Skip to main content
. 2011 Apr 6;31(14):5504–5511. doi: 10.1523/JNEUROSCI.6316-10.2011

Table 2.

Qualities of behavioral fits of both models

Direct actor Q-learning
−LL 107.6 114.3
Pseudo-R2 0.3534 0.3131
Number of parameters 4 5
BIC 118.6 128.0

Q-learning and policy-gradient models were fit to 20 subjects individually. Average quantities are reported. −LL, Negative log likelihood; BIC, Bayesian information criterion.