Impact Analysis Tutorial
In this tutorial we will use the Asia sample network, which is included with Bayes Server, to demonstrate how to use the Impact Analysis tool.
The Impact Analysis tool enables us to evaluate how various sets of evidence influence a specific target variable, often called the "Hypothesis." This approach helps researchers and decision-makers gauge the effect of particular data points on an outcome, providing valuable insights into which pieces of evidence are most significant in determining the result.
The Impact Analysis tool supports both discrete and continuous variables.
Open the Asia sample network included with Bayes Server, either from the Start Page or from the File menu, click Open.
Set evidence on the following variables, by clicking the respective bar in the user interface:
- Has Tuberculosis = True
- XRay Result = Abnormal
- Dyspnea = True
The Asia network with evidence set should look like this...
Select the Has Bronchitis node, without setting evidence on it.
From the Analysis menu, click Impact to launch the Impact analysis dialog.
Ensure that Has Bronchitis is selected as the target variable (V).
Ensure that True is selected as the target state (S).
Ensure that Include is selected as the Subset method.
The options page should look as follows:
- Click Run
The results page should look as follows:
Since we set the Subset method to Include, we can see that the target variable (V) has been queried for each combination of evidence, without any other evidence present.
This allows us to evaluate how each data point affects the target variable (V).
With Subset method set to Include one of the key statistics reported is **Distance (V) to None. This tells us how different the target (V) distribution is with this piece of evidence set.
We can see that Dyspnea has the biggest impact on Has Bronchitis when considering each piece of evidence in isolation (Subset Method = Include).
- Return to the Options page
- Change the Subset method to Exclude
The options page should now look as follows:
- Click Run
The results page should look as follows:
In this case we are more interested in the Distance (V) from All statistic, and we can see that the absence of Has Tuberculosis has the biggest impact on Has Bronchitis.
- Verify this by closing the Impact analysis dialog, and testing how Has Bronchitis changes with the absence of each piece of evidence.
In addition to only include a single piece of evidence, or excluding a single piece of evidence, you can adjust the Max evidence subset option, to include or exclude pairs of evidence or other subset counts.