Getting Started with Decision Tree Software

Welcome to the world of decision analysis. SpiceLogic Decision Tree Pro is a decision tree maker and analyzer for modeling choices, uncertainty, payoffs, risk, utility functions, cost-effectiveness analysis, Markov models, Monte Carlo simulation, and more.

This getting-started page does not try to explain every feature. The other documentation pages cover those details. Here, we will build a simple decision tree with payoffs and run an expected-value analysis. That is one of the most common and useful ways to start using the software.

The example is medical, but the software is not limited to healthcare. The same decision-analysis ideas apply to business, engineering, finance, operations, and many other fields. Healthcare students often use this software, so a medical example is a helpful way to demonstrate the workflow.

Suppose you are modeling a medical decision with two first options: "Do a test" and "No test". If you do the test, the result can be positive or negative. In either case, you then choose between "Treatment A" and "Treatment B". Each treatment can succeed or fail, and each outcome has a life-year payoff for the patient.

Using a decision tree, we want to calculate the expected value and identify which option gives the better result. The final model we are going to build is shown below.

Sample SpiceLogic Decision Tree Analyzer model comparing Test versus No Test treatment paths with chance events, probabilities, and life-year payoffs.

One more note: Rational Will includes the Decision Tree module. If you came here from Rational Will, this documentation still applies. The standalone Decision Tree software and the Decision Tree module inside Rational Will use the same workflow.

Start SpiceLogic Decision Tree Pro. You will see the start window below. You will also see this same window if you click "Decision Tree" from the Rational Will start screen.

SpiceLogic Rational Will start page with buttons to begin a Decision Tree from a Decision Node, Chance Node, or Markov Model, plus options to set up criteria first or use Machine Learning.

In the center, you will see "Begin your Decision Tree with" and several buttons: "A Decision Node", "A Chance Node", and "A Markov Chance Node". These buttons choose the root node of your tree.

In most regular decision models, the root node is a Decision Node. Use a Chance Node when you want to start with uncertainty or perform probability analysis. Use a Markov Chance Node when you are building a Markov chain or Markov decision process, which is covered on a separate page.

For this tutorial, click "A Decision Node".

Creating a Decision Node and Actions

Newly created Decision 1 root node selected on the canvas with a red callout pointing to the Add an Action Node flyover button.

After the decision node is created, notice the fly-over menu. When you select a node, the software shows a small menu with the child node types that are valid for that node. This keeps the controls close to the diagram.

Click the fly-over menu button for creating an action node. Click it twice to create two actions.

Decision 1 root node with two default-labeled Action 1 and Action 2 children added from the flyover menu.

Now edit the action text. Double-click an action node to turn it into an editable text box.

First Action node turned into an inline editable text box and renamed to No Test, with the flyover menu visible above it.

You can also right-click a node to open its context menu. The context menu includes commands such as Edit and Delete.

Right-click context menu on an Action node with Edit Text highlighted, plus Cut, Copy, Paste, Set Payoff, Delete, and Move Up options.

After you rename the two actions as "No Test" and "Test", select the "No Test" action node. From its fly-over menu, click the square symbol to add a decision node as a child of that action.

Flyover menu on the No Test Action node with the square symbol highlighted to add a Decision node as its child.

You do not need to memorize every symbol. Decision Tree Pro shows tooltips throughout the diagram. Hover over a button to see what it does.

Tooltip Add a Decision Node shown when hovering over the square button in the flyover menu on the No Test action.

Click the button shown above to add a child decision node.

Tree after a child Decision node has been added under the No Test action, ready for its own action children.

As before, add two more actions under this new decision node and name them "Treatment A" and "Treatment B".

Child Decision node with two Action node children renamed Treatment A and Treatment B.

Decision nodes are named Decision 1, Decision 2, and so on by default. You can double-click a label to rename it to something more meaningful.

Pop-up edit box on Decision 2 with the new name Treatments entered and a Show on Diagram checkbox ticked.

If the node labels make the tree feel crowded, hide them from the View menu.

View menu button in the Decision Tree Analyzer that hides all Decision and Chance node labels at once to keep the tree clean.

For this tutorial, hide the labels to keep the tree clean.

Adding a Chance Node and Events with Probabilities

Select the "Treatment A" action under the new decision node. The fly-over menu appears again. This time, click the circle symbol to add a chance node.

Flyover menu on an Action node with the circle button highlighted, used to add a Chance node child.

After you add the chance node, select it to show its fly-over menu. From that menu, add two event nodes.

Flyover menu on a Chance node with the button highlighted that adds two Event node children at once.

The software only shows valid child node types in the fly-over menu.

After adding the two events, double-click the event nodes and rename them "Success" and "Failure". The question mark on the edge means the probabilities are still unknown.

When probabilities are unknown, the software uses the principle of indifference for calculation. If there is no reason to believe one event is more likely than another, the events are treated as equally likely. With two events, that means each event has a probability of 0.5. The tooltip explains this as well.

Success and Failure event edges showing question marks, indicating their probabilities are unknown and the principle of indifference will apply.

Click the question mark to specify a probability. A message box appears. Click "I know the probability".

Probability prompt dialog with the I know the probability button used to assign an explicit probability to an event edge.

Setting Success probability to 0.7 with the complementary Failure probability auto-calculated as 0.3 on the sibling event edge.

Set the probability of Success to 0.7. The Failure probability is then calculated as 0.3.

Copying and Pasting

Now comes the useful part. Select the chance node and right-click to open its context menu. Choose Copy. You can also use the keyboard shortcut Ctrl+C.

Right-click context menu on a Chance node with the Copy option highlighted, equivalent to keyboard shortcut Ctrl plus C.

Select the "Treatment B" node and right-click to open its context menu. This time, choose Paste. You can also use Ctrl+V. After pasting the chance node, the tree will look like this.

Tree after the copied Success and Failure chance subtree has been pasted onto the Treatment B action node.

After pasting, update the pasted probabilities to 0.8 for Success and 0.2 for Failure.

Now select the "Test" action node and add a chance node. Add two events under it: "Result Positive" and "Result Negative". Set the probability of Result Positive to 0.4 and Result Negative to 0.6.

You may be wondering where these numbers come from. They are simply the sample values from the decision tree shown at the beginning of this tutorial. They are not fixed rules. For your own model, use values that fit your situation, data, or expert judgment.

If you followed the steps, your tree should now look like this.

Third chance node added under the Test action with Result Positive (0.4) and Result Negative (0.6) event probabilities set.

If the Result Positive and Result Negative nodes do not show their full text, the node width is too narrow. Expand the Diagram Navigator tab and use the node-width slider. This panel also includes controls for horizontal spacing, vertical spacing, zoom, and other diagram layout options.

Diagram navigator panel with sliders for node width, horizontal spacing, vertical spacing, and zoom level used to fit long node text on screen.

After increasing the width, the full node text is visible. Now copy the decision node that contains "Treatment A" and "Treatment B", then paste it under each event of the result chance node: Result Positive and Result Negative.

Copied Treatment A and Treatment B decision subtree pasted onto each event of the Result chance node to grow the tree.

At this point, the tree should match the sample decision tree shown at the beginning of the tutorial. Now update the probabilities. Because the nodes were copied and pasted, their probabilities came from the original subtree. Select each edge and set the probabilities from the sample diagram.

After the probabilities are corrected, the decision tree should look like this.

Completed treatment decision tree with all chance event probabilities updated to match the reference diagram before adding payoffs.

Setting Payoff

Now add payoffs. In this example, the payoff is measured in life-years received by the patient after treatment.

Click the top "Success" node and look at its fly-over menu.

Flyover menu on a Success event node showing the payoff button used to assign a numeric payoff in life-years.

Click the payoff button. The first time you add a payoff, the software asks what type of payoff you want to use. It can be numeric, subjective, Boolean, or cost-effectiveness based.

For this tree, the payoff is a simple numeric criterion: life-years. Choose "Regular single or multiple criteria based analysis".

First-time payoff intent window letting the user choose between Regular single or multiple criteria based analysis and Cost-Effectiveness Analysis in Healthcare.

After clicking the first button, you will see the next setup window.

Objective criterion setup step shown after choosing numeric payoff, where the maximize or minimize objective is named.

Select "Number Type" for this criterion.

Criterion type selector for the Life-year criterion, with Numerical Type chosen over Subjective Type.

Set the unit to "Years". In this example, the minimum possible life-year value is 0 and the maximum is 10. Enter those values and click "Proceed".

Number criterion configuration with unit set to Years and minimum 0, maximum 10 for the life-years payoff range.

After clicking "Proceed", the wizard asks whether you want to add another criterion. Choose "No". You will return to the decision tree, and the top Success node will show the payoff editor.

Set the payoff to 9 years, using the value from the reference diagram.

Inline payoff editor on the top Success node where a value of 9 years is entered from the reference decision tree.

Set the payoff for all Success and Failure nodes using the values from the reference diagram. After the payoff type is configured, clicking the payoff button opens the payoff editor directly.

Terminals

Select each Success and Failure node and add a terminal node from the fly-over menu. The terminal node displays the total payoff at that point.

Flyover menu on a Success event node with the Terminal Node button highlighted to mark the end of that branch.

The terminal node also shows useful metrics in its tooltip.

Tooltip on a Terminal node displaying total payoff and other useful metrics aggregated along that path.

Analyzing Results and Policy

After the payoffs are set, the decision tree calculates the result immediately. Expected values are displayed on the nodes.

The recommended path is highlighted in green. In this example, the expected value for the "Test" node is 7.44 life-years, compared with 6.6 life-years for "No Test". So the root decision recommends "Test".

The green path is also shown for later decisions, such as the action to take after a positive or negative test result. This is called a policy: the recommended action for each situation in the tree.

Solved decision tree with expected values on each node and the recommended Test path highlighted in green (7.44 life-years versus No Test 6.6), with red callouts explaining the policy path.

The tooltip for each node also shows metrics such as expected value, total payoff at that point, and expected probability.

Tooltip on a node showing expected value, total payoff at that point, expected probability, and other metrics.

Expand the Options Analyzer tab to view charts and additional metrics.

Options Analyzer panel expanded to show the result charts and metrics summarising the solved treatment decision tree.

From the View menu, click "Policy/Ruleset" to see the policy as text instructions.

Policy and Ruleset panel opened from the View menu, listing the recommended strategy for the tree as readable text instructions.

Planning Point

Now suppose you want to analyze the tree from a later point. For example, what if the "No Test" path has already happened and you now need the best policy from that point?

You can do this by changing the planning point. The current planning point is the root node. Select the decision node that comes after the "No Test" action. In its fly-over menu, click the flag button to set that node as the planning point.

Flyover menu on the No Test sub-decision showing the flag button used to set that node as the planning point.

After setting the planning point, the tree is recalculated from that node. The green policy path starts from the new planning point, and the metrics and charts update accordingly.

Decision tree recalculated from the new planning point with a fresh green policy path and updated metrics and charts.

Conclusion

This tutorial only scratches the surface, but it covers the basic workflow: create decision and chance nodes, enter probabilities, set payoffs, read the expected value result, and inspect the recommended policy. Continue through the other documentation pages to explore risk metrics, utility functions, sensitivity analysis, Markov models, cost-effectiveness analysis, and more.