Food Rejection Taxonomy
In this Section, I discuss how food rejections are not all alike. I attempt to replicate and build upon the methodology and results of Rozin & Fallon (1980; Fallon & Rozin, 1983) in devising a taxonomy of food rejection.
In their two papers, Paul Rozin and April Fallon introduced a new methodology whereby we ask multiple questions of subjects about their food rejections. The data so derived can then be used to construct a taxonomy of food rejection behaviours, showing how they group and how those groups interrelate. Rozin & Fallon (1980) initially suggested that there are three basic forms of food rejection: 'distaste', 'disgust' and 'danger'. I have conducted a questionnaire study to test this tripartite division, expanding on the earlier work. I also apply some novel statistical techniques to aid the interpretation of results.
As well as replication, I hope to demonstrate the taxonomy's value by seeing where potentially contentious food rejection phenomena fit. Pliner & Pelchat (1991) have already used the methods of Rozin and Fallon in demonstrating the special status of novel foods of animal origin, something I also replicate. Further, I investigate whether the food rejection behaviour in major food scares of recent years (concerning BSE and Salmonella) fit the model as we would expect. Other rejection phenomena are also considered.
2.1 The Need for a Taxonomy
The very beauty of the list heuristic is also its limitation. By asking one simple question about each food and averaging over many foods, the list heuristic allows us to see the wood despite the trees. However, food rejections are a multi-species forest.
Food rejections are not all of the same type. Problems in Section 1 over category restrictive diets and applying the list heuristic cross-culturally have already demonstrated that one cannot always ignore such differences. Although I judged the differences between neophobic and non-neophobic responses and between responses to different food types insufficient to damage the list methodology's utility, there are still differences to explain.
There is a tremendous range of behaviours involving food rejection. My personal distaste for celery seems very different to a colleague's fearful attitude towards tomatoes, apparently a much generalised haemophobia. A colleague will no longer eat pork after her fireman husband told her that burning human bodies smell like cooking pork; that seems very different to a rejection of pork based on the Mosaic food laws. The uncertainty of the average person faced with sushi in a Japanese restaurant seems very different to the behaviour of someone suffering anorexia nervosa.
Some of these responses may be more common than others, but the less frequent ones may still be of great importance or give us greater insight into the underlying psychology. Different food rejection phenomena may also be of interest for different reasons. Food manufacturers want to know about neophobia. Politicians may want to know about food scares. Medical professionals want to know about health-related choices. Of particular note are what we might call pathological food rejection behaviours. Certain anorexias appear to be psychogenic, notably anorexia nervosa and cancer anorexia. In these, food rejection behaviours are inappropriately applied: what is nutritious and should be ingested is wrongly spurned.
Given this diversity, why does the list heuristic work at all? As I suggested in 1.6, the list heuristic perhaps works because most food rejections are of one type (neophobia) and because it implicitly concentrates on one type of food rejection by accepting cultural definitions of what is food in the first place. Further, what might work on a population level is not necessarily appropriate on an individual level, be the individuals and population of people or foods. Anyway, the list heuristic does not necessarily work in many situations. I have discussed the problem of category restrictive diets and experimenters have implicitly avoided extreme cases, as in pathological food rejection behaviours.
If it is too simplistic to lump all of these behaviours together, how then should we study this gamut of food rejection? Clearly much research has considered many of these behaviours separately. The literature on anorexia nervosa, cancer anorexia, religious food taboos, learned aversions, bitter sensitivity and so on are all considerable. Yet all these differences still exist within a commonality that the behaviours involve not eating things. Is there some way of recognising the diversity of food rejection behaviours while still studying food rejection behaviours in general? If so, one would hope that lessons from across the spectrum of behaviours could aid research into any specific one.
2.2 The Method of Rozin & Fallon (1980)
If we have a range of different food rejection behaviours, a sensible first step would be to place these in some sort of taxonomy, which is what Rozin and Fallon set out to do in their 1980 paper "The psychological categorization of foods and non-foods: A preliminary taxonomy of food rejections" and in a follow-up study (Fallon & Rozin, 1983). They devised a method to allow the categorisation of food rejections based on a crucial paradigmatic shift.
Instead of simply measuring overall preference (be that a direct preference question, a willingness to eat or a consumption measure), Rozin and Fallon had subjects answer a range of questions about each food, or, specifically, about their reactions to each food. Thus, subjects were asked to distinguish between, for example, the taste of the item, whether the item makes them nauseous and whether they fear any post-ingestive consequences from the item. Subjects also answered questions that placed the food in different circumstances to further investigate different aspects of the behaviour.
Instead of describing a food as rejected, we can describe it as being unpalatable and nauseating; or as palatable but considered to cause hazardous post-ingestive consequences; or as unpalatable, but not contaminating. We can then go on to identify specific patterns of aversion, like 'disgust' or 'danger' responses.
The realisation that a food rejection behaviour could be multi-faceted was certainly not new. In Section 3, I consider the debate about the nature of the aversion in learned food aversions. What was innovative about the two Rozin and Fallon studies was the idea to ask a standardised set of questions about multiple items. This allows us to construct a profile for each food and multivariate statistical technique can then be applied to interpret the relationships between profiles.
Rozin & Fallon (1980) had subjects answer True, False
or Uncertain to a list of eight questions. The proportion answering
per question gave a profile for each food. For the 1983 Fallon &
Rozin paper, subjects answered twelve questions on Likert scales (with
only ten questions used in the main analysis). The questions used in the
two papers are given in Table 2.1.
|Rozin & Fallon (1980)||Fallon & Rozin (1983)|
|I dislike the taste, smell or texture of this food.||The taste of (substance).|
|The thought of eating this food makes me nauseous.||The thought of eating (substance) produces nausea.|
|I dislike the idea of having this food in my stomach.||The presence of (substance) in your stomach.|
|I dislike the appearance of this food.||The appearance of (substance).|
|I would dislike it if some of this food got on my hands.||The presence of (substance) on your hands. [Not used in main analyses.]|
|I would dislike any dish that contained even the tiniest amount of this food, even if I could not taste, smell, feel or see it.||A dish of food you ordinarily like, that contained the tiniest amount of (substance), even if you could not taste, smell, feel or see it.|
|I dislike this food because of the idea of what it is or where it comes from.||As a food, the idea of what (substance) is or where it comes from.|
|I feel that this food might contain something that even in modest amounts might physically endanger my body.|
|The amount of temporary discomfort (other than taste or psychological discomfort) that eating (substance) would produce.|
|The amount of permanent damage to your body that eating (substance) would produce.|
|Your attitude towards eating (substance). [Not used in main analyses.]|
|The extent of social (e.g. embarrassment) problems produced by others knowing that you eat (substance).|
|The extent of personal-moral (e.g. guilt) problems produced by eating (substance).|
with related items in Fallon & Rozin (1983, Table I) matched.
Rozin & Fallon (1980) hypothesised three categories: 'distaste', 'disgust' and 'danger'. They also speculated on a fourth, 'inappropriate', included in Fallon & Rozin (1983). As these words all have more general meanings than specifically in the context of forms of rejection behaviour, when referring to them as food rejection prototypes, I will use single quotes about them.
'Distaste' is when we simply do not like the taste or other sensory properties of a food. The food is not objectionable in other ways: for example, we do not believe it will do us any harm. It is quite acceptable if we can avoid tasting it, so its presence in the environment or hidden in a liked food is not problematic. It is not contaminating.
'Disgust' appears to be a basic, innate human emotion (e.g. Phillips, Senior, Fahy & David, 1998; Darwin, 1872/1998). The yuk! reaction, disgust goes beyond food rejection to emotions about our and others' bodies, sex and many other things. 'Disgust' is contaminatory and ideational. The tiniest amount of a disgusting food will spoil otherwise liked foods. 'Disgust' is not something sensed in a food, but based on knowledge about it. Moreover, it is special kind of knowledge and resistant to more rational thought. 'Disgust' is a term used widely in psychology and psychoanalysis (consider Freud, 1910; 1974), if not always very consistently. A role in anorexia and bulimia nervosa and various other psychopathologies has been discussed (Phillips et al., 1998).
Confusion between "disgust" and "distaste" is easy with "disgust" often being seen as just an extreme form of "distaste". The Merriam-Webster online dictionary (Merriam-Webster, Incorporated, 1999) gives a definition for "disgust" of "marked aversion aroused by something highly distasteful". The two words are etymologically similar with "disgust" using the Latin root gust* meaning "taste". A meaning for "disgust" as more extreme than and not simply synonymous with "distaste" is only a recent usage. Thus, it is important to note that Rozin and Fallon's usage of 'disgust' is qualitatively different from 'distaste'.
'Danger' is the rational rejection response. We know that eating a food is unadvisable, so we choose not to do so, even though the food may be wholly palatable. It is a conscious awareness of negative post-ingestive consequences, specifically those for one's health.
'Inappropriate' items are those which are simply not considered as food, but which do not evoke the same strong, negative affective response that 'disgust' does.
Rozin & Fallon (1980; Fallon & Rozin, 1983) chose items as prototypes of their suggested categories and further items to test their taxonomy. Their results, including multidimensional scaling analyses (see 2.6), supported the four-fold categorisation suggested.
In Section 1, I discussed the central role of neophobia to everyday food rejection behaviours. The two Rozin and Fallon papers did not dwell on neophobic responses, but one other key study has significantly developed the ideas proposed by Rozin and Fallon in looking at this area. Pliner & Pelchat (1991) drew a distinction between the reaction to novel animal-derived foods and novel plant-derived foods. Novel foods of animal origin, they found, tend to provoke a 'disgust' reaction, while novel foods of botanical origin tend to provoke a 'distaste' or 'inappropriate' reaction.
In this section, a study based on the methodology of Rozin and Fallon is presented. I have three main goals from the study. Firstly, to attempt to replicate the results of Rozin & Fallon (1980; Fallon & Rozin, 1983) and Pliner & Pelchat (1991). Secondly, to improve on the design and analysis of the methods introduced by Rozin and Fallon. Thirdly, I wish to use the taxonomy to tell us about other food rejection behaviours.
Rozin & Fallon (1980; Fallon & Rozin, 1983) chose items as category prototypes, while including further items where the categorisation was uncertain to test their taxonomy. I have included more such items, which will be discussed in 2.8 seq. These items test not only the validity of the taxonomy, but its utility. The value of a taxonomy comes when it provides insight into our studies of food rejection behaviour. By understanding that a certain food produces a certain form of rejection, we can then apply our more general knowledge about that category of food rejection behaviour. Instead of being more interested in the taxonomy, I am now more interested in the items.
2.3 The Importance of Nausea
In considering different dimensions of aversive response, one special symptom is nausea; or, from the point of view of the food items, nauseatingness—a propensity to induce nausea. Nausea is important in learning processes involving food rejection. Nausea after eating a novel food usually leads to an aversion to that food. This learned food aversion (LFA) has been a central model in food psychology for food preference change and is also associated with a number of important clinical phenomenon. Section 3 of this thesis concerns the nature of the aversion in LFAs.
Nausea is also important beyond the LFA. 'Disgust' items are nauseating, one of the things that separates them from the other Rozin and Fallon categories.
While we are all familiar with nausea, the precise nature of nausea is rather hard to specify. Although nausea and emesis are obviously linked, they are not prerequisites of each other. Emesis without nausea is possible (as seen in bulimics). Nausea without emesis is common. The function of emesis is clear: to expel the potentially toxic stomach contents. However, what is the function of nausea? As a warning of impending emesis? As an analogue of pain, a gastric 'slap on the wrist'? To prevent further eating? Is nausea a psychological correlate of any physiological processes distinct from emesis (or the preparation for emesis)?
Uncertainty about the nature of nausea makes measurement of nausea difficult, an issue discussed in 3.4.2. Self-report has largely to be relied on.
2.4.1 The Food Rejection Indices and its Design
I shall refer to the standard set of rejection questions as a Food Rejection Index (FRI). I re-designed the FRI in a number of ways as compared to the sets of questions used by Rozin & Fallon (1980; Fallon & Rozin, 1983). An example of what I used is in 5.2.1.
Visual analogue scales:
Rozin & Fallon (1980) used a binary (True vs. Uncertain or False) response for answers, although they report similar results using a ternary scale (True, Uncertain, False). They concluded that there is a need for "scaled ratings of intensity of dislike (or "truth")" (p. 200) and used five- and six-point Likert scales in their 1983 paper. Likert scales are still quantised and are inconvenient for some statistical procedures, so I have used visual analogue scales (101mm) for this study.
The questions used are given in Table 2.2.
|1||Your overall attitude towards eating [item]||like / dislike|
|2||The taste and other sensory properties, like texture and smell, (or expected taste etc.) of [item]||like / dislike|
|3||As a food, the idea of what [item] is or where it comes from||like / dislike|
|4||A dish of food that you ordinarily like, that you were informed contained the tiniest amount of [item], even if you could not taste, smell, feel or see it||like / dislike|
|5||The presence of [item] in your stomach||like / dislike|
|6||The idea of giving [item] to a child of yours||like / dislike|
|7||How much, do you think, would the ?average man or woman in the street? like to eat [item]||like / dislike|
|8||The extent of nausea that eating [item] would produce||none / extreme|
|9||The amount of temporary physical discomfort other than nausea that eating [item] would produce. (Don't include either taste or psychological discomfort.)||none / extreme|
|10||The amount of permanent damage to your body that eating [item] would produce||none / extreme|
|11||The extent of social problems (e.g. embarrassment) produced by others knowing that you have eaten [item]||none / extreme|
|12||The extent of personal-moral problems (e.g. guilt) produced by eating [item]||none / extreme|
|13||Presuming nothing else was available, how hungry would you have to be before you would willingly eat [item]||not at all / extremely|
These questions are largely (2-5, 8-12) based on Rozin & Fallon (1980; Fallon & Rozin, 1983). Questions 1 and 13 were included as two alternate overall measures of aversiveness. Questions 6 and 7 were included to add a greater social context to food rejection behaviour.
As each FRI is lengthy, this limited the number that could be presented.
Items were chosen to cover the basic rejection forms as described in 2.2
and to investigate specific further phenomena. Details of each item are
given with their discussions, which make up the end of this section (2.7-2.11).
Abbreviations for the items, which will be used hereafter, are presented
in Table 2.3.
|FRI item||Abbreviation in text||Abbreviation in figures||Equivalents used by Rozin and Fallon|
|A food that you strongly dislike because of the taste. [Subject specified]||Strong distaste||STRG.DSL||F&R83 strong distaste|
|A food that you mildly dislike because of the taste. [Subject specified]||Mild distaste||MILD.DSL||F&R83 mild distaste|
|Horseradish paste. [Presented]||Wasabi||WASABI||F&R83 chili|
|Lemon juice. [Presented]||Lemon juice||LEMONJ|
|Tonic water concentrate. [Presented]||Quinine||QUININE||Quinine water was a suggested example for R&F80 distaste|
|A food or substance that you dislike because it is disgusting. [Subject specified]||Disgusting||DISGUST||R&F80 or F&R83 disgust|
|A fruit that you know has been sprayed with a known, potent, tasteless carcinogen.||Carcinogen||CARCINGN||F&R83 carcinogen|
|A fruit that you know contains a tasteless and odourless toxin that will produce stomach cramps for about half an hour, but no other effects.||Cramps||S.CRAMPS||F&R83 discomfort|
|A fruit that you know has been infected with bacteria that can cause food poisoning.||Food poisoning||F.POISON||R&F80 danger|
|A piece of cardboard.||Cardboard||CARDBORD|
|Pance - a salt-water animal, related to crabs and lobsters, with a soft blue shell, valued for the meat in its claws, which is usually steamed.||Pance||n/a|
|Skoikos - a mild, fresh yellow cheese, usually eaten with a spoon.||Skoikos||n/a|
|Gurnoe root - a pale green root vegetable, usually boiled and sliced for serving.||Gurnoe||GURNOE|
|Water buffalo meat.||Water buffalo||WBUFFALO|
|Clover (as in a four-leafed clover)||Clover||CLOVER|
|Steak, from a farm where a number of animals have been found to have bovine spongiform encephalitis (BSE or "mad cow disease").||BSE||BSE|
|A soft-boiled egg, taken from a farm where a number of the chickens have been found to be infected with Salmonella.||Salmonella||SALMNLLA|
|A fruit that unexpectedly tasted different from normal and you don't know why.||Differs||DIFFERS|
|A fruit.||Fruit||n/a||R&F80 meat|
The prototypes of distaste were "a food that you strongly dislike because of the taste" and "a food that you mildly dislike because of the taste", questions based on Rozin & Fallon (1980). Two levels were used to investigate different quantitative levels of distaste. Also, a number of items that are used as examples of the rejected tastes were actually presented to subjects: quinine solution for bitter, lemon juice for sour and wasabi (Japanese horseradish paste) for 'hot'. These were presented to the subject as "tonic water concentrate", "lemon juice" and "horseradish paste" respectively: see 2.7.1.
Disgust was represented by "a food or substance that you dislike because it is disgusting", although other items were predicted to elicit disgust, following Rozin & Fallon (1980). Danger was represented by "a fruit that you know has been sprayed with a known, potent, tasteless carcinogen"; "a fruit that you know contains a tasteless and odourless toxin that will produce stomach cramps for about half an hour, but no other effects" and "a fruit that you know has been infected with bacteria that can cause food poisoning": see 2.7.2. (See below for how the fruit was chosen.) The first two of these items were based on Rozin & Fallon (1980; Fallon & Rozin, 1983).
To continue Pliner & Pelchat's (1991) work and to study neophobia, a number of novel foods were used. Invented food items were again used as guaranteed novel items (see Section 1). One plant-based food ("gurnoe root - a pale green root vegetable, usually boiled and sliced for serving"), one animal-based food ("pance - a salt-water animal, related to crabs and lobsters, with a soft blue shell, valued for the meat in its claws, which is usually steamed") and one dairy product ("skoikos - a mild, fresh yellow cheese, usually eaten with a spoon") were used. Gurnoe and pance are from Section 1. It has proven difficult to invent cheeses that some people do not claim to have eaten. In piloting here and for Section 1, brynza and another item name from Pliner & Pelchat (1991), gietast, proved too believable. Thus, the name skoikos was invented (with the same description as brynza), which proved more successful in piloting.
Real animal- and plant-derived items were also used as novel foods. For each category, two items at opposite ends of acceptability were chosen (partly based on results in 188.8.131.52). For animal-based foods, these were "cockroach" and "water buffalo meat"; for plant-based foods, these were "moss" and "clover". See 2.8 for discussion of all the novel foods.
Reactions to the risks of BSE and Salmonella were investigated with the items "steak, from a farm where a number of animals have been found to have bovine spongiform encephalitis (BSE or "mad cow disease")" and "a soft-boiled egg, taken from a farm where a number of the chickens have been found to be infected with Salmonella". The non-specific nature of the descriptions was to make for a more real-world scenario (compare the 'Danger' item from Rozin & Fallon, 1980): see 2.10. "Steak" and "soft-boiled egg" were included for comparison.
"Cardboard" was an attempt to further probe Fallon & Rozin's (1983) 'inappropriate' category (2.7.4). Fallon & Rozin (1983) suggested that novel plant-derived foods would elicit an 'inappropriate' response. Other items were already considering reactions to various novel items, so cardboard was chosen in an attempt to find a familiar item to fit this category.
"A fruit that unexpectedly tasted different from normal and you don't know why" was to investigate a result found by Sullivan & Birch (1990) and is discussed in 2.11. "Cigarette (tobacco)" investigated the use of the FRI in a different setting (2.12).
The item "steak" can provide data on vegetarianism. The responses of vegetarians in this study were collected and have been reported elsewhere (Potts, 1995), but as only twelve vegetarians took part in the study, the results can only be considered as preliminary observations and are not reported herein. Potts (1995) suggested that vegetarians generally show a 'disgust' response to all meats, a view also supported by earlier work (Rozin & Fallon, 1980; Fallon & Rozin, 1983). Potts (1995) also makes the distinction between a proximal rejection mechanism—the immediate psychological processes on confronting the rejected item—and distal ones—one's reason for choosing to make a dietary choice. Proximal mechanisms may be cognitive strategies to achieve a distal goal. Such ideas can be applied to vegetarianism and also to other food rejection phenomena, as with strategies used by those with anorexia nervosa to aid starvation. In vegetarianism, the distal rejection mechanism may be certain ethical beliefs, whereas the proximal mechanism often appears to be 'disgust'. Potts (1995) also argues that this display of 'disgust' to all meats shown by many vegetarians is related to the tendency of everyone to find novel meats disgusting. One can similarly argue that religious food taboos, which are predominantly against eating certain meats, or the proximal mechanism of religious food taboos, are again part of the same class of food rejections. Descriptions of the Mosaic food laws in Leviticus are very reminiscent of the 'disgust' formulation. Of course, in a society that has widely adopted a particular religious stricture against eating a certain meat, that meat will necessarily be novel and, thus, like all novel meats, be disgusting. Further research into these matters is warranted.
FRI item details—item descriptions:
When presented to the subject, the descriptions given above, which were printed on the sheets given to the subject, description were read and sometimes further details were given as specified below. When subjects expressed doubt about the nature of the preparation of any item, they were told to imagine any particular version they wished.
"A food that you strongly dislike because of the taste." [Subject specified]
For these items, subjects were asked to pick an example and then to answer the FRI questions for that particular item. Several subjects had difficulty thinking of items. For them, it was explained that "dislike because of the taste" could also include disliked smell, texture or other such property. Subjects were encouraged to pick a food, but beverages were allowed. Subjects occasionally picked items that would seem more at home in some other category; however, these were allowed to stand (see 184.108.40.206)."A food that you mildly dislike because of the taste." [Subject specified]
Again, subjects were asked to pick an example and then to answer the FRI questions for that particular item (220.127.116.11). They were told that they could pick any item, not necessarily something usually thought of as a food."A food or substance that you dislike because it is disgusting." [Subject specified]
A "carcinogen" was explained as something that can cause cancer. For the second item, it was stressed that it would cause no other symptoms."A fruit that you know has been sprayed with a known, potent, tasteless carcinogen."
"A fruit that you know contains a tasteless and odourless toxin that will produce stomach cramps for about half an hour, but no other effects."
These two are quite conceptually difficult and were carefully explained. The steak or egg is not necessarily infective, but may be. Subjects who asked about the dangers of either BSE or Salmonella were told that I (the experimenter) was unable to discuss such matters until the end of the experiment, when they were debriefed."Steak, from a farm where a number of animals have been found to have bovine spongiform encephalitis (BSE or "mad cow disease").
"A soft-boiled egg, taken from a farm where a number of the chickens have been found to be infected with Salmonella."
Subjects completed a number of FRIs as part of another study (3.5.1). Here, those parts of the procedure relevant to the FRIs only are described. The loss of textual linearity is because the bulk of the detail of this study concerns learned food aversions (Section 3); but that study uses the FRI as a tool, requiring its introduction first. Subject recruitment, exclusion criteria and timing are dealt with in 18.104.22.168.
At the beginning of the experimental procedure, subjects were asked their name, age and Is there anything that you can?t eat or refuse to eat? (subjects being prompted with examples of vegetarianism or kosher if they did not understand the question). If subjects answered yes to this last question, details were determined as to the exact extent and reason for their rejections, e.g. if they said they were vegetarian, the precise boundaries were determined: do they eat fish etc.? Subjects were then asked their smoking status; their smoking history (apart from the first 20 subjects); whether they have any children; and their height and weight. The exclusion criteria were then checked and a consent form was signed.
The next phase of the experiment involved the attempted creation of a learned food aversion and will not be explained here. The procedure did involve familiarisation with the use of visual analogue scales. It also involved, for some subjects, a procedure to evoke nausea, although this actually proved somewhat unsuccessful (22.214.171.124.1). During this time, subjects filled out a hunger rating (101mm VAS) and state STAI (Spielberger et al., 1983). Various criteria were used to ensure that subjects had fully recovered from any symptoms when completing the FRIs and other questionnaires (see 126.96.36.199). A few subjects' data were excluded because they still had minor symptoms.
Subjects completed a number of paper-and-pencil measures: a liking rating for four fruits; short-form MCSD (Reynolds, 1982); EPQ "E" (Eysenck & Eysenck, 1975); Anglicized version of FNS (see 188.8.131.52 and 5.1.2); Absorption subscale of the Multidimensional Personality Questionnaire (Tellegen, 1978/1982; 1982; 1992); trait STAI (Spielberger et al., 1983) and DEBQ (van Strien et al., 1986). Those subjects with category-restrictive diets were asked to answer the FNS excepting that fact, i.e. as if the whole world conformed with their diet. This was done to avoid the problems discussed in Section 1.
What was described as "the main part of the study", the FRIs, was introduced. The experimenter helped the subject to complete the initial FRI, for "cat meat" (as in meat derived from a cat). This contained explanatory details about each FRI question and the experimenter also worked through each question with the subject. The total number of FRIs to be completed, that they would be in a random order, and that some of them may seem strange was all explained.
The FRIs, except for those to be tasted, were randomised in order and split into two halves (the latter half having one more FRI), except for "strong..." and "mild distaste". "Strong distaste" was always in the first half and "mild distaste" in the second half. The first 20 subjects did not receive "food poisoning" or "cardboard". Between the two halves, a second hunger rating and state STAI were given. Each FRI was introduced verbally, as detailed previously (2.4.1). The liking rating for four fruits was used to select for each subject a liked fruit. This was then inserted into some of the FRI items (designated "fruit" in the descriptions in 2.4.1).
After these, ratings of tasted stimuli were performed. The subject first rated kśrrta (see 184.108.40.206) with a sample present, but before actually tasting any. This FRI sheet having been removed, they then tasted the sample and completed another FRI. They then tasted and rated the samples of lemon juice, wasabi and quinine. Subjects were told to taste as much or as little of each sample as they wanted. This point was stressed again for the quinine sample given its highly unpalatable nature. Between each tasting, the subject had some carbonated water. Subjects were allowed to spit out any food items (a receptacle for this purpose being provided). All the foods were tasted at room temperature. Kśrrta was tasted first so that the tasting of the other items did not interfere with its ratings. The other items were presented in order of increasing unpalatability so as to minimise carry-over effects. In particular, the taste of the quinine solution lingers requiring it be tasted last. Subjects were offered some biscuits afterwards to help dispel the taste of the quinine.
Debriefing details are also in 220.127.116.11.
The lemon juice was as reconstituted concentrate (J. Sainsbury plc, London). The quinine solution was made by dissolving 2 ? 200 mg tablets of quinine sulphate in 100ml of water (4g/l). All samples were presented in 15ml pots: kśrrta samples filled about 3/4 of a pot; lemon juice and quinine solution samples filled about 1/2 a pot. About 1/2 cm3 of wasabi paste was used.
I tried to present the items in a familiar way as possible to exclude any confound with neophobia. This was probably only partly successful, but note that neophobia may not have an effect once an item has been tasted. Wasabi (Japanese horseradish) was used as a 'hot' taste as opposed to the more usual capsaicin for a variety of reasons. There are practical advantages to using wasabi, but also sinigrin (the active component of horseradish) produces a 'cleaner' heat, one that does not linger. The wasabi was described as horseradish to minimise neophobia. (A few subjects were familiar with it and recognised it as specifically wasabi.) Concentrated lemon juice was used as a sour taste. It was merely described as "lemon juice". A quinine solution was used for the bitter taste. This was described as tonic water concentrate, a not wholly inaccurate description as quinine is the main flavouring of tonic water.
2.5 Basic Results
2.5.1 Description of Subjects
The mean (standard deviation) age was 27.0 (7.7). Ages ranged from 20 to 51, but the mode was 20, with ten subjects being of that age. Age is confounded by occupation, with younger subjects practically all being undergraduates. There were 21 women to 22 men. Six subjects had children.
Vegetarians were split into five categories as shown in Table 2.4.
|... will also eat invertebrate seafood (e.g. prawns)||
|... will also eat fish||
|... will also eat poultry||
|not at all vegetarian||
For further analyses, all twelve vegetarians have been grouped together.
Twenty subjects were non-smokers and nine regular smokers. Thirteen subjects were categorised as occasional smokers (following Hajek, West & Wilson, 1995). These occasional smokers were a heterogeneous group, including 'social smokers', cigar smokers and cannabis users. One subject had given up smoking the day before and was not placed in any category. For the twenty-three subjects for whom a smoking history was taken, six had never smoked, eleven had smoked occasionally in the past and six had smoked regularly in the past.
One subject was excluded from further analyses as there was reason to believe that he had clinically disordered eating behaviour.
Table 2.5 gives descriptive statistics for the various background measures.
|state STAI (time 1)||
|state STAI (time 2)||
|hunger rating (time 1)||
|hunger rating (time 2)||
The two hunger ratings were significantly different on a paired t-test (t40 = 2.98, p = 0.005), but significantly correlated (r = 0.81, p < 0.001, one-tailed). As would be expected, the second hunger rating is higher.
The two state STAI ratings are also significantly different (t41
= 6.42, p < 0.001) and significantly correlated (r = 0.80,
< 0.001, one-tailed). The first rating was higher, as expected as it
comes just before an anxiogenic procedure (see 18.104.22.168) about which
the subject has received many warnings. Despite this, compared to published
norms (Spielberger et al., 1983), both state STAI ratings are very
low. The MCSD figure agrees with those obtained by Reynolds (1982).
p = 0.015
p = 0.001
|state STAI (time 1)||
p = 0.001
|state STAI (time 2)||
p = 0.034
|Table 2.6: Correlations with FNS.|
The relationship of FNS to other variables in previous work is muddled, as discussed in 1.7. In this sample (Table 2.6), we see significant positive correlations with all three STAI measures, a result found by Pliner & Hobden (1992) but not otherwise replicated (Section 1), and an unexpected significant negative correlation with MCSD that could be mediated by trait STAI.
In summary, the subject population broadly conformed to expected norms on the various psychometric tests given. The subject population is unusual in having a high proportion (about a quarter) of vegetarians. This may reflect the increasing frequency of vegetarianism amongst student populations in the country as reported elsewhere (Sanders, 1994) or a recruitment bias. While obviously the practicalities of subject recruitment mean that the group here are far from being random, it is reassuring that they respond as expected on the background measures.
2.5.2 Results: FRIs
VAS data frequently suffers from floor and ceiling effects and both are common in the current data. Subjects seem to interpret distance near the ends of the scales as corresponding to a greater change than distance near the centre of the scale. There is little published on this problem (although see McCormack, Horne & Sheather, 1988, for review), but the arcsine root transformation was chosen on pragmatic grounds as it tends to give a more normal result. Thus, VAS score data was transformed using the arcsine root transformation,
f(x) = sin-1(x-˝)
having first been rescaled to be in the (0, 1) interval.
In 5.2.3, for each FRI item, the profile of results will be shown both in terms of means and standard deviations on the transformed scale and medians and interquartile ranges on the untransformed scale. For the purposes of analyses, the transformed values are always used. Larger values always mean more rejected.
Data were missing for a large variety of reasons. In some instances, no data were available; in others, data was ignored.
Two items, "cardboard" and "food poisoning", were only introduced halfway through the study and so are coded as missing for the subjects in the first half of the study.
Vegetarian subjects' rejection responses to certain items involving animal products will be confounded between their vegetarianism and the response shown also by non-vegetarians (be it neophobia, fear of infection or whatever). Thus, for certain items, vegetarians' responses were excluded. The items were: "water buffalo", "Salmonella" (as involving egg), "BSE", "pance" and "skoikos". Details about the extent of individuals' vegetarianism was collected, so it would have been possible to include, say, ovo-lacto-vegetarians for "Salmonella" and "skoikos". However, I felt that whatever the level of vegetarianism reported, it could confound the response to any item involving animal products. The one exception to this rule is that vegetarian subjects' responses to "cockroach" were included as I decided, on the basis of subjects' behaviour in the study, that the strong reaction to the idea of eating a cockroach would override any opposition from a vegetarian perspective.
For the fictional items designed to investigate neophobia ("gurnoe", "pance" and "skoikos"), data from subjects who reported having tasted these items were excluded. Clearly, these reports for the three fictional items cannot have been accurate, but if the subject had identified the item as something else erroneously, that will still have the effect of invalidating their response.
Few of the food items in this study are necessarily aversive to every subject. Some items for some people may be perfectly acceptable. As I am interested in rejection phenomena, I exclude FRI responses for accepted items, as did Rozin & Fallon (1980; Fallon & Rozin, 1983). Acceptance was defined by an arbitrary cutoff of >20mm on the FRI question 1. The summary tables and the MDS analyses (2.6 seq.) exclude these data, although some other analyses later in the section will include data for all the subjects.
There was no question 13 for the item "cigarette".
Occasional methodological and procedural errors, chiefly subjects accidentally missing questions or items, accounted for further missing values.
2.6 Multidimensional Scaling Results
I wish to attempt to confirm and refine the results of Rozin & Fallon (1980; Fallon & Rozin, 1983) and validate a general taxonomy of food rejection behaviour. Further, I wish to apply this taxonomy to more ambiguous items. Following Rozin and Fallon, I primarily use multidimensional scaling (MDS) analyses. MDS is a technique to produce a graphical representation of the relationships between items.
In sections 2.7 seq., I will be making predictions about where specific items may lie in the taxonomy and thus desire to perform something akin to hypothesis testing. However, the statistical methods used in Rozin & Fallon (1980; Fallon & Rozin, 1983) are more exploratory and ill-suited to such an approach. To solve this problem, I will use a novel technique in applying bootstrapping to MDS output to generate pseudo-confidence regions (Potts, 1996). In order to develop these bootstrapping techniques, Potts (1996) used part of this data set and also considered some of the initial data processing described below. For this doctoral work, the bootstrapping techniques were applied to the entire data set.
2.6.1 Preliminary Data Processing
In order to carry out multidimensional scaling, it is necessary to express the data as pairwise similarity (or dissimilarity) indices between the items. That is, we need to generate a number representing the similarity (or dissimilarity) of each food item to each other food item. Specifically, MDS algorithms requires dissimilarities. The less alike two items are, the greater the dissimilarity. A dissimilarity can be considered as a measure of distance between two items.
Rozin & Fallon (1980) compared for pairs of items the proportions of True responses over their population per question using a chi-squared statistic as a measure of dissimilarity. Fallon & Rozin (1983) used the correlation between items as a similarity measure. That is, the average scores across the questions for an item gives a vector of numbers; the numbers in two vectors can be correlated with each other. That some of their questions were answered on five-point scales (0-4) and some on six-point scales (0-5) effectively gives slightly higher weight to those questions answered on the six-point scale, although their paper does not comment on whether this was a deliberate or unforseen consequence of their design. Here, all the questions were answered on the same VAS scale.
Nunnally (1978) discusses the analysis of profiles; figures 1 in Rozin & Fallon (1980) and Fallon & Rozin (1983) shows how items? scores can be represented as a profile. A graph is drawn of score (Fallon & Rozin, 1983) or proportion answering True (in Rozin & Fallon, 1980) against question (in an arbitrary but fixed order). The points are connected to give a line, or profile. A profile contains three types of information: the level, dispersion, and shape. The level here is the average height of the line and represents the overall magnitude of the rejection evoked, which does not interest us. It is unclear whether the dispersion (the variance around the line) carries any information, but Rozin and Fallon's interest clearly lies with what Nunnally calls the shape. Different shapes correspond to different aspects of the food rejection behaviour being more or less important, the basis for our taxonomy. Thus, the (dis)similarity index used should measure differences in profile.
Zegers and ten Berge (1985; Zegers, 1986) and Fagot & Mazo (1989) showed the correlation coefficient to be an association index that respects only the profile shape and that it is a member of a family of association indices with desirable statistical properties. Fallon & Rozin's (1983) choice of the correlation coefficient as the similarity measure can be vindicated.
Rozin & Fallon (1980; Fallon & Rozin, 1983) do not give details of their dissimilarity calculations. There is a variety of ways in which the correlation coefficient, a measure of similarity, can be used to derive a measure of dissimilarity, as required for MDS. Here, the correlation coefficient, r, was converted into a dissimilarity measure, delta, following Mardia, Kent & Bibby (1979, p. 402):
delta = (1 - r)-˝
Thus, we can calculate all the pairwise dissimilarities between the food items. The correlation across the thirteen questions between the two food items is calculated and then converted to a dissimilarity. This procedure is repeated for every pairwise comparison to generate a full dissimilarity matrix.
Two subjects gave the same VAS rating on every FRI question for certain items. This always occurred with maximum scores being given across all the items, occurring for one subject for "disgust" and for another for "Salmonella", "BSE" and "food poisoning". With a constant FRI profile, a correlation coefficient is undefined and so the associated parts of the dissimilarity matrix for these subjects had to be coded as missing. For the latter subject, however, with three items coded wholly at ceiling, the correlation between these three items was coded as being 0.99999999. A unity value is not possible as the software used does not allow coincident points in the MDS analysis; a correlation of 1 gives a dissimilarity of 0, implying that the two items lie at the same position..
If individual FRI questions are missing (as always with question 13 for "cigarette" by design), then a correlation coefficient can still be calculated over the remaining questions and the dissimilarity matrix is unaffected. If an item is missing, the corresponding row and column will be missing in the dissimilarity matrix.
Rozin & Fallon (1980; Fallon & Rozin, 1983) generated item profiles by averaging over subjects. They then calculated a dissimilarity matrix and performed MDS. However, it would be more informative if we could take account of variation between subjects as well.
Subjects can be expected to interpret the FRI questions and the scales differently. The visual analogue scale was originally intended only for within-subject comparisons (McCormack, Horne & Sheather, 1988). Thus, it makes sense to calculate dissimilarities for each subject and then to compare subjects' dissimilarity matrices. We can then average at the level of the dissimilarity matrices and apply MDS to each subject's individual dissimilarity matrix or to this averaged dissimilarity matrix.
While MDS can be applied to each subject's data individually, separate MDS output for each subject are of limited value. We would like to have some sort of averaged MDS output of all the subjects' data and some sort of representation of intersubject variation. This latter point is approached using bootstrapping below (2.6.3), while one way of achieving the former is to apply MDS to the averaged dissimilarity matrix.
2.6.2 Overall Solution
I first consider the results from the averaged dissimilarity matrix. A variety of multivariate techniques were applied to the data, including MDS as used by Rozin & Fallon (1980; Fallon & Rozin, 1983). Different MDS techniques exist and, as the resulting MDS solution is only a representation of relationships and not a real map, non-metric MDS is an appropriate approach. I follow common practice in using Kruskal's algorithm. A two-dimensional solution for non-metric MDS is shown in figure 2.1. This is the core analysis on which the bootstrapping work (2.6.3) will be based.
Apart from "cigarette", which is a clear outlier, there are two groups apparent. "BSE", "carcinogen", "Salmonella", "food poisoning" and "cramps" form a 'danger' group. The other group is more complicated: a core ("gurnoe", "clover", "differs", "lemon juice", "quinine", "wasabi", "moss", "cardboard", "mild dislike", "water buffalo") appears to be a 'distaste'/'inappropriate' group. "Strong dislike" lies at the edge of the group in one direction, while "cockroach" and "disgust" form a small 'disgust' group at another edge. Section 2.7 seq. will consider these results in further detail.
We have no a priori reason to expect the MDS solution to be of a particular dimensionality. That is, we do not know whether the pattern of relationships between the items can be represented by the position of points on a flat plane or whether they need to be represented in three or even more dimensions. MDS gives the best fit to the data for a given number of dimensions but does not automatically point to a particular number of dimensions as correct. Clearly, two-dimensional solutions are easiest to display, but that does not mean they are necessarily supported by the data. Thus, three- and higher dimensional non-metric MDS results were also investigated.
Figure 2.1: Overall multidimensional scaling analysis output (two dimensional). [Figures not yet available on the web.]
The outlying "cigarette" appears to dominate the third dimension in a three-dimensional result. Items lie roughly in a plane in the same pattern as the two-dimensional solution, but with cigarette displaced along the third dimension. Higher dimensional solutions are very difficult to interpret by their nature.
The goodness of fit of an MDS output to the data, measured by a statistic called STRESS for non-metric MDS, can be plotted against dimensionality in a scree plot. This hopefully displays an 'elbow' in the data showing when additional dimensions are needed to explain true structure and when they are not. As often happens, however, no clear 'elbow' is apparent here: there is some suggestion that three dimensions is optimal, although this may be a result strongly dependent on the "cigarette" item. Thus, two dimensions seems satisfactory.
There are many alternative procedures in MDS. Both metric MDS and Sammon mapping, an alternate non-metric form of multidimensional scaling, gave similar results to Kruskal's non-metric MDS. A variety of cluster analytic results were also investigated, which again largely agreed with the non-metric MDS result. Cluster analysis also works on a dissimilarity matrix, but displays a hierarchical grouping of items as opposed to a map of results. MDS was preferred as the primary analytic technique because it can handle intermediate items better. In a map, an item that evoked both 'danger' and 'disgust' can lie between those two groups. Cluster analysis forces such an item to be associated with one or other group.
This consistency of results across techniques is reassuring: it suggests that the results are inherent in the data and not artefacts of the statistical techniques applied.
Let us again consider individual subjects' dissimilarity matrices. Individual MDS analyses on these—to considerably varying degrees—support the results from the averaged dissimilarity matrix, showing similar groupings. Each of the subjects' dissimilarity matrices can be thought of as coming from a population of dissimilarity matrices. The averaged dissimilarity matrix represents a population mean and we desire some sort of standard error-like statistic of that mean to be graphically represented on the MDS output map derived from it. There is no analytical solution to this problem, but we can turn to the empirical method of bootstrapping instead.
If we know the population distribution of some variable, but analytic techniques are not available or practical for deriving the sampling properties of some statistic, we can use simulation to estimate them. Consider a simple case that can also be solved analytically. Say we have data from a normal population and we wish to know the sampling properties for the median so that we can quote confidence intervals for it. We could simulate those properties by repeatedly drawing a sample from the population distribution and calculating the median value. Doing this a few thousand times will show us the sampling properties of the median and allow us to give a confidence interval.
Unfortunately, it is usually the case that we do not know the true population distribution, so we cannot simulate from it. However, we can estimate the population distribution. The maximum likelihood estimator—for the purposes of this description, what one could consider the best estimator—of the population distribution is the actual distribution of the data we have, known as the empirical distribution function. We can then simulate as before, but from this empirical distribution function. This procedure is known as bootstrapping.
Individual subjects' dissimilarity matrices represent a sample from an unknown population of dissimilarity matrices and that sample, therefore, forms an empirical (multivariate) distribution function. We can bootstrap from this and re-calculate the MDS output for each bootstrapped averaged dissimilarity matrix. This produces an MDS output for each bootstrap. However, whereas this would suffice for a scalar statistic (one represented by a single number), this does not work for our purposes.
MDS output is invariant under rotation, reflection and translation. The resulting map can be rotated, reflected or shifted along and still fit the data equally well. There is no correct orientation to the map. Each bootstrapped MDS output thus has an arbitrary rotation, reflection and translation. These arbitrary components prevent us from combining the outputs.
In order to compare between resulting plots, one solution is to hold all but one or a few items constant in this procedure. Instead of bootstrapping the complete MDS solution, we hold most of that solution constant and just re-fit an individual item (or a small number of items) for each bootstrap, conditioning on the rest of the map being constant (Potts, 1996).
Repeating this a suitably large number of times generates a point cloud. This point cloud (examples given as even numbered figures throughout the rest of this Section) represents the intersubject variability for each item, under certain constraints. By comparing different point clouds for different items, we can see that there is greater uncertainty in the position of some items (e.g. "water buffalo"—figure 2.24, "cardboard"—figure 2.20) than others (e.g. "quinine"—figure 2.6).
With a univariate statistic, we will often quote an alpha% confidence interval. It is desirable to do the same here, but in two dimensions (and the issue generalises to higher dimensions): how do we construct a confidence region? We could construct a circle centred on the original estimate that contains alpha% of the points in the point cloud, but to do so is to assume a certain shape for the confidence region. In fact, we have no a priori reason to expect the confidence regions to be circular, convex or even contiguous! How do we convert the information in the point cloud into a confidence region without pre-defining a shape to the region?
The bootstrapped point cloud can be considered to represent samples from a multivariate distribution of all possible bootstraps. With sufficient sample points, we can estimate that distribution using two-dimensional kernel density estimation. (The process of kernel density estimation is like plotting the data in a histogram and then smoothing the histogram boxes a continuous distribution.) Once we know (or have estimated) such a distribution, we can draw isocontour lines joining points of equal probability. We can then select an isocontour that contains alpha% of the volume of probability under the distribution giving an alpha% confidence region.
While this is theoretically possible, it is computationally prohibitive. As an approximation of this process, we can instead choose the alpha percentile bootstrapped point by height on the distribution and draw an isocontour at that point. Even this is very computationally intensive, so the heights on the kernel density estimate of each bootstrapped point are estimated instead of being calculated directly. The result of this approximation is to make the regions very slightly larger (Potts, 1996).
While, say, 1000 bootstraps is often more than sufficient for bootstrapping a univariate statistic, for our two-dimensional statistic (position of a point on a two-dimensional map), a larger number is needed because of the inherent sparseness of higher dimensions. (Given a k-dimensional solution, the number of points needed rises with k as the exponent.) The isocontour depends on the kernel density estimate in an area where the points are comparatively sparse. To achieve a reasonable density of points near the isocontour requires a large number of simulations.
Simulation methods are computer intensive. Those here where each bootstrap procedure is itself computationally expensive, as MDS is, are particularly slow to perform. The whole procedure was implemented in S-Plus using standard routines and programmes written in S-Plus and C++ specifically for this project (Potts, 1996). For the results here, 5000 simulations were calculated, which study suggested was a minimum sufficient number. On a 166MHz Pentium PC with 32Mb of RAM, the bootstrapping procedure took several hours per item. An alternative software and/or programming implementation may improve efficiency.
As there is much missing data in the subjects? individual dissimilarity matrices in the form of missing rows/columns—corresponding to items not rejected by that subject or otherwise missing—the bootstrap replicates may have a very large amount of missing data for certain rows/columns. Because bootstrapping samples from the actual data (strictly, from the empirical distribution function), some bootstrap samples will, by chance, include considerable missing data. This would be undesirable, inflating the confidence regions, and so these were excluded. However, by doing this, we bias towards the data from subjects with more complete dissimilarity matrices. To avoid this bias, bootstrap replicates with symmetrically particularly low numbers of missing values were also filtered out (Potts, 1996). This ensures that the bootstrap samples are roughly similar to the original data in terms of missing data.
The calculated confidence regions can aid in determining what structure in the averaged result reflects real differences and what is no more than random variation. For each item considered, the bootstrapped point cloud and the confidence region are shown as separate plots in figures 2.2 seq.
A theoretical basis to much of the statistical procedures described above is absent; this is virgin territory at the meeting of computer-intensive and multivariate statistics. An empirical approach has been taken, but while the method appears to produce sensible results, the output should strictly be considered only as pseudo-confidence regions.
We have imposed a very strong conditioning in the bootstrap procedure by fixing all items but one at a time. This produces noticeable artefacts in the confidence regions. For example, the region for "water buffalo" (figure 2.25) shies away from the items "disgust", "cockroach", "clover" and "differs". This is correct in the sense that the data consistently show that "water buffalo" is not very close to any of the other items. However, the result is unsatisfactory for we have no special confidence in the exact placement of the other items. In reality, there is uncertainty in the position of all the items and moving any one item affects all the others. In practice, we have to consider each item separately. Care in interpretation is needed.
Note also that power considerations are as relevant here as ever in statistics. Clearly, no power calculations could be done, but 43 subjects is not a huge number. Subtle, but real, differences may go unnoticed.
Potts (1996) calculated confidence regions for two items to develop, test and demonstrate the techniques developed. Here, I have applied the bootstrapping procedure to the whole data set.
2.7 Prototypes for the Four Rejection Categories
Following Rozin & Fallon (1980; Fallon & Rozin, 1983), subjects selected items themselves for "mild..." and "strong distaste" to produce two items prototypical for "distaste". Details of the items picked are given in 22.214.171.124. The use of a dissimilarity measure based on a correlation should remove overall magnitude effects in subjects' FRI responses, thus "mild..." and "strong distaste" should be close to each other in the solution. However, subjects tended to pick items towards which they had extreme feelings for "strong distaste". This may make it a bad prototype, both because an extreme response may involve more than the usual response and because the analysis is in danger of ceiling effects on the VASs. More generally, with subjects selecting items themselves, it is possible for them to misinterpret the instructions and the items they select may not fall into our theoretic 'distaste' group. All subjects' choices were allowed, although some seemed dubious (see 126.96.36.199). This may lead to larger confidence intervals for these items.
As alternative prototypes for 'distaste', subjects actually tasted three items. These were chosen for three basic tastes which appear to be innately unpalatable: bitter ("quinine"), sour ("lemon juice") and 'hot' ("wasabi").
Although perhaps the most obvious of Rozin and Fallon's categories, 'distaste' is poorly understood. Novel foods are often considered distasteful, even though that is necessarily supposition. Further discussion of reactions to novel foods follows in 2.8. Anecdotally, most individuals' 'distastes' of which they actually have experience appear to be idiosyncratic. However, there is yet ubiquity among 'distastes': bitter and other tastes appear to be innately unpalatable. The relationship between a common dislike of bitterness and individuals' personal 'distastes' is unclear, although they are usually considered to be the same. One school of thought argues that bitterness underlies many 'distastes' and interpersonal variability can be explained in terms of differing sensitivities, i.e. to different perceived bitterness (e.g. Bartoshuk & Duffy, 1994). Others argue that bitterness sensitivity only accounts for a small proportion of 'distastes'.
Note that there are different possible sensory modalities involved in these different examples. Bitter is sensed by specialised taste buds, while many 'distastes' may—despite the word's etymology—be reactions against flavours sensed and identified by olfaction. Other senses involved in "taste" in the wider meaning may also be involved. Our conscious perception of "taste" conflates the actions of the tongue, the nose and, to a lesser extent, the trigeminal sense of irritation and texture. Everyday language gets confusing: taste can either mean the full spectrum of gustatory experiences; or just the action of the taste buds. In this study, instructions to subjects have conflated all sensory modalities, thus the FRI question "The taste and other sensory properties, like texture and smell, (or expected taste etc.) of item". When picking items for "mild..." and "strong distaste", it was verbally explained to subjects that they could also pick items that they disliked because of the smell or texture. Are these differences between the 'distaste' of a bitter substance—quinine here—and personal 'distastes' important when considering the food rejection behaviour elicited? While probable, it is not obvious that a bitter tasting substance will be close to the "mild..." and "strong distaste" items.
Sour is another taste specifically sensed by a class of taste buds and it has been theorised to be innately unpalatable, although it has been little studied compared to bitterness. Here, it is represented by "lemon juice". I expect this item to be close to "quinine".
Certain oral irritants are sensed by receptors linked to the trigeminal nerve; most notably, this is how we sense 'hot' (as in spicy) foods. 'Hot-ness' seems to have a slightly more distinct identity than other senses involved in tasting and is also innately unpleasant. Again, while it is probable that the 'hot' item used here, "wasabi", will group with the other 'distaste' items, it may not.
There is some risk of a confound for the three tasted items here as all may be somewhat novel, provoking neophobic responses. The items were selected to be as familiar as possible and their novelty was downplayed by how they were described. The quinine solution was introduced as "concentrated tonic water", a small falsehood as quinine is the primary ingredient of tonic. Lemon juice is very familiar anyway. The wasabi was introduced as horseradish paste, which it is albeit of a form less familiar to the British palate.
Thus, we have five items that should form a 'distaste' group: "mild distaste", "strong distaste", "quinine", "lemon juice" and "wasabi". There may be a difference between subject-identified items ("mild..." and "strong distaste") and items tasted which were chosen as innately unpalatable ("quinine", "lemon juice" and "wasabi"). There may also be differences depending on the underlying sensory modality: bitter taste for "quinine", sour taste for "lemon juice", trigeminal irritation for "wasabi" and unknown, but possibly flavour (smell) for "mild..." and "strong distaste".
The results broadly match prediction: see figures 2.2-2.11. The three experienced items are very close, with "quinine" and "wasabi" being the closest two items on the whole MDS solution. There is no evidence that sensory modality differences are relevant here.
"Strong distaste" is remote from the other items, as was feared because of the extreme nature of items picked. This suggests a lesson for future studies. "Mild distaste" lies between "strong distaste" and the experienced items. While "strong distaste" is somewhat remote from the other 'distaste' items, it remains off in a different direction to 'disgust' or 'danger' items, suggesting there is no problem with the basic taxonomy.
As with "mild..." and "strong distaste", I followed Rozin & Fallon (1980; Fallon & Rozin, 1983) and allowed subjects to choose their own items. Details of items picked are given in 188.8.131.52. Again, as with "strong distaste", subjects often selected extreme or surprising items (e.g. paint, alcohol). This may make "disgust" a less than ideal prototype for the "disgust" group.
In the results (figures 2.12 and 2.13), "disgust" shows a clear separation from the "distaste" and "danger" groups, supporting the basic taxonomy.
Figure 2.2: Point cloud for "mild distaste". [Figures not yet available
on the web.]
Figure 2.3: Confidence region for "mild distaste". [Figures not yet available on the web.]
Figure 2.4: Point cloud for "strong distaste". [Figures not yet available on the web.]
Figure 2.5: Confidence region for "strong distaste". [Figures not yet available on the web.]
Figure 2.6: Point cloud for "quinine". [Figures not yet available on the web.]
Figure 2.7: Confidence region for "quinine". [Figures not yet available on the web.]
Figure 2.8: Point cloud for "lemon juice". [Figures not yet available on the web.]
Figure 2.9: Confidence region for "lemon juice". [Figures not yet available on the web.]
Figure 2.10: Point cloud for "wasabi". [Figures not yet available on the web.]
Figure 2.11: Confidence region for "wasabi". [Figures not yet available on the web.]
Figure 2.12: Point cloud for "disgust". [Figures not yet available on the web.]
Figure 2.13: Confidence region for "disgust". [Figures not yet available on the web.]
The items "carcinogen" and "cramps" were taken from Rozin & Fallon (1980; Fallon & Rozin, 1983) as prototypes for the 'danger' group. They are intended to evoke strong and mild forms of the 'danger' reaction respectively. Both items are in the form of a chemical poisoning threat, so a third item, "food poisoning", was added. This bacteriological threat is perhaps closer to the real threats people face in everyday life. It is a closer parallel to the recent food safety scares from Escherichia coli O157, Salmonella (also considered directly, see 2.9), Listeria, cholera etc.
The real dangers in a biological threat like food poisoning are different from those with the "carcinogen" and "cramps" items. If the threat is different, then perhaps the reaction is too. Davison (1989) has previously suggested that the threat of food poisoning may provoke specific responses.
Note that the largest ever E. coli O157 scare, involving 20 deaths in Scotland, and some other notable outbreaks postdated this study (Mead & Griffin, 1998). Such a food poisoning event, which had extensive media coverage, may have changed attitudes in the population.
In figures 2.14-2.19, we see that all three items form a clear 'danger' cluster, with "carcinogen", perhaps because of the particular fears associated with cancer, furthest out.
Fallon & Rozin (1983) considered a fourth category of 'inappropriate'. Although this may also be relevant to certain neophobic responses (2.8), to consider it directly, I used the item 'cardboard' as something familiar but fitting the definition of the category. 'Cardboard' lies centrally in the final solution, although clearly closest to the 'distaste' items. The confidence region is large and does not show clear differentiation from the 'distaste' group (figures 2.20 and 2.21). The large confidence region for 'cardboard' suggests that there may be wide interpersonal variation in people's reactions to it as a potential food stuff. As such, it makes for a poor prototype for the category. Thus, there is no evidence here to support the quadripartite division of Fallon & Rozin (1983) over the tripartite one shown in the results of Rozin & Fallon (1980). However, with low power and perhaps a poor prototype, this cannot be considered strong negative evidence.
Figure 2.14: Point cloud for "carcinogen". [Figures not yet available
on the web.]
Figure 2.15: Confidence region for "carcinogen". [Figures not yet available on the web.]
Figure 2.16: Point cloud for "cramps". [Figures not yet available on the web.]
Figure 2.17: Confidence region for "cramps". [Figures not yet available on the web.]
Figure 2.18: Point cloud for "food poisoning". [Figures not yet available on the web.]
Figure 2.19: Confidence region for "food poisoning". [Figures not yet available on the web.]
Figure 2.20: Point cloud for "cardboard". [Figures not yet available on the web.]
Figure 2.21: Confidence region for "cardboard". [Figures not yet available on the web.]
The basic tripartite taxonomy suggested by Rozin & Fallon (1980) holds in these results, with the three groups lying approximately equidistant from each other. The position of "cardboard" does not support the existence of an 'inappropriate' category.
The more peripheral positions of "strong distaste" and "carcinogen" suggest that items that provoke extreme reactions make poor group prototypes. It is a common tendency to pick extremes for experimental purposes to exaggerate the effects under study. However, extremes are, by their nature, not typical of most of a category.
In Section 1, I concluded that food neophobia is a major determinant of food rejection behaviour. It is perhaps the major determinant of food choice behaviour in general, but its importance as such is not immediately obvious on a personal level. We consider food choice from among a set of food items without considering that neophobia has dictated what is in that set. Food neophobia cannot explain whether I choose to have corned beef or mushroom pâté in my sandwiches tomorrow, but it does explain why I never have beetles or most of the myriad of possible foods that exist.
The expression of neophobia through culture also masks its importance as a determinant of food choice. Particularly since the industrialisation of the food industry, if a potential food is not consumed within a culture, the culture does not recognise the substance concerned as a food and it is unavailable for anyone to obtain. Neophobia as an ultimate cause for the rejection of, say, sheep's eyes, becomes replaced by a proximal cause: you cannot buy them in the local supermarket.
It is important to think of neophobia as opposed to familiarity being liked. Familiarity is not a sufficient factor for liking a food: even after repeated presentations, a strong quinine solution remains unpalatable. Once a food is familiar, it can be judged on its own effects (physiological or imagined); neophobia means that, in the absence of any information, we err on the side of caution and assume the worst.
How do neophobia responses fit into the taxonomy? Pliner & Pelchat's 1991 paper is the most important previous study, with results showing that novel animal- and plant-based foods evoke very different responses, animal-based foods evoking 'disgust' and plant-based foods, 'distaste'.
Seven items were chosen to study the range of neophobic responses. Three imaginary foods, based on those used in 1.5.1, were chosen, representing a plant-based item (gurnoe), animal-based item (pance) and dairy product (skoikos). However, pance and skoikos proved insufficiently aversive (perhaps as they were insufficiently novel) and too few people demonstrated rejection of these items on the FRIs. These items are thus ignored, leaving just gurnoe.
Four items were chosen which are real items of which people would be aware and which are potentially edible, but which are not usually eaten. There were two plant-based items (clover, moss) and two animal-based items (cockroach, water buffalo). Within each pair, one item was chosen to be very unpleasant (moss, cockroach) and the other to be nearly acceptable (clover, water buffalo), this choice being informed by data collected in the first list heuristic study (1.5.1).
Pliner & Pelchat (1991) would predict that "cockroach" and "water buffalo" fall with "disgust", while "moss" and "clover" fall with the various 'distaste' items. Fallon & Rozin (1983) would predict that "moss" and "clover" would evoke an 'inappropriate' response and lie with "cardboard". "Gurnoe" should behave like "clover".
2.8.1 MDS Results
Pliner & Pelchat's (1991) result was broadly confirmed, with "cockroach" and "water buffalo" lying near "disgust"; and the other three items associated with the large 'distaste' cluster (figures 2.22-2.31). Results also suggest that further resolution of the structure of the 'distaste' cluster is meaningful.
"Clover" and "gurnoe" are near each other and near the "lemon juice"/"quinine"/"wasabi" cluster; all the respective confidence regions overlap considerably (figures 2.26-2.29). However, they lie on the other side of this cluster to "mild..." and "strong distaste" and confidence regions suggest that this difference is significant.
Figure 2.22: Point cloud for "cockroach?. [Figures not yet available
on the web.]
Figure 2.23: Confidence region for "cockroach". [Figures not yet available on the web.]
Figure 2.24: Point cloud for "water buffalo". [Figures not yet available on the web.]
Figure 2.25: Confidence region for "water buffalo". [Figures not yet available on the web.]
Figure 2.26: Point cloud for "clover". [Figures not yet available on the web.]
Figure 2.27: Confidence region for "clover". [Figures not yet available on the web.]
Figure 2.28: Point cloud for "gurnoe". [Figures not yet available on the web.]
Figure 2.29: Confidence region for "gurnoe". [Figures not yet available on the web.]
Figure 2.30: Point cloud for "moss". [Figures not yet available on the web.]
Figure 2.31: Confidence region for "moss".
"Moss" lies very close to "cardboard", but with a much smaller confidence region (figure 2.31). However, it still has overlapping confidence regions with "clover", "quinine" and "wasabi". While there is considerable overlap of confidence regions among all these 'distaste' items—"gurnoe", "clover", "moss", "cardboard", "lemon juice", "quinine", "wasabi", "mild..." and "strong distaste"—there does appear to be meaningful structure within this group. The results are equivocal, but there appears to be a bottom left-to-top right dimension, going from "mild..." and "strong distaste" through the three tasted items to the neophobia items and "cardboard". It is unclear to what this dimension corresponds. In an orthogonal direction, there again appears to be meaning, going from "gurnoe" to "clover" to "moss" (and "cardboard") and beyond even through "water buffalo" to "cockroach". This would seem to suggest that there is some evidence for an 'inappropriate' category centred on "moss" and perhaps a novelty/'disgust' dimension.
"Cockroach" lies close to "disgust" and their confidence regions greatly overlap (figure 2.23). The result here is clear and unsurprising. "Water buffalo" lies between the 'distaste' cluster and "cockroach" and "disgust" and sports a very large confidence region (figure 2.25). Much of the confidence region for "water buffalo" lies in the area of the 'disgust' cluster, while much also overlaps with the central area and "moss" and "cardboard", what we might see as the 'inappropriate' cluster. This suggests that attitudes to the "water buffalo" item are very mixed in the population. Clearly for many, "water buffalo" is evoking the same 'disgust' response as "cockroach", but this is not always the case. It is possible that some subjects did not know what a water buffalo is, with their varied misunderstandings adding to the variance.
2.8.2 Suggested Ethology
The utility of neophobia as a survival strategy has long been commented upon in the literature (Rozin & Vollmecke, 1986). Ingestion exposes the organism to many threats and it is wise to avoid the unknown. However, why should animal and plants provoke different types of aversive reaction?
From the point of view of evolutionary psychology, a logical answer is that animal and plants involve different types of threats from ingestion whereby a 'distaste' or 'inappropriate' response counters the threat from novel plant-based foods and 'disgust' counters the threat from novel animal-based foods.
Plants, unable to flee their predators, have to rely on other, static forms of defence. (I use "predator" here to mean any organism that eats other organisms, a heterotroph, thus including herbivores.) Three options exist for plants: decreasing nutritional worth; physical defence (notably on the cellular level with the cellulose-rich cell wall) and chemical defence. Chemical defence is through a number of toxic secondary compounds. Strategies to counter these by the herbivore exist in the form of physiological pathways or symbiotic gut flora to cope with such compounds. However, most organisms also use a behavioural strategy of avoidance.
For both the plant defending itself and the herbivore avoiding these defences, the strategy requires the herbivore to detect the toxins. It is in the prey's interests to advertise its toxicity clearly and rapidly and in the predator's interest to sense this. The plant usually does not benefit from killing its predator—that could merely lead to the predator being replaced by another naďve predator; better a knowledgable one that will not eat you—so the threat should be detectable before it does (too much) harm. Thus, detecting chemical defences is usually fairly easy.
Plants generally rely on taste to warn off the predator. Both bitter and sour taste systems have been theorised specifically to exist to detect plant toxins.
Thus, it makes sense to be wary of novel plants, but the danger is usually obvious and is not too great. Toxic plant material is easy to detect and relatively safe, in that you can usually afford to consume a small amount.
Many animal species—but few amniotes—also use chemical defences and similar arguments apply. However, while a plant can afford the loss of a leaf, an animal cannot usually afford the loss of a limb, thus a more extreme defence is usually involved. Chemical defences aside, there are very considerable dangers to ingesting animal-based foods for other reasons. Most notably there is the danger of pathogens. Plants too, of course, have their own pathogens, but only a handful are known to be transmissible to any tetrapods. Numerous pathogens can be caught from ingesting other animals: from prions (e.g. BSE), through viruses (e.g. hepatitis A) and bacteria (e.g. Salmonella) to various eukaryotes, notably worms (e.g. pork tapeworm).
Pathogens will have evolved to be largely undetectable. Very small amounts of meat can be infectious and infection can be transferred via other surfaces. When we speak of contamination in terms of a rejected food item's ability to make other things aversive, the word comes from the behaviour of pathogens.
We are considering specifically neophobia. While novel meats are no more likely to be infectious, there are more likely to be novel pathogens, i.e. ones to which the organism has no antibodies.
Another danger from meat products, which I have not found considered in the literature, is immunological. The cost of having a sophisticated immune system, poised to deal swiftly with infection (or cancer), is that it is also very delicate; thus the numerous auto-immune diseases that afflict us. It is known that meat products—containing proteins similar to our own—may be responsible for stimulating such auto-immune diseases. Meat products that we—and therefore our immune systems—are familiar with pose no threat, but novel meat products do. An immunological danger has no evolutionary reason to be obvious (like plant toxins) or hidden (like pathogens).
Both biological and immunological dangers pose risks in minute quantities that chemical poisons do not, thus explaining the need for, say, the contamination phenomenon in 'disgust'.
2.8.3 Relationship with Anxiety
Section 1 discussed at length the possible relationships between neophobia and anxiety, albeit with less than conclusive results. FRI responses to the novel items here constitute an alternate measure of neophobia, which should show consistency with the FNS. If a relationship with anxiety holds, it will further be interesting to see for what items the relationship holds.
Data was included for all subjects, except those who had reported trying
the items (even the fictional ones).
The scores on question 1 of the FRIs, an overall measure of aversiveness,
for each item were correlated with FNS (Table 2.8).
We would expect positive correlations between the FRI question 1 score and FNS, but none show a significant effect. These results are disappointing, although the power is low.
We can also perform canonical correlational analyses between all the
FRI questions at once and FNS (Table 2.9).
|"water buffalo"||13, 25||< 1||0.9|
|"cockroach"||13, 26||< 1||0.7|
As before, FNS again shows no relationship.
Similarly, we can also perform canonical correlational analyses with
STAI trait anxiety (Table 2.10).
|"water buffalo"||13, 25||1.55||0.2|
|"cockroach"||13, 26||< 1||0.8|
No relationships with trait anxiety are apparent.
These results show no evidence of FNS or trait anxiety predicting responses to individual novel foods on the FRIs.
2.9 Food scares—BSE and Salmonella
Food scares have become a major theme of the nineties in the UK. Major scares have had profound effects on patterns of consumption, on the food industry and on political events. There is much that seems irrational about these food scares. They erupt at odd times. Scares can lead to very significant changes in consumption patterns, but old patterns usually bounce back regardless of whether the risks have changed. They may even be associated with cases of mass sociogenic illness, as with the worries over dioxin contamination of poultry in Belgium and the subsequent Coca-Cola scare (Nemery, Fischler, Boogaerts & Lison, 1999).
Situations with little real risk can generate big scares, while situations with greater risk go ignored. Food scares are clearly more to do with the public?s perceptions than with real risks and politicians have increasingly come to respond to 'consumer confidence' above and beyond what scientific advisers may say. This is food rejection psychology on a mass scale. Even as I write this thesis, political turmoil reigns concerning genetically-modified foods. Commenting on this latest worry, Riddell (1999) makes a number of important points about scares:
Next to sex scandals, food scares are the trickiest political problems for any government to handle. They are often a recipe for irrationality, fear of the unknown and irresponsible and exaggerated political and media reactions. [...]
In a MORI poll [Hornsby, 1999; Elliott, 1999] for the Better Regulation
Task Force in the Cabinet Office, more than two-thirds of those interviewed
were afraid of the long-term effects of chemicals in food, and more than
half were concerned about the production of genetically modififed food,
about BSE and about food poisoning generally.
The bovine spongiform encephalopathy (BSE) epidemic in cattle and the concomitant risk of human disease from the ingestion of beef products has been perhaps the most significant food scare in the world ever. It has had profound effects on society?s relationship with food, affecting subsequent health scares as with current debate in the UK over genetically modified organisms, and major health, economic and political consequences. In itself, it is clearly a worthy subject of research for food psychology, but there are further special reasons why a thesis on food rejections should be interested in BSE as considered below. Although of a much smaller magnitude, the Salmonella in eggs scare of some years before was something of a dry run for the BSE crisis. It still had significant repercussions for food manufacturers and government policy and it serves as a useful comparator with BSE. Since when the data herein presented was collected, Escherichia coli O157 has received far more media coverage following its worst ever recorded outbreak anywhere with 20 deaths in Scotland (Mead & Griffin, 1998). However, public concern for Salmonella remains greater than for E. coli O157 or for the commonest cause of food poisoning, Campylobacter (Hornsby, 1999; Elliott, 1999). Items relating to both BSE and Salmonella were included in this study.
The ?danger? prototypes used by Rozin & Fallon (1980; Fallon & Rozin, 1983) involve chemical poisoning, yet many threats in real life are biotic, involving infection by pathogens. Public and media attention in the UK has concentrated on such risks: infection by Listeria, E. coli O157, cholera and Salmonella have all made the headlines, while BSE has dwarfed any other food scare, although chemical threats inspire scares too, as with the recent Belgian dioxin crisis.
I have already introduced the item ?food poisoning? in the study (see 2.7.3); this was used to test whether this different form of risk produces a different mode of rejection to Rozin & Fallon?s (1980; Fallon & Rozin, 1983) ?danger? prototypes. ?Food poisoning? showed no distinction from ?cramps? and ?carcinogen?. ?BSE? and ?Salmonella??served as more specific examples of a biotic ?danger? rejection.
The economic consequences of Salmonella and especially BSE have been considerable, largely through the resultant major shifts in consumer choices. Decisions by individuals about whether to eat eggs and then beef products were able to affect the macroeconomy and such decisions clearly fall under the remit of this thesis. UK beef sales fell by 30% after the 1996 scare (Blackstock, 1998). Meanwhile, some 170,000 cattle have died or been slaughtered (Brown & Bradley, 1998).
If we wish to study food rejection behaviour, then it is often useful to study changes in food rejection behaviour. We can consider these at an individual level, for example, when someone adopts vegetarianism or with learned food aversions. We can also consider changes at a population level: for example, with vegetarianism again, we can consider its increasing popularity in the population as a whole. We can also consider the increased acceptance of foods previously novel to a society, as with many fruits and vegetables in the UK over the last few decades. These population changes are comparatively gradual, however, compared to the items considered here. Food scares represent very dramatic changes in a population?s food rejection behaviour when a previously accepted food comes to be considered dangerous.
I begin by describing the biology behind BSE and Salmonella and the history of their associated scares. However, particularly with BSE, attention must also be paid to what our understanding of the disease has been at different points through time.
184.108.40.206 History and Description of BSE
Bovine spongiform encephalopathy (BSE) is the bovine form of a group of diseases seen in several species, little understood and all fatal. The human spongiform encephalopathy exists in a number of forms, notably as Creutzfeldt-Jakob disease (CJD). Spongiform encephalopathies can be transmitted between individuals through exposure to, including ingestion of, certain infected tissues. Unlike most transmissible diseases, the infectious agent seems to be not a biological organism, but a ?prion? protein, known as PrP. The most popular theory is that a misshaped PrP molecule is transmitted, which triggers other proteins of the same type to change from their normal shape to match the misshape. Details of how this happens remain unclear however; and it is only in recent years that the prion hypothesis has been widely accepted. We await a theory to fully explain the spongiform encephalopathies and thus remain hampered in our attempts to either cure the disease or prevent its occurrence. Note also that spongiform encephalopathies can arise without infection; CJD is a rare sporadic disease, while other forms are genetic. Following the BSE epidemic, the literature on spongiform encephalopathies has exploded; Brown & Bradley (1998) is a recent review, but only one of many, on this subject.
The transmission of spongiform encephalopathies by ingestion within species is commonly seen. In humans, the spongiform encephalopathy kuru seems to have been transmitted by cannibalism, probably the ingestion of the brain and spinal column of diseased relatives. Experiments show that trans-species transmission via ingestion is possible between certain species with certain spongiform encephalopathy types. When and when not transmission is possible and why remains unclear.
BSE was discovered in the UK cattle population in 1986, already at epidemic levels. Given the long, asymptomatic progression of the disease and delays before measures were initially taken (in 1989) and later rigorously enforced (in 1995) to prevent contaminated material entering the human food chain, the crucial question is whether BSE is transmissible to humans by this route and what the risks might be.
Initial thought was that the transmission to humans was unlikely. BSE is thought to have arisen when cattle ate infected material from diseased sheep following changes in the animal rendering process around 1980 that had previously destroyed the infectious agent. Sheep are known to have carried their spongiform encephalopathy, scrapie, for centuries, with no transmission to humans. Cows and sheep are closely related, so BSE could be viewed as just scrapie in cows. The scrapie analogy and hope for other reasons in a trans-species barrier preventing BSE transmission to humans led to an orthodox view, as promulgated by the government, that the risk to people was practically zero. With confidence in measures to prevent infected material entering the human food chain, at the time of this experiment, most experts believed that British beef in the shops was perfectly safe. Further, most believed that even eating meat from an infected cow was probably safe. Yet there was significant heterodoxy; with so little known about BSE, it was easy to argue that the risks were far greater.
In 1996, evidence emerged of a handful of human cases exhibiting a variant of CJD (Will, Ironside, Zeidler, Cousens, Estibeiro, Alperovitch, Poser, Pocchiari, Hofman & Smith, 1996). The causative agent of variant CJD is the same as that for BSE (Hill, Desbruslais, Joiner, Sidle, Gowland & Collinge, 1997), suggesting that it is the result of BSE transmission to humans. However, what route transmission takes has not yet been firmly established, although ingestion of infected beef products remains the most plausible explanation (e.g. Cousens, Linsell, Smith, Chandrakumar, Wilesmith, Knight, Zeidler, Stewart & Will, 1999). The resulting scare eclipsed earlier worries about BSE, but, unfortunately, took place after data collection for this thesis had ended.
The number of identified cases of variant CJD is still small, but with antemortem diagnosis difficult and incubation periods long (e.g. Hill, Butterworth, Joiner, Jackson, Rossor, Thomas, Frosh, Tolley, Bell, Spencer, King, Al-Sarraj, Ironside, Lantos & Collinge, 1999), it is impossible to tell whether they will always remain so or whether these are just the beginnings of an epidemic. The latest numbers before the printing of this thesis were of 40 definite and probable cases of variant CJD in the UK up to the end of March 1999 (Department of Health, 1999) and 1 in France (Hill et al., 1999). Public worries remain about BSE: an opinion poll in the UK at the beginning of 1999 showed that 54% are worried about BSE (Hornsby, 1999; Elliott, 1999).
The transmissible agent in spongiform encephalopathies is found in certain organs of the diseased individual, notably the brain and spinal column. In cows, these organs, known as specified bovine offal (SBO), are removed from all carcasses and all diseased individuals are destroyed. Infection from other parts of the animal (or milk) may be possible, but most scientists think that these are probably risk-free and certainly present a much lower risk. Thus, flesh meat (steak, prime mince etc.) should be largely safe. One facet of the peculiar nature of the BSE agent is that it is highly resistant to damage. In particular, heat treatment ? usually known as cooking ? does not destroy infectivity.
In the wake of the 1996 BSE scare, following the announcement of variant CJD, the Government frequently suggested that measures to tackle the problem sought to restore ?consumer confidence?, but were superfluous on health grounds. While this may partly have been doublespeak, shifting the blame on to others for creating panic as opposed to admitting that mistakes had been made in handling the issue, there is much truth in what was said. Politics ultimately responds to public perceptions of dangers, not the real dangers involved. The focus on consumer confidence in the BSE scare was perhaps the first significant time that politicians so overtly considered the psychology of risk above the reality of risk. All this merely reiterates the importance of work such as presented here in investigating how people react to such threats.
It is also worth noting that some potential scares have not occurred. The possibility of BSE re-infecting sheep and leading to a more infectious form of scrapie and of the BSE agent being found in (cow?s) milk (Warden, 1998) have not (yet) elicited any food scares.
220.127.116.11 History and Description of Salmonella
Salmonella enteritidis is a far more normal pathogen than the BSE agent. It is a common bacterium and the second most common cause of food poisoning in the UK. Salmonella infection causes only a few days of comparatively mild illness in most adults and is easily treatable with antibiotics. However, immunosuppressed individuals, notably the elderly, can suffer more serious consequences and infection can be fatal. Epidemic rates in chickens prompted a scare over the safety of eggs in 1988, leading to the resignation of junior Health Minister, Edwina Currie. Little has changed since then with about one third of all chickens in the UK infected, however this news no longer commands the headlines of Edwina Currie?s days. The public remains concerned: an opinion poll in the UK at the beginning of 1999 showed that 51% feel at risk from Salmonella (Hornsby, 1999; Elliott, 1999).
As with all bacteria, sufficient cooking will kill Salmonella bacteria. However, eggs are frequently used raw or only partially cooked. The soft-boiled egg is the an obvious example of danger and was used here for the item ?Salmonella?.
18.104.22.168 Comparison Between BSE and Salmonella
BSE and Salmonella represent two very different food poisoning threats. Salmonella is very well-understood by medical science, while the spongiform encephalopathies remain one of the most mysterious groups of disease. Salmonella is highly infective, but the consequences of infection are minor, particularly among healthy, young adults, who formed most of the experimental subjects here. We still do not know for certain how infectious the BSE agent might be; even in the light of newer evidence, the risk seems low and what consensus there was as the time of the experiment would have said the risk was non-existent or near enough so. The ?Salmonella? item concerns a soft-boiled egg, a prime source for infection, while the ?BSE? item concerns a piece of steak, probably the safest part of a cow to eat and possibly perfectly safe. However, presuming if one does contract variant CJD, the consequences are certainly fatal within a few years.
Thus, these two items present an extremely low risk of infection, but where the consequence of infection is extremely high (death) and about which there is great uncertainty, and a very high risk of infection, but where the consequence is mild and the danger well understood. Such a statement is based on the medical knowledge at the time of the study; the dangers from BSE are received to be greater now. I, therefore, am assuming that people are well informed about the dangers and that they trust official sources of information. This is clearly unlikely and actual perceptions will cloud the differences I have laid out.
The big Salmonella scares in the UK came at the end of the eighties. At the time of the study, Salmonella was remembered, but far from the public eye. Timing with respect to the BSE story was unfortunate, coming during a lull between the first big scares at the discovery of the BSE epidemic in the British cattle population and the biggest scare afterwards when variant CJD and the first clear evidence of possible transmission to humans was announced. With little published on the details of such food scares, it is difficult to know how reactions vary through the time course of a scare. It seems obvious that the magnitude of reactions would vary and reactions at the height of a scare might be so extreme as to ?top out? on the FRI. On the other hand, the more distant from a scare, the more either the ?BSE? or ?Salmonella? items become just like the ?food poisoning? item.
2.9.2 ?Danger? or ?disgust??
Although different in some ways to the chemical poisoning used as examples by Rozin & Fallon (1980; Fallon & Rozin, 1983), infection examples still fit their definition for the ?danger? category. Is there any reason to expect any alternate response to these items? I wish to suggest that a ?disgust? reaction could be involved as well.
On reflection, what is surprising about food scares is that they occur at all. The risks involved are usually very small compared to known risks in our lives. Tell people they face considerable risk from eating eggs and steak as part of a high-fat diet and they ignore you. Tell people that there is a very minor risk from Salmonella or BSE and there are dramatic responses with sales plummeting. Further, such scares often show a short lifespan, even though the risks may remain. Egg consumption rose again not long after the main Salmonella scare even though rates of Salmonella infection in chickens were little changed. While the BSE scare has shown more constancy, perhaps because the extent of the risk has only slowly emerged, there has been large increases in beef consumption in between outbursts of media attention. While there are some rational explanations of such behaviour, it is taken for granted that consumers are behaving somewhat irrationally. There are many possible explanations for this, most concentrating on the salience of the information (and misinformation) presented. Perhaps another part of the explanation is that these scares somehow trigger more than just a ?danger? response. If these scares caused ?disgust? responses, which we know to be very strong and irrational, that might be able to explain their course.
Further, I have observed that the language of ?disgust? ? and even the very word ? have repeatedly cropped up in discussion of BSE. It is widely thought that the BSE epidemic was spread by the feeding of protein supplements including cattle remains to other cattle, an enforced cannibalism. The beginning of BSE is also thought to have arisen via a similar route, with cattle eating infected material from diseased sheep, an enforced carnivory. Public reaction to this unsuspected carnivory and cannibalism by cows was strong. People were disgusted. That these things should be seen as ?disgusting? perhaps makes sense as an anthropomorphisation of our own feelings about cannibalism and novel meats. Beef and lamb may be familiar and acceptable meats to us, but constitute cannibalism and a novel meat to a cow. Further, the offal involved in rendering processes is probably ?disgusting? to most people as well.
Examples of criticism of the processes are commonplace. Consider these comments from farmers in national newspapers:
Now we find that even in the past few months we may have been giving
our sheep and hens food based on the muck from a poultry house floor. [Aldous,
The public used these same themes; letters to major newspapers included:
It is widely acknowledged that the bovine form of this disease originated
with the feeding to cattle of products composed largely of the remains
of other animals. It is noteworthy that these potentially, and now actually,
controversial feedstuffs were commonly marketed under such descriptive
names as ?meal?, ?nuts? and ?cake?, all terms which stress a lack of any
relation to meat or animal products. It seems likely that the manufacturers
of these feeds were at least aware of possible public disgust, if not potential
risks, involved in feeding products containing processed carcasses to animals
biologically adapted to consume only vegetable matter. [Shelley, 1995]
Over the past fortnight correspondents to newspapers and callers on phone-ins have queued up to denounce this practice, and many have added a simple observation: if farmers and the food industry are prepared to do something so unnatural, it is no wonder that a terrible disease should result.
This is the same gut reaction [.??] feeding meat products to animals offends against nature. So how did it begin? When and why did we cross the ethical threshold? And if BSE is a judgment on us, how did it start?
We have BSE now because we ceased to process these animal products with
sufficient care. Is it therefore a purely technical issue, rather than
an ethical one? Sir Richard Southwood, who has monitored these issues closely,
links the two. ?When you are doing something unnatural and peculiar,? he
says, ?you?ve got to be particularly careful.? [Watts, 1996]
While the cattle food practices described above clearly evoked a certain vicarious ?disgust?, could those reactions have fed through to people?s reactions to beef products? Did the ?disgusting? origins of the disease contaminate the food stuffs under suspicion?
The reactions to Salmonella were never as extreme and its origins and mechanism of spread are well known and familiar. However, here too, Davison (1989) has suggested that underlying attitudes to eggs, including a sensitivity to ?disgust?, could have contributed to the public reaction in the Salmonella scare.
?BSE? and ?Salmonella? are clearly in the ?danger? cluster (figures 2.32-2.35). Confidence regions of the various ?danger? items comfortably overlap. There is no indication that ?BSE? and ?Salmonella? show responses at all like ?disgust?.
Figure 2.32: Point cloud for ?BSE?. [Figures not yet available on the web.]
Figure 2.33: Confidence region for ?BSE?. [Figures not yet available on the web.]
Figure 2.34: Point cloud for ?Salmonella?. [Figures not yet available on the web.]
Figure 2.35: Confidence region for ?Salmonella?.
The items ?BSE? and ?Salmonella? were worded to be uncertain, to better match a real life situation where people have to make risk judgments on incomplete information. Away from specifically reactions to foods, cognitive theories of anxiety predict that more anxious individuals will perceive ambiguous risks to be greater (Butler & Mathews, 1983; Mathews, 1990). A number of experimental paradigms have demonstrated such effects in clinically anxious populations (e.g. Eysenck, Mogg, May, Richards & Mathews, 1991; Butler & Mathews, 1983); however, reports of such effects in non-clinical populations are less common (e.g. Eysenck, MacLeod & Mathews, 1987) and may be experimental design artefacts (Mogg, Bradley, Miller, Potts, Glenwright & Kentish, 1994). In particular, Mogg et al. (1994) suggest that ?social desirability? (Crowne & Marlowe, 1964) may be underlying the effect. Social desirability is the tendency to behave in a socially desired manner. It is, thus, also predictive of a subject?s tendency to be influenced by response biases in an experiment. Further, individuals scoring high on social desirability tend to produce low scores on anxiety measures; not necessarily because they are unanxious but because anxiety symptoms are seen as socially undesirable. Thus, a social desirability effect leads to an apparent anxiety effect.
While the work mentioned above tried to consider general behaviour, there is little evidence either way for how anxiety predicts more specific behaviours. For example, in health psychology, Saidi, Sutton & Bickler (1995) failed to show a relationship between a general measure of anxiety (STAI) and specific behaviours relating to breast cancer screening, although some other studies have. The mixed evidence in Section 1 that anxiety may be related to food neophobia could be interpreted as a further example or counter-example of this effect.
Thus, I make a weak prediction that highly anxious individuals will be more negative about the ?BSE? and ?Salmonella? items, though social desirability may mediate such an effect. Anxiety was measured by STAI trait anxiety and social desirability by the Marlowe-Crowne Social Desirability score, see 2.4 for details.
On a canonical correlation analysis of all thirteen questions for ?BSE? against trait anxiety, F13, 24 < 1, p = 0.6; and against MCSD, F13, 24 = 1.08, p = 0.4. For ?Salmonella? against trait anxiety, F13, 27 = 1.52, p = 0.2; and against MCSD, F13, 27 = 1.29, p = 0.3.
Thus, this study provides no evidence for an anxiety/perceived threat relationship in cognitions about BSE and Salmonella, nor for any relationship with MCSD. This could be because such relationships are inconsistent in non-clinical populations (Mogg et al., 1994), that general psychometric measures are unhelpful when trying to predict behaviours relating to specific, single foods or simply that this particular study had insufficient power.
2.10 Unusual Variants
2.10.1 Sullivan & Birch (1990)
This section was inspired by a paper by Sullivan & Birch (1990) and its surprising result. Their study was designed to investigate how children acquire preferences for added sugar and salt. 39 children (aged 44-71 months) were given repeated experience (15 exposures over 2 weeks) with one of three forms of a novel food, processed tofu: either plain, with 2g/100g added salt or with 14g/100g added sucrose. (Their use of processed tofu was one inspiration for the design of kśrrta, used elsewhere in this thesis.) Ratings for all the forms of tofu were made at preexposure; after 8 exposures; one week later with no further exposures in between; and after seven more exposures. For the ratings, the children had to place the food sample in front of one of three faces: like (a smiling face); just okay (a neutral face) or dislike (a sad face). Within each category, the child then rank ordered the foods. Such a technique had been shown to be a reliable predictor of consumption (Birch, 1979).
Sullivan & Birch (1990) predicted that liking for the exposed form would increase with the number of exposures, as has been found in many other studies (e.g. Birch & Marlin, 1982), but that this increased liking would not generalise to the other forms. In fact: ?Preference for the exposed version increased with exposure, whereas preference for the unfamiliar flavoured versions actually declined? (Sullivan & Birch, 1990: p. 549, my italics). This also means that, after exposure, the unfamiliar forms were liked less than when they were totally novel foods.
Their results were not wholly consistent. For children exposed to the plain version, preferences for the flavoured versions decreased, against prediction. For children exposed to a flavoured version of tofu, their preference for the other (unexposed) flavour remained unchanged while that for the unexposed, plain version of tofu decreased. Nonetheless, Sullivan & Birch (1990) say ?[the learned safety] interpretation of mere exposure effects implies that such exposure effects should be specific to foods actually ingested and should not generalize to other similar but untried and potentially dangerous substances. However, this interpretation does not account for the observed decreases in preference we noted for the unexposed, differently flavoured versions of the same food? (p. 50).
This result has yet to be replicated as far as I know. However, anecdotal evidence lends support to the phenomenon occurring in adults. People show aversions to foods that ?taste funny?. However, this observation is confounded by the fact that such items are usually potentially dangerous, having ?gone off?. Then again, perhaps this gives us a clue to the cause of such rejections as I suggest in 2.10.3 below.
In an attempt to learn more about this phenomenon, a FRI item was designed: ?a fruit that unexpectedly tasted different from normal and you don?t know why?. Many readers voiced the reasoning behind their answers, saying that they would be suspicious about the fruit. To what extent such cognitions were present in Sullivan & Birch?s (1990) children is hard to tell. Further, to what extent such cognitions were ?post hoc? over an instinctive reaction or not is also unknown. That many subjects did have difficulty with understanding the item means that the results can not be considered so reliable.
?Differs? lies between the ?distaste? (specifically, the novel plant-based items) and the ?danger? cluster. The confidence region is large and overlaps with ?distaste? items, but not the ?danger? items (figure 2.37). The unusual taste appears to relegate the item to being novel again, although the results are inconclusive with subjects having difficulties with the description given and the large confidence interval.
2.10.3 Suggested Ethology
What might the adaptive value of this model be whereby exposure to one form produces a decrease in preference for other (unfamiliar) forms of that food? A novel food is one that the individual knows nothing about. However, if an individual has other information about a food,
Figure 2.36: Point cloud for ?differs?.
Figure 2.37: Confidence region for ?differs?.
then a better judgement can be made. An unfamiliar form of a familiar food has some information connected with it—it would be surprising if this was not considered. If the new form is similar enough to the familiar form, then the learned safety of that form should generalize. It makes sense for this generalization to be minimal because, in a natural setting, foods often come in multiple forms with widely differing toxicities and nutritional values. However, I suggest, that not only do most foods come in multiple forms, but often, if one of these forms is safe to eat, the others are likely to be dangerous. Consider, for instance, fruit: though the ripe form may be highly nutritious, underripe fruit or overripe, rotting fruit are both potentially dangerous.
Yet why should this be the case? To consider fruit again, there is only a ?window of opportunity? when the fruit is ripe during which it is edible. A fruit is a mechanism for seed dispersal, which is why ripe fruit is palatable: animals take the fruit to eat but reject the seed, discarding it away from the mother plant. However, it is against the plant?s interests for the fruit to be eaten before its time, so fruiting plants have evolved to make underripe fruits inedible, either physically or through toxic secondary compounds. At the other end of the fruit?s lifetime, fruit tends to rot. If the fruit has not been eaten in a certain time, it is no longer of use to the plant and, therefore, there is no reason for the plant to sustain the fruit?s edibility. Meanwhile, various microorganisms (typically fungi and bacteria) may infect the fruit, leading to rotting. It is in their interests to make the fruit unpalatable to macro-organisms, with whom they are in competition for the resources contained in the fruit (Janzen, 1977). Janzen?s (1977) suggestion that it is interspecific competition that causes rotting food to become unpalatable he applied equally to other food sources, like seeds moulding or carrion spoiling.
The suggested model also works for other food sources. Plants usually protect their leaves (and the rest of their bodies) through the use of toxins. However, the cost of producing these versus the benefit in lowered predation will vary over the lifetime of a leaf. Thus, at certain stages, a leaf can be a rich source of nutrients, and, at other times, highly toxic.
Thus, there are possible ecological explanations for why foods may have forms worth avoiding as well as forms which are nutritious. This suggests why a special aversion towards unfamiliar forms of a prefered food could be adaptive.
Inhalation has many parallels to ingestion, although inhalation choices in life are rather limited in comparison. One such choice of great importance to public health is whether to smoke. Research has usually looked at why people choose to smoke. Chronically, the answer is nicotine addiction, but much research has gone into the more vexed question of why people start smoking. However, we can also ask why do other people not smoke?
Can the use of the methodology used here on food rejection tell us anything about the choice not to smoke? A FRI was adapted to cover smoking a cigarette as opposed to eating an item. (For the MDS analyses, FRI data was only included if the item was rejected sufficiently on the first question, so the responses analysed here will be of people who dislike smoking.) If this matter is comparable to food rejection behaviours,?cigarette? can be expected to fall in the ?danger? cluster. However, the very strong public health message, clearly zealously embraced by some members of the public, perhaps make cigarettes different from the usual food threats. Oft refered to as a ?disgusting habit?, the ?cigarette? item could lie in the ?disgust? cluster.
2.11.1 MDS Results
?Cigarette? stands apart as a clear outlier, remote from all the other items (figure 2.39). In a three-dimensional MDS solution (see 2.6.2), ?cigarette? dominates the third dimension, placing it away from everything else. It does not seem useful to compare ?cigarette? to the other items in this manner.
2.11.2 Use of FRI in a Different Context
Beyond looking at the relationships between different items with FRIs, the FRI also leads us to think about rejection behaviours as multi-faceted. Negative attitudes towards smoking may involve a fear of the long-term health risks or a short term nausea (produced by nicotine),
Figure 2.38: Point cloud for ?cigarette?. [Figures not yet available on the web.]
Figure 2.39: Confidence region for ?cigarette?. [Figures not yet available on the web.]
finding the process unpleasurable, or finding it disgusting. We can compare regular smokers (n = 9), occasional smokers (n = 13) and non-smokers (n = 19) on the FRI data using a MANOVA with a within-subject factor with 12 levels of the FRI questions and a between subject factor of smoking status to see whether smoking behaviour reflects differing attitudes. The interaction between these two was significant: Wilks? ? = 0.26, F22, 52 = 2.24, p = 0.009. Thus, not only does smoking status affect negative attitudes in a quantitative manner, as could be expected, but it does so differently across various questions.
Means (of transformed data) are shown for the FRI questions by smoking
status in Table 2.11.
|Non-smoker||Occasional Smoker||Regular Smoker|
|5 Lung (?Stomach?)||
|7 Average person||
|9 Temporary discomfort||
|10 Permanent damage||
As would be expected, smokers are less negative about smoking. However, differeces between the groups are not constant for each FRI question, with questions referring to the hedonic experience varying most.
Is this methodology potentially of use to the smoking researcher? Some results seem potentially of interest. For example, on the question of long-term damage caused by smoking, the occasional group score lower than the non-smokers or the regular smokers: F9, 36 = 3.59, p = 0.038, occasional smokers different on least significant difference post hoc tests. As this study was not designed to investigate differences in smoking behaviour and involves small numbers, I do not wish to claim this result as conclusive, but it suggests an approach to smoking research which may be of value. Just as I am suggesting that a paradigm shift in the consideration of food rejection phenomena on multiple dimensions may be of use, so too might it be of use in considering other behaviours.
In this Section, I have introduced the idea that food rejection is a multi-faceted phenomenon which requires more complex methodologies than, for example, the list heuristic of the previous Section. Food can be unpalatable, nauseating, dangerous or all three and so on. By using a battery of questions, here called a FRI, we can consider these many aspects.
Once we have described food rejection in a multidimensional manner, a logical first step is to construct a taxonomy. Once a taxonomy exists, the position of new items in the taxonomy is of interest. Backed up by a more developed statistical approach and more varied items, I have replicated the tripartite taxonomy of Rozin & Fallon (1980). Fallon & Rozin?s (1983) suggestion of a fourth category (?inappropriate?) is not supported, but the possibility of some sort of substructure for the ?distaste? category encompassing similar ideas is suggested. Pliner & Pelchat?s (1991) observations on different reactions to novel plant- and animal-derived foods is also replicated.
Crises involving BSE and, earlier, Salmonella were situations where dramatic changes in food rejection behaviour had significant impact on the country?s economics and politics. While at first sight classic examples of the ?danger? category, an argument can be constructed for suspecting that these items might provoke a ?disgust? response. However, the results show the rejection responses evoked by BSE and Salmonella in this study to fall squarely in the ?danger? group. Whether these results would be replicated at other points in the natural history of these food scares is not known. It is unfortunate that this study took place during something of a lull in BSE worries.
A familiar item with an unexpected taste may be commonly disliked. Here, it seems that the rejection response is similar to that for wholly novel items, suggesting something more about this little researched phenomenon. Smoking does not fit into our food taxonomy, but perhaps the methodology may still be useful to researchers in that field.
While I have explored the methodologies here, unresolved issues remain. As with the list heuristic, arbitrary choices have had to be made. Here, most notably, the choice of questions in the FRI remains difficult. However, and again as with the list heuristic, common findings across changes in the methodology suggest the results are robust.
Unfortunately, the FRIs are complicated to use and multiple questions per FRI for multiple rejected items is very time-consuming. Can we take the message of this section—of food rejection being multidimensional—and apply it to specific questions in food psychology without going through the rigmarole of devising a full taxonomy? I do that in Section 3 when I look at one particular distinction in food rejection—unpalatability versus nausea—in one particular food rejection phenomenon—learned food aversions. That example will help demonstrate the importance of Section 2.