Methamphetamine-dependent participants and controls earned similar rewards on a laboratory choice task, but computational models suggested different ways of learning and choosing. Equal final scores therefore did not establish that the underlying decision processes were identical.
Harlé and colleagues compared 16 methamphetamine-dependent adults with 16 healthy controls. Participants repeatedly selected between two options with fixed but initially unknown reward rates. Each outcome provided both a possible reward and information about which option might pay better later. Researchers separated learning about reward rates from the policy used to choose between options.
The groups did not differ significantly in overall earnings (p = 0.93) or general response latency (p = 0.50). Both groups became faster with task experience. These behavioral results matter: the study did not demonstrate that every participant with dependence made slower decisions or obtained fewer rewards.
Control participants’ choices were best described by a policy that incorporated estimated reward values. Methamphetamine-dependent participants were more likely to be described by a simpler win-stay/lose-shift rule. Among those best fit by the value-based model, the dependence group showed lower initial reward expectations and less tendency to choose the option estimated to be most rewarding. A structural-imaging association also linked a model-derived choice tendency with thalamic gray-matter volume.
The model parameters were inferred from choices; they were not direct measurements of pessimism, motivation or a person’s beliefs outside the task. The small observational sample cannot show that methamphetamine caused the inferred strategy differences. Results also depend on which learning rules and decision policies were considered. They do not establish intoxication, financial incapacity or inevitable poor choices in everyday life. The study’s contribution is to show why a total task score may miss differences that become visible only when the sequence of choices is examined.
For a separate question, see explicit rule switching versus learning from feedback.