1. Ecology
  2. Evolutionary Biology
Download icon

Positive and negative incentive contrasts lead to relative value perception in ants

  1. Stephanie Wendt  Is a corresponding author
  2. Kim S Strunk
  3. Jürgen Heinze
  4. Andreas Roider
  5. Tomer J Czaczkes
  1. University of Regensburg, Germany
  2. University of Passau, Germany
Research Article
  • Cited 2
  • Views 778
  • Annotations
Cite this article as: eLife 2019;8:e45450 doi: 10.7554/eLife.45450

Abstract

Humans usually assess things not according to their absolute value, but relative to reference points – a main tenant of Prospect Theory. For example, people rate a new salary relative to previous salaries and salaries of their peers, rather than absolute income. We demonstrate a similar effect in an insect: ants expecting to find low-quality food showed higher acceptance of medium-quality food than ants expecting medium quality, and vice versa for high expectations. Further experiments demonstrate that these contrast effects arise from cognitive rather than mere sensory or pre-cognitive perceptual causes. Social information gained inside the nest can also serve as a reference point: the quality of food received from other ants affected the perceived value of food found later. Value judgement is a key element in decision making, and thus relative value perception strongly influences which option is chosen and ultimately how all animals make decisions.

https://doi.org/10.7554/eLife.45450.001

eLife digest

We make many decisions every day, often by comparing options and choosing the one with the greatest profit. But how much we value something often does not depend solely on our needs. Instead, this value may depend on our expectations or other arbitrary reference points. For example, how satisfied you are with your income might depend on how much your colleagues or friends earn.

Animals, including insects, also make decisions when feeding, choosing a partner, or finding a nesting site. Sometimes animals behave in ways that look like disappointment. For example, monkeys may reject a cucumber as a reward if they have seen another monkey get a grape for completing the same task. But it is hard to tell if this behavior reflects a value judgment.

To investigate whether insects evaluate their options against their expectations, Wendt et al. offered black garden ants sugar water over multiple trials. Some ants grew to expect low quality sugar water (containing little sugar); some expected medium quality; and others expected high quality sugar water (containing a high concentration of sugar). Ants that expected to find low quality sugar water were more likely to accept medium quality options than ants that expected the medium quality sugar water. Similarly, ants that expected high quality sugar water were less likely to accept lower quality sugar water. Further experiments confirmed that the ants were not using physical cues such as satiation to guide their behavior.

Furthermore, Wendt et al. found that ants that returned to the nest after foraging passed on information that altered the expectations of the next group of foragers about nearby food. This suggests that the value that ants place on food sources depends both on individual experiences and on information gained from others.

Studies of decision making in humans can be difficult to perform and interpret, because volunteers may try to second-guess what the experimenters want to find, and culture and education may also influence choices. Studying ants instead could help to avoid these pitfalls, as the results presented by Wendt et al. suggest they make decisions in similar ways to humans. Future work building on these findings could also help researchers to predict how insects behave, particularly in rapidly changing environments.

https://doi.org/10.7554/eLife.45450.002

Introduction

We all compare options when making both large and small decisions, ranging from career choices to the choices of an evening’s entertainment. Understanding how options are compared has thus been central to the study of behavioural economics. Theories explaining the mechanisms by which options are compared and decisions are made have a long tradition (Vlaev et al., 2011), with Expected Utility Theory (EUT) being the most widely used theory in economic models (Mankiw, 2011; von Neumann and Morgenstern, 1944). EUT suggests that decisions are made by evaluating and comparing the expected utility from each option. A rational decision maker then chooses the option resulting in the best end state: the option providing the greatest utility (von Neumann and Morgenstern, 1944).

However, over the past decades economic research on human decision making has started to shift away from a view of (absolute) utility maximisation towards more nuanced notions of relative utility, such as reference-dependent evaluations. Kahneman and Tversky (1979) made a major contribution to this shift by introducing Prospect Theory, suggesting that decision making is not based on absolute outcomes, but rather on relative perceptions of gains and losses. In contrast to EUT, the utility attributed to options being evaluated is determined relative to a reference point, such as the status quo or former experience (Vlaev et al., 2011; Kahneman and Tversky, 1979; Parducci, 1984; Tversky and Kahneman, 1992; Ungemach et al., 2011). Various examples of relative value perception have been described. For example, satisfaction gained from income is perceived not absolutely, but relative to the income of others in the social reference group – such as one’s colleagues (Boyce et al., 2010). Overall, Prospect Theory has enriched our understanding of human decision making by conceptualising it as more nuanced and less rational than previously assumed (Tversky and Kahneman, 1974; Tversky and Kahneman, 1981).

The concept of malleable value perception is not just relevant to humans. Value judgments in animals are also influenced by factors apparently independent of the absolute value of options. For example, capuchin monkeys refuse otherwise acceptable pay (cucumber) in exchanges with a human experimenter if they had witnessed a conspecific obtain a more attractive reward (grape) for equal effort (Brosnan and De Waal, 2003). Rats, starlings, and ants, like humans, place greater value on things they work harder for (Aw et al., 2011; Czaczkes et al., 2018a; Lydall et al., 2010), and starlings, fish and locusts demonstrate state-dependent learning, wherein they show a preference for options experienced when they were in a poor condition (Aw et al., 2009; Pompilio et al., 2006; Schuck-Paim et al., 2004). Roces and Núñez (Roces, 1993; Roces and Núñez, 1993) aimed to show that in leaf cutting ants perceived value can be influenced by other ants. Ants recruited to higher quality food sources ran faster, deposited more pheromone, but cut smaller leaf fragments, even if the food source the recruits find is replaced by a standardised food source (Roces, 1993; Roces and Núñez, 1993). However, in these experiments, the absolute value and nature of the reference remains unclear, and indeed pheromone presence may have caused the observed behaviours without influencing the ants’ expectations or value perception at all.

Healey and Pratt (2008) showed that colonies of the house-hunting ant species Temnothorax curvispinosus move into a nest of mediocre quality faster when they were previously housed in a high-quality nest compared to colonies which were previously housed in a poor-quality nest (Healey and Pratt, 2008). In contrast, Stroeymeyt et al. (2011) showed that colonies of Temnothorax albipennis developed an aversion towards mediocre-quality nests available in their environment when they were housed in a high-quality nest, whereas colonies housed in a low-quality nest did not, and thus show an experience-dependent flexibility in nest choice (Stroeymeyt et al., 2011). However, critically missing from the existing works is a systematic description of value judgment relative to a reference point.

‘Value distortion by comparison’ effects have been studied for decades using the successive contrasts paradigm, in which animals are trained to a quality or quantity of reward which is then suddenly increased (positive incentive contrast) or decreased (negative incentive contrast) (Bentosela et al., 2009; Bitterman, 1976; Couvillon and Bitterman, 1984; Crespi, 1942; Flaherty, 1982; Flaherty, 1999; Mustaca et al., 2000; Weinstein, 1970a). Many mammals, including apes, monkeys, rats, and dogs (Brosnan and De Waal, 2003; Bentosela et al., 2009; Crespi, 1942; Flaherty, 1999; Mustaca et al., 2000; Pellegrini and Mustaca, 2000; Weinstein, 1970b) have been shown to respond to successive negative contrast by disrupting their behaviour compared to control animals which had not experienced a change in reward. The animals display behaviour akin to disappointment – slower running speeds to a reward (Bower, 1961), depressed licking behaviour (Flaherty et al., 1985; Vogel et al., 1968), or reward rejection (Tinklepaugh, 1928).

Contrast effects were also successfully described in invertebrates (Bitterman, 1976; Couvillon and Bitterman, 1984; Richter and Waddington, 1993). Bitterman (1976) found negative incentive contrast effects in honey bees which were trained to a high-quality feeder and then received a downshift to a lower quality feeder. In contrast, bees which experienced an upshift in feeder quality did not show any feeding interruptions (Bitterman, 1976; Couvillon and Bitterman, 1984). While negative successive contrast effects – akin to disappointment – have been well described in animals, positive successive contrast effects – akin to elation – have often proved elusive (Bower, 1961; Black, 1968; Capaldi and Lynch, 1967; Dunham, 1968; Papini et al., 2001). There are several factors which may prevent positive contrast effects from being detected. Firstly, ceiling effects may occur when the performance of animals receiving a large reward is at or near a physical limit. The absence of positive contrast effects may then not be due to the absence of perceived positive contrast, but rather due to an artefact of experimental design (Bower, 1961; Campbell et al., 1970). Secondly, neophobia counteracts positive contrast effects: animals may be reluctant to eat a novel food – even if the food is of higher quality than normal (Flaherty, 1999; Oberhauser and Czaczkes, 2018). Finally, generalisation decrement may prevent stronger responses to positive contrast. Generalisation decrement occurs when animals are trained under one set of stimuli and then tested under another. The strength of the tested response may decrease with increasing differences between the training and testing stimuli (Kimble, 1961), which may then result in weaker positive contrast effects following a reward shift. Thus, the reward change itself may lead to a decrease in responding just as would any other change in context, such as a change in the brightness of the runway or scent of the food (Oberhauser and Czaczkes, 2018; Capaldi, 1978; Premack and Hillix, 1962).

Even though positive contrast effects proved to be hard to demonstrate in laboratory experiments, there are good theoretical reasons for expecting both positive and negative contrast effects to evolve (McNamara et al., 2013). Incentive contrasts have also been demonstrated for rewards other than food. Females become more (or less) likely to accept a mate of given quality if they have prior experience of better (or worse) mates. Such mate quality contrast effects are reported in both vertebrates (Collins, 1995) and invertebrates (Dukas, 2005; Reid and Stamps, 1997; Wagner et al., 2001).

In this study, we investigate positive and negative contrast effects using the successive contrasts paradigm, and, in addition to demonstrating positive and negative contrast effects, define the first relative value curve in an invertebrate; the ant Lasius niger. We conduct a critical control experiment to rule out physiological or psychophysical effects which may lead to the same pattern (see experiment 2) and thus provide strong evidence for a purely cognitive relative value effect in a non-human animal. Furthermore, we demonstrate that information flowing into the nest can influence value perception in outgoing foragers. This suggests that food sources are not only valued based on individual experiences, but also based on social information gained inside the nest. The perceived value of a food source influences social information dissemination, by affecting the strength of pheromone trails which then lead further ants to the food source. Thus, the way in which value is judged is likely to strongly affect the foraging mechanics of a whole colony.

Materials and methods

Study animals

Request a detailed protocol

Eight stock colonies of the black garden ant Lasius niger were collected on the University of Regensburg campus. The colonies were kept in 30 × 30 × 10 cm foraging boxes with a layer of plaster covering the bottom. Each box contained a circular plaster nest box (14 cm diameter, 2 cm height). The colonies were queenless with around 1000–2000 workers and small amounts of brood. Queenless colonies still forage and lay pheromone trails and are frequently used in foraging experiments (Devigne and Detrain, 2002; Dussutour et al., 2004). The colonies were fed with ad libitum 0.5M sucrose solution and received Drosophila fruit flies once a week. Water was available ad libitum.

One sub-colony of 500 individuals was formed from each stock colony, and these eight fixed-size sub-colonies were used for our experiments. Sub-colonies were maintained identically to the stock colonies, but did not receive any Drosophila fruit flies to prevent brood production, and were starved 4 days prior to the experiments in order to achieve a uniform and high motivation for foraging (Mailleux et al., 2006; Josens and Roces, 2000). During starvation, water was available ad libitum. Any ants which died or were removed from the sub-colonies were replaced with ants from the original stock colonies.

General setup, ant selection, and monitoring

Request a detailed protocol

The general setup used for all of our three experiments was identical and consisted of a 20 × 1 cm long paper-covered runway which was connected to the sub-colony’s nest box via a 40 cm long drawbridge (Figure 1A). A 5 mm diameter drop of sucrose solution (Sigma-Aldrich) was placed on an acetate feeder at the end of the runway (60 cm from the nest). The molarity of the sucrose droplet depended on the experiment, treatment and on the ants’ number of visit to the food source.

Experimental Setups.

(A) General setup used for all presented experiments. The 20 cm long runway is connected to the nest box via a 40 cm long drawbridge. The droplet of sucrose solution is placed at the end of the runway (60 cm distance to the nest). (B) Y-maze used on the 10th visit of experiment 2. All arms were 10 cm long. The arm connected to the nest box was covered with unscented paper overlays while the other two arms were covered with lemon and rosemary scented paper overlays (one scent on each side). Visual cues (landmarks) were placed directly behind the two scented arms. The first decision line was located 2.5 cm from the Y-maze centre and marked the initial decision of an ant while the second decision line was located 7.5 cm from the centre and marked the final decision.

https://doi.org/10.7554/eLife.45450.003

To begin an experiment, 2–4 ants were allowed onto the runway, and the first ant to reach the feeder was marked with a dot of acrylic paint on its gaster. This procedure may select for the more active foragers, but does not introduce any selection bias between treatments. The marked ant was allowed to drink to repletion at the food source, while all other ants were returned to the nest.

Food acceptance scores as a measure of perceived value were noted for each ant and visit as follows: Full acceptance (1) was scored when the ant remained in contact with the drop from the moment of contact and did not interrupt drinking within 3 s of initial contact (see Video 1). Partial acceptance (0.5) was scored if feeding was interrupted within 3 s after the first contact with the food source, but the ant still filled its crop within 10 min (as can be seen by the distention of the abdominal tergites). Ants which interrupt feeding within the first seconds after contacting the food usually show successive feeding interruptions and generally show a rather ‘impatient’ behaviour compared to ants which show a food acceptance score of 1 (see Video 2). Lastly, rejection (0) was scored if the ant refused to feed at the sucrose solution and either returned to the nest immediately or failed to fill its crop within 10 min.

Video 1
Ant displaying food acceptance score 1.

It shows no food interruptions within the first seconds of feeding.

https://doi.org/10.7554/eLife.45450.004
Video 2
Ant displaying food acceptance score 0.5.

It interrupts feeding within the first seconds of feeding and repeatedly interrupts feeding, but still feeds at the food source (an ant displaying food acceptance score 0 would refuse to feed at the sucrose solution and either return to the nest immediately or fail to fill its crop within 10 min).

https://doi.org/10.7554/eLife.45450.005

When the ant had filled its crop or decided not to feed at the sucrose droplet, it was allowed to return to the nest. Inside the nest, the ant unloaded its crop to its nestmates and was then allowed back onto the runway for another visit. The drawbridge was now used to selectively allow only the marked ant onto the runway.

In addition to measuring food acceptance, we also measured pheromone deposition. Individual pheromone deposition behaviour correlates with the (perceived) quality of a food source (Beckers et al., 1993; Hangartner, 1970; Czaczkes et al., 2015). Individual ants can adapt the strength of a pheromone trail by either depositing pheromone or not, or varying the intensity of a pheromone trail through number of pheromone depositions (Beckers et al., 1993; Hangartner, 1970). Pheromone deposition behaviour in L. niger is highly stereotypic. To deposit pheromone, an ant briefly interrupts running to bend its gaster and press the tip of the gaster onto the substrate (Beckers et al., 1992). This allows the strength of a pheromone trail to be quantified by counting the number of pheromone depositions over the 20 cm runway leading to the feeder. Pheromone depositions were measured each time the ant moved from the food source back to the nest (inward trip), and each time the ant moved from the nest towards the food source (outward trip). Because L. niger foragers almost never lay pheromone when they are not aware of a food source (Beckers et al., 1992), we did not measure pheromone depositions for the very first outward trip (visit 1). The presence of trail pheromone on a path depresses further pheromone deposition (Czaczkes et al., 2013). Thus, each time an ant had passed the 20 cm runway, the paper overlay covering the runway was replaced by a fresh one every time the ant left the runway to feed at the feeder or returned to the nest.

All experimental runs were recorded with a Panasonic DMC-FZ1000 camera to allow for later video analysis. Each tested ant was observed until all experimental runs were finished and then discarded from the colony before switching to the next ant. If an ant did not return before finishing all experimental runs, we waited for 15 min, then discarded it from the colony and moved to the next ant.

Statistical analysis

Request a detailed protocol

Statistical analyses were carried out in R v. 3.4.1 (R Development Core Team, 2016) using Generalized Linear Mixed Models (GLMMs) in the LME4 package (Bates et al., 2014) to analyse pheromone depositions data and Cumulative Link Mixed Models (CLMMs) in the ordinal package (Christensen, 2015) to analyse food acceptance scores. CLMMs were used to analyse the acceptance data since we used an ordered factor with three levels (1 = full acceptance, 0.5 = partial acceptance, 0 = rejection).

As multiple ants were tested per colony, colony identity was added as a random effect to each model. GLMMs were tested for fit, dispersion and zero inflation using the DHARMa package (Hartig, 2017). The model predictors and interactions were defined a priori, following Forstmeier and Schielzeth (2011). All p-values presented were corrected for multiple testing using the Benjamini–Hochberg method (Benjamini and Hochberg, 1995). A total of 1070 ants were tested, with 829 in experiment 1, 73 in experiment 2 and 168 in experiment 3 (Supplementary file 1). Sample sizes were set ahead of time by deciding how much time we will invest in data collection (1 day per treatment per colony).

Food acceptance data

Request a detailed protocol

Depending on the experiment, we either used treatment (experiment 1 and 3 = Reference Molarity; experiment 2 = expected molarity triggered by a scented runway and the odours presented on the runway) or an interaction between treatment and visit number, and the odours presented on the runway (training visits of experiment 2) or trophallaxis time (experiment 3) as fixed factors. The interaction between expected molarity and visit number in the training runs of experiment 2 was added, because experience with a food source is likely to affect the behaviour at a food source. The odours presented on the runway were added as fixed factors to test for odour preferences regardless of sucrose molarity. The interaction between trophallaxis time and reference molarity in experiment 3 was added because trophallaxis time may affect food acceptance through crop load and information gained through trophallaxis (for the effects of trophallaxis time on food acceptance see Figure 5—figure supplement 1, and Table S4 in Figure 5—source data 1). Because individual ants were tested multiple times in experiments 1 and 2, we included AntID nested in colony as a random factor for statistical analyses of the training visits.

We used the following general model formula (this formula varied depending on experiment as described above):

FoodAcceptancetreatment+(randomfactor:colony)

Pheromone deposition data

Request a detailed protocol

As the pheromone deposition data is count data, they were analysed using a GLMM with a Poisson distribution.

Depending on the experiment, we either used treatment (experiment 1 = Reference Molarity; experiment 2 = expected molarity triggered by a scented runway and the odours presented on the runway) or an interaction between treatment and visit number (training visits of experiment 2) as fixed factors. The interaction between expected molarity and visit number in the training runs of experiment 2 was added, because experience with a food source is likely to affect the behaviour at a food source. The odours presented on the runway were added as fixed factors to test for odour preferences regardless of sucrose molarity. Because individual ants were tested multiple times in experiment 2, we included AntID nested in colony as a random factor for statistical analyses of the training visits.

For statistical analysis of experiment 1, we also added a variable indicating if ants deposited more or less pheromone compared to the average to correct for individual strength of pheromone depositions and overdispersion. The variable was calculated as follows:

Differencetoaverage=((NumberPheromoneDepositions1stvisitmeannumberPheromoneDepositions1stvisit)+(NumberPheromoneDepositions2ndvisitmeannumberPheromoneDepositions2ndvisit))/2

We used the following model formulae in the model:

Experiment 1:

NumberPheromoneDepositionstreatment+Differencetoaverage+(Differencetoaverage)2+(randomeffects:colony/AntID)

Experiment 2:

NumberPheromoneDepositionsscentassociatedtomolarity+(randomeffects:colony)

Other analyses

Request a detailed protocol

The number of drinking interruptions was quantified via video analysis in experiment 2 (see below). This was analysed statistically in a manner identical to number of pheromone depositions.

Trophallaxis time in seconds in experiment three were used in full seconds and treated as count data. We performed a GLMM with Poisson distribution and Reference Molarity as a fixed effect, while colony identity was added as a random factor:

TrophallaxisTimesecondsReferenceMolarity+(randomeffects:colony)

Experiment 1 – Defining a relative value perception curve

The aim of this experiment was to test whether Lasius niger ants value a given absolute sucrose solution concentration relative to a reference point or based on its absolute value. We used a range of twelve molarities as reference points in order to describe a value curve. To exclude effects of the researcher’s expectations on the data, the data for this experiment were collected blind to treatment (Holman et al., 2015).

Experiment 1 - Methods

Ants made two initial training visits to a feeder at the end of a runway in order to set their reference point (Figure 1A). The quality of the sucrose solution was varied between ants, with each ant receiving the same quality twice successively. Twelve different molarities were used: 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1, 1.5 or 2M (also referred to as pre-shift solution or reference point). Lasius niger workers learn the quality of a feeder within two visits (Wendt and Czaczkes, 2017). On the third visit (test visit), the food source was replaced by a 0.5M sucrose solution droplet for all ants (also referred to as post-shift solution). Thus, ants trained to qualities < 0.5M experienced a positive successive contrast, ants trained to > 0.5M experienced a negative successive contrast, and the ants trained to 0.5M constituted the control (no contrast). 97% of ants successfully finished the training procedure and participated in the test visit (third visit).

Experiment 1 - Results

Ants seemed to value sucrose solution droplets relative to their reference point (Figure 2—figure supplement 1). In the training visits, acceptance scores increased significantly with increasing molarity of the reference quality (CLMM: estimate = 1.97, z = 9.65, p<0.001, Figure 2). However, in the test (contrast) visit, acceptance scores decreased significantly with increasing molarity of the reference quality (CLMM: estimate = −2.59, z = −13.57, p<0.001, Figure 2). Ants which were trained to the lowest molarity (0.1M: p<0.001) showed significantly higher acceptance of 0.5M sucrose than control ants, while ants trained to high molarities (1.5M: p<0.001, 2M: p<0.001) showed lower acceptance of 0.5M than the control group (see Table S1 in Figure 2—source data 1 for all pairwise comparisons).

Figure 2 with 1 supplement see all
Food acceptance shown in experiment one for the two training visits (visit 1 and 2) in which ants received one of 12 molarities and the test visit (3) in which all ants received 0.5M (sample sizes: 0.1M: 57; 0.2M: 80; 0.3M: 76; 0.4M: 66; 0.5M: 77; 0.6M: 65; 0.7M: 73; 0.8M: 66; 0.9M: 72; 1M: 55; 1.5M: 72; 2M: 70).

Shown are the mean food acceptance (points) and the 95% confidence intervals (coloured ribbons) for each reference molarity and visit. Data was normalised to show the mean food acceptance of the control group (received 0.5M on each visit) at 0 for all three visits. For a non-normalised graph of the data see Figure 2—figure supplement 1.

https://doi.org/10.7554/eLife.45450.006

A similar pattern was found for pheromone deposition behaviour on the way back to the nest (Figure 3). In the training visits, number of pheromone depositions increased significantly with increasing molarity of the reference solution (GLMM: estimate = 0.86, z = 13.87, p<0.001). By contrast, on the test visit pheromone depositions decreased significantly with increasing molarity of the reference solution (GLMM: estimate = −0.82, z = −9.75, p<0.001, Figure 3). Ants which deposited more pheromone during the training visits generally deposited more pheromone on the test visit compared to ants which deposited less pheromone during the training visits (GLMM: estimate = 0.16, z = 15.99, p<0.001). Ants which were trained to a low molarity (0.2M: p<0.01) deposited significantly more pheromone in the test visit than control ants, while ants trained to high molarities (1M: p<0.001, 1.5M: p<0.001, 2M: p<0.001) deposited less pheromone than the control group (see Table S2 in Figure 3—source data 1 for pairwise comparisons).

Figure 3 with 1 supplement see all
Pheromone depositions on the way back to the nest shown in experiment one for the two training visits (visit 1 and 2) in which ants received one of 12 molarities and the test visit (3) in which all ants received 0.5M (sample sizes: 0.1M: 57; 0.2M: 80; 0.3M: 76; 0.4M: 66; 0.5M: 77; 0.6M: 65; 0.7M: 73; 0.8M: 66; 0.9M: 72; 1M: 55; 1.5M: 72; 2M: 70).

Shown are the median number of pheromone depositions (points) and the 95% confidence intervals (coloured ribbons) measured on a 20 cm track right behind the food source for each reference molarity and visit. Data was normalised to show the median number of pheromone depositions of the control group (received 0.5M on each visit) at 0 for all three visits. For a non-normalised graph of the data see Figure 3—figure supplement 1.

https://doi.org/10.7554/eLife.45450.009

These results are consistent with relative value perception stemming from the psychological effects of successive contrasts. We could further define a relative value perception curve similar to that described in Prospect Theory, as well as showing positive contrast effects for both food acceptance and number of pheromone depositions.

However, there is another possible explanation for these results: non-random selection of individuals with different acceptance thresholds. Different individuals from the same colony may have different acceptance thresholds. Animals with lower acceptance thresholds may readily exploit low-quality food sources while animals with higher thresholds may not. When training to lower molarity sucrose, ants with high thresholds may not have completed training, leaving only a non-random subset of ants with low acceptance thresholds at the test phase (Robinson et al., 2009). Thresholds may also be influenced by experience, by which animals use the best experienced option as a threshold for accepting a new option or not (Stroeymeyt et al., 2011; Robinson et al., 2011). However, we can exclude this possibility, as the proportion of ants not completing training was uniformly low and did not vary with treatment (see Supplementary file 1).

Experiment 2 – ruling out alternative explanations using scent training

Alternative hypotheses could also explain the results from experiment one and lead to the same behavioural patterns observed. Five possible ‘lower-level’ mechanisms must be excluded: sensory satiation, ingested sucrose changing haemolymph-sugar levels, psychophysical sensory contrast effects, the fact that ants may expect pre-shift solutions to return in later visits, and non-random selection of individuals with different food acceptance thresholds in different treatments.

Sensory satiation

This may occur in ants which were trained to higher molarity food due to the blocking of more sweetness receptors compared to low molarity sucrose. The more sweetness receptors are blocked by a sweet reference solution, the fewer receptors will fire when confronted with a post-shift reward, thus making solutions taste less sweet for ants trained to high-molarity solutions, and sweeter for ants which were trained to low molarities (Bitterman, 1976).

Haemolymph-sugar levels

Ants may not only have stored sucrose solutions in their crop while foraging, but may also have ingested small amounts of it, leading to an increase of haemolymph-sugar levels. Higher blood-sugar levels negatively affect sweetness perception in humans (Mayer-Gross and Walker, 1946; Melanson et al., 1999), and a similar effect could cause a post-shift solution to taste less sweet to animals trained on high sucrose concentrations.

Psychophysical sensory contrasts

The contrast effects shown in experiment one could also derive from simple psychophysical mechanisms (Fechner, 1860; Zwislocki, 2009), and thus arise from sensory perceptual mechanisms rather than higher level cognitive processing of value. Sensory judgements are usually made relative to reference points and through constant comparisons with former stimuli (Vlaev et al., 2011; Helson, 1964). Thus, identical stimuli may be perceived differently depending on the context they are presented within. The position of the reference point in the range of stimuli may thus bias how the stimulus, and thus the value, of a post-shift reward is perceived (Zwislocki, 2009). For example, the sweetness of a sucrose solution may be perceived as much stronger when the reference point to which it is compared is low. Psychophysical sensory contrasts are physiological or low-level cognitive phenomena, found in all animal taxa studied, and even potentially in bacteria (Akre and Johnsen, 2014; Kojadinovic et al., 2013; Mesibov et al., 1973).

Future expectations

Animals may rationally expect the pre-shift reward to be available in the future again and therefore rationally show lower acceptance towards the post-shift reward, because they are waiting for the pre-shift reward to reoccur.

All these alternative factors would lead to the same behavioural patterns found in experiment one without relative value perception necessarily being present. Experiment two was designed to rule out these alternative explanations.

Experiment 2 - Methods

To rule out the alternative non-psychological explanations for the contrast effects we described above, we needed to change the expectations of the ants while exposing all ants to identical training regimes. This would provide a reference point for testing relative value perception while keeping sensory saturation, haemolymph-sugar levels, psychophysical effects, future expectations, and ant subsets the same until the testing phase.

Ants were allowed to make eight training visits. The quality of the sucrose solution offered at the end of the runway alternated each visit, always beginning with the low-quality solution. The solutions were scented using either rosemary or lemon essential oils (0.05 µl essential oil per ml sucrose solution, rosemary: Rosmarinus officinalis; Lemon: Citrus limon, Markl GbR, Grünwald). In half the trials the 1.5M solution was scented with lemon and the 0.25M with rosemary, and vice versa for the other trials. In addition, to support learning and to allow solution quality anticipation, we also scented the paper overlays covering the runway leading to the feeder. Paper overlays were scented by storing them for at least 1 day in an airtight box containing a droplet of essential oil on filter paper in a petridish. Finally, in addition to odours cuing sucrose molarity, visual cues were also provided. These consisted of printed and laminated pieces of paper (22 × 16.5 cm, displayed in Figure 1B) displayed at the end of the runway, directly behind the sucrose droplet.

On the 9th (test) visit, the odour of the runway and the visual cue signified either 1.5M or 0.25M, while the sucrose solution provided was unscented and of intermediate (0.5M) quality. Runway scents in the test visit were varied systematically between ants, but each ant was confronted with only one of the two runway scents coupled with unscented 0.5M sucrose. While the ant fed at the sucrose droplet, the scented runway overlay was replaced with an unscented overlay in order to eliminate possible effects of scent association on pheromone deposition behaviour. Previous work has shown that L. niger foragers can form robust expectations of upcoming reward quality based on runway odour after four visits to each odour/quality combination (Czaczkes et al., 2018b). Nonetheless, to ensure that learning had taken place, on the 10th visit, we carried out a memory probe. The linear runway was replaced with a Y-maze (Figure 1B), with two 10 cm long arms and a 10 cm long stem. The Y-maze stem was covered with an unscented paper overlay while one arm was covered with the 1.5M-associated odour overlay, and the other with the 0.25M-associated odour overlay. The matching visual cues were placed directly behind the relevant Y-maze arms. Trained ants were allowed to walk onto the Y-maze and their arm choice was noted. We used two decision lines to define arm choice – an initial decision line (Figure 1B, 2.5 cm after the bifurcation) and a final decision line (7.5 cm after the bifurcation). After testing on the Y-maze, the ants were permanently removed from the colony.

97.2% of ants successfully finished the training procedure and participated in the last test visit.

Additionally to the other measures, on the 9th (test) visit of this experiment we counted the number of food interruptions made by an ant from the moment of first hitting the food source until it had finished feeding at the sucrose droplet. The number of food interruptions are likely to reflect and support the behaviour encoded in food acceptance scores and was thus investigated to give stronger support for the results of this experiment.

Experiment 2 - Results

During training, ants behaved as expected, showing higher acceptance and pheromone deposition for 1.5M compared to 0.25M on all but the very first visit to 0.25M (Food acceptance: CLMM: estimate = −7.34, z = −8.9, p<0.001; pheromone depositions outward journey: GLMM: estimate = 0.23, z = 2.89, p<0.01; pheromone depositions inward journey: GLMM: estimate = −2.49, z = −19.46, p<0.001, Figure 4A,C & E). Furthermore, food acceptance and pheromone depositions both on the outward and inward journeys decreased with increasing experience with the 0.25M feeder and increased with increasing experience with the 1.5M feeder (Food acceptance: CLMM: estimate = −2.84, z = −3.63, p<0.001; pheromone depositions outward journey: GLMM: estimate = −0.94, z = −10.00, p<0.001; pheromone depositions inward journey: GLMM: estimate = −0.53, z = −4.41, p<0.001).

Figure 4 with 1 supplement see all
Food acceptance (A and B) and number of pheromone depositions towards the food source (C and D) and towards the nest (E and F) in experiment 2.

The left panels (A, C, E) show behaviour over the eight training trials, in which ants received 0.25M coupled with one scent and 1.5M coupled with another scent on alternating visits. The right panels (B, D, F) show behaviour on the test visit, in which ants always received unscented 0.5M sucrose solution, but the runway leading towards the food source was impregnated with one of the learned scents, causing ants to expect either a high or low reward. 40 ants were induced to expect a high reward, and 32 to expect a low reward. A and B show the mean food acceptance (points) and the 95% confidence intervals (error bars) for each visit; CF show the median number of pheromone depositions on a 20 cm track leading to the food source and the 75%/25% quantiles for each visit.

https://doi.org/10.7554/eLife.45450.012

On the outward journey of the 9th (test) visit, ants walking towards the feeder while exposed to 1.5M sucrose-associated cues deposited more pheromone (median = 15, Figure 4D) compared to ants exposed to 0.25M-associated cues (median = 2, GLMM: estimate = −1.31, z = −12.94, p<0.001). Moreover, in the learning probe, 87% of ants chose the 1.5M associated arm. This demonstrates that ants formed a robust expectation of food molarity based on the cues learned during training.

Ants exposed to 1.5M-associated cues during the 9th visit showed significantly lower food acceptance towards the unscented 0.5M feeder than ants exposed to 0.25M-associated cues (CLMM: estimate = 1.04, z = 2.049, p<0.05, Figure 4B, Supplementary file 1). Although ants exposed to high molarity associated cues – presented through scented runways on the way to the food – showed a significantly higher number of pheromone depositions on their return journey than ants confronted with low molarity scent (GLMM: estimate = −1.65, z = −3.03, p<0.01, Figure 4E & F), the number of pheromone depositions decreased drastically for both treatments compared to training visits (median 1.5M = 0, median 0.25M = 0, Figure 4E and F, Supplementary file 1).

Even after controlling for alternative explanations, ants still show contrast effects depending on the quality of the post shift solution. This is in spite of all ants undergoing identical training experiences. The only difference between the groups was the odour of the runway on the 9th (test) visit. It is thus unlikely that sensory saturation, increased haemolymph-sugar levels, simple psychophysical effects or ants expecting pre-shift solutions to return can fully explain the behaviour of the ants in our experiments. All videos were re-analysed by a naive scientific assistant and this blind analysis of the ants behaviour confirmed the stated results (CLMM: estimate = 1.42, z = 2.35, p=0.019), and also found that ants interrupted drinking significantly more often when expecting high rather than low food qualities (GLMM, estimate = 0.36, z = 2.74, p=0.006, see Figure 4—figure supplement 1—source data 1 and Figure 4—figure supplement 1).

Non-random selection of individuals with different acceptance thresholds can also be excluded for the results of this experiment as the proportion of ants not completing training was again uniformly low (see Supplementary file 1) and all ants had to taste both low and high molarities in order to complete training.

Experiment 3 – expectation setting via trophallaxis: the nest as an information hub

Ants receive information about available food sources, such as food odour and palatability, through food exchanges (trophallaxis) inside the nest (Provecho and Josens, 2009; Josens et al., 2016). An ant beginning a food scouting bout may not have direct information about the quality of the food sources available in the environment, but nonetheless must make a value judgement on their first visit to a food source. The aim of this experiment was to ascertain whether information about sucrose concentrations gained through trophallaxis in the nest affected the perceived value of food sources found outside the nest.

Experiment 3 - Methods

An ant was allowed to feed at an unscented sucrose solution droplet of either 0.16, 0.5 or 1.5M (also referred to as pre-shift solution or reference point) and return to the nest to unload its crop via trophallaxis. When trophallaxis began, we noted the time spent in trophallaxis with the first trophallactic partner. When trophallaxis stopped, the receiving trophallactic partner (receiver) was gently moved from the nest and placed onto the start of a runway offering unscented 0.5M sucrose solution at the end (also referred to as post-shift solution). As the receiver fed, we noted its food acceptance.

Experiment 3 - Results

Acceptance scores of receivers towards 0.5M decreased with increasing molarity of the sucrose solution received through food exchanges inside the nest (CLMM: estimate = −0.57, z = −3.07, p<0.01). The interaction of reference molarity and trophallaxis time significantly predicted acceptance (CLMM: estimate = −0.48, z = −2.33, p=0.02, Figure 5) and longer trophallaxis times led to lower food acceptance in ants as well (CLMM: estimate = −0.70, z = −3.62, p<0.001). Ants which received 0.16M inside the nest showed significantly higher acceptance of 0.5M sucrose than ants which received 1.5M (p<0.01, see Table S3 in Figure 5—source data 1 for pairwise comparisons). The time spent in trophallaxis with the receiver increased significantly with increasing molarity (GLMM: estimate = 0.13, z = 4.79, p<0.001, see Figure 5—source data 1).

Figure 5 with 1 supplement see all
Food acceptance shown in experiment 3 for the receivers which received either 0.16, 0.5 or 1.5M through trophallaxis in the nest and then found 0.5M at the end of the runway (sample sizes: 0.16M 63; 0.5M: 52; 1.5M: 53).

Shown are the mean food acceptance (points) and the 95% confidence intervals (error bars) for each reference molarity.

https://doi.org/10.7554/eLife.45450.021

Ants valued a standard quality food source relative to the molarity which they received from a returning forager inside the nest. This suggests that information about the quality of a food source received through trophallactic interactions inside the nest can be used by naive foragers when evaluating new food sources outside the nest. Thus, the nest serves as an information hub in which information about available food sources can be gathered, processed, and disseminated.

Discussion

The introduction of Prospect Theory (Kahneman and Tversky, 1979) contributed to a major shift in economic research by suggesting that humans do not perceive value in absolute terms, but relative to reference points. Here, we demonstrate parallel findings in an insect. To the best of our knowledge, we provide the first detailed description of relative value perception in an invertebrate based on individual experience, but also induced by social information. Furthermore, we demonstrate the elusive positive contrast effects in ants which were trained to low molarities (Figure 2 and 3).

Similar results in house-hunting ants were explained by a simple threshold rule (Stroeymeyt et al., 2011; Robinson et al., 2009; Robinson et al., 2011) which suggests that individuals have different acceptance thresholds and ants with lower thresholds accept lower quality options. The higher the quality of the option, the more often it exceeds the acceptance threshold of individual ants, and thus the option is accepted more readily. This could have potentially affected our results in experiment 1, as we would expect fewer individuals to accept very low reference points. Ants which did not accept the low-quality sucrose would thus not be tested. Therefore, at low reference points, we would only select individuals with very low acceptance thresholds, while no threshold selection would occur at high reference points. When confronting ants with medium-quality food after training, the differently selected acceptance thresholds may lead to the same pattern as we observed. However, 97% of all ants finished both the training and the test phases and no higher proportion of cancelled training can be seen at lower reference molarities (see Supplementary file 1). It is thus unlikely that a simple threshold rule leads to the results shown in experiment 1 (Figure 2 and 3).

While a second major prediction of Prospect Theory, that ‘losses loom larger than gains’ (Tversky and Kahneman, 1992), is not supported by the data of our main experiment, it is also not ruled out. We believe ants do overemphasise losses, but, due to limitations in the experimental design and physiological limitations of the animals, we cannot make strong claims about this (Collins, 1995; Dukas, 2005; Reid and Stamps, 1997). The lack of strong evidence for losses being overemphasised may stem from the psychophysics of our study system: a basic tenant of psychophysics is that the Just Noticeable Difference (JNDs) between two stimuli is a function of the relative difference between the stimuli (Fechner, 1860; Zwislocki, 2009; Stevens, 1957). Thus, ants shifted from 0.1M to medium (0.5M) quality experience a fivefold increase in molarity, while those down-shifted from 0.9M to 0.5M experience less than a twofold decrease, although the absolute change was of the same magnitude. This would predict larger shift-changes, in terms of absolute molarity change, for gains than for losses. Indeed, the fact that this is also not seen may imply that losses are indeed – relatively speaking – looming larger than gains for the ants. Finally, it must be kept in mind that acceptance scores are unlikely to be linear, and that pheromone deposition behaviour shows large variation (Beckers et al., 1992), making it difficult to use either of these factors to quantitatively test for over- and undervaluation of gains and losses.

The results of experiment 2 allow us to exclude all but a cognitive relative value effect (Figure 4). This cognitive effect is subjectively very familiar to humans, and its presence in an invertebrate is at first glance surprising. However, insects have been shown to display many cognitive traits in parallel with humans (Brosnan and De Waal, 2003; Aw et al., 2011; Premack and Hillix, 1962; Dukas, 2005; Reid and Stamps, 1997), and contrast effects are likely selected for (McNamara et al., 2013).

The smaller effect size in experiment two is presumably driven either by the exclusion of the additional driving factors (see experiment two description), or the additional complications involved in an extensive training regime, or both. Specifically, the expectations leading to contrast effects in experiment two were driven by differential learning of odour-quality associations, rather than a simple one-component memory of food quality as may have been the case in experiment 1. This may have weakened the observed effect.

Another possible explanation for smaller effect sizes may be that in experiment 2 ants had access to two reference points (0.25M and 1.5M) to use for value judgement of the medium-quality food in the control experiment, while in experiment 1 they only had one reference. Thus, while the odour cue may have overemphasised the role of the associated quality as reference, the competing reference quality may have been acting as a second reference. Additional reference points are likely to affect the scale post-shift rewards are compared to (Zwislocki, 2009). This possibility is supported by the acceptance data collected during training in experiment 2. On the first training visit, all ants encountered low quality food and showed a high food acceptance towards the feeder (Figure 4A). However, as soon as ants had experienced a high-quality sucrose solution, the previously acceptable low-quality food became unattractive, and food acceptance scores decreased from a mean of 0.99 to 0.39. This strongly suggests that the ants were valuing the training solutions in relation to each other, and may therefore have used both reference points to judge the value of an unscented medium-quality food source. It is possible that the ants may have calculated an average from both reference points, and used the average as comparison to judge the value of the post-shift reward (Flaherty, 1999), as shown in rats (Peters and McHose, 1974). However, the fact that medium quality elicited different food acceptance scores depending on the runway scent makes it unlikely that this would be the only factor affecting acceptance scores.

Lastly, masking effects may also explain the smaller contrast effects of experiment 2 compared to experiment 1: learning theory suggests that neutral cues associated to positive stimuli will elicit positive responses even when no reward is given and vice versa (Rescorla and Wagner, 1972). Therefore, since ants were confronted with the scent associated to high-quality food, food acceptance may have been affected by the scent itself, leading to an elevated food acceptance compared to ants tested in experiment 1 which did not receive a positive cue, but only medium-quality food.

The reduced pheromone deposition seen in the final return in experiment 2 may be due to the change in environment (scented runways to unscented runways) causing a disruption in recruitment behaviour, perhaps due to generalisation decrement (Kimble, 1961; Capaldi, 1978) or neophobia (Barnett, 1958; Johnson, 2000; Mitchell, 1976; Pliner and Loewen, 1997). Furthermore, since only the scented paper overlays were replaced by unscented ones, but not the runways themselves, it is possible that small portions of the odours were still present, driving the ants to deposit pheromone according to the remaining odours, with higher deposition rates for the high-quality associated odour. In a separate experiment, such pheromone deposition directly related to quality-associated odours on runways was clearly demonstrated (Wendt and Czaczkes, 2019). This would explain why pheromone depositions were higher for ants returning to the nest from a high molarity scent than in ants returning from a low molarity scent.

Information about sucrose concentrations gained through trophallactic interactions inside the nest can affect the way newly discovered food sources are valued outside the nest (Figure 5), as well as providing other information (Provecho and Josens, 2009; Josens et al., 2016; LeBoeuf et al., 2016). By taking into account information gained inside the nest, recruited workers are able to evaluate newly discovered food sources in relation to other food sources available in the environment. Ants could thus forego food sources which are of lower quality than the average available food sources (Wendt and Czaczkes, 2017). Even though higher trophallaxis times led to lower acceptance scores and trophallaxis times were higher at high reference molarities, this does not necessarily imply that ants ingested more sucrose at higher references and were thus less hungry or motivated. Higher sucrose solutions are more viscose and thus ants take longer to ingest the same amount of sucrose compared to low molarities (Josens et al., 1998). If, however, more sucrose solution was transferred between the returning forager and the recruit at longer trophallaxis times, it is likely that information input increases and food acceptance decreases. The longer the trophallaxis time, the more the recruit can fill its crop through trophallaxis and therefore the food acceptance may decrease, because the recruit is less starved than an ant which showed a short trophallaxis time. However, even if more food was transferred, the food acceptance scores are a measure of the first assessment of ants at a food source, not the ingested volume. Thus, while some ants may have had less space in their crop left, this may not necessarily affect the food acceptance score, while it is very likely to affect ingested volume after trophallaxis. Additionally, if longer trophallaxis times lead to more ingested sucrose solution, it is also more likely that a higher amount of information about the past food quality is transferred. Thus, more transferred food during trophallaxis may have led to better informed ants reaching the post-shift solution and thus stronger contrast effects. Since the data shows clear effects of both trophallaxis time and reference solution on the food acceptance of 0.5M sucrose, longer trophallaxis times cannot be the only factor driving the contrast effects found in this experiment (see Figure 5—source data 1 and Figure 5—figure supplement 1). Even at high trophallaxis times, ants with a 0.16M reference showed no low food acceptance scores, unlike ants with high reference solution after long trophallaxis times.

Ultimately, we see the nest serving as an information hub, in which information about currently available food sources can be collected, synthesised, and fed back to outgoing foragers. Relative value perception can therefore be expected to have strong effects not only on the individual behaviour of animals, but also on the collective behaviour of insect colonies. For example, colonies of house-hunting ants developed an aversion towards mediocre nests when housed in high-quality nests, but not when they were housed in low-quality nests. Such mediocre nests are then avoided when colonies have to find a new nest site while newly discovered mediocre nests are readily accepted (Stroeymeyt et al., 2011; Robinson et al., 2011). However, while in house-hunting the reference resource is directly experienced by scouts only, we demonstrate that information brought back to the nest can set a reference point for ants which have not directly experienced the resource in situ.

A broad range of behaviours relevant to behavioural economics have been described in invertebrates (Czaczkes et al., 2018a; Pompilio et al., 2006; Wendt and Czaczkes, 2017; Czaczkes et al., 2018b; Cheng et al., 2002). We propose that invertebrates make attractive models for a broader understanding of behavioural economics in humans. The benefits of an interdisciplinary approach will likely flow both ways. Using animal models allows researchers to avoid pitfalls associated with studies on humans, such as cultural and educational differences (Carter and Irons, 1991; Guiso et al., 2006), second-guessing of experimenters, and non-relevant reward sizes (Levitt and List, 2007) as well as relaxing ethical concerns. The game-like designs of many economic experiments are highly artificial and the incentive magnitudes that can be provided are limited (Kahneman and Tversky, 1979; Levitt and List, 2007). While there has been much progress in field studies on humans to clearly measure causal relationships (Harrison and List, 2004), the usefulness of these new techniques is constrained by the range of questions and settings to which they can be applied. Hence, while behavioural studies on invertebrates also have their limitations (for example, in that inducing expectations is more of a challenge), they can be easily designed to be ecologically meaningful, and offer rewards which are in line with the real-life budgets under which the animals operate. Finally, due to human complexity, building economic models which accurately predict human behaviour is challenging. Insect economic behaviours are demonstrably similar to that of humans, but likely simpler. We therefore propose that economic models to predict invertebrate decision making may be a complementary step on the way to predicting human behaviour.

There is a well-developed tradition of integrating economics and biology (Aw et al., 2011; Czaczkes et al., 2018a; Lydall et al., 2010; Aw et al., 2009; Wendt and Czaczkes, 2017; Cheng et al., 2002; Evans and Westergaard, 2006). Here we provide a systematic description of value judgment relative to a reference point in ants, define a relative value curve as described in Prospect Theory, and provide some of the first strong evidence for a purely cognitive element to relative value judgement. Reference points can not only be set by individual experiences but also through social information such as pheromone trails or through trophallactic contacts inside the nest. We feel a critical mass of evidence is now available to consider comparative behavioural economics as a relevant discipline for both biologists and economists.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
    Elemente Der Psychophysik [Elements of Psychophysics]
    1. GT Fechner
    (1860)
    Leipzig, Germany: Breitkopf und Härtel.
  34. 34
  35. 35
  36. 36
    Incentive Relativity
    1. CF Flaherty
    (1999)
    Cambridge University Press.
  37. 37
  38. 38
  39. 39
  40. 40
    Field experiments
    1. GW Harrison
    2. JA List
    (2004)
    Journal of Economic Literature 42:1009–1055.
    https://doi.org/10.1257/0022051043004577
  41. 41
    Residual Diagnostics for Hierarchical (Multi-Level / Mixed) Regression Models
    1. F Hartig
    (2017)
    Residual Diagnostics for Hierarchical (Multi-Level / Mixed) Regression Models, http://florianhartig.github.io/DHARMa/.
  42. 42
  43. 43
    Adaptation-Level Theory: An Experimental and Systematic Approach to Behavior
    1. H Helson
    (1964)
    New York: Harper and Row.
  44. 44
  45. 45
  46. 46
    Nectar feeding by the ant camponotus mus: intake rate and crop filling as a function of sucrose concentration
    1. R Josens
    2. WM Farina
    3. RB Josens
    (1998)
    Journal of Insect Physiology 44:579–585.
  47. 47
  48. 48
  49. 49
  50. 50
    Hilgard and Marquis’ “Conditioning and Learning”
    1. GA Kimble
    (1961)
    East Norwalk, CT US: Appleton-Century-Crofts.
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
    Principles of Economics
    1. NG Mankiw
    (2011)
    South-Western College Pub.
  57. 57
    Taste and selection of food in hypoglycaemia
    1. W Mayer-Gross
    2. JW Walker
    (1946)
    British Journal of Experimental Pathology 27:297–305.
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65
    Value Judgments: Toward a Relational Theory of Happiness
    1. A Parducci
    (1984)
    In: J. R Eiser, editors. Attitudinal Judgment. New York: Springer. pp. 3–21.
  66. 66
  67. 67
  68. 68
  69. 69
  70. 70
  71. 71
  72. 72
    R: A language and environment for statistical computing 
    1. R Development Core Team
    (2016)
    R Foundation for Statistical Computing, Vienna, Austria.
  73. 73
  74. 74
    A Theory of Pavlovian Conditioning: Variations in the Effectiveness of Reinforcement and Nonreinforcement
    1. RA Rescorla
    2. AR Wagner
    (1972)
    In: A. H Black, W. F Prokasky, editors. Classical Conditioning II: Current Research and Theory. New York: Appleton Century Crofts. pp. 64–99.
  75. 75
  76. 76
    Do ants make direct comparisons?
    1. EJH Robinson
    2. FD Smith
    3. KME Sullivan
    4. NR Franks
    (2009)
    Proceedings of the Royal Society B: Biological Sciences 276:2635–2641.
    https://doi.org/10.1098/rspb.2009.0350
  77. 77
  78. 78
  79. 79
  80. 80
  81. 81
  82. 82
  83. 83
  84. 84
  85. 85
  86. 86
  87. 87
  88. 88
  89. 89
  90. 90
    Theory of Games and Economic Behavior
    1. J von Neumann
    2. O Morgenstern
    (1944)
    Princeton, NJ: Princeton University Press.
  91. 91
  92. 92
  93. 93
  94. 94
  95. 95
  96. 96

Decision letter

  1. Diethard Tautz
    Senior and Reviewing Editor; Max-Planck Institute for Evolutionary Biology, Germany
  2. Nathalie Stroeymeyt
    Reviewer; University of Fribourg, Switzerland
  3. Stephen Pratt
    Reviewer; Arizona State University, United States

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

[Editors’ note: a previous version of this study was rejected after peer review, but the authors submitted for reconsideration. The first decision letter after peer review is shown below.]

Thank you for submitting your work entitled "Positive and negative incentive contrasts lead to relative value perception in ants" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and a Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Nathalie Stroeymeyt (Reviewer #1); Stephen Pratt (Reviewer #3).

Our decision has been reached after consultation between the reviewers. Based on these discussions and the individual reviews below, we regret to inform you that your work will not be considered further for publication in eLife.

While the three reviews found value to the study, they raised concerns about the paper lacking important information (all three reviewers), possible alternative explanations for the results (reviewers 1 and 2) and lack of references to previously published work. Given the extent of the revisions necessary, we are returning the paper to you; however, we would be willing to consider a new submission with a revised manuscript if you can adequately address the comments from the reviewers.

Reviewer #1:

This manuscript presents experimental evidence that the likelihood of an ant worker accepting food of a given quality is influenced by its prior experience with food of higher or lower quality. They authors argue that the underlying mechanism is a cognitive process akin to relative value perception in humans, which depends directly on expectations, with individuals being more likely to reject a food source of a given quality if they had high expectations, and vice versa. Overall, I find the study well-thought-out and carried out to a high scientific standard, and the manuscript unusually well-written. If true, the finding that expectations influence value judgment in ants is both novel and exciting, as prior studies which have highlighted the effect of prior experience on decision-making in insects have not attempted to study the underlying cognitive mechanisms. However, I think that an alternative hypothesis could also explain the experimental results presented in the manuscript, and I would like the authors to provide additional information so this alternative explanation may be (hopefully) ruled out. Furthermore, some crucial information should be moved from the Supplementary material to the main text to help the readers understand the experimental design and judge the validity of the results without having to constantly refer to the SI. Finally, the results should also be discussed in light of some relevant studies on ant decision-making which are currently not cited in the paper. If all these points can be addressed, I believe there would be a strong argument for publication in eLife.

1) The results from experiment 1 and 2 could be explained by an alternative hypothesis in which each ant has a fixed, experience-independent threshold for food acceptance. (This alternative hypothesis is inspired from the following work on house-hunting in ants: Robinson et al., 2009, Robinson et al., 2011, Stroeymeyt et al., 2011).

In experiment 1, the authors explain that they allow 4 ants on the drawbridge, and wait until the first ant finds the food. That ant becomes the focal experimental ant and her behavior is monitored during that visit plus two subsequent trips, then is discarded from the colony. However, no information is given regarding whether any ant had to be discarded from the experiment because they did not perform three trips. One could imagine a scenario in which ants are more likely to lose motivation to forage when the food source they found is of lower quality than their internal, fixed acceptance threshold. Under that scenario, when trained with a food source of very high quality, all or most ants would keep motivation to forage, whereas when trained with a food source of low quality, only ants with a low acceptance threshold would keep motivation to forage. Thus during the third, test visit, the populations of tested foragers would not be equivalent between treatments but would show a sampling bias: in the first treatment (high quality training food), tested foragers would present a broad variation of acceptance threshold and would thus be more likely to reject an average-quality food source during the third visit, whereas in the second treatment (low quality training food), tested foragers would only include ants with low acceptance threshold, which would thus all accept an average-quality food source. In order to reject this scenario, it is necessary that the authors present data on the proportion of focal ants that did not do three trips. If that proportion does not increase with decreasing quality of the training food, then the alternative hypothesis I presented can be safely rejected, which would provide more support for the authors' interpretation.

Similarly, in experiment 2, it is possible that ants with a high acceptance threshold make a U-turn and return to the nest when reaching the scented portion of the bridge when it carries the smell associated with the low-quality food, while all ants may cross the bridge and reach the food when the bridge carries the smell associated with the high-quality food. Under that scenario, the two test populations would again not be equivalent, and the results would be explained by self-organized sampling bias rather than by a cognitive expectation effect. I would thus like the author to include data on the proportion of ants that performed U-turns when reaching the scented portion of the bridge in the two treatments. If there is no difference between the two treatments, then the alternative hypothesis can again be safely rejected and this would provide more support for the authors' interpretation.

Finally, I am not convinced that the results from experiment 3 cannot be solely explained by the different crop loads of the receiver ants in the three treatments: as the receiver ants receive more food from donor ant when the original food source is of higher quality, and as they were not given any opportunity to unload before being confronted with the test food, then one would predict them to drink less during the test (and so to have a lower food acceptance score) simply because they have less space in their crop. I am not sure that the statistical analyses presented by the authors are enough to demonstrate an effect beyond this simple mechanical effect. Instead of Figure S6.3, can the author present a scatterplot of mean acceptance score as a function of trophallaxis time for each of the three treatments and elaborate on the interpretation of the statistical analyses presented in the Supplementary material? Given that the main effect 'Reference Molarity' is non-significant and that the sign of the estimate (+) is opposite to that expected based on the author's hypothesis (higher reference molarity should result in lower food acceptance), it seems that the conclusions drawn from experiment 3 hold may not be supported by the data. Please elaborate and modify the text if relevant.

2) When reading the main text, I had trouble evaluating the solidity of the authors' claim because of the lack of key information which is only provided in the Supplementary material. In particular, please provide the following information in the main text:

- definition of the food acceptance score

- information that each ant was uniquely identified owing to the design of the experiment (each experimental ant monitored fully then discarded from the colony before switching to the next ant)

- more detail on the alternative scenarios at the beginning of experiment 2

- more detail on the experimental ant selection procedure, to show that your experimental design does not introduce any a priori selection bias between treatments

3) Previous work on house-hunting in ants has already shown that experience with a nest of high- or low-quality influences the collective likelihood of accepting another nest: specifically, Healey and Pratt, 2008, showed that colonies housed in a low-quality nest for months accepted an average-quality new nest more readily than colonies housed for months in a high-quality nest. Furthermore, Stroeymeyt et al., 2011, showed that colonies housed in a high-quality nest developed an aversion towards low-quality available nests available in their environment, whereas colonies housed in a low-quality nest did not; and they developed an explanatory model similar to the alternative scenario I described in my first major comment. Please add these studies to the Introduction and discuss your results in light of these previous studies (individual mechanism unknown, but similar colony-level benefits as the one you refer to in the second paragraph of the Discussion).

Reviewer #2:

This is a very interesting study raising an original issue. The paper is well written. Unfortunately, while I appreciate the amount of work done in this study, in my opinion the paper suffers from several major issues that need to be addressed or at least discussed further. As it is the results do not yet support entirely the authors hypothesis.

- Experiment 1: one control group for each concentration (or a sub sample) would have been better than a single control as the results obtained might reflect the ant nutritional state (ants might be hungry when they received low concentration solutions while ants might be satiated when they received high concentration solutions). It is essential to show that the acceptance level would be the same between the 2nd and the 3rd visit for each concentration tested and not only for 0.5M. That would strengthen the results considerably and provide more conclusive evidences than experiment 2.

- Experiment 2: The results are not convincing enough.

First ants exposed to high molarity (1.5M) associated cues on their outwards journey showed a significantly higher number of pheromone depositions on their return journey after feeding on a 0.5M solution than ants confronted with low molarity scent (0.25M). Following the authors’ hypothesis and the results from experiment 1 the opposite is expected. Why this counterintuitive results? The author discussed the "experimental disruption effect" (Note 5 in supplementary material) but it can't explain why pheromones deposition is higher when ants are experiencing negative contrast effect.

Thus, the only result supporting the authors hypothesis is shown in Figure 3B. They show that ants exposed to 1.5M-associated cues showed significantly lower food acceptance towards the unscented 0.5M feeder than ants exposed to 0.25M-associated cues (P-Value 0.03). However the model run to compare the data does not include the factor "scent/molarity association". The authors used two different scents (each scent was associated with either the low or the high concentration). Even if the design is perfectly well balanced (well done!), the association ought to be added as a factor. When exploring the results (Excel file provided by the authors), ants exposed to high molarity associated cues facing a 0.5M solution showed an acceptance score of 0.65 (Lemon) or 0.74 (Rosemary) while ants confronted with low molarity scent facing a 0.5M solution showed an acceptance score of 0.92 (Lemon) or 0.77 (Rosemary). I understand that the authors want to focus on the difference between acceptance scores obtained when ants encountered high molarity associated cues or low molarity associated cues. However, while the difference is there, the absolute acceptance score for a solution associated with rosemary remain similar no matter the treatment while the score acceptance for Lemon varies. Thus, I think the authors should add the "scent/molarity association" to their model. In addition, as discussed in Note 5, in experiment 2 the acceptance score for a 0.5M solution (0.7) obtained when the ant encountered the high molarity associated cues (1.5M) is almost twice as high as the acceptance score (0.4) obtained when an ant was proposed a 0.5M after being offered 1.5 M. The authors discuss this results but did not test any of the hypothesis enunciated. They could for example measure the acceptance score for a 0.5M solution associated with Lemon or Rosemary (with an alternance between odors or no alternance between odors). I think it's crucial to understand why acceptance score vary so widely as it is the main behavioral response explored in this paper.

- Experiment 3: In ants and honeybees, It is well known that transfer rate decreased with increasing viscosity (Farina and Josens work) so it is not surprising that trophallaxis time is higher for high molarity solution. Thus, the authors cannot assert that trophallaxis time is due entirely to food quality. They need to discuss this effect or control for viscosity by adding different amounts of tylose (see Josens and Farina, 2001). Again here, the nutritional state could be an alternative hypothesis for the acceptance scores obtained (an ant that received a 1.5M solution via trophallaxis might be less hungry and less motivated than an ant that received a 0.16 solution via trophallaxis).

- I found the acceptance score (the main behavioral response used in this paper) a bit too qualitative. I don’t see a large difference between full acceptance (the ant remained in contact with the drop from the moment of contact and did not interrupt drinking within 3 seconds of initial contact) and partial acceptance (feeding was interrupted within 3 seconds after the first contact with the food source, but the ant still filled its crop). How does an interruption in feeding translate in acceptance? Is there any literature to support such evidence?

The authors then state “rejection was scored if the ant refused to feed at the sucrose solution and either returned to the nest immediately or failed to fill its crop within 10 minutes”.

=> so, from what I understand (and I might be wrong!), an ant half full or an ant that do not feed are given the same acceptance score (0) while ants that filled their crop completely but with or without interruption in the first 3 sec are given different scores (1 and 0.5 respectively). A better variable could be time spend feeding or actual volume ingested (see Mailleux et al., 2010). As the authors recorded their experiment with a camera, that could possibly be done.

Reviewer #3:

This is an interesting and generally well-designed and presented study on the application of prospect theory to ant foraging decisions. It convincingly shows (barring one issue described below) that an individual ant's assessment of a food source is strongly affected by the source's quality compared to a recently experienced reference. The main result is quite robust, and the authors performed a useful further experiment that at least partly discounts explanations based on sensory factors or nutritional state. I have a few critiques of the organization, presentation, and relationship to previous work.

1) The manuscript ignores highly relevant literature about the effects of recent female experience on mate choice. In essence, these studies report that a female becomes more (or less) choosy about a mate of given quality if she has prior experience of better (or worse mates). This literature includes studies both of vertebrates and invertebrates (a few examples: Collins, 1995; Dukas, 2005; Reid and Stamps, 1997; Wagner et al., 2001).

2) I am concerned about the behavioral definitions of food acceptance. The difference between full acceptance and partial acceptance seems slight and arbitrary. This may be a consequence of too brief a description.

3) As written, the results (especially for experiment 2) are very difficult to understand without first reading the Materials and methods. It would be better to preface each experiment's results with a brief description of its general design and rationale. This could most easily be done by transferring to the Results much of the experimental descriptions currently in the Materials and methods.

4) There is much information in the supplementary material that belongs in the main text. Notes 1 through 5 give important background and motivation for experimental designs, as well as discussion of key results. Some of the supplementary methods are in fact repeated in the main text, but there is much else that should be moved there. The supplementary material is best used for alternative analyses and graphical displays, details of statistical models, and very fine-grained method descriptions. The basic methods needed to understand the experiments well should be in the main text.

5) The legend to Figure 3 is hard to read. The opening sentence is simply too long and difficult to follow.

[Editors’ note: what now follows is the decision letter after the authors submitted for further consideration.]

Thank you for submitting your article "Positive and negative incentive contrasts lead to relative value perception in ants" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by Diethard Tautz as the Senior and Reviewing Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Nathalie Stroeymeyt (Reviewer #1); Stephen Pratt (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

The authors have adequately addressed the major comments. The clarity of the manuscript has also been much improved, as the authors moved crucial information from the Supplementary materials to the main text. However, some remaining problems have still been identified that need a further revision.

Essential revisions:

1) The paragraph discussing how social insects can be used as a model to study behavioural economics in humans (Discussion, tenth paragraph) is still a bit hand-wavy, and the additional sentences inserted by the authors are not so helpful (""Direct damage, or the scent of damaged or freshly killed ants, or predators, could, for example, be used to quantify the economic cost of death or massive physical damage. This would allow direct economic quantification of factors which cannot be inflicted on human subjects.") Given that the main text has significantly increased in length due to the addition of important information from the supplementary material, we recommend removing this paragraph.

2) The authors justify their choice of 3 seconds uninterrupted drinking as a cut-off between full and partial food acceptance based on their 'extensive experience' with the ants' behaviour, and provide two videos to illustrate the difference between the two scores. However, the videos are not included – they need to be made accessible. Furthermore, the justification of this apparently arbitrary score remains unclear. Could one simply have counted the number of interruptions until the ants' crop was filled and used this as a continuous measure instead? The authors also say that "a 2-second or 4-second window would probably have given the same results", but they have not actually checked. Repeating the analyses using these thresholds and finding that the results are indeed qualitatively the same would go a long way towards justifying the authors' choice.

3) In experiment 2 the authors state: "40 ants were induced to expect a high reward, and 33 to expect a low reward." (while in reality it is 42 and 32 respectively). Looking at the table uploaded on dryad, in total 74 ants were trained and 2 had missing data for the test (visit 9). Thus, 72 observations should have been included in the analysis. However, the number of observations indicated in the model output is 73 (subsection “Statistical analysis”).

As the main results of experiment 2 hold on this particular statistical test (P=0.03), it should be run properly.

4) I reiterate my comment: in the Cumulative Link Mixed Model it is indicated that two measures were performed on the same ant (visit 1 and visit 2) but the factor, "visitnr" is treated as a fixed factor and the ant ID is not included in the model.

5) Throughout the statistics, the authors often added the random factors as (1 | Colony) + (1 | AntID) but some other time they added them as (1 | Colony/AntID) i.e. (AntID nested in colony). A justification is required why this has been done.

https://doi.org/10.7554/eLife.45450.029

Author response

Reviewer #1:

[…] 1) The results from experiment 1 and 2 could be explained by an alternative hypothesis in which each ant has a fixed, experience-independent threshold for food acceptance. (This alternative hypothesis is inspired from the following work on house-hunting in ants: Robinson et al., 2009, Robinson et al., 2011, Stroeymeyt et al., 2011).

Subsection “Non-random selection of individuals with different acceptance thresholds”. This alternative hypothesis was added to the introduction of experiment 2 and discussed in light of our results (subsection “Experiment 2 – Methods”). Our data and experimental design allow us to rule out this hypothesis (see detailed response below).

“Non-random selection of individuals with different acceptance thresholds: different individuals of the same colony may have different acceptance thresholds. […] However, we can exclude this possibility, as the proportion of ants not completing training was uniformly low and did not vary with treatment (see Supplementary file 1).”

“Similar results in house-hunting ants were explained by a simple threshold rule (Stroeymeyt et al., 2011; Robinson et al., 2009, 2011) which suggests that individuals have different acceptance thresholds and ants with lower thresholds accept lower quality options. […] It is thus unlikely that a simple threshold rule led to the results shown in experiment 1 (Figure 2 and 3).”

In experiment 1, the authors explain that they allow 4 ants on the drawbridge, and wait until the first ant finds the food. That ant becomes the focal experimental ant and her behavior is monitored during that visit plus two subsequent trips, then is discarded from the colony. However, no information is given regarding whether any ant had to be discarded from the experiment because they did not perform three trips. One could imagine a scenario in which ants are more likely to lose motivation to forage when the food source they found is of lower quality than their internal, fixed acceptance threshold. Under that scenario, when trained with a food source of very high quality, all or most ants would keep motivation to forage, whereas when trained with a food source of low quality, only ants with a low acceptance threshold would keep motivation to forage. Thus during the third, test visit, the populations of tested foragers would not be equivalent between treatments but would show a sampling bias: in the first treatment (high quality training food), tested foragers would present a broad variation of acceptance threshold and would thus be more likely to reject an average-quality food source during the third visit, whereas in the second treatment (low quality training food), tested foragers would only include ants with low acceptance threshold, which would thus all accept an average-quality food source. In order to reject this scenario, it is necessary that the authors present data on the proportion of focal ants that did not do three trips. If that proportion does not increase with decreasing quality of the training food, then the alternative hypothesis I presented can be safely rejected, which would provide more support for the authors' interpretation.

This is a very insightful hypothesis, and one which had not occurred to us. Indeed, the reviewer is completely correct that this could present a critical flaw in our experiment. However, luckily, the proportion of ants, which did not participate in all three training visits did not increase with decreasing quality of the training food. Supplementary file 1 shows the complete data of ants, which completed all three visits, and ants, which cancelled foraging before training and data collection was complete.

Two columns (Number of ants excluded and proportion finished) were added to Supplementary file 1 and a mean of all successfully finished trials added to the Materials and methods section of experiment 1: “97% of ants successfully finished the training procedure and participated in the test visit (third visit).”.

Similarly, in experiment 2, it is possible that ants with a high acceptance threshold make a U-turn and return to the nest when reaching the scented portion of the bridge when it carries the smell associated with the low-quality food, while all ants may cross the bridge and reach the food when the bridge carries the smell associated with the high-quality food. Under that scenario, the two test populations would again not be equivalent, and the results would be explained by self-organized sampling bias rather than by a cognitive expectation effect. I would thus like the author to include data on the proportion of ants that performed U-turns when reaching the scented portion of the bridge in the two treatments. If there is no difference between the two treatments, then the alternative hypothesis can again be safely rejected and this would provide more support for the authors' interpretation.

In experiment 2, there were no ants which returned to the nest on the test visit. All ants tasted the food at the end of the runway regardless of motivation when reaching the runway scent, and food acceptance scores were noted for each ant which completed training. Furthermore, all ants received the same training procedure and had to taste each sucrose droplet during training. Motivation generally dropped when confronted with a scent which was associated to a low molarity and ants performed U-turns during later training visits. However, the sucrose droplet had to be tasted on each visit in order to complete training. Thus, during the test visit, we also get food acceptance scores from ants which performed more U-turns and showed lower motivation compared to others.

However, even if a filter was in place and we were only testing the high motivation ants (which seems unlikely, given the information in Table 1), this would be the case for both treatments, because treatments were identical until the final test visit, in which only the odour of the runway changed. While it is in principle possible that on the test visit ‘high threshold’ ants would be filtered out, no ants which completed training returned to the nest during the testing trial and did not complete testing.

Finally, I am not convinced that the results from experiment 3 cannot be solely explained by the different crop loads of the receiver ants in the three treatments: as the receiver ants receive more food from donor ant when the original food source is of higher quality, and as they were not given any opportunity to unload before being confronted with the test food, then one would predict them to drink less during the test (and so to have a lower food acceptance score) simply because they have less space in their crop. I am not sure that the statistical analyses presented by the authors are enough to demonstrate an effect beyond this simple mechanical effect. Instead of Figure S6.3, can the author present a scatterplot of mean acceptance score as a function of trophallaxis time for each of the three treatments and elaborate on the interpretation of the statistical analyses presented in the Supplementary material? Given that the main effect 'Reference Molarity' is non-significant and that the sign of the estimate (+) is opposite to that expected based on the author's hypothesis (higher reference molarity should result in lower food acceptance), it seems that the conclusions drawn from experiment 3 hold may not be supported by the data. Please elaborate and modify the text if relevant.

This is a reasonable point. Firstly, it may not be that ants which showed higher trophallaxis times also ingest more sucrose solution during trophallaxis. Higher molarities may lead to longer trophallaxis times simply due to the higher viscosity of the solution while the amount of transferred food stays the same for all three molarities (Josens, Farina, and Roces, 1998; Tezze and Farina, 1999). Even if more food was transferred, the food acceptance scores are a measure of the first assessment of ants at a food source, not the ingested volume. Thus, while some ants may have had less space in their crop left, this may not necessarily affect the food acceptance score, while it is very likely to affect ingested volume after trophallaxis.

Additionally, if longer trophallaxis times lead to more ingested sucrose solution, it is also more likely that a higher amount of information about the past food quality is transferred. Thus, longer trophallaxis times may also be expected to lead to stronger contrast effects due to more information being transferred through trophallaxis.

With all these caveats, the comments of the reviewer are still valid, and so we carried out the requested analyses. The data shows clear effects of both trophallaxis time and reference solution on the food acceptance towards 0.5M sucrose. Even at high trophallaxis times, ants with a 0.16M reference showed no low food acceptance scores, unlike ants with high reference solution after long trophallaxis times.

The fixed effects of the model indeed seemed a bit odd and as the data seemed to show a very different pattern, so we ran the model again with scaled continuous variables as suggested for GLMMs. While the effect of the interaction stayed the same in the updated model, we also see a clearer effect of the fixed effects, confirming that both trophallaxis time and reference molarity affect the food acceptance at the post-shift solution – as shown in Author response image 1 and 2. Thank you once again for the suggestion!

Author response image 1
Author response image 2

We now add this analysis and figures to Figure 5—source data 1, and discuss this hypothesis briefly in the Discussion: “If, however, more sucrose solution was transferred between the returning forager and the recruit at longer trophallaxis times, it is likely that information input increases and food acceptance decreases. […] Even at high trophallaxis times, ants with a 0.16M reference showed no low food acceptance scores, unlike ants with high reference solution after long trophallaxis times.”

2) When reading the main text, I had trouble evaluating the solidity of the authors' claim because of the lack of key information which is only provided in the Supplementary material. In particular, please provide the following information in the main text:

- definition of the food acceptance score

This was added to the Materials and methods section of experiment. For clarity, we will also add videos for the scores 1 and 0.5 as additional supplements.

“Food acceptance scores as a measure of perceived value were noted for each ant and visit as follows: As the ant drank at the droplet it was given one of three food acceptance scores. […] Lastly, rejection (0) was scored if the ant refused to feed at the sucrose solution and either returned to the nest immediately or failed to fill its crop within 10 minutes.”

- information that each ant was uniquely identified owing to the design of the experiment (each experimental ant monitored fully then discarded from the colony before switching to the next ant)

This was added to the Materials and methods section: “Each tested ant was observed until all experimental runs were finished and then discarded from the colony before switching to the next ant.”

- more detail on the alternative scenarios at the beginning of experiment 2

This was added to the introductory part of experiment 2 (subsection “Experiment 2 – ruling out alternative explanations using scent training”).

- more detail on the experimental ant selection procedure, to show that your experimental design does not introduce any a priori selection bias between treatments

This was added to the ant selection and monitoring section: “This procedure may select for the more active foragers, but does not introduce any selection bias between treatments.”

3) Previous work on house-hunting in ants has already shown that experience with a nest of high- or low-quality influences the collective likelihood of accepting another nest: specifically, Healey and Pratt, 2008, showed that colonies housed in a low-quality nest for months accepted an average-quality new nest more readily than colonies housed for months in a high-quality nest. Furthermore, Stroeymeyt et al., 2011, showed that colonies housed in a high-quality nest developed an aversion towards low-quality available nests available in their environment, whereas colonies housed in a low-quality nest did not; and they developed an explanatory model similar to the alternative scenario I described in my first major comment. Please add these studies to the Introduction and discuss your results in light of these previous studies (individual mechanism unknown, but similar colony-level benefits as the one you refer to in the second paragraph of the Discussion).

This was added to the general Introduction, introduction of experiment 2 (subsection “Non-random selection of individuals with different acceptance thresholds”), and the Discussion of the main manuscript.

“Healey and Pratt, 2008, showed that colonies of the house-hunting ant species Temnothorax curvispinosus move into a nest of mediocre quality faster when they were previously housed in a high-quality nest compared to colonies which were previously housed in a poor-quality nest (Healey and Pratt, 2008). In contrast, Stroeymeyt et al., 2011, showed that colonies of Temnothorax albipennis developed an aversion towards mediocre-quality nests available in their environment when they were housed in a high-quality nest, whereas colonies housed in a low-quality nest did not, and thus show an experience-dependent flexibility in nest choice (Stroeymeyt et al., 2011).”

“Non-random selection of individuals with different acceptance thresholds: different individuals of the same colony may have different acceptance thresholds. […] However, we can exclude this possibility, as the proportion of ants not completing training was uniformly low and did not vary with treatment (see Supplementary file 1).”

“Similar results in house-hunting ants were explained by a simple threshold rule (Stroeymeyt et al., 2011; Robinson et al., 2009, 2011) which suggests that individuals have different acceptance thresholds and ants with lower thresholds accept lower quality options. […] It is thus unlikely that a simple threshold rule led to the results shown in experiment 1 (Figure 2 and 3).”

“For example, colonies of house-hunting ants developed an aversion towards mediocre nests when housed in high-quality nests, but not when they were housed in low-quality nests. […] However, while in house-hunting the reference resource is directly experienced by scouts only, we demonstrate that information brought back to the nest can set a reference point for ants which have not directly experienced the resource in situ.”

Reviewer #2:

This is a very interesting study raising an original issue. The paper is well written. Unfortunately, while I appreciate the amount of work done in this study, in my opinion the paper suffers from several major issues that need to be addressed or at least discussed further. As it is the results do not yet support entirely the authors hypothesis.

- Experiment 1: one control group for each concentration (or a sub sample) would have been better than a single control as the results obtained might reflect the ant nutritional state (ants might be hungry when they received low concentration solutions while ants might be satiated when they received high concentration solutions). It is essential to show that the acceptance level would be the same between the 2nd and the 3rd visit for each concentration tested and not only for 0.5M. That would strengthen the results considerably and provide more conclusive evidences than experiment 2.

As all tested colonies were starved 4 days prior to experiments and foragers collect food for the whole colony, it is very unlikely that the single forager or even the whole colony was satiated from a maximum of 30 crop loads during a testing day. We can provide data for a 1M food source from our lab at which food acceptance on the second visit is at 0.88, and for the third visit at 0.87 (n=353 for both visits), suggesting no change in food acceptance from the second to third visit. While adding this information would indeed add elegance, this would require an effective doubling of the number of trials. As experiment 1 required over two months to perform, this seems to us excessive.

Finally, this issue was addressed with experiment 2 where all ants received the same training procedure and were thus supposed to be at a similar hunger level. Collecting more data on other molarities would also not rule out other explanations as experiment 2 does.

- Experiment 2: The results are not convincing enough.

First ants exposed to high molarity (1.5M) associated cues on their outwards journey showed a significantly higher number of pheromone depositions on their return journey after feeding on a 0.5M solution than ants confronted with low molarity scent (0.25M). Following the authors’ hypothesis and the results from experiment 1 the opposite is expected. Why this counterintuitive results? The author discussed the "experimental disruption effect" (Note 5 in supplementary material) but it can't explain why pheromones deposition is higher when ants are experiencing negative contrast effect.

This is indeed an important point, and we thank the reviewer for raising it and asking us to think carefully about it again. After careful consideration of the methods and data, we believe we have uncovered the reason for this pattern in the pheromone deposition data:

The scented paper overlays were replaced with unscented ones. However, we did not replace the runways on which the paper overlays were placed. It is thus possible that small portions of the previous odour was still present. The ants were thus likely reacting to the remaining odour as associated in the training visit (that is, depositing more pheromone for the ‘good’ smell and less for the ‘poor’ smell), overlaid over a major change and reduction in deposition due to a disruption effect.

Comparing the pheromone data with data collected in a separate experiment currently in preparation supports this hypothesis. In this experiment the scented overlays were not replaced after the last visit. We see the same pattern of pheromone deposition in the current experiment. Ants were confronted with scented overlays and scented sucrose solution also in the test visit (0.387M post-shift). We show this data in Author response image 3: panel A shows the number of pheromone depositions during 8 training visits with 1.5M as high molarity and 0.1M as low molarity. Panel B shows pheromone depositions on the 9th visit with 0.387M scented sucrose solution as post-shift reward and scented paper overlays on the way back to the nest. Panel C shows the data from the 9th visit of the current experiment 2, for comparison. Note that B and C are very similar. However, since both training molarities and test molarities differed between the experiments, and the data was collected in different years and mostly with different colonies, direct comparisons should be made with caution.

We now discuss the reason for the unexpected pheromone deposition pattern: “Furthermore, since only the scented paper overlays were replaced by unscented ones, but not the runways themselves, it is possible that small portions of the odours were still present, driving the ants to deposit pheromone according to the remaining odours, with higher depositions rates for the high-quality associated odour. […] This would explain why pheromone depositions were higher for ants returning to the nest from a high molarity scent than in ants returning from a low molarity scent.”

Author response image 3

Thus, the only result supporting the authors hypothesis is shown in Figure 3B. They show that ants exposed to 1.5M-associated cues showed significantly lower food acceptance towards the unscented 0.5M feeder than ants exposed to 0.25M-associated cues (P-Value 0.03). However the model run to compare the data does not include the factor "scent/molarity association". The authors used two different scents (each scent was associated with either the low or the high concentration). Even if the design is perfectly well balanced (well done!), the association ought to be added as a factor. When exploring the results (Excel file provided by the authors), ants exposed to high molarity associated cues facing a 0.5M solution showed an acceptance score of 0.65 (Lemon) or 0.74 (Rosemary) while ants confronted with low molarity scent facing a 0.5M solution showed an acceptance score of 0.92 (Lemon) or 0.77 (Rosemary). I understand that the authors want to focus on the difference between acceptance scores obtained when ants encountered high molarity associated cues or low molarity associated cues. However, while the difference is there, the absolute acceptance score for a solution associated with rosemary remain similar no matter the treatment while the score acceptance for Lemon varies. Thus, I think the authors should add the "scent/molarity association" to their model. In addition, as discussed in Note 5, in experiment 2 the acceptance score for a 0.5M solution (0.7) obtained when the ant encountered the high molarity associated cues (1.5M) is almost twice as high as the acceptance score (0.4) obtained when an ant was proposed a 0.5M after being offered 1.5 M. The authors discuss this results but did not test any of the hypothesis enunciated. They could for example measure the acceptance score for a 0.5M solution associated with Lemon or Rosemary (with an alternance between odors or no alternance between odors). I think it's crucial to understand why acceptance score vary so widely as it is the main behavioral response explored in this paper.

This is a fair point. The scent/molarity association was added to the model as suggested by the reviewer. There is still a significant effect of expected molarity, but however no significant effect of the odour used (see model summary below).

Cumulative Link Mixed Model fitted with the Laplace approximation

formula: newFoodAcceptability ~ Scent.Molarity * HighLowMolarityscent + (1 | Colony)

data: visit9FA

link threshold nobs logLik AIC niter max.grad cond.H

logit flexible 72 -57.86 127.72 281(385) 1.62e-05 3.8e+04

Random effects:

Groups Name Variance Std.Dev.

Colony (Intercept) 0.003578 0.05982

Number of groups: Colony 6

Coefficients:

Estimate Std. Error z value Pr(>|z|)

Scent.MolarityRosemary 0.3924 0.6280 0.625 0.5321

HighLowMolarityscentLow 1.9828 0.9007 2.201 0.0277 *

Scent.MolarityRosemary:HighLowMolarityscentLow -1.5018 1.0953 -1.371 0.1703

---

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘’ 1

Threshold coefficients:

Estimate Std. Error z value

0|0.5 -2.3120 0.6309 -3.665

0.5|1 0.3579 0.4580 0.781

Author response image 4

We do not have data on acceptance scores for scented 0.5M sucrose. However, we can provide acceptance data of ants feeding on a 1M sucrose solution associated to Lemon over successive visits (visit1: 0.99, visit 2: 0.89, visit 3: 0.89) or Rosemary scent (visit 1: 0.97, visit 2: 0.86, visit 3: 0.84), showing that acceptance scores vary from the first to second visits but then remain rather constant when confronted with the same odour and molarity. This data was collected by Oberhauser and Czaczkes and partly published in Biology Letters (Oberhauser and Czaczkes, 2018).

Understanding why acceptance scores vary so widely is a huge undertaking, and is effectively attempting to answer the question “what drives relative value perception?” Our manuscript covers one important driver of relative value perception: expectations. However, fully answering such a broad question could fill an entire career. We hope we have begun to make good inroads into answering this question, as have others before us, but we are still far from a comprehensive answer.

- Experiment 3: In ants and honeybees, It is well known that transfer rate decreased with increasing viscosity (Farina and Josens work) so it is not surprising that trophallaxis time is higher for high molarity solution. Thus, the authors cannot assert that trophallaxis time is due entirely to food quality. They need to discuss this effect or control for viscosity by adding different amounts of tylose (see Josens and Farina, 2001). Again here, the nutritional state could be an alternative hypothesis for the acceptance scores obtained (an ant that received a 1.5M solution via trophallaxis might be less hungry and less motivated than an ant that received a 0.16 solution via trophallaxis).

The fact that higher viscosities also lead to longer trophallaxis times has been made clearer in the manuscript. However, while this effect is well known, our attention is on the effect of the quality of reference solution on the food acceptance when confronted with medium quality sucrose, not on the trophallaxis times. We completely agree that experiment 3 cannot rule out alternative explanations such as different hunger levels. Only experiment 2 was designed to rule these alternatives out.

“Even though higher trophallaxis times led to lower acceptance scores and trophallaxis times were higher at high reference molarities, this does not necessarily imply that ants ingested more sucrose at higher references and were thus less hungry or motivated. […] Even at high trophallaxis times, ants with a 0.16M reference showed no low food acceptance scores, unlike ants with high reference solution after long trophallaxis times.”

- I found the acceptance score (the main behavioral response used in this paper) a bit too qualitative. I don’t see a large difference between full acceptance (the ant remained in contact with the drop from the moment of contact and did not interrupt drinking within 3 seconds of initial contact) and partial acceptance (feeding was interrupted within 3 seconds after the first contact with the food source, but the ant still filled its crop). How does an interruption in feeding translate in acceptance? Is there any literature to support such evidence?

Videos have now been added (Video 1 and Video 2) to illustrate the foraging behavior for acceptance scores 1 and 0.5. Ants which displayed score 1 were generally very calm while feeding, while in contrast ants which displayed score 0.5 tend to show multiple feeding interruptions and show ‘impatience’ while feeding.

The authors then state “rejection was scored if the ant refused to feed at the sucrose solution and either returned to the nest immediately or failed to fill its crop within 10 minutes”.

=> so, from what I understand (and I might be wrong!), an ant half full or an ant that do not feed are given the same acceptance score (0) while ants that filled their crop completely but with or without interruption in the first 3 sec are given different scores (1 and 0.5 respectively). A better variable could be time spend feeding or actual volume ingested (see Mailleux et al., 2010). As the authors recorded their experiment with a camera, that could possibly be done.

This is a reasonable comment. Unfortunately, time spent feeding is apparently not a very good variable here, because higher viscosity leads to higher drinking times in ants (Josens, Farina, and Roces, 1998) and increasing sucrose molarities lead to higher viscosity of the solutions. While we attempted to extract a measure of ingested volume from the videos, it was not possible to do this reliably due to limitations in video resolution.

We hope that the videos now make the difference in food acceptance scores clearer. We think the scores delineate biologically reasonable points in terms of acceptance: a 0 score almost always represents ants which originally showed score 0.5, but then cancelled the foraging trip without filling their crop. It is worth nothing that scores of 0 were rather rare.

Finally, we reiterate that the scoring was performed blind to treatment, which suggests that this is a reliable indicator that acceptance correlates with value.

Reviewer #3:

This is an interesting and generally well-designed and presented study on the application of prospect theory to ant foraging decisions. It convincingly shows (barring one issue described below) that an individual ant's assessment of a food source is strongly affected by the source's quality compared to a recently experienced reference. The main result is quite robust, and the authors performed a useful further experiment that at least partly discounts explanations based on sensory factors or nutritional state. I have a few critiques of the organization, presentation, and relationship to previous work.

1) The manuscript ignores highly relevant literature about the effects of recent female experience on mate choice. In essence, these studies report that a female becomes more (or less) choosy about a mate of given quality if she has prior experience of better (or worse mates). This literature includes studies both of vertebrates and invertebrates (a few examples: Collins, 1995; Dukas, 2005; Reid and Stamps, 1997; Wagner et al., 2001).

This was added to the Introduction of the main manuscript. Many thanks for pointing our attention towards these relevant studies.

“Incentive contrasts have also been demonstrated for rewards other than food. Females become more (or less) likely to accept a mate of given quality if they have prior experience of better (or worse) mates. Such mate quality contrast effects are reported in both vertebrates (Collins, 1995) and invertebrates (Dukas, 2005; Reid and Stamps, 1997; Wagner, Smeds, and Wiegmann 2001).”

2) I am concerned about the behavioral definitions of food acceptance. The difference between full acceptance and partial acceptance seems slight and arbitrary. This may be a consequence of too brief a description.

Videos and a more detailed description were added to the main manuscript. See also our response to points by reviewer 2.

“Food acceptance scores as a measure of perceived value were noted for each ant and visit as follows: when the ant drank at the droplet it was given one of three food acceptance scores. […] Lastly, rejection (0) was scored if the ant refused to feed at the sucrose solution and either returned to the nest immediately or failed to fill its crop within 10 minutes.”

3) As written, the results (especially for experiment 2) are very difficult to understand without first reading the Materials and methods. It would be better to preface each experiment's results with a brief description of its general design and rationale. This could most easily be done by transferring to the Results much of the experimental descriptions currently in the Materials and methods.

We agree, and have completely restructured the manuscript to increase clarity. We now describe the methods and results of each experiment together continuing to the next experiment.

4) There is much information in the supplementary material that belongs in the main text. Notes 1 through 5 give important background and motivation for experimental designs, as well as discussion of key results. Some of the supplementary methods are in fact repeated in the main text, but there is much else that should be moved there. The supplementary material is best used for alternative analyses and graphical displays, details of statistical models, and very fine-grained method descriptions. The basic methods needed to understand the experiments well should be in the main text.

The supplementary notes 1 to 5 were added to the main manuscript. Many methodological details have also now been moved to the main manuscript.

5) The legend to Figure 3 is hard to read. The opening sentence is simply too long and difficult to follow.

Agreed. This has now been reworded.

[Editors’ note: what now follows is the decision letter after the authors submitted for further consideration.]

Essential revisions:

1) The paragraph discussing how social insects can be used as a model to study behavioural economics in humans (Discussion, tenth paragraph) is still a bit hand-wavy, and the additional sentences inserted by the authors are not so helpful (""Direct damage, or the scent of damaged or freshly killed ants, or predators, could, for example, be used to quantify the economic cost of death or massive physical damage. This would allow direct economic quantification of factors which cannot be inflicted on human subjects.") Given that the main text has significantly increased in length due to the addition of important information from the supplementary material, we recommend removing this paragraph.

These sentences have now been removed from the manuscript.

2) The authors justify their choice of 3 seconds uninterrupted drinking as a cut-off between full and partial food acceptance based on their 'extensive experience' with the ants' behaviour, and provide two videos to illustrate the difference between the two scores. However, the videos are not included – they need to be made accessible. Furthermore, the justification of this apparently arbitrary score remains unclear. Could one simply have counted the number of interruptions until the ants' crop was filled and used this as a continuous measure instead? The authors also say that "a 2-second or 4-second window would probably have given the same results", but they have not actually checked. Repeating the analyses using these thresholds and finding that the results are indeed qualitatively the same would go a long way towards justifying the authors' choice.

The videos had been made accessible via the offered Dryad link. We apologize for not making this clearer and easier to access them. However, the videos have now been updated directly through the submission form as video 1 (showing food acceptance score 1) and video 2 (showing food acceptance score 0.5).

The results of experiment 2 have been re-analysed from the videos in order to get data on the number of drinking interruptions until the crop is filled, and to be able to compare the 2, 3 and 4 second windows of food acceptance. This also represents an independent blind assessment of the ants’ behaviour, as the analysis was performed by a naïve analyser unfamiliar with the study. Due to the very large amount of videos and time limitations, it was not feasible for us to re-analyse the data of all three experiments from the videos. We thus chose to focus on experiment 2, as it was the focus of much of reviewer 2’s concerns. Unfortunately, two of the videos were missing, resulting in a slightly lower sample size (68 instead of 70).

We found a slightly stronger significant effect of the expected molarity based on the runway scent in the 3- and 4-second windows of food acceptance than that presented in the main manuscript (3 seconds CLMM: estimate=1.42, z=2.35, p=0.019, 4 seconds CLMM: estimate=1.27, z=2.28, p=0.023). While the same trend is maintained for the 2-second window, the groups were not significantly different using this definition (estimate=0.88, z=1.32, p=0.19). However, pleasingly, the number of food interruptions until the crop was filled was significantly lower when ants expected to find low molarity food, but got medium quality (positively surprised) compared to when they expected high molarity food but got medium (disappointed) (GLMM: estimate=-0.35, z=-2.74, p=0.006, see Author response image 5).

We furthermore analysed first interruption times of the tested ants on the last (9th) visit. Surprisingly, we found no significant effect of ant’s expectations on the first interruption time (GLMM: estimate=-0.34, z=-1.07, p=0.28), although a visual inspection of the data suggested it should be there (see Author response image 6). There was, however, a highly significant effect of the used odour on the first interruption time (GLMM: estimate=1.19, z=3.7, p< 0.001). Author response image 6 shows that first interruption times were higher in ants which expected to receive low molarity food compared to ants which expected high molarity food. Since we found a significant effect of odour, we split the data according to odour presented on the 9th visit and analysed the subsets once again, which nicely explained the results: Author response image 7 clearly shows a strong effect of expectations on first interruption times when the lemon odour was used (GLMM: estimate=1.74, z=3.92, p<0.001), but a weaker effect, although still in the same direction when rosemary odour was used (GLMM: estimate=0.63, z=1.44, p=0.15).

Unfortunately, we cannot explain the differences between lemon and rosemary odours in these experiments. Lemon and rosemary odours are used in our conditioning experiments, because ants did not show a preference for any of the odours in previous experiments. It is possible that ants are more able to associate food quality (either high or low) with lemon odour than rosemary odour.

We now briefly mention the blind reanalysis of acceptance scores using the 3 second window, and the analysis of interruption times, in the main text (subsection “Experiment 2 – Results”). We hope that the data collected from the re-analysis of the second experiment, as well as the videos showing food acceptance scores 1 and 0.5, now sufficiently justify the use of our 3-second food acceptance scores as a measure of the attractiveness of a food source in these experiments.

Graph and statistical analysis for the number of food interruptions:

Author response image 5
Number of food interruptions on the last (9th) visit depending on the ant’s expectations until the crop was filled.

## Generalized linear mixed model fit by maximum likelihood (Laplace

## Approximation) [glmerMod]

## Family: poisson (log)

## Formula: Pauses ~ HighLow + Scent + (1 | Colony)

## Data: alldat

## Control: glmerControl(optCtrl = list(maxfun = 10000))

##

## AIC BIC logLik deviance df.resid

## 321.6 330.5 -156.8 313.6 64

##

## Scaled residuals:

## Min 1Q Median 3Q Max

## -2.1614 -0.8313 -0.3108 0.6991 5.2410

##

## Random effects:

## Groups Name Variance Std.Dev.

## Colony (Intercept) 4.975e-17 7.053e-09

## Number of obs: 68, groups: Colony, 6

##

## Fixed effects:

## Estimate Std. Error z value Pr(>|z|)

## (Intercept) 1.3896 0.1055 13.167 < 2e-16 ***

## HighLowLow -0.3551 0.1295 -2.741 0.00613 **

## ScentRosemary 0.1520 0.1273 1.194 0.23248

## ---

## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

##

## Correlation of Fixed Effects:

## (Intr) HghLwL

## HighLowLow -0.391

## ScentRosmry -0.665 -0.095

Graph and statistical analysis for the first interruption times including the whole data of experiment 2:

Author response image 6
First interruption times [seconds] on the last (9th) visit depending on ant’s expectations for the complete data.

## Generalized linear mixed model fit by maximum likelihood (Laplace

## Approximation) [glmerMod]

## Family: poisson (log)

## Formula:

## as.integer(FirstInterruption) ~ HighLow + Scent + (1 | Colony/AntID)

## Data: alldat

## Control: glmerControl(optCtrl = list(maxfun = 10000))

##

## AIC BIC logLik deviance df.resid

## 568.8 579.9 -279.4 558.8 63

##

## Scaled residuals:

## Min 1Q Median 3Q Max

## -2.46587 -0.20136 -0.01499 0.07549 1.20787

##

## Random effects:

## Groups Name Variance Std.Dev.

## AntID:Colony (Intercept) 2.616e+00 1.617e+00

## Colony (Intercept) 2.071e-10 1.439e-05

## Number of obs: 68, groups: AntID:Colony, 67; Colony, 6

##

## Fixed effects:

## Estimate Std. Error z value Pr(>|z|)

## (Intercept) 1.5870 0.3531 4.495 6.95e-06 ***

## HighLowLow -0.3400 0.3175 -1.071 0.284268

## ScentRosemary 1.1931 0.3218 3.708 0.000209 ***

## ---

## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

##

## Correlation of Fixed Effects:

## (Intr) HghLwL

## HighLowLow -0.663

## ScentRosmry -0.737 0.498

Graph and statistical analysis for the first interruption times for data subsets of lemon and rosemary odours:

Author response image 7
First interruption times [seconds] on the last (9th) visit depending on ant’s expectations.

Data for ants being confronted with lemon and rosemary odours on the runways were split.

GLMM for the Lemon Odour:

## Generalized linear mixed model fit by maximum likelihood (Laplace

## Approximation) [glmerMod]

## Family: poisson (log)

## Formula: as.integer(FirstInterruption) ~ HighLow + (1 | Colony/AntID)

## Data: lemonsubset

## Control: glmerControl(optCtrl = list(maxfun = 10000))

##

## AIC BIC logLik deviance df.resid

## 239.5 245.1 -115.7 231.5 26

##

s## Scaled residual:

## Min 1Q Median 3Q Max

## -1.06113 -0.30094 0.01654 0.08866 0.26499

##

## Random effects:

## Groups Name Variance Std.Dev.

## AntID:Colony (Intercept) 1.276e+00 1.129e+00

## Colony (Intercept) 6.094e-10 2.469e-05

## Number of obs: 30, groups: AntID:Colony, 30; Colony, 6

##

## Fixed effects:

## Estimate Std. Error z value Pr(>|z|)

## (Intercept) 1.5550 0.2940 5.289 1.23e-07 ***

## HighLowLow 1.7416 0.4435 3.927 8.60e-05 ***

## ---

## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

##

## Correlation of Fixed Effects:

## (Intr)

## HighLowLow -0.658

GLMM for the Rosemary Odour:

## Generalized linear mixed model fit by maximum likelihood (Laplace

## Approximation) [glmerMod]

## Family: poisson (log)

## Formula: as.integer(FirstInterruption) ~ HighLow + (1 | Colony/AntID)

## Data: rosemarysubset

## Control: glmerControl(optCtrl = list(maxfun = 10000))

##

## AIC BIC logLik deviance df.resid

## 292.9 299.4 -142.4 284.9 34

##

## Scaled residuals:

## Min 1Q Median 3Q Max

## -1.02190 -0.31663 -0.03179 0.09659 0.19368

##

## Random effects:

## Groups Name Variance Std.Dev.

## AntID:Colony (Intercept) 1.606e+00 1.267e+00

## Colony (Intercept) 5.220e-10 2.285e-05

## Number of obs: 38, groups: AntID:Colony, 38; Colony, 6

##

## Fixed effects:

## Estimate Std. Error z value Pr(>|z|)

## (Intercept) 1.7210 0.3163 5.441 5.31e-08 ***

## HighLowLow 0.6292 0.4360 1.443 0.149

## ---

## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

##

## Correlation of Fixed Effects:

## (Intr)

## HighLowLow -0.721

3) In experiment 2 the authors state: "40 ants were induced to expect a high reward, and 33 to expect a low reward." (while in reality it is 42 and 32 respectively). Looking at the table uploaded on dryad, in total 74 ants were trained and 2 had missing data for the test (visit 9). Thus, 72 observations should have been included in the analysis. However, the number of observations indicated in the model output is 73 (subsection “Statistical analysis”).

As the main results of experiment 2 hold on this particular statistical test (P=0.03), it should be run properly.

Many thanks for bringing these two mistakes to our attention. The data was re-checked and it seems that two doubles were included in the data file, probably when the data was re-analysed through the videos. We thus re-ran all statistics for this experiment and corrected the mistakes in the manuscript, Figure 4—source data 1-6 and the Raw Data File. The results from experiment 2 remain unchanged, as do their significance (p = 0.04).

4) I reiterate my comment: in the Cumulative Link Mixed Model it is indicated that two measures were performed on the same ant (visit 1 and visit 2) but the factor, "visitnr" is treated as a fixed factor and the ant ID is not included in the model.

AntID was now included in the Cumulative Link Mixed Model (see Figure 2 – Source Data 1). The results of the model remain largely unchanged.

5) Throughout the statistics, the authors often added the random factors as (1 | Colony) + (1 | AntID) but some other time they added them as (1 | Colony/AntID) i.e. (AntID nested in colony). A justification is required why this has been done.

This had been done in order to achieve a better model fit. This change did however not affect the model fit and was now, in the process of re-analysing the whole experiment 2 section, changed to the correct version (AntID nested in colony) throughout the whole dataset.

The same random effects were added in the GLMMs for pheromone depositions of the last visits of experiment 2 in order to correct for overdispersion and obtain a better fit for the models. This was also done in the GLMM for the inbound pheromone depositions on visit 3 of experiment 1, also to correct for overdispersion.

https://doi.org/10.7554/eLife.45450.030

Article and author information

Author details

  1. Stephanie Wendt

    Animal Comparative Economics Laboratory, Institute of Zoology and Evolutionary Biology, University of Regensburg, Regensburg, Germany
    Contribution
    Conceptualization, Data curation, Formal analysis, Validation, Investigation, Methodology, Writing—original draft, Writing—review and editing
    For correspondence
    wendtstephanie@outlook.de
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-8950-2845
  2. Kim S Strunk

    School of Business, Economics and Information Systems, Chair of Management, People and Information, University of Passau, Passau, Germany
    Contribution
    Validation, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-2896-814X
  3. Jürgen Heinze

    Institute of Zoology and Evolutionary Biology, University of Regensburg, Regensburg, Germany
    Contribution
    Resources, Validation, Writing—review and editing
    Competing interests
    No competing interests declared
  4. Andreas Roider

    Department of Economics, University of Regensburg, Regensburg, Germany
    Contribution
    Validation, Writing—review and editing
    Competing interests
    No competing interests declared
  5. Tomer J Czaczkes

    Animal Comparative Economics Laboratory, Institute of Zoology and Evolutionary Biology, University of Regensburg, Regensburg, Germany
    Contribution
    Conceptualization, Resources, Formal analysis, Supervision, Funding acquisition, Validation, Methodology, Writing—original draft, Project administration, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-1350-4975

Funding

Deutsche Forschungsgemeinschaft (Emmy-Noether-Grant (Grant number CZ 237/1-1))

  • Stephanie Wendt
  • Tomer J Czaczkes

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Flavio Roces for helpful comments on this work, Florian Hartig for advice concerning statistical analysis of our data, and Nathalie Stroeymeyt, Stephen Pratt, and an anonymous reviewer for comments on an earlier version of this manuscript. We also thank the DFG (Deutsche Forschungsgemeinschaft) which funded SW and TJC with an Emmy Noether grant to TJC, grant number CZ 237/1–1.

Ethics

Animal experimentation: All animal treatment guidelines applicable to ants under German law have been followed.

Senior and Reviewing Editor

  1. Diethard Tautz, Max-Planck Institute for Evolutionary Biology, Germany

Reviewers

  1. Nathalie Stroeymeyt, University of Fribourg, Switzerland
  2. Stephen Pratt, Arizona State University, United States

Publication history

  1. Received: January 23, 2019
  2. Accepted: May 30, 2019
  3. Version of Record published: July 2, 2019 (version 1)

Copyright

© 2019, Wendt et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 778
    Page views
  • 69
    Downloads
  • 2
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)