1. What inspired this wish list request?
The inspiration for this request comes from a very common and fundamental question in experimental design that is currently difficult to answer prospectively in JMP: "How many samples do I need to collect to detect a meaningful correlation between two continuous variables?"
Currently, when designing a simple experiment to test the relationship between a continuous factor (e.g., Temperature) and a continuous response (e.g., Strength), JMP's powerful DOE platforms are optimized for categorical factors or more complex, multi-factor models. The existing tools to answer this question are either not intuitive or designed for retrospective analysis:
-
Fit Model > Effect Leverage: This is a powerful tool for retrospective power analysis after data has been collected. It is not suitable for prospective planning before an experiment is run.
-
DOE > Sample Size and Power: While this platform has calculators for various scenarios, they are not framed in the way a scientist or engineer typically asks the question.
This leads users to rely on external software, like G*Power, or to simply guess at a sample size, often resulting in experiments that are either underpowered (failing to detect a real effect) or overpowered (wasting time and resources). An example in G*Power is the following settings:
- Test family: F tests
- Statistical test: Linear multiple regression: Fixed model, R^2 deviation from zero
- Type of power analysis: A priori: Compute required sample size given alpha, power, and effect size
Example Use Case: I want to design an experiment to see if <SPAN class="selected">catalyst concentration</SPAN> affects <SPAN class="selected">product yield</SPAN>. Before I run the experiment, I need to know how many samples to collect. I want to be 90% sure (power) that I can detect a relationship if the true correlation accounts for at least 15% of the variation (an R² of 0.15). I also have an estimate of the process noise (RMSE) from historical data. Right now, there is no straightforward, interactive platform in JMP to help me determine the required sample size for this simple scenario.
2. What is the improvement you would like to see?
I propose the creation of an interactive sample size profiler specifically for designing an experiment with a single continuous factor and a single continuous response. This could be a new personality within the <SPAN class="selected">DOE > Sample Size and Power</SPAN> platform.
The profiler would be visually intuitive and function similarly to other profilers in JMP.
Key Features:
Mock-up of the Proposed Feature:
This interface would allow a user to instantly see the trade-offs. For example, they could see how a smaller R² or higher process noise would dramatically increase the required sample size, all before committing resources to the experiment.
3. Why is this idea important?
Implementing this idea would provide significant value by:
-
Closing a Critical Gap: It addresses one of the most common experimental design questions that is currently underserved by JMP's prospective tools. It would prevent users from having to leave the JMP ecosystem for external tools.
-
Promoting Good Statistical Practice: It makes power and sample size determination accessible and intuitive, encouraging users to properly plan experiments. This leads to better, more efficient science and engineering, saving money and resources by avoiding underpowered or overpowered studies.
-
Enhancing JMP's "Thinking Tool" Value: JMP excels at being an interactive tool for statistical discovery. This profiler fits that philosophy perfectly, allowing users to explore "what-if" scenarios for their experimental design in a visual way, rather than just plugging numbers into a static calculator.
-
Ease of Use: It frames the problem in the language that scientists and engineers use (<SPAN class="selected">R²</SPAN>, <SPAN class="selected">RMSE</SPAN>), lowering the barrier to entry for robust experimental design. It is a must-have for anyone doing even basic process characterization or validation.
