The feeling of hunger is a ubiquitous phenomenon. Every day we face the cycle of hunger when meal time is approaching and satiation following food consumption. Hunger is affecting almost every human behavior (Loewenstein, 1996), ranging from our internal physiology, through our moods, to different aspects in the process of decision-making. For example, gastric ghrelin production, which is high during hunger, is believed to regulate both food intake (Sarker et al., 2013) and induces adrenocorticotrophin hormone (ACTH) secretion (Kluge et al., 2011), which in turn increases cortisol release and can promote anxiety-like responses (Chuang and Zigman, 2010). Hunger motivates food seeking behavior and food consumption (Nederkoorn et al., 2009). Hunger also induces an approach bias, is associated with an impairment of response inhibition, and affects attention allocation in human subjects (Loeber et al., 2013). Moreover, eating in the absence of hunger is considered an index of disinhibited eating (Francis et al., 2007) and higher impulsivity (Hou et al., 2011; Farrow, 2012).

The hunger cycle has a considerable effect on decision-making. For example, the percentage of favorable rulings by judges drops gradually from ≈65% to nearly zero before a food break and return abruptly to ≈65% afterward (Danziger et al., 2011). This demonstrates that hunger levels presumably affect how judges decide to rule. However, the full effect of changing hunger levels on decision-making yielded somewhat contradicting results. Our aim in the current study is to clarify these inconsistencies using both objective (physiological markers) and subjective (self-reports) measurements of hunger levels, and conduct several behavioral economic tasks within the same individual.

The fundamental finding in previous studies is that the value of a deprived reward, such as food when hungry, is increased in animals (Herrnstein and Loveland, 1974; Epstein et al., 2003; Raynor and Epstein, 2003; Pompilio et al., 2006; Aw et al., 2009) as well as in humans (Hill and McCutcheon, 1975; Spiegel et al., 1989; Drobes et al., 2001; Epstein et al., 2003). Food deprivation also increased the activity in value-related brain areas following presentation of high caloric foods (Siep et al., 2009). Furthermore, a previous study showed that delayed choices were influenced by current hunger as well as future hunger (Read and van Leeuwen, 1998). Hungry subjects chose more unhealthy snacks than did non-hungry subjects, and they chose more unhealthy snacks for immediate consumption than for delayed consumption. However, the study did not directly measure hunger levels for each subject but rather examined all subjects at two time points (separated by 1 week) during the day (after lunch and late afternoon), assuming similar hunger levels for all subjects. Another study demonstrated the link between energy depletion and reduced self-control, and found hunger to promote aggressive behavior (DeWall et al., 2011).

Hunger has also implications on consumer behavior, such that individuals tend to purchase more products (Nisbett and Kanouse, 1969; Mela et al., 1996; Tal and Wansink, 2013) and products higher in caloric content (Tal and Wansink, 2013) when hungry compared to when they are satiated. In addition, financial dissatisfaction results in an increased desire for food energy (Briers and Laporte, 2013). In the social decision-making domain, hungry subjects donated less to charity and to other players in a “give-some game” than did their sated peers, and they ate more chocolate M&Ms after imagining winning 25 than after imagining winning 25,000 (Briers et al., 2006).

It seems that hunger affects risk preferences as well, but the exact nature of this effect is still under debate in the literature (Houston, 1991; Mcnamara and Houston, 1992; Bateson and Kacelnik, 1996; Bednekoff, 1996; Smallwood, 1996; Bateson, 2002; Stephens, 2008). On the one hand, there are studies demonstrating an increase in risk-seeking when an organism is hungry compared to satiated. The idea is that in situations of extreme hunger, which can lead to death, the organism must become risk-seeking and look for food at the expense of increased danger. However, others show the opposite effect and reasoning (Mcnamara and Houston, 1992; McNamara, 1996; Kacelnik and Bateson, 1997; Stephens, 2008). For example, studies of Muslim stock markets indicate that overall market volatility declines sharply during the month of Ramadan, when people fast throughout the day for a month (Seyyed et al., 2005; Bialkowski et al., 2009).

The question that arises is then, why are the effects of hunger on decision making not consistent? We suggest that part of the reason is that most previous studies that examined the effect of hunger on decision-making in humans used subjective reports rather than objective physiological measurements to estimate subjects’ actual level of hunger. The use of questionnaires for estimating hunger levels is, by definition, subjective, and tends to bare various biases. First of all, when using self-report questionnaires, there is a strong assumption that subjects adhere to the experimental manipulation, which usually instructs subjects to refrain from eating for a specific amount of time. Second, it is well documented that subjective reports, such as the visual analog scale (VAS) that is prevalent in hunger studies, can be biased, inaccurate, sensitive to demand effects, and influenced by the social desirability bias (Fisher, 1993; Neeley and Cronley, 2004). Hence, previously reported effects could cause unwarranted theoretical or practical conclusions about consumers’ psychological traits (Campbell, 1950; Peltier and Walsh, 1990), purchase motivations (Levy, 1981), attitudes, intentions, and behaviors (Mensch and Kandel, 1988; Peltier and Walsh, 1990; Podsakoff et al., 2003), as well as contradicting conclusions regarding the effect of hunger levels on risk preferences (Levy et al., 2013).

In addition, when using the VAS there is a strong assumption that hunger levels increase linearly. Moreover, usually scientists treat the VAS as a “true” quantitative ratio scale, which is inaccurate. Assuming a ratio scale to scores that do not behave in such a manner may lead to a biased estimation of the measured variable (Narens, 1981; Alper, 1985; Luce, 2001), and thus may lead to inaccurate (and in some cases opposite) conclusions regarding the measured effect. Even more problematic is that some studies (e.g., Briers et al., 2006) rely solely on the experimental manipulation and examine average group effects, ignoring the large variability in hunger levels across subjects.

Therefore, there is a need for an objective and unbiased measurement in order to better understand the effect that different hunger levels have on decision-making, and it needs to be on a subject-specific basis. The most straightforward method is to rely on physiological measurements, which are objective and are very hard to control or intentionally bias without active external interference. A common method for estimating hunger levels is to draw blood from subjects at several time points during the experiment and examine glucose levels. However, using glucose levels does not always give a reliable or consistent result for hunger levels (for a review see de Graaf et al., 2004). Other, more reliable signals of hunger levels are ghrelin – both for short and long term levels (Tschop et al., 2000; Cummings et al., 2001; Nakazato et al., 2001; Wren et al., 2001; Druce et al., 2005; Klok et al., 2007) and leptin – for long-term energy balance (Sawchenko, 1998; Klok et al., 2007). Although using blood samples to estimate subjects’ hunger levels is relatively accurate when measuring ghrelin levels, there are several drawbacks to applying this method in field studies or to use in marketing research in the industry. Drawing blood is complicated, invasive, expensive, and requires expertise and extensive approvals. Hence, we propose to use saliva samples instead of blood samples in order to estimate objective physiological levels of hunger in humans. Saliva samples are easy and cheap to gather, do not require expert personnel, easily conducted anywhere, can be analyzed relatively fast, and are far less invasive than blood draws. However, ghrelin and leptin are not well detected in saliva; therefore, an alternative marker is needed.

In the current paper, we are using a physiological marker that was only recently associated with hunger levels – alpha-amylase (AA) (Harthoorn and Dransfield, 2008). Salivary AA is one of the most important enzymes in saliva that catalyzes the hydrolysis of starch and glycogen (Yang et al., 2015), and is involved in defense against pathogens (mainly food-born) (Petrakova et al., 2015). AA secretion increases as a result of two main events: one, AA has a pronounced and distinct diurnal rhythm with a strong drop in activity in the first hour after awakening, followed by a steady increase toward the evening (Nater et al., 2007). Note that this diurnal cycle is not related to food consumption or hunger levels. Second, AA secretion increases after food consumption. AA is secreted into the oral cavity from the acini cells of the parotid gland (O’Donnell et al., 2010), which is innervated by efferent sympathetic nerves (Vanstegeren et al., 2006). Thus, AA secretion following food intake is a direct response to the activity of the sympathetic nervous system (Nater et al., 2006; Harthoorn and Dransfield, 2008; Jayasinghe et al., 2014). Previous studies demonstrated the link between salivary AA and food intake, especially its role in metabolism of carbohydrates (Squires, 1953; Jayasinghe et al., 2014). The increase in salivary AA following food consumption appears to be related to the nutrient content of the consumed food (Squires, 1953; Mejean et al., 2015), as well as to the consumed amount and, importantly, to perceived fullness, i.e., not being hungry (Harthoorn, 2008; Harthoorn et al., 2009). The most straightforward evidence for the link between salivary AA and hunger comes from a study that explicitly examined the relationship between food intake, self-reported levels of satiety and hunger, and levels of salivary AA. The authors found a positive correlation between AA increase after a meal and satiety and fullness, alongside a negative correlation with hunger and the desire to eat. Another study (Harthoorn, 2008) reported two relevant findings: (1) a negative correlation between the amount of food consumed and the level of AA before consumption, in line with the assumption that hungrier subjects would eat more and (2) a positive correlation between the amount of food consumed and the level of AA after consumption, suggesting that subjects who ate more felt less hungry.

Therefore, based on these previous studies and extensive preliminary examinations conducted in our lab, we chose to use salivary AA as our indicator for subjects’ shift from hunger to not-being hungry following food intake. Importantly, an objective subject-specific measurement in each session has the additional benefit of annulling individual differences in baseline hunger levels due to differences in food consumption on the day before the experiment.

To conclude, in the current study our aim was twofold: First, to examine how a change in hunger levels affect choices in different tasks. Second and more importantly, we directly contrasted the effect of objective (AA levels) with subjective (VAS scores) measurements of an internal human state and examined which of them could reliably explain some of the behavioral changes in choices. In order to do so, subjects came to the lab twice – once hungry and once satiated. We define hunger as the physiological state after 12 h of fasting, corresponding to skipping one’s breakfast. In each session, subjects completed three choice tasks and we estimated their objective and subjective levels of hunger by measuring AA levels in collected saliva samples as well as collecting their subjective reports on a VAS.

We had several hypotheses. First, we examined if we would replicate previous findings (Levy et al., 2013) that, on average, less hungry subjects (compared to hungry subjects) are less risk averse. Second, that hungry subjects will tend to reject more (compared to when being less hungry) unfair offers in the ultimatum game. Third, that hungry subjects will be less consistent in their choices. Fourth, and most importantly, that the behavioral effects will be moderated by the objective, but not the subjective, measurements of hunger.

We present here data that corroborate our first two hypotheses but not the third one. Additionally, we clearly demonstrate that only objective measurements, but not subjective reports of hunger, moderate the effect on subjects’ choices, thereby corroborating our forth hypothesis. We show that hunger levels affect subjects’ choices and that when estimating the effects of internal state, it is crucial to use a physiological objective measurement because it is more reliable, accurate, and less biased than subjective measurements based on self-report.

In the current study, we set out to explore the ways in which hunger affects a number of human economic behaviors: risk-taking, social preferences, and consistency. Notwithstanding, we had another major aim in mind – introduce a new method to objectively and reliably measure levels of hunger. Using three separate behavioral paradigms, we found that hunger tips subjects’ risk preferences toward more risk-seeking, modulates their sensitivity to fairness, but evidently has no detectable effects on consistency of economic choices. More importantly, we show that our objective measure of hunger moderates the behavioral effects of hunger, while the commonly used subjective self-report measure fails to do so.

Our first finding, that hungrier subjects are more inclined to choose the lottery option over a safe amount of money, is in accordance with some previous work (Symmonds et al., 2010; Levy et al., 2013), that showed a greater tendency to be less risk-averse when hungry. These previous findings and our current results are in line with the notion that organisms tend to become less risk averse (and even risk seeking) under extreme hunger conditions because they need to take more risks in order to find food for survival. Presumably, although our subjects were not starving, this tendency of hungry individuals to become less risk averse is evidence of how evolutionary pressure from the past is still influencing our behavior. However, as mentioned above, there are contradictory findings as for the direction of the effect of hunger on risky behaviors. While some studies report an increase in risk-seeking behavior under deprivation conditions, others find the opposite effect, with subjects becoming more risk-averse as resources become scarce. A possible explanation to the discrepancy in results may arise from individual differences across the measured population, with different members of a species reacting differently to the same changes in environment. This phenomenon was demonstrated in a previous work, in which the authors found that hunger has a converging effect on a population – individuals who were highly risk-averse when satiated became less so when hungry, while risk-seeking individuals became more risk-averse (Levy et al., 2013). Another possible explanation for the discrepancy is that different studies manipulate hunger to varying degrees, which suggests that the modulation of risk preferences depends on the severity of deprivation. Furthermore, several previous studies that showed contradicting findings relied on self-reports for estimating hunger levels. As we clearly demonstrated in the current study, hunger levels as estimated using self-reports were not associated with any of our choice tasks, and they were not a reliable indication of the “true” hunger levels but rather a mix of hunger levels, desirability bias, and experimental demand effects. Moreover, animal studies showed that not all hungry animals (even to the same extent) behave the same way. Only animals that were under a negative budget rate environment (i.e., the rate of food arrival/consumption will not be enough for the animal to survive) demonstrated risk-seeking behavior while animals with a positive budget rate did not (Caraco et al., 1980). Interestingly, we find a difference in risk preferences only under the Hunger condition. This suggests that only a true state of hunger influences risk attitudes. That is, when subjects were in the Non-Hunger session, although differences in AA between subjects were present, they are not as hungry as they are in the Hunger session. Only when faced with substantial hunger, as is experienced in the Hunger session, subjects’ attitude toward risk change – and this change is moderated by the individual differences in the level of physiological hunger, such that the hungriest subjects exhibit the biggest risk-seeking behavior. Our objective measure of the physiological hunger state was able to capture the subject-by-subject variability in hunger levels and therefore serve as a good proxy for the effect that individual hunger levels have on individual risk preferences. In any way, careful and systematic study of humans as well as other animals will be needed to tease apart the contribution of environment and individual differences. Notwithstanding, the use of physiological tools to measure the objective levels of hunger would be crucial for such studies, and as we show here, the AA enzyme is a prime candidate to study the mediating effect of hunger on risk preferences.

Our second finding is that subjects under extreme physiological states – either very hungry or very not hungry – exhibit greater sensitivity to fairness, by rejecting more unfair offers. A moderate internal state, such as a mild hunger, however, renders people less vindictive, with similar acceptance rates for fair and unfair offers. Much like studies of hunger and risk, previous studies that examined the effect of deprivation on social behavior struggled to reach a consensus – it appears that under various conditions and states, and for different individuals, deprivation causes different and opposite behaviors. For example, one study found hungry subjects to be less inclined to donate to charity, i.e., deprivation reduced social preferences (Briers et al., 2006). Another study, however, reported a negative association between consumption of sugary drinks with support for social welfare, i.e., deprivation was actually linked to increased social preferences (Aaroe and Petersen, 2013). Note that the latter study found no change in actual sharing behavior, only in the stated self-reported behavior, demonstrating yet again the feebleness of subjective measures. Other studies have examined the effect of stress on subjects’ prosocial behavior, and found that stress increases it (Buchanan and Preston, 2014). However, at least one study reports a temporal dependency for this effect, showing that only 90 min after exposure to acute stress subjects exhibit an increase in acceptance rates, but not immediately after (Vinkers et al., 2013). Furthermore, it seems that the prosocial effect of stress is dependent upon the affect it evokes – individuals who perceive the stress as positive (a challenge) become more altruistic, whereas individuals who perceive it as a threat become more selfish (Buchanan and Preston, 2014). We therefore propose, that the u-shaped association between levels of hunger and levels of fairness-sensitivity that we report here hints to the dynamic nature of hunger effects on prosocial preferences, similar to that of stress. That is, under mild deprivation human subjects tend to become less sensitive to fairness (increasing prosocial behavior), whereas under extreme hunger they demonstrate higher degrees of vindictiveness. It is important to note that higher acceptance rates of unfair offers may represent either prosocial behavior (acting altruistically toward the proposing partner) or an increase in selfishness (acting to increase self-gain) (Mancini et al., 2011). Our paradigm cannot distinguish between the two motivations, but it may be that the decrease in sensitivity to fairness we observe in extremely hungry subjects stems in fact from a self-preserving motivation, and not a change in social preference. Further research will be needed to specifically address this hypothesis.

In our third and final behavioral task, we could not identify an effect of hunger on choice consistency. From this null result, we must not make any general statements that hunger levels do not have an effect on choice consistency. All we can conclude is that under 12 h of fasting, using the GARP task and our measurement tool of hunger levels, we could not detect any changes in choice inconsistency as a function of hunger levels. Strengthening this possibility is a recent study that showed that marmosets display more consistent behavior as they become hungrier (Yamada, 2017). Hence, because our subjects demonstrated a relatively high level of consistency throughout the task and in both conditions, our failure to detect differences in consistency as a function of hunger levels may be due to a ceiling effect. It is also possible that in order to detect changes in the level of choice consistency subjects need to be a lot hungrier. We plan to examine this in future studies.

The most important finding of the current paper is that only an objective measurement of hunger levels, but not subjective reports, could accurately measure subjects’ hunger levels and serve as a reliable indication of how subject-by-subject variability in hunger levels influences choice behavior. Moreover, we demonstrated that an objective reliable measurement of hunger levels is crucial for the understanding of the complex effects hunger has on choice behavior. We show that looking at group averages alone or examining only the main effect of the manipulation (which was not significant) might lead to wrong conclusions regarding the effect of hunger on choice behavior.

Any conclusion drawn from a study first and foremost depends on the reliability of the measures it employs. Self-reports, common in social sciences, were often found to be biased by personality traits such as social-desirability or social-approval biases (Podsakoff et al., 2003). For example, subjects report inaccurately their height and weight (Gorber et al., 2007), their dietary intake (Freedman et al., 2014), as well as their attitudes and beliefs about others (Greenwald et al., 1998). We therefore consider subjective hunger ratings as malleable and imprecise. Although self-report measures may have value in and of themselves for studying the effects of a psychological state of hunger on behavior, which cannot be studied using physiological markers, we argue that their contamination with non-relevant information renders them unsuitable for our study. That is not to say that all self-report measures are wrong, but that their implementation must be highly controlled, in order to preclude subjects from adjusting their response to appease the experimenters. Furthermore, when studying the effects of physiological hunger on behavior, as we are here, the use of subjective tools is tricky at best. In our current work, we present both the need for an objective measure of an organism’s internal state, and put forward the novel use of a salivary enzyme called alpha-amylase as such a measure. Naturally, the most direct, accurate, and straightforward way of physiologically measuring hunger levels is via a blood sample, as have several previous studies done (Symmonds et al., 2010; Aaroe and Petersen, 2013). Some of these studies focused on glucose and/or insulin levels. However, in healthy humans the blood-level changes of these markers are rapid, happening in a relatively short time scale (Dinneen, 1997), and do not change very much due to the natural process of homeostasis (Anderson et al., 1967; Alzaid et al., 1994; Cohen, 2006). Hence, glucose or insulin levels are not good predictors of hunger levels after 12 h of fasting (when, in general, glucose levels are back to baseline) but rather an indication of the current number of calories entered into the body, and how fast can the body adjust to the rise in glucose (as is done in a glucose tolerance test, (Ernsberger and Koletsky, 2012; Zhou et al., 2016). Furthermore, there are contradicting results regarding the effect of sucrose levels on decision-making (Molden et al., 2012) (for a review, see Orquin and Kurzban (2016)].

An alternative measure is ghrelin and/or leptin, both of which are regulatory hormones taking part in the hunger/satiety cycle (Klok et al., 2007), which were shown to be relatively reliable indications for current hunger levels (ghrelin) and long-term fat storage (leptin) (Klok et al., 2007), as well as moderate risk preferences to some extent (Symmonds et al., 2010). However, although blood measures are arguably most accurate in assessing an organism’s internal state, including a blood draw procedure in an experiment may have undesirable consequences. First, it can potentially bias the selection of subjects, by excluding individuals with needle-anxiety. Second, drawing blood may have a stressogenic effect on subjects, which may alter their behavior (Buckert et al., 2014; Bendahan et al., 2017). Third, the procedural implications are plentiful: drawing blood requires skilled personnel, expansive equipment, and special requirements from ethics committees. On the other hand, using saliva samples is relatively cheap, easy-to-use, does not induce stress, and minimizes the selection bias.

Our finding hinders the traditional and wide use of the visual analog scale as a measure of individuals’ hunger levels, and suggests that it is unreliable and may lead researchers to wrong conclusions about the manner in which hunger affects behavior. We strongly believe that future researchers interested in the interaction of physiology with psychology should employ objective physiological tools, such as the AA enzyme. The method we introduce here is perfectly tailored for studies in behavioral economics and marketing, due to its ease of use and high applicability. Especially in a field of study of such abundant implications and interest, alongside highly controversial and contradicting finding thus far, researchers should employ the most meticulous and precise methods as possible.

DL designed the behavioral aspects of the experiment. MS-S designed the biological aspects of the experiment. MS-S and AS collected the data. MS-S, AS, and TS analyzed the data. MS-S, AS, TS, and DL wrote and edited the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer LS and handling Editor declared their shared affiliation.