Toolkit

Choosing Scientific Methods

Scientific Methods
Decision Support Tools
Data Analysis

Common scientific methods used in the decision support tools and how to choose the most appropriate one for your project.

Cover description

Overview

Who is this guidance for?

This page provides developers, designers, and product owners with an overview of scientific methods commonly used in decision support tools, along with guidance on selecting the most appropriate method for their specific use case.

Summary

  • Problem and Decision Type Matter: Start by clearly outlining the problem your decision support tool (DST) will address and the type of decision it will support (operational, tactical, strategic).
  • Tailor to Your Data: analyse the data available (structured vs. unstructured, complexity, etc.) as this influences suitable scientific methods.
  • Output and User Interaction: Determine what output the DST should provide (e.g., recommendations, visualisations) and how much user control is needed.
  • Match Method to Requirements: Common decision support methods include rule-based systems, statistical methods, machine learning, simulation, and optimisation. Select the one that aligns best with your problem, data, and output needs.
  • Technical Factors and User Expertise: Consider your technical resources and assess the technical sophistication of potential users to ensure the DST’s complexity is appropriate.

How to Choose?

1. Define the Problem and Decision Type

  • Problem Clarity: Clearly outline the problem the DST is intended to address. Is it complex, straightforward, highly structured, or unstructured?
  • Decision Type: Identify the nature of the decision:
    • Operational: Routine, day-to-day decisions.
    • Tactical: Mid-level decisions focusing on resource allocation and planning.
    • Strategic: High-level, long-term decisions impacting the organization’s direction.

2. Consider the Data Landscape

  • Data Availability: Evaluate the type and quality of data you have access to (structured vs. unstructured, internal vs. external).
  • Data Complexity: Assess the volume and intricacy of the data. Will it involve simple calculations or complex analysis?

3. Determine the Desired Output and Level of User Interaction

  • Output Needs: What type of output does the DST need to provide (recommendations, reports, visualisations, etc.)?
  • User Interaction: How much control and interactivity do users require? Should it be a simple recommendation engine or a system that allows for exploration and what-if scenarios?

4. Match Methods to Your Requirements

Here are common decision support techniques and when they are most suitable:

  • Rule-Based Systems: Ideal for structured problems with clear if-then logic.
  • Statistical Methods: Effective for analyzing historical data, identifying trends, and forecasting (e.g., regression analysis).
  • Machine Learning: Suited for complex problems where patterns aren’t easily defined by rules. Can adapt as new data becomes available (e.g., classification, clustering, deep learning).
  • Simulation: Useful for testing “what-if” scenarios and understanding the impact of potential decisions in a risk-free environment.
  • Optimisation: Helps find the best solution when there are multiple constraints and objectives.
  • Physics-based or Environmental Models: Leverage the laws of physics and domain-specific knowledge to simulate real-world systems and their behaviours. These models are valuable for complex environmental decision-making where cause-and-effect relationships are well understood and need to be explicitly represented (e.g., assessing the impact of infrastructure projects on ecosystems).

5. Consider Your Assumptions and User Expertise

  • Transparency of Assumptions and Uncertainty: Explicitly communicate the assumptions underlying models or analysis. Provide users with a clear understanding of potential uncertainties and their impacts on the decision-making process.
  • User Sophistication: Assess the technical expertise of potential users. The DST’s complexity should match their skills.

Important Considerations

  • Explainability: Some methods (like rule-based systems) are transparent, while others (like deep neural networks) can be more of a “black box.” Consider the need for explainability in your decision-making process.
  • Bias: Watch out for potential bias in your data, as this might carry over into your DST’s outputs.
  • Technical Resources: Do you have in-house expertise or need external development or science support? Are there computational limitations?s
  • Validation: Rigorously test and validate your DST using real-world data to ensure it yields accurate and helpful results.

Example

A product owner wants to develop a DST to support mineral exploration decisions.

  • Problem: Prioritise areas for further exploration based on the likelihood of finding valuable mineral deposits.
  • Data: Historical geological surveys, geochemical data, geophysical measurements, remote sensing data, etc.
  • Output: Risk/potential score indicating the probability of a commercially viable mineral deposit within the area.
  • Method: A machine learning model (e.g., random forest or neural network) might be appropriate as it can analyse complex relationships between geological variables and known mineral deposit locations.

Remember: The best method for your decision support tool is the one that most effectively addresses the specific nature of your decision problem, available data, and desired outputs.