Reward & decision
Conditioned place preference
Preference for a chamber that was paired with a drug or reward.
Arena. Rectangle covering the whole apparatus, with the chambers side by side from left to right in the image. Phase: Pre-test, Conditioning or Test.
Settings
| Setting | Default | Meaning |
|---|---|---|
| Apparatus | 2 chambers | 2 or 3 chambers |
| Divider position (2 chambers) | 50 % from left | Where the two chambers meet |
| Neutral chamber width (3 chambers) | 20 % of arena | Width of the middle chamber, centered |
| Paired (drug / reward) side | Left | Which side chamber was paired |
| Zone entry criterion | 0.5 s | Minimum stay for a chamber entry |
| Metric | Unit | Definition |
|---|---|---|
| Preference score | s | Paired − unpaired chamber time |
| Paired-side preference | % | Paired / (paired + unpaired) × 100. Chance is 50 %. |
| Paired, unpaired chamber time | s | Time in each side chamber |
| Neutral chamber time | s | Time in the middle chamber (3-chamber apparatus) |
| Paired chamber entries | count | Entries into the paired chamber |
| Total distance | m |
Analysis. The usual CPP measure is the change from pre-test to test. In the Groups tab choose Change between phases, From Pre-test To Test, measure Preference score. Each animal's test − pre-test difference is then compared between groups. Paired measures (paired vs unpaired side) tests the preference within each group.
Two-armed bandit
Trial-by-trial choices between two options with probabilistic rewards, often with reversals of which option is better. The data come from an operant box or a T-maze. They are entered trial by trial (Enter Trials) or, more conveniently, imported from a trial CSV.
Columns (one row per trial): subject, group, optional day or phase (session / block), trial, choice (L/R, left/right or 0/1), reward (0/1), and optionally p_left, p_right (the reward probabilities on that trial) and rt (reaction time in s). See the sample file.
| Metric | Unit | Definition |
|---|---|---|
| High-value choices | % | Choices of the option with the higher reward probability (needs p_left/p_right). Chance is 50 %. |
| Reward rate | % | Rewarded trials / trials × 100 |
| Win-stay | % | P(same choice as the previous trial | previous trial rewarded) × 100 |
| Lose-shift | % | P(other choice | previous trial unrewarded) × 100 |
| Switch rate | % | P(choice differs from the previous trial) × 100 |
| Right choices | % | Side bias. Chance is 50 %. |
| Learning rate α | — | Q-learning fit (see below) |
| Inverse temperature β | — | Q-learning fit |
| Model pseudo-R² | — | 1 − NLL / (n · ln 2), where 0 means chance-level prediction |
| Trials | count | |
| Mean reaction time | s | Mean of the rt column |
Series: the probability of choosing the new best option from 5 trials before to 20 trials after each reversal, available when reward probabilities are given. The result screen also plots the choices across the session.
The Q-learning model. Each option has a value Q, starting at 0.5. After each trial, the chosen option is updated:
Q ← Q + α (r − Q), with r = 1 if rewarded, else 0.
The choice probability follows a softmax of the value difference:
P(right) = 1 / (1 + exp(−β (Qright − Qleft))).
α (0–1) and β (0–50) are fitted by maximum likelihood, using a grid search refined by a local search, on each animal's full trial sequence. At least 10 trials are needed. The negative log-likelihood and BIC are listed under the result's parameters.