Rats exhibit similar biases in foraging and intertemporal choice tasks

  1. Gary A Kane  Is a corresponding author
  2. Aaron M Bornstein
  3. Amitai Shenhav
  4. Robert C Wilson
  5. Nathaniel D Daw
  6. Jonathan D Cohen
  1. Princeton University, United States
  2. Harvard University, United States
  3. University of California, Irvine, United States
  4. Brown University, United States
  5. University of Arizona, United States
7 figures, 2 tables and 3 additional files

Figures

Figure 1 with 1 supplement
Rat foraging behavior.

Rat foraging behavior in the (A) Travel Time, (B) Depletion Rate, (C) Scale, and (D) Pre-vs-Post Experiments. In (A), points and error bars represent mean ± standard error. In (B-D), points and …

https://doi.org/10.7554/eLife.48429.003
Figure 1—source data 1

Trial-by-trial foraging behavior.

https://doi.org/10.7554/eLife.48429.005
Figure 1—figure supplement 1
Diagram of the foraging task.

Rats press a lever to harvest reward from the patch then receive reward in an adjacent port following a pre-reward delay (handling time or HT). After receiving reward, there is a post-reward delay …

https://doi.org/10.7554/eLife.48429.004
Post-reward delay foraging and intertemporal choice behavior.

(A) Rat behavior in the Post-Reward Delay Experiment. Points and lines represent behavior of individual rats. Red line indicates optimal behavior (per MVT). (B) Rat behavior in the intertemporal …

https://doi.org/10.7554/eLife.48429.007
Figure 2—source data 1

Trial-by-trial intertemporal choice behavior.

https://doi.org/10.7554/eLife.48429.008
Figure 3 with 5 supplements
Model predictions for foraging tasks.

(A-E) Predictions of the best fit quasi-hyperbolic discounting model to all foraging tasks. Points and error bars represent mean ± standard deviation of the means for each individual rat; lines and …

https://doi.org/10.7554/eLife.48429.010
Figure 3—figure supplement 1
State space diagram of the foraging task.

State space diagram for the semi-markov model of the foraging task. Decisions to stay vs. leave are made in Decision states. A decision to stay causes a transition to the handling time, then reward, …

https://doi.org/10.7554/eLife.48429.011
Figure 3—figure supplement 2
Predictions of the best fit subjective cost and nonlinear reward utility models.

Predictions of the best fit subjective cost and nonlinear reward utility models (power law = util pwr; constant relative risk aversion = util CRRA). Black points and error bars represent mean ± …

https://doi.org/10.7554/eLife.48429.012
Figure 3—figure supplement 3
Predictions of the best fit biased time perception models.

Predictions of the best fit models of overestimation of pre-reward delays (pre-delay), linear underestimation of post-reward delays (post-delay), and nonlinear underestimation of post-reward delays …

https://doi.org/10.7554/eLife.48429.013
Figure 3—figure supplement 4
Predictions of the best fit discounting models.

Predictions of the best fit exponential discounting model (disc-exp), hyperbolic discounting model (disc-hyp), and constant sensitivity discounting model (disc-cs). Points and error bars are the …

https://doi.org/10.7554/eLife.48429.014
Figure 3—figure supplement 5
iBIC for each model for each foraging experiment.

Cost = subjective cost model, util-pwr and util-crra = nonlinear reward utility with power and CRRA function respectively, pre-del = linear overestimation of pre-reward delays, post-del = linear …

https://doi.org/10.7554/eLife.48429.015
Figure 4 with 2 supplements
Model predictions for the intertemporal choice task.

(A) Quasi-hyperbolic model predictions for the intertemporal choice task. Points and error bars represent the mean ± standard error of individual rat behavior; lines and ribbon represent mean ± …

https://doi.org/10.7554/eLife.48429.016
Figure 4—figure supplement 1
State space diagram of the intertemporal choice task.

Decisions made in Decision states cause transition to the Delay, Reward, and ITI states for the option chosen (either SS or LL), then back to the next Decision state. The model consisted of 10 …

https://doi.org/10.7554/eLife.48429.017
Figure 4—figure supplement 2
Comparison of all-future horizon and one-trial horizon discounting models.

(A) iBIC for the full horizon and one-trial horizon discounting models (B) Measured log-transformed discount factors for the full horizon and one-trial horizon discounting models. Bars and errorbars …

https://doi.org/10.7554/eLife.48429.018
Cross-task model predictions.

(A) Predicted foraging behavior for quasi-hyperbolic model parameters fit to either the foraging task (red line) or delay discounting task (DD; blue line). Black points and error bars represent mean …

https://doi.org/10.7554/eLife.48429.019
Discount function of the μAgent hyperbolic discounting model vs. standard hyperbolic discounting.

Points represent the standard hyperbolic discounting function, 1/(1+k*time). Lines represent the μAgent discount function in which the discount factor for each of the 10 μAgents was equal to the 5–95% …

https://doi.org/10.7554/eLife.48429.020
Author response image 1

Tables

Table 1
Parameters for each of the first four foraging experiments.

Harvest time = time to make a decision to harvest + pre-reward delay + post-reward delay (inter-trial interval). To control reward rate in the patch, the post-reward delay was adjusted relative to …

https://doi.org/10.7554/eLife.48429.006
ExperimentConditionStart RewardDepletion RatePre-Reward DelayHarvest TimeTravel Time
travel time10 s60, 90, or 120 μL−8 μL0 s10 s10 s
30 s30 s
depletion rate−8 μL90 μL−8 μL0 s12 s12 s
−16 μL−16 μL
scale90 μL/10 s90 μL−8 μL0 s10 s10 s
180 μL/20 s180 μL−16 μL20 s20 s
handling time0 s90 μL−8 μL0 s15 s15 s
3 s3 s
post-reward delay*3 s90 μL−8 μL0 s5–8 s**10 s
12 s13–16 s**
  1. Rats encountered all three patch types in both conditions.

    *One group of rats (n = 8) was tested on the first four experiments, but a separate group (n = 8) was tested on this final foraging experiment.

  2. **In this experiment, the harvest time was not held constant — the post-reward delay was always 3 s or 12 s regardless of the time to make a decision.

Table 2
Description of the hypotheses for overharvesting with general, qualitative predictions for the degree of overharvesting in each experiment.

Quantitative predictions depend on the exact formalization of each model and its specific parameters.

https://doi.org/10.7554/eLife.48429.009
HypothesisExperimental predictions
Subjective CostsA cost term c reduces the value of leaving a patch. Predicts greater overharvesting with higher c. Not affected by specific manipulations to reward or time.Rats will follow qualitative predictions of the Marginal Value Theorem, but exhibit an equal degree of overharvesting across conditions in each experiment.
Nonlinear Reward UtilitySubjective value increases sublinear to reward magnitude. Predicts greater overharvesting with steeper utility functions with larger rewards.Rats will exhibit an equal degree of ovarharvesting in all experiments except for the Scale experiment. In the Scale experiment, rats will overharvest more in the conditions with larger rewards.
Biased Time Perceptioni) Post-reward delays perceived as shorter, ii) pre-reward delays perceived as longer, or iii) longer delays (irrespective of their placement) perceived as shorter. All three hypotheses predict greater overharvesting with longer delays.Rats will exhibit a greater degree of overharvesting in the condition with longer delays in the Scale environment, in the condition with the longer post-reward delay in the Pre-vs-Post experiment, and in the condition with longer post-reward delay in the Post-Reward Delay experiment
Temporal DiscountingValue of future rewards discounted due to delay to receive them. Predicts greater overharvesting with greater levels of discounting and with longer delaysRats will overharvest to a greater degree in the conditions with longer delays in the Scale and Post-Reward Delay experiments and they will leave patches earlier due to the longer pre-reward delay in the Pre-vs-Post experiment.

Additional files

Download links