While classic models of animal decision-making assume that individuals objectively assess the absolute value of options, decades of research have shown that rewards are often evaluated relative to prior experience, creating ‘contrast effects’. Contrast effects are often assumed to be purely sensory, yet consumer psychology tells us that label-based expectations can affect value perception. However, this has rarely been tested in non-model systems. Bumblebees forage on a variety of flower types that vary in their signals and rewards and show clear contrast effects when rewards are lower than their immediate previous experience. Here we manipulated bees’ expectations of a stimulus’ quality, before downshifting the reward to induce incentive contrast. We found that contrast effects were not solely driven by prior experience with a better reward, but also influenced by experience with associated stimuli. Bees were faster to accept the lower-quality reward when it was paired with a novel rather than a familiar stimulus. We then explored the boundaries of these label-based expectations by testing bees along a stimulus gradient and found that expectations generalized to similar stimuli. Such reference-dependent evaluations may play an important role in bees’ foraging choices, with the potential to impact floral evolution and plant community dynamics.

Data collection:

Data were collected between November 2020 to May 2022 at the University of Texas at Austin. Experiments were conducted by CTH.

Experiment 1:

We used worker bumblebees Bombus impatiens from commercially-reared colonies (n=3) (Koppert, USA). We used 20 bees per treatment (10 for each colour combination), with treatments equally represented across colonies (Table S2). Individuals were trained to a colour (blue or yellow) paired with a high-quality reward (8ml of 50% w/w sucrose) over three consecutive trials spaced 5 minutes apart. Within each trial, individuals visited ~10 rewarding flowers and consumed all rewards; the number of flowers visited did not differ across treatments (for additional information see Supplementary Material). Following these three training trials, bees were presented with a lower quality reward in a ‘test’ trial (8ml of 30% w/w sucrose) paired with either the familiar or a novel colour (Figure 1a). Our experimental nectar concentrations were designed to match natural variation found in bumblebee-visited flowers [33]. In the test trial, we recorded bees’ responses to downshifted rewards over their first 20 visits to flowers. We chose 20 successive visits based on our expectation that bees would change their acceptance of the downshifted reward over this timescale. Acceptance was measured as bees consuming the solution, while rejection was characterized by bees probing the solution and exiting without imbibing (video S1).

Experiment 2:

To determine how similar stimuli needed to be to bear the cost of higher expectations, we tested bees using a range of colour stimuli. Individuals (n= 85 from 5 colonies; Table S1) were trained to a blue stimulus paired with a high-quality reward across three training trials (Figure 1b). We then presented individuals with one of five possible colours. Four of these stimuli ranged from blue to green and were increasingly different from the originally-trained colour whilst still being discriminable to foragers [34] (chromatic contrasts calculated in the bee colour space model [35,36]; Table S1). The fifth colour was yellow, serving as a ‘novel’ stimulus as in Experiment 1. We used a slightly different blue stimulus in Experiment 2 than Experiment 1, while the yellow was the same across both experiments. Again, we measured individuals’ acceptance of the downshifted reward over their first 20 visits.

Data analysis

We used generalized linear models (GLMs) and linear mixed-effect models (GLMMs) with the glm() and glmer() functions in the lme4 package. We first analysed the differences in initial acceptance of the downshifted reward in the first visits between treatments using binomial GLMs with ‘acceptance’ as the main response and ‘stimulus type’ as the explanatory variable. We also looked at acceptance over the first five visits using binomial GLMMs with the response variable ‘accept’ or ‘reject’, the explanatory variables ‘stimulus type’ (different for each experiment) and ‘visit number’ (continuous variable), and ‘bee’ and ‘colony’ as random factors. To determine whether the number of visits before acceptance differed across treatments, we carried out GLMs with a quasi-Poisson distribution using ‘number of visits until acceptance’ as the response variable and ‘stimulus type’ as the explanatory variable. Finally, to analyse acceptance across all 20 visits we ran binomial GLMMs with the response variable ‘accept’ or ‘reject’, the explanatory variables ‘stimulus type’ (different for each experiment) and ‘visit number’ (continuous variable), and ‘bee’ as a random factor. We also included ‘colony’ as a random factor in Experiment 2, but not Experiment 1 due to a singularity issue.