Calculate SST using MATLAB

An interactive tool and guide for Significance Sensitivity Testing (SST) in MATLAB.

SST Calculator for MATLAB

Formula Used:

The Significance Sensitivity Test (SST) quantifies how much an output changes in response to a small change in an input parameter. It is calculated as the ratio of the relative change in the output to the relative change in the input parameter. A higher SST value indicates greater sensitivity.

SST = ( (Output_Perturbed – Output_Baseline) / Output_Baseline ) / ( (ParameterA_Perturbed – ParameterA_Baseline) / ParameterA_Baseline )

Simplified, if ParameterA_Baseline is the original input and ParameterA_Perturbed = ParameterA_Baseline + PerturbationValue:

SST = (Relative Change in Output) / (Relative Change in Input)

Where: Relative Change in Output = (Perturbed Result – Baseline Result) / Baseline Result

And: Relative Change in Input = (Perturbed Parameter A – Input Parameter A) / Input Parameter A = Perturbation Value / Input Parameter A

SST Data Table

SST Calculation Data

Input Parameter	Value	Unit	Role
Parameter A (Baseline)	–	N/A	Original Input Value
Parameter A (Perturbed)	–	N/A	Input Value + Perturbation
Perturbation Value	–	N/A	Delta for Sensitivity
Baseline Result	–	Output Units	Model Output at Baseline
Perturbed Result	–	Output Units	Model Output at Perturbed
Relative Change in Output	–	%	(Perturbed Result – Baseline Result) / Baseline Result
Relative Change in Input	–	%	(Perturbed Parameter A – Parameter A Baseline) / Parameter A Baseline
SST Value	–	N/A	Sensitivity Metric

What is Significance Sensitivity Test (SST) in MATLAB?

The Significance Sensitivity Test (SST) is a crucial analytical technique used in various fields, including engineering, finance, and scientific modeling, to understand how sensitive a model’s output is to small changes in its input parameters. When implemented in MATLAB, SST allows researchers and developers to quantify this sensitivity, identify critical input variables, and assess the robustness of their models. In essence, it answers the question: “If I slightly change this input, how much will the output change?”

Who should use SST?

• Model Developers: To understand which input parameters have the most significant impact on their model’s predictions or behavior.
• Risk Analysts: To quantify uncertainty and potential deviations in financial models or system performance.
• Engineers: To assess the tolerance of a design to variations in material properties, environmental conditions, or operational parameters.
• Researchers: To validate hypotheses and understand the underlying mechanisms driving simulation results.
• Anyone using MATLAB for complex modeling: If the reliability and stability of model outputs are important, SST is an invaluable tool.

Common Misconceptions about SST:

• SST is the same as error analysis: While related, SST specifically measures the *rate of change* of the output with respect to the input, whereas error analysis typically focuses on the absolute or relative magnitude of the error itself.
• A high SST always means the input is “bad”: Not necessarily. A high SST simply indicates a high degree of influence. It means careful control and accurate measurement of that input parameter are vital for reliable results.
• SST is only for linear systems: While easier to interpret in linear systems, SST is highly effective for non-linear systems too, often revealing localized sensitivity around specific operating points.
• MATLAB’s `gradient` or `diff` functions are sufficient: These functions compute derivatives numerically. SST specifically uses the *relative* changes, making it a normalized measure that is often more interpretable across different scales of input and output. Our calculator utilizes the core concept of relative changes for a robust SST.

This guide will walk you through calculating SST using MATLAB principles and our practical calculator.

Jump back to the SST Calculator to experiment.

SST Formula and Mathematical Explanation

The core idea behind the Significance Sensitivity Test (SST) is to measure the relative change in the output of a function or model in response to a relative change in one of its input parameters. This normalized approach makes the sensitivity measure comparable across different models and scales.

Let \( f(x) \) be a function or model where \( y = f(x) \). Here, \( x \) is the input parameter and \( y \) is the output.

1. Baseline: We start with a baseline input value, \( x_{base} \), and observe the corresponding baseline output, \( y_{base} = f(x_{base}) \).
2. Perturbation: We introduce a small change (perturbation) to the input parameter, \( \Delta x \). The new input value becomes \( x_{pert} = x_{base} + \Delta x \).
3. Perturbed Output: We calculate the output using the perturbed input value: \( y_{pert} = f(x_{pert}) \).
4. Relative Change in Input: This measures the fractional change in the input parameter relative to its baseline value.
   $$ \text{Relative Change}_{Input} = \frac{x_{pert} – x_{base}}{x_{base}} = \frac{\Delta x}{x_{base}} $$
   This is often expressed as a percentage.
5. Relative Change in Output: Similarly, this measures the fractional change in the output relative to its baseline value.
   $$ \text{Relative Change}_{Output} = \frac{y_{pert} – y_{base}}{y_{base}} $$
   This is also often expressed as a percentage.
6. Significance Sensitivity Test (SST): The SST is the ratio of the relative change in output to the relative change in input.
   $$ \text{SST} = \frac{\text{Relative Change}_{Output}}{\text{Relative Change}_{Input}} = \frac{(y_{pert} – y_{base}) / y_{base}}{(x_{pert} – x_{base}) / x_{base}} $$
   $$ \text{SST} = \frac{(y_{pert} – y_{base}) / y_{base}}{(\Delta x) / x_{base}} $$
   A higher SST value indicates that the output \( y \) is highly sensitive to changes in the input \( x \).

In our calculator, Input Parameter A corresponds to \( x_{base} \), Perturbation Value corresponds to \( \Delta x \), Baseline Result corresponds to \( y_{base} \), and Perturbed Result corresponds to \( y_{pert} \).

Variables Table:

Variables in SST Calculation

Variable	Meaning	Unit	Typical Range/Notes
\( x_{base} \) / Input Parameter A	Baseline value of the input parameter being tested.	Depends on parameter (e.g., concentration, temperature, price)	Positive, realistic value for the system.
\( \Delta x \) / Perturbation Value	The magnitude of the change applied to the input parameter.	Same unit as \( x_{base} \)	Small, positive value (e.g., 1-10% of \( x_{base} \)). Crucial for meaningful sensitivity.
\( x_{pert} \) / Perturbed Parameter A	The input parameter value after perturbation (\( x_{base} + \Delta x \)).	Same unit as \( x_{base} \)	\( x_{base} + \Delta x \)
\( y_{base} \) / Baseline Result	The output of the model/function at the baseline input \( x_{base} \).	Depends on output (e.g., yield, profit, stress)	Non-zero value.
\( y_{pert} \) / Perturbed Result	The output of the model/function at the perturbed input \( x_{pert} \).	Same unit as \( y_{base} \)	\( f(x_{base} + \Delta x) \)
Relative Change in Input	The fractional change in the input parameter.	Unitless (often expressed as %)	\( \Delta x / x_{base} \)
Relative Change in Output	The fractional change in the output.	Unitless (often expressed as %)	\( (y_{pert} – y_{base}) / y_{base} \)
SST	Significance Sensitivity Test value. Measures output change per unit of input change (relative).	Unitless	Positive values indicate direct relationship; negative values indicate inverse. Magnitude indicates sensitivity level.

Understanding the interdependencies between these variables is key to effective model validation in MATLAB.

Practical Examples (Real-World Use Cases)

Let’s illustrate the SST calculation with practical scenarios using our calculator.

Example 1: Chemical Reaction Yield Sensitivity

A chemical engineer is modeling the yield of a reaction based on the concentration of a catalyst (Catalyst A). The baseline concentration is 10 mol/L, and at this concentration, the yield is 85%. The engineer wants to know how sensitive the yield is to a small increase in catalyst concentration.

• Input Parameter A (Baseline Catalyst Concentration): 10 mol/L
• Perturbation Value (Increase in Concentration): 0.5 mol/L
• Baseline Result (Yield): 85%
• Perturbed Result (Yield with 10.5 mol/L Catalyst): Let’s assume the simulation shows a yield of 88%.

Calculation using the calculator:

• Input Parameter A: 10
• Perturbation Value: 0.5
• Baseline Result: 85
• Perturbed Result: 88

Results:

• Relative Change in Input: (0.5 / 10) = 0.05 (or 5%)
• Relative Change in Output: (88 – 85) / 85 = 3 / 85 ≈ 0.0353 (or 3.53%)
• SST Value: 0.0353 / 0.05 ≈ 0.706

Interpretation: An SST of 0.706 suggests that for every 1% increase in catalyst concentration (relative to the baseline), the yield increases by approximately 0.706%. This indicates moderate sensitivity. The engineer might decide that precise control of Catalyst A concentration is important but not critically so if minor fluctuations are acceptable.

Example 2: Financial Model – Profit Sensitivity to Material Cost

A business analyst is evaluating a product’s profitability. The baseline material cost is $50 per unit, leading to a profit of $150 per unit. The analyst wants to assess the impact of a small increase in material cost.

• Input Parameter A (Baseline Material Cost): $50
• Perturbation Value (Increase in Cost): $2
• Baseline Result (Profit): $150
• Perturbed Result (Profit with $52 Material Cost): Let’s assume the profit drops to $146.

Calculation using the calculator:

• Input Parameter A: 50
• Perturbation Value: 2
• Baseline Result: 150
• Perturbed Result: 146

Results:

• Relative Change in Input: (2 / 50) = 0.04 (or 4%)
• Relative Change in Output: (146 – 150) / 150 = -4 / 150 ≈ -0.0267 (or -2.67%)
• SST Value: -0.0267 / 0.04 ≈ -0.6675

Interpretation: An SST of -0.6675 indicates an inverse relationship. For every 1% increase in material cost (relative to the baseline), the profit decreases by approximately 0.6675%. This suggests a significant sensitivity. The business might need to explore cost-saving measures for materials or hedging strategies to mitigate profit volatility. This insight is crucial for making informed decisions within a financial planning framework.

These examples highlight how SST helps in understanding and quantifying the impact of input variations on model outcomes, which is a fundamental aspect of robust MATLAB model development.

How to Use This SST Calculator for MATLAB

Our interactive SST calculator is designed to be intuitive and provide quick insights into your model’s sensitivity. Follow these steps to utilize it effectively:

Step-by-Step Instructions:

1. Identify Your Parameters: Determine the specific input parameter (e.g., a variable in your MATLAB script, a coefficient) and the corresponding output you want to analyze.
2. Determine Baseline Values: Find the standard or typical operating value for your input parameter. This is your ‘Input Parameter A (Baseline)’. Also, determine the output value when the input is at this baseline. This is your ‘Baseline Result’.
3. Choose a Perturbation: Decide on a small, realistic change you want to apply to the input parameter. This is your ‘Perturbation Value’. It should represent a plausible fluctuation (e.g., a small increase or decrease). A positive value will be added to the baseline.
4. Simulate/Calculate Perturbed Output: Determine the output value of your model when the input parameter is changed by the ‘Perturbation Value’. This is your ‘Perturbed Result’. If your MATLAB script calculates this, run it with the adjusted input.
5. Enter Values into the Calculator:
   • Input ‘Input Parameter A (Baseline)’
   • Input ‘Perturbation Value’
   • Input ‘Baseline Result’
   • Input ‘Perturbed Result’
   • (Optional) Input Parameter B: This field is included for completeness but is not used in the standard SST formula calculation shown here, which focuses on a single parameter’s sensitivity.
6. Click ‘Calculate SST’: The calculator will instantly compute and display the SST value, along with key intermediate metrics like relative changes.

How to Read Results:

• SST Value: This is the primary metric.
  • A large positive SST means the output increases significantly when the input increases.
  • A large negative SST means the output decreases significantly when the input increases (or increases significantly when the input decreases).
  • An SST close to zero suggests the output is not very sensitive to changes in this particular input parameter.
• Relative Change in Output: Shows the percentage change in your model’s output relative to its baseline value.
• Relative Change in Input: Shows the percentage change applied to your input parameter.
• Perturbed Parameter A: Displays the actual value of the input after the perturbation was applied.
• Data Table: Provides a detailed breakdown of all input values and calculated metrics for clarity.
• Chart: Visually compares the relative change in input versus the relative change in output, offering a quick graphical understanding.

Decision-Making Guidance:

• High SST: If the SST is high (either positive or negative), it indicates that this input parameter is critical. Ensure its value is accurately measured or controlled in your system or model. Small errors in this input can lead to large deviations in the output. Consider adding more robust validation or control mechanisms for this parameter in your MATLAB simulation setup.
• Low SST: If the SST is close to zero, changes in this input parameter have a minimal impact on the output. This parameter might be less critical for model accuracy or system stability, allowing for more tolerance in its measurement or setting.
• Context is Key: Always interpret the SST value within the context of your specific application and acceptable output ranges. What constitutes “high” or “low” sensitivity depends on the domain and the consequences of output variation.

Use the ‘Reset’ button to clear the fields and start a new calculation. The ‘Copy Results’ button helps you easily transfer the computed values for documentation or further analysis.

Key Factors That Affect SST Results

Several factors can influence the calculated SST value, making it essential to consider them for accurate interpretation and reliable modeling in MATLAB. Understanding these factors helps in refining your analysis and improving your MATLAB model validation process.

1. Magnitude of Perturbation (\( \Delta x \)):
   The SST formula relies on the assumption of a *small* change in the input parameter. If the perturbation (\( \Delta x \)) is too large, the relationship between input and output might become non-linear within that range, and the calculated SST might not accurately represent the sensitivity at the baseline point. Using a very small, non-zero perturbation is generally recommended.
2. Baseline Value of Input Parameter (\( x_{base} \)):
   Since the relative change in input is \( \Delta x / x_{base} \), the baseline value significantly impacts the denominator. A small \( x_{base} \) with a fixed \( \Delta x \) will result in a larger relative input change, potentially affecting the SST. Conversely, a large \( x_{base} \) will yield a smaller relative input change. Ensure your baseline is representative.
3. Baseline Value of Output (\( y_{base} \)):
   Similarly, the relative change in output is calculated as \( (y_{pert} – y_{base}) / y_{base} \). If the baseline output \( y_{base} \) is very close to zero, the relative change can become extremely large or undefined, leading to unreliable SST values. This is a critical edge case to watch for.
4. Non-Linearity of the Model/Function:
   The SST provides a measure of *local* sensitivity at the baseline point. If the function \( f(x) \) is highly non-linear, the sensitivity might change drastically with small shifts away from \( x_{base} \). The SST calculated with one perturbation value might differ significantly from the sensitivity measured with a different perturbation or around a different baseline. Advanced techniques like Sobol indices might be needed for global sensitivity analysis in highly non-linear systems.
5. Scale and Units of Input and Output:
   The use of relative changes makes SST inherently scale-invariant to some extent. However, the *choice* of units can still influence the practical interpretation. A large SST value might seem alarming, but if the baseline values themselves are very small (e.g., dealing with micro-quantities), the absolute impact might still be negligible. Always consider the absolute values alongside the relative sensitivity.
6. Variability and Noise in Data/Model:
   If the underlying data or the model itself contains significant noise or random variability, it can obscure the true sensitivity. Small, genuine changes in the input might be masked by random fluctuations in the output. Robustness checks and potentially smoothing techniques might be necessary before performing SST, especially when working with real-world experimental data in MATLAB.
7. Interactions with Other Parameters:
   SST typically analyzes the sensitivity to one parameter at a time, assuming others are held constant. In complex systems, parameters often interact. A high SST for parameter A might be observed, but its practical impact could be mitigated or amplified by changes in parameter B. For such cases, more advanced methods like variance-based sensitivity analysis (e.g., using MATLAB toolboxes) are recommended.

Careful consideration of these factors is vital for obtaining meaningful results from your SST calculations in any MATLAB project.

Frequently Asked Questions (FAQ)

Q1: What is the difference between SST and a simple derivative?

The derivative \( f'(x) \) measures the instantaneous rate of change of the output with respect to the input at a specific point. SST, however, uses *relative* changes \( (\Delta y / y) / (\Delta x / x) \), making it a normalized measure. This normalization makes SST more comparable across different models with varying scales and units. While related, SST provides a sensitivity index that is often more directly interpretable in practical contexts than a raw derivative.

Q2: How do I choose the ‘Perturbation Value’ (\( \Delta x \))?

The perturbation value should be small enough to assume the relationship between the input and output is approximately linear within that small range, but large enough to yield a measurable change in the output that is distinct from noise. A common practice is to use a percentage of the baseline input value, such as 1% or 5%. Experimenting with slightly different perturbation values can help confirm if the sensitivity is stable.

Q3: What does an SST value of 1 mean?

An SST value of 1 means that the relative change in the output is equal to the relative change in the input. For example, if the input increases by 10% (relative change = 0.1), the output also increases by 10% (relative change = 0.1). This indicates a proportional relationship in terms of relative changes.

Q4: Can SST be used for categorical inputs?

The standard SST formula, as implemented here, is designed for continuous numerical inputs where a perturbation (\( \Delta x \)) can be meaningfully defined. For categorical inputs, different sensitivity analysis techniques are required, such as comparing outputs across different categories or using specialized methods that can handle discrete variables.

Q5: My baseline result is zero. What happens?

If your baseline result (\( y_{base} \)) is zero, the calculation for ‘Relative Change in Output’ involves division by zero, making the SST undefined. In such cases, you might need to adjust your baseline to a non-zero value if possible, or use a different sensitivity metric that doesn’t rely on relative changes, or use a shifted output (e.g., \( y – y_{min} + \epsilon \)) if appropriate for your model.

Q6: How can I implement SST in MATLAB code?

You can implement SST in MATLAB using basic arithmetic operations. Define your function, specify baseline and perturbation values, calculate the outputs for both, and then apply the formula:
% Example MATLAB implementation
x_base = 10; % Input Parameter A (Baseline)
delta_x = 0.1; % Perturbation Value
y_base = func(x_base); % Baseline Result
x_pert = x_base + delta_x;
y_pert = func(x_pert); % Perturbed Result

rel_change_input = delta_x / x_base;
rel_change_output = (y_pert – y_base) / y_base;

sst_value = rel_change_output / rel_change_input;

fprintf(‘SST Value: %.4f\n’, sst_value);
Replace `func` with your actual MATLAB function or model calculation.

Q7: What does a negative SST indicate?

A negative SST indicates an inverse relationship between the input parameter and the output. As the input parameter increases, the output decreases, and vice versa. For example, increasing the cost of raw materials (input) typically decreases the profit (output), resulting in a negative SST.

Q8: Is SST useful for optimization problems in MATLAB?

Yes, SST can be very useful. By identifying parameters to which the objective function is highly sensitive, optimization algorithms can focus their search efforts more effectively or prioritize constraints related to those sensitive parameters. It helps understand the landscape around an optimal solution.

Related Tools and Internal Resources