%Mini-projects
In the section on explaining away we saw how gaining new evidence could lead to many patterns of correlation between previously un-correlated variables. Common forms of interaction between causes in giving rise an effect, such as a noisy-or, lead to anti-correlation when the effect is observed---this is the classic "explaining away" effect. However, other forms of interaction can lead to linear correlation or more complex patterns. We might call the latter "generalized explaining away."
How would you look for evidence of generalized explaining away in human reasoning? Your task in this mini-project is to design an experiment to explore (generalized) explaining away. Can you provide evidence that in some situations previously uncorrelated variables become correlated? What judgements from people would be evidence for this? Can you provide evidence that in different situations the form of correlation can be different (e.g. anti-correlation in one situation, but correlation in another)?
Some things to consider in designing your experiment:
- What instructions will you give participants?
- How will you set up the situation that the participant is asked to reason about (this is often called the "cover story")?
- What will you manipulate to generate the evidence (i.e. what are the independent variables)?
- What questions will you ask participants (i.e. what is the dependent measure)?
- What is a probabilistic model for predicting judgements? Is there a competing model?
- What are your predictions?
- Are there confounds or missing information (such as parameters) that would be needed to get model predictions?
- How will you analyze the data? Are you looking for qualitative or quantitative effects?