have brought together all of the session data extractions. In addition, you will find that this data has been enhance by some Meta-Data taken from other sources.
When ready you can add the data-set to your LLM.
Key questions to explore
There are so many questions you might want to ask about this data-set but here are few suggestions to get you going.
- “Which class had the most Items in the top 20%?”
- “Are there standout performers or outliers?”
- “What is the spread of scores across different classes?”
LLM Tabulated Data Output Example
LLM Response Output Example
Interpretation of Relative Performance
- Highest Overall Median: Class 1 has the highest median score (74.72), suggesting the central tendency of this class performed best on the assessment.
- Highest Score Dispersion (IQR): Class 1 also has the widest interquartile range (IQR = 82.28), indicating the largest spread in performance among the middle 50% of students.
- Lowest Overall Median: Class 4 has the lowest median score (34.75), suggesting the central tendency of this class struggled most.
- Highest Upper Quartile (Q3): Class 6 features the highest third quartile (96.90), meaning a quarter of their students achieved scores between 96.90 and 98.61, indicating very strong high-end performance.
- Highest Minimum Score: Class 7 has the highest minimum score (18.29) and the highest first quartile (36.95), suggesting that while their very top performance is not as high as others, their performance baseline is considerably higher, with fewer students scoring in the extremely low range.
Generating Visualisations with an LLM
Example prompt:
"Using the uploaded CSV file, create a box and whisker chart to show the relative performance of each Class using the scaled scores.
- The class information is in under the 'School' column.
- The scaled scores are under the 'Params as Score' column."
LLM Chart Output Example
Interpretation, Reporting, and Actionable Insights
- What does this data show?
- What further questions might James ask in his role as Head of Department?
- There are 7 classes. What key actions should each class teacher consider (tabulate)?
LLM Response Example
Reflection Prompts for Lesson 7.2.1
1. Interpreting the Data
- Which class or cohort showed the strongest average performance? Which showed the weakest?
- Was there greater variation within certain classes, and what might explain this pattern (e.g., teaching style, cohort composition, attendance)?
2. Spotting Opportunities for Intervention
- Did you identify any underperforming groups or outliers?
- What might be the causes (e.g., support needs, question misfit, curriculum gaps) and what first steps might you suggest for intervention?
3. Informing Future Teaching Practice
- How could these insights influence how teachers approach the next topic or the targeting of resources?
- Could high-performing approaches from one class be shared or scaled to others?
4. Student & Stakeholder Communication
- How would you communicate these results to teachers, students, or leadership without demotivating or wrongly singling out individuals?
- How could this data be used in feedback meetings, departmental reviews, or development planning?
5. Closing the Loop
- Do you see new questions emerging from this analysis that you could investigate with another data slice or future session?
- How might you track improvement over time using repeated RM Compare sessions?
What this means for you:
This lesson should help you to better understand the current strengths and limitations of LLM technology. Things are changing fast as the models improve, however the output always needs human inspection, interpretation and correction. The LLM is your assistant not your replacement.