Defining mood

Affect science distinguishes ‘mood’ from ‘emotion’. Emotions are strong affective responses that usually have visible behavioural effects, such as changes in facial expression⁽ Reference Greimel, Macht and Krumhuber ^{5 )} and are fundamentally communicative acts⁽ Reference Hammersley and Reid ^{4 )}. One way of defining emotion is according to facial expression – happiness, sadness, surprise, fear, disgust, anger⁽ Reference Ekman, Dalgleish and Power ^{6 )} – although this may not cover all emotions. Moods are weaker phenomenological experiences that may not have behavioural effects⁽ Reference Parkinson, Totterdell and Briner ^{2 )}.

It is also important to distinguish ‘protracted mood’ over a period of hours or days, which more readily can be assessed by questionnaire⁽ Reference Ekkekakis ^{1 ,} Reference Parkinson, Totterdell and Briner ^{2 )}, from ‘transient mood’, which fluctuates⁽ Reference Hammersley and Reid ^{4 )}. Most research with ‘mood’ in the title is about protracted, usually depressed, mood. Reports and ratings of protracted mood draw upon and somehow average information from episodic memory to generate the phenomenon of a relatively stable state from the underlying momentary flux of feelings. For example, depressed individuals exhibit specific memory and attention biases towards reporting negative events and thoughts⁽ Reference Elliott, Rubinsztein and Sahakian ^{7 )}, although in reality they also experience positive events and thoughts. Thus, although depression is probably caused by a complex of metabolic, neurological and cognitive factors⁽ Reference Graham, Salimi-Khorshidi and Hagan ^{8 ,} Reference Preiss, Brennan and Clarke ^{9 )} and has behavioural effects as well as phenomenological ones, depression as a protracted mood is cognitively formed and involves selective attention to different aspects of experience, as for instance in rumination⁽ Reference Huffziger, Ebner-Priemer and Eisenbach ^{10 )}.

Transient mood, being much shorter, is inherently variable. Additionally, the determinants of transient mood may be unconscious, or only briefly accessible to conscious and rapidly forgotten⁽ Reference Ericsson and Simon ^{11 )}. A transient mood does not dominate consciousness, for if it does then either it becomes strong enough to manifest in behaviour as an emotion, or it becomes a protracted mood, which permeates consciousness and/or behaviour for some time.

It may be inaccurate to conceptualise transient mood as a ‘state’⁽ Reference Hammersley and Reid ^{4 )}. By definition, transient mood is relatively brief and prone to change; also ‘state’ implies something that dominates consciousness and has content that can be introspected. Some affect research uses ‘valence’ to mean a tendency to orient towards some types of stimuli rather than others, or respond in certain ways, as in the attentional biases of depression⁽ Reference Elliott, Rubinsztein and Sahakian ^{7 )}. In animal models ‘value’ is sometimes used in a similar way⁽ Reference Grabenhorst and Rolls ^{12 )}.

As will be reviewed, transient mood effects may not always involve an overt orientation of this kind. Hammersley & Reid⁽ Reference Hammersley and Reid ^{4 )} suggest the use of ‘tint’. A transient mood can tint consciousness, but awareness of the tint itself varies, and the tint subtly emphasises some aspects of phenomenological experience and de-emphasises others. Valence and value are two examples of the effects of a mood tint, but there may be others, such as cognitive biases of various kinds.

Research on the effects of single acts of ingestion usually hypothesises changes in tint, rather than effects on protracted mood, or emotion. For example, high-carbohydrate breakfasts might tint consciousness with happiness⁽ Reference Micha, Rogers and Nelson ^{13 )}, without causing the happy facial expressions commonly shown in advertising.

Transient mood is caused by cognitive processes, along with physiological ones⁽ Reference Hammersley and Reid ^{4 )}. Asking individuals to evaluate, rate or report their mood tends to orient individuals to the affective tinting of their conscious state and also to change the contents of consciousness, filling it with the relevant mood assessment task. At the extreme, individuals may be unaware of their transient mood unless they are asked about it, because transient mood can exist without full awareness⁽ Reference Hammersley and Reid ^{4 )}. Affective priming involves both fully conscious and less conscious processes⁽ Reference Murphy and Zajonc ^{14 )}. One example is irritable mood, when an individual may feel normal and show no overt signs of irritability, but nonetheless have more propensity to exhibit irritation to stimuli that might not irritate them normally.

Affective priming is one cause of transient mood, where prior events and thoughts cause a tendency to be oriented towards cognitions of specific emotional significance. For example, mood-salient words embedded in a task can influence subsequent mood ratings without the subject noticing their systematic presentation⁽ Reference Chartrand, van Baaren and Bargh ^{15 )}. Similarly, foods or drinks with learned associations to specific moods might trigger the mood. Sweet taste consistently causes a transient facial expression of happiness, whereas bitterness or bitter–sweet tastes have less consistent effects⁽ Reference Greimel, Macht and Krumhuber ^{5 )}.

Even strong and immediate affect experiences are cognitively mediated⁽ Reference Smith, Haynes and Lazarus ^{16 ,} Reference Scherer ^{17 )}, so it is unsafe to assume that rated mood represents the unmediated physiological effects of ingestion on subjective state. Research participants' feelings can be influenced by their prior outcome expectancies⁽ Reference Hammersley, Finnigan and Millar ^{18 –} Reference Yeomans, Lartamo and Procter ^{20 )}. It is also necessary to consider the demand characteristics of the experiment, including the nature and demeanour of the experimenter⁽ Reference Green, Sandall and Phelps ^{21 –} Reference Walach, Schmidt and Biehr ^{25 )}. To minimise experimenter effects⁽ Reference Rosenthal ^{26 )}, social priming effects⁽ Reference Bargh ^{27 )} and expectancy effects⁽ Reference Wardell, Read and Curtin ^{28 ,} Reference Dawkins, Shahzad and Ahmed ^{29 )}, ideally all research with individuals should be double-blind whenever feasible.

Most procedures reporting on mood involve retrospective reconstructing of recent feelings, so by the time individuals rate or otherwise report on a small mood change, they may have forgotten or distorted its initial nature. Modern conceptualisations of consciousness, such as the working memory framework⁽ Reference Repovs and Baddeley ^{30 )} and fast v. slow processes⁽ Reference Kahneman ^{31 )}, regard it as having multiple components with different functions and properties, as well as indistinct boundaries from unconscious processes, rather than it being a simple state that can be examined and reported upon without problem. Many studies of the effects of ingestion on mood have assumed that rated mood represents conscious state, caused by ingestion. Then, research sometimes ‘back-calculates’ the physiological effects of ingestion from mood ratings, which is unsafe because of the above complexity and the relationships between physiological arousal and conscious state are complex⁽ Reference Hammersley and Reid ^{4 )}. For example, below will be reviewed the challenge of demonstrating that carbohydrate can improve mood due to elevating tryptophan levels in the brain and enhancing serotonin metabolism. Carbohydrate often apparently affects transient mood, but this mechanism applies only amongst individuals who crave carbohydrates and habitually ingest pure carbohydrate deliberately to improve mood⁽ Reference Corsica and Spring ^{32 )}.

However, sweetness produces a strong positive affect experience⁽ Reference Macht and Mueller ^{33 )} and a happy facial expression⁽ Reference Greimel, Macht and Krumhuber ^{5 )}, so it may have direct effects on subjective state. As well as in adult humans, this occurs also in young infants, primates and other animals⁽ Reference Berridge ^{34 ,} Reference Steiner, Glaser and Hawilo ^{35 )}. Depending on how one defines and distinguishes ‘reward’ from ‘reinforcement’ this may suggest that sweetness is a primary positive reinforcer with the potential to cause a positive mood without cognitive processing. Some tastes also trigger disgust reactions⁽ Reference Rolls ^{36 )}. However, there is no convincing evidence that any of the other properties of food are primary reinforcers. Indeed, some orosensory properties that tend to be associated with energy-dense foods, such as viscosity, are represented independently of their reinforcement value⁽ Reference Rolls ^{37 )}. Research on conditioned taste aversions suggests that near-instantaneous responses to foods are often learned⁽ Reference Konnecke, Wiesner and Ferstl ^{38 )}.

Research on the effects of carbohydrates on transient mood

This section reviews experiments that examine the effects of food and drink on transient mood. Excluded are experiments using caffeine and other drugs, studies that assess the protracted mood effects of various diets across several days or longer, and studies where mood is assessed but is not reported as a main focus of the study. Of remaining experiments there have been most on the effects of carbohydrates, including studies of sugar, breakfast cereal and breakfast bars. There are also a few studies of protein without comparing it with carbohydrate, a few on the effects of other meals, and some on the effects of miscellaneous foodstuffs that possibly have mood benefits. This section will focus on carbohydrates, because there are too few of the other studies to be able to judge the consistency of their findings, even before considering the challenges of mood rating.

In studies on simple carbohydrates of under 24 h, effects on mood have been found inconsistently⁽ Reference Micha, Rogers and Nelson ^{13 ,} Reference Brody and Wolitzky ^{39 –} Reference Thayer, Peters and Takahashi ^{61 )}. Not all studies have blinded participants to what they ingest⁽ Reference Blundell, Rogers and Hill ^{56 ,} Reference Hammersley, Reid and Duffy ^{58 )}, which may help explain variations in findings⁽ Reference Reid and Hammersley ^{45 )}. Another difficulty is that many studies use multiple measures of mood, and sometimes multiple questionnaires, making it difficult to interpret changes on a small number of mood items. The questionnaires themselves are not optimised for this purpose (see below).

Other difficulties in this literature include that eating has generic mood effects, releasing endorphins⁽ Reference Benton and Donohoe ^{62 )}, alleviating hunger and the cognitive effects of fasting⁽ Reference Benton and Brock ^{63 )}, and reducing dehydration⁽ Reference Ganio, Armstrong and Casa ^{64 )}. Consequently, showing that a meal of specific composition changes mood compared with fasting cannot be attributed to the nutrient content of the meal. To demonstrate this requires comparison with a control meal of systematically varied composition.

Breakfast compared with fasting improves cognitive performance, often without affecting rated mood⁽ Reference Micha, Rogers and Nelson ^{13 ,} Reference Defeyter and Russo ^{65 –} Reference Widenhorn-Mueller, Hille and Klenk ^{68 )}. When mood is affected by breakfast, then alertness tends to increase and fatigue to decrease⁽ Reference Micha, Rogers and Nelson ^{13 ,} Reference Pasman, Blokdijk and Bertina ^{41 ,} Reference Lloyd, Rogers and Hedderley ^{53 ,} Reference Veasey, Gonzalez and Kennedy ^{66 ,} Reference Smith and Stamatakis ^{67 ,} Reference Maridakis, Herring and O'Connor ^{69 –} Reference Smith, Kendrick and Maben ^{71 )}, but, as shall be reviewed, there are difficulties with inconsistent mood measurement techniques. It is also conceivable that breakfast improves cognitive function after an overnight fast, then consequently better cognitive function improves rated mood.

Five papers explicitly compare high-carbohydrate breakfasts with high-fat or high-protein breakfasts of otherwise similar content. For this comparison, there may be issues regarding the different speeds at which different meals are digested. Two have found no distinct mood effects of different meals using the Visual Analogue Scales (VAS) and the Profile of Mood States (POMS)⁽ Reference Lemmens, Born and Martens ^{42 ,} Reference Probst, Humpeler and Heinzl ^{43 )}. Experiments finding specific ‘significant’ mood effects of one meal type compared with another are difficult to interpret, because effects found differ. Carbohydrate may reduce fatigue and dysphoria (VAS)⁽ Reference Lloyd, Rogers and Hedderley ^{53 )}, or increase fatigue (VAS)⁽ Reference Pasman, Blokdijk and Bertina ^{41 )}, or increase alertness and happiness whilst reducing nervousness and thirst (VAS)⁽ Reference Micha, Rogers and Nelson ^{13 )}. It is impossible to tell whether reduced ‘fatigue’ and increased ‘alertness’ are the same thing, or how VAS ‘fatigue’ can be translated into POMS scales or those of another instrument. Nor can one tell whether one instrument is more sensitive than another.

There have also been four experiments looking at giving glucose drinks during vigorous exercise, compared with only water. Such drinks had no mood (all POMS) or performance effects on recreational cycling⁽ Reference O'Neal, Poulos and Wingo ^{72 )}. Playing soccer, compared with placebo they increased ratings of activation and of perceived exertion⁽ Reference Backhouse, Ali and Biddle ^{73 )}. They improved vigilance, reduced confusion and increased vigour amongst military personnel during sustained aerobic activity⁽ Reference Lieberman, Falco and Slade ^{74 )}. In real desert training glucose drinks improved energy intake but did not influence mood⁽ Reference Cline, Tharion and Tulley ^{75 )}. As will be discussed, the extrinsic situation can dominate mood and make it unlikely that food will influence it. Again, there are issues of mood measurement, and the possibility that improved cognitive function causes mood.

However, there is a plausible mechanism whereby simple carbohydrates could affect transient mood, by altering tryptophan levels in the brain⁽ Reference Kroes, van Wingen and Wittwer ^{76 )}, although even small amounts of protein consumed simultaneously prevent this, so the mechanism may be uncommon in everyday living⁽ Reference Benton ^{77 )}. Moreover, the dose–response relationships to tryptophan for cognitive functioning and for mood appear to be different, and mediated by the prior neurochemical state of the brain⁽ Reference Hulsken, Martin and Mohajeri ^{78 )}, again raising the question of whether tryptophan sometimes improves mood by improving cognitive efficiency.

It has been hypothesised that some individuals are prone to carbohydrate craving, making them atypically vulnerable to mood changes and prone to using relatively pure carbohydrate to manage their mood⁽ Reference Wurtman and Wurtman ^{79 –} Reference Ventura, Santander and Torres ^{81 )}. According to this hypothesis, only carbohydrate cravers should exhibit mood changes after carbohydrate. There is evidence of a correlation between craving and mood or mood management⁽ Reference Christensen and Pettijohn ^{82 )}, and there has been a blind, randomised controlled trial of responses to a carbohydrate-rich food, given after an induced low mood. This suggested that carbohydrates can reverse induced low mood amongst individuals who score highly on ratings of carbohydrate craving, and, critically, also regularly consume pure carbohydrates that could affect brain serotonin⁽ Reference Corsica and Spring ^{32 )}. This trial used the POMS, but only looked at effects on dysphoria (hedonic tone) and vigour (arousal), finding reduced dysphoria but no effect on arousal.

Another study relevant to the tryptophan hypothesis⁽ Reference Markus, Panhuysen and Tuiten ^{83 )} fed high- and low-neuroticism-scoring participants high-carbohydrate or high-protein diets across breakfast, a snack and lunch and gave them a stressful task. High-neuroticism participants who had eaten the high-carbohydrate diet did not show the stress-induced rise in depression, decline in vigour (POMS) and elevated cortisol levels that they showed after the high-protein diet. This suggests that carbohydrate may moderate or protect against a stressed mood response, rather than changing normal mood.

In summary, it appears that carbohydrates can affect mood but it remains difficult to specify the relevant conditions. Nutritional status, stress, and the habitual use of carbohydrates to affect mood may be important, but findings are obscured by some designs that do not blind participants to what they are ingesting, and by lack of a consistent and planned approach to mood measurement. It is possible that the POMS and VAS measure somewhat different things, and their sensitivity to nutritional manipulations may also differ.

Causes of transient mood rating

Mood rating may only take from a few seconds to a few minutes, depending on how many items are used, but this is ample time for complex unconscious and conscious cognitive processes to affect mood rating⁽ Reference Poulton ^{84 )}. Priming effects may be particularly important, where prior stimuli affect subsequent responses although the individual does not consciously recall the stimuli, and the stimuli may be non-salient and incidental to the response.

Analysis of mood ratings

Mood ratings repeated over time generate large quantities of data. There is the potential for type I error, so data analysis needs to have a plan to minimise the number of statistical comparisons. Such a plan ideally includes having clear hypotheses about how ingestion should influence mood, regarding which components of mood should be affected, at what time, and for how long. Research should still assess mood comprehensively, to avoid item availability bias, but analysis should focus on the components where effects are hypothesised.

It is not always possible to form clear hypotheses about how ingestion should influence mood, particularly as mood is often assessed as an adjunct to other research questions. In such cases it is important that a minimal number of mood components are analysed, and a conservative approach to analysis is taken that considers the problem of multiple comparisons. Without a considered approach to analysis, there is a risk of the ad hoc identification of isolated ‘significant’ but meaningless differences, on one mood rating at one point in time, or a few ratings at different points in time. This may partially explain inconsistent and not readily replicated findings regarding the effects of ingestion on mood.

In analysis and interpretation, it is important to not lose sight of the fact that mood ratings are ordinal ratings of subjective state, not accurate interval measurements of anything. The absolute rating values are relatively meaningless and analysis should focus instead on change and comparison between groups or conditions and over time. It is advisable to consider individual differences in rating scale use, for example by transforming raw data into z-scores.

Requirements of a transient mood-rating system

1. (1) It should cover the theoretically considered main dimensions of mood (arousal and affect).

2. (2) It should also assess common physical feelings such as hunger, thirst, pain and illness, so that changes in these do not manifest as changes in arousal or affect.

3. (3) It should use as few questions as possible, given requirements 1 and 2, to: facilitate repeated administration; reduce test fatigue; reduce non-conscious priming effects, which increase with repetition; reduce other biases; minimise the duration of the rating process in order to capture ‘mood’ across a relatively brief moment in time; facilitate data analysis.

4. (4) Individual rating items should be structured as polar opposites.

5. (5) The granularity of each scale should be fit for purpose, usually either an unnumbered line, or a five- or seven-point scale.

6. (6) The instrument should be straightforward and have face validity because mood is inherently subjective. What are the implications of ‘subtle’ mood effects averaged across multiple items? If someone is subjectively experiencing a mood they should be able to report it with an appropriate instrument administered in a timely fashion.

7. (7) Assessing mood too often may radically alter the natural phenomenological condition. For example, fifty or more questions rated repeatedly across a period of hours may dominate consciousness more than any effects of the initial food. Alternative approaches include making more use of between-subjects designs with mood measured but once in each group, and ecological momentary assessment methods⁽ Reference Moskowitz and Young ^{103 )} where participants do not know when they will be asked to rate mood.

The putative mood effects of substances are usually only one of a range of things to be assessed when measuring the short- or long-term effects of what individuals ingest. Many studies have relied on either very basic rating scales, or one or more standard rating scales discussed below. In analysis and interpretation of data, mood-rating data are generally treated as unproblematic measures of internal subjective state. Indeed, abstracts often fail to report how mood was rated, as if for example ‘fatigue’ was the same in every rating system.

Are widely used rating systems fit for purpose?

Table 1 summarises the most commonly used rating systems. The Visual Analogue Mood Scale⁽ Reference Folstein and Luria ^{104 )} is too simple, with just one rating item of best to worst mood, and it has not been properly validated for use with transient mood, so it is not recommended.

Table 1 The mood-rating systems most commonly used in ingestion research

VAMS, Visual Analogue Mood Scale; POMS, Profile of Mood States; VAS, Visual Analogue Scales; ADACL, Activation–Deactivation Adjective Check-List.

Most popular in nutrition research⁽ Reference Spring, Maller and Wurtman ^{59 ,} Reference Backhouse, Ali and Biddle ^{73 ,} Reference Gobbi, Slater and Boucher ^{105 –} Reference D'Anci, Watts and Kanarek ^{111 )} is the POMS⁽ Reference Micha, Rogers and Nelson ^{13 ,} Reference Pasman, Blokdijk and Bertina ^{41 ,} Reference O'Neal, Poulos and Wingo ^{72 ,} Reference Lieberman, Falco and Slade ^{74 ,} Reference Markus, Panhuysen and Tuiten ^{83 ,} Reference McNair, Lorr and Droppleman ^{112 –} Reference Achten, Halson and Moseley ^{114 )}, which was originally developed to assess how individuals ‘have been feeling recently, including today’, so it was validated to measure protracted mood. It originally asked participants to rate seventy-two mood adjectives on a 1–5 scale of ‘not at all’ to ‘extremely’, generating the six scales shown in Table 1, and it has developed over the years⁽ Reference Reid and Hammersley ^{45 ,} Reference Benton and Donohoe ^{62 –} Reference Veasey, Gonzalez and Kennedy ^{66 )}. Now updated to the POMS-2, with sixty-five items, or sixty items in the youth form, or thirty-five items in the short form⁽ Reference Heuchet and McNair ^{113 )}, giving six scales plus ‘friendliness’ (which is rated separately), labelled somewhat differently to POMS-1.

Widespread use does not of itself validate a measure⁽ Reference Ekkekakis ^{1 )}. The POMS never has been properly validated as a measure of transient mood, and it may be too complex for purpose, particularly for repeated administration. Also, it is entirely empirically derived and it is unclear how the six scales map on to the two primary mood dimensions of arousal and affect: arousal might be measured with ‘vigor’ or ‘fatigue’, while bad mood might be measured with ‘anger’, ‘tension’ or ‘depression’. If POMS results in nutrition research were highly consistent, then one might be more assured of its utility.

The VAS⁽ Reference Herbert, Johns and Doré ^{115 )} comprise eighteen (or sometimes sixteen) rating items producing two, or three, factors. Although VAS factors are similar to the main dimensions of mood, they are empirically derived and the variation in the number of factors raises questions about the stability of the underlying factor structure and whether those particular questionnaire items are the most appropriate ones⁽ Reference Ekkekakis ^{1 )}. However, there is some evidence that the VAS are sensitive to changes over periods of less than 24 h, especially to the effects of breakfast⁽ Reference Lloyd, Rogers and Hedderley ^{53 ,} Reference Defeyter and Russo ^{65 –} Reference Smith and Stamatakis ^{67 ,} Reference Herbert, Johns and Doré ^{115 )}.

The evidence that the Activation–Deactivation Adjective Check-List (ADACL) has this sensitivity is more limited because it has not been widely used other than by its author⁽ Reference Thayer ^{48 ,} Reference Thayer, Peters and Takahashi ^{61 ,} Reference Thayer, Newman and Mcclain ^{116 –} Reference Thayer ^{119 )}. The ADACL was theoretically derived according to a two-dimensional mood structure as is commonly found in affect research⁽ Reference Parkinson, Totterdell and Briner ^{2 ,} Reference Thayer, Newman and Mcclain ^{116 )}. Dimensions are arousal or energy level and hedonic tone (positive or negative mood) and it is also possible to cross-culturally map emotion words on to these dimensions⁽ Reference Russell ^{120 )}. The fifty-item version of the ADACL is too long for repeated administration, but the twenty-item short version could be used in this way.

In choosing a method, a caution is that the validated VAS are sometimes cited as justification for using other, non-validated, sets of visual analogue scales, as if it were the scale design rather than the content that had been shown to measure mood. English includes hundreds of mood-relevant words⁽ Reference Frijda ^{121 )}, and the choice of which to use needs to be theoretically or empirically derived. There is a risk of arbitrary items being chosen⁽ Reference Ekkekakis ^{1 )}.

Not commonly used in nutrition research, but worth considering, are the ‘Visual Analogue Mood Scales’⁽ Reference Arruda, Stern and Hooper ^{122 )}, which are widely used in medical research, require minimal cognitive or verbal skills, and measure eight mood dimensions: afraid, confused, sad, angry, energetic, tired, happy and tense. As with the POMS, these dimensions are not theoretically derived and do not appear to have been shown to be sensitive to changes over less than 24 h. Also worth considering are the Positive and Negative Affect Scales⁽ Reference Watson, Clark and Tellegen ^{123 ,} Reference Hartmann, Rief and Hilbert ^{124 )}; here the dimensions are negative arousal – annoyed/anxious v. contented/serene – and positive arousal – euphoric/elated v. tired/bored. This maps the same space as arousal/hedonic tone, emphasising different axes. Another method is the Geneva Emotion Wheel, which maps emotion words on a circle with axes of hedonic tone (called emotional valence) and perceived low to high control, which is hypothesised to influence mood⁽ Reference Scherer ^{17 )}.

A final consideration is how many dimensions of mood to measure. Two-dimensional models of mood are usefully simplifying, but using a two-dimensional scale begs the question of whether food might affect components of mood other than arousal or hedonic tone. Moreover, often in nutrition research phenomenological state is not assessed just with a single mood questionnaire, but also with questions regarding subjective physical well-being, and sometimes also with multiple mood questionnaires, including many others in wide use in health-related research not shown to be sensitive to changes over less than 24 h. As discussed above, these different ratings can potentially interact in ways that are not yet understood, but with multiple questionnaires to complete, it is not unusual for designs to require participants to complete several hundred questions in total, leading to test fatigue.