General Approach

• Logistic regression models: What features should we put in?
• Eventually: Want nonparametric. For now, let’s explore parametric ones

Treating Forced and Free Trials Separately

• Logistic regression model with separate terms for forced and free choice trials
• Reveals equal effects on reward learning, different effects on perseveration

• Model comparison shows that this difference is important

Looking for Exponential Decay

• Exponential pattern predicted by reinforcement learning models
• These plots look kinda exponential. How exponential are they really?
• Fit exponential curve to points 2:n

• Exponential is a good approximation, not perfect
• It’s better if we let the first point be free
• Let’s try this as a model. Seven parameters:
  • Fit exponential for 2:n to reward-seeking (2 params)
  • Fit exponential for 2:n to choice perseveration (2 params)
  • Fit lag 1 reward-seekend: win-stay/lost-shift (1 param)
  • Fit lag 1 perseveration (1 param)
  • Fit bias (1 param)