Bootstrapped decision-making models: How
Posted: October 25, 2011 | Author: admin | Filed under: tools | Tags: bootstrap, bootstrapping, decision criteria, linear regression, model | Leave a comment »__________________________________________________________
Quick and dirty:
- Build exhaustive criteria list for a decision
- Insert criteria values
- Elicit predictions from a human for an outcome value in each case
- Run linear regression
- Use significant criteria weights in a formula to predict outcomes
Long Description:
Bootstrapped models employ a hybrid qualitative/quantitative approach to create a decision-making aid. In “bootstrapping” a model, one seeks to quantitatively emulate the qualitative decision-making approach used by a human.
To create such a model, first build an exhaustive list of all possible and practical decision criteria. In what specific ways does the one think about the decision? Keep in mind that each of these decision criteria needs to be quantified, so think in terms of numbers, binary values, etc. The next step is to populate the criteria fields with values that reflect real data. Repeat the act of populating the criteria fields to create an expanded data set.
Here’s the crux: Have the person (that you are seeking to bootstrap) look at each instance in the data set and use their qualitative judgment of the values across criteria to predict what the outcome values will be. No calculations or formulas – just have the person use qualitative judgment to produce an outcome value. Now run a linear regression with the predicted outcome value as the dependent variable and the criteria as independent variables.
Use the resulting weights for each significant independent variable to build a formula, which will become the bootstrapped model. To use it, you simply have to input the values for decision criteria to obtain a predicted outcome value that will closely match the value that the human subject you bootstrapped would have predicted.