The List Heuristic
Over the last half century, a methodology has been used to investigate predictors of eating behaviour which involves the use of a list of foods, with subjects answering some preference question(s) about each food. For each person, the number of foods which elicit a particular response, usually of rejection, is used as a psychometric measure of their eating behaviour. I refer to this as the list heuristic. There has been great variability in the exact design used (see 1.4.1) and comparing across studies is difficult. I tackle this issue in a number of ways, including by comparing ‘sublists’ within one study. The previous literature suggests a distinction between neophobia and general ‘pickiness’ (see 1.3) and this issue is also considered.
Previous list heuristic studies have found a link between behaviour and both anxiety and sensation-seeking (1.4.3). This is the first review to consider all these studies together. Food choice behaviours can also be studied through psychometric questionnaires of more traditional design (1.4.2) and such studies have produced similar results (1.4.3).
In this research, the list heuristic is used in two studies (1.5), representing the first applications outside North American populations. My results are summarised in 1.6. They are notably different from previous studies in finding no relationships with anxiety; weak relationships with sensation-seeking are confirmed. Further, results concerning neophobia measures (list heuristic and questionnaire) suggest problems in differentiating food neophobia from more general food rejection behaviour.
I begin in 1.1 by explaining the importance of the list heuristic in the wider context of studying food rejection behaviour and food choice behaviour in general.
1.1 The Problem: Learning About Food Rejections
We can approach the subject of food rejections from many different angles. We can observe that some people are more ‘rejecting’ of food than others. Some people are more ‘picky’ about what they eat than others. Let us assume for the moment that there is a unitary personality trait behind such behaviour, which I shall call ‘pickiness’. As Rozin, Fallon & Mandell (1984) say in the introduction to their paper: "Although there is much variation in human food attitudes and preferences, very little of this variance has been explained" (p. 309). If we can explain the variation between people in this trait, we should learn what causes this trait.
I am hesitant about using the terms ‘picky’ and ‘pickiness’ for they have been used in alternate ways previously in the literature and they may have unintended connotations. An alternate term would be ‘finicky’. The Chambers Dictionary’s definition of "finicky" does seem particularly appropriate: "particular about trifles" (Macdonald, 1972). Pelchat & Pliner (1986) and Pliner & Hobden (1992) have used finickiness "to refer to a tendency to dislike the taste of foods", although what this means in practice is unclear. Specifically, in Pliner & Hobden’s (1992) study, this was measured as a sensitivity to bitter adulteration (following, e.g., Schachter, 1971), mean palatability ratings for foods (either familiar or novel) or expected palatability ratings for novel foods. The reader should thus note that ‘pickiness’ herein is being used as a strict term referring to a tendency to reject many types of food.
There are obvious problems in assuming that there is a unitary trait behind all food rejection behaviour, as we shall soon discover in considering the details of the list heuristic (1.4.1). Why should we assume that an orthodox Jew’s rejection of pork is at all similar to a coeliac disease sufferer’s rejection of pizza or to my own rejection of celery on the grounds of taste? (Individuals with coeliac disease suffer an adverse reaction to gluten (wheat protein), such as would be found in a pizza base.) Section 2 takes this as its starting point for a wholly different approach to the study of food rejection behaviour. Until then, the methods described in Section 1 must stand by their own record. Of course, it is not necessary that all food rejections have a common nature as long as certain forms are dominant. If that is the case, any general method can be used to study those dominant forms.
1.2 Methodological Approaches
In order to investigate pickiness, we must be able to quantify the trait. What I have christened the list heuristic presents itself as an obvious tool; so obvious that it has been independently invented at least three times. Let us present individuals with a set list of foods and ask them to indicate which they would reject. The total number would then stand for the latent variable of pickiness. Variations on this theme also suggest themselves. We could ask a different question about each food or choose certain types of food for the list.
This procedural flexibility in the list heuristic is both useful and problematic. The problems will be considered below, but the utility is that the method can be adapted to consider various aspects of food choice behaviour. Some of these go beyond the remit of this thesis, food rejection, but some of these will be considered for what they tell us about the list heuristic.
Traditionally, psychology has used psychometric measures to study personality traits. The individual is given a set of questions describing themself. Some overall score is calculated and used as a measure of the underlying trait. There is no foreseeable obstacle to constructing such questionnaires about food rejection behaviour. The most studied area of food volume rejection—restrained eating—has spawned numerous such questionnaires (some of which we will meet below). Other food-related questionnaires have also been designed: for example, Steptoe, Pollard & Wardle’s (1995) Food Choice Questionnaire and Westcombe & Wardle’s (1997) questions on attitudes to fat.
Pickiness or neophobia should be equally open to this technique and various such measures are considered below.
We can make a theoretical distinction between food neophobia—rejection of a novel food, that is, one of which we have no experience—and rejection of a familiar food. I refer to the rejection of familiar foods as food ‘pickiness’. Note that ‘familiar’ should be taken to operate at a purely individual level. It seems reasonable to posit both common and different psychological processes involved in the two. Smith, Powell & Ross (1955b), for example, suggested, "those Ss who have many aversions for foods which they have tasted will generalize this aversion to foods which they have never tasted" (p. 145). On the other hand, rejection of novel foods may be driven by processes that are not associated with day-to-day food choices.
Of the list heuristic studies (to be described in 1.4), Smith et al. (1955b) was the first to explicitly consider that one possible cause for rejecting a food is neophobia. Neophobia may even define what is food and what is not food. A role for neophobia in pickiness is uncertain. In considering interpersonal variation in pickiness, the fact that we all consider certain items as not normally considered food (which may be an aspect of neophobia) is not relevant. On the other hand, pickiness has long been associated with neophobia.
Further consideration of neophobia will come in Section 2 (2.8).
1.4 Literature Review
The first use of a list heuristic is in a 1939 paper by Tussing. This led to a series of papers from the USA until 1955: Wallen (1943; 1945; 1948); Gough (1946); Altus (1949); Smith, Powell & Ross (1955a; b). The method has recently re-emerged in three places independently of each other: a presentation at Food Choice Conference-1 by Walsh (1993); a series of studies from Cincinnati (Frank & van der Klaauw, 1994; van der Klaauw, Raudenbush, Frank & Howe, 1994; Raudenbush, van der Klaauw & Frank, 1995) and the work presented here (and also as Potts & Wardle, 1998). There are various procedural differences between these other studies, which are discussed in 1.4.1, along with a discussion of how I have attempted to study the effects of some of these differences. 1.4.2 considers related psychometric-style questionnaires relating to food choice behaviour. Finally, 1.4.3 then considers what these and other studies have found concerning the relationship of food choice behaviours to other variables.
1.4.1 Methods of Earlier Studies
The two earliest studies (Tussing, 1939; Wallen, 1943), both considering sex differences, used an alternate statistic to that described above. Each food on the list was classified by whether more men or women rejected it. Wallen (1943) also introduced the more usual sum method.
Exactly what question is asked about each food has varied from study to study. Most of the first wave of papers, all from the USA, concerned themselves, as I do, with food aversions: subjects, for example, were asked to mark those foods that "you DISLIKE so much that you refuse to eat them Remember to mark only those that you don't like at all" (Wallen, 1948, p. 310). The total number of items marked, i.e. of food aversions, was summed. It is worth noting that a food not being disliked is not equivalent to its being favoured.
Previous studies have approached the distinction between neophobia and pickiness in different ways. In the early studies, the list items were all fairly common foods. Wallen (1943; 1945) excluded untried items from the total number of aversions, while Smith et al. (1955b) calculated two scores, including and excluding untried rejected foods. Excluding untried foods, however, produces a bias on a sum score as those subjects who have tried fewer foods are effectively using a shorter list, decreasing their possible score.
Other studies have allowed more than just a Yes/No answer. In Walsh’s (1993) work, subjects responded to a list of 100 foods in terms of like, don’t like, don’t like—never tried or don’t know, with the number of foods in each group giving four scores per subject. The most recent series of studies have used the Food Attitudes Survey (FAS), a list comprising 217 foods or more (Frank & van der Klaauw, 1994; van der Klaauw et al., 1994; Raudenbush et al., 1995). Here, subjects respond to each food with one of five responses. The total number of foods for each of three responses—1: I really like this food, I think it tastes good (or like); 2: I dislike this food, it tastes awful (or dislike) and 3: I've never tried this food, and I never intend to try it (or won’t try)—gives three scores for each person. (The other two responses were not used.) Each of such measures as used in the FAS and by Walsh (1993) clearly bears some relationship to the simpler measure of the early studies.
The food aversion sum of the 1943-1955 studies, Walsh’s (1993) don’t like sum and the FAS dislike sum would appear to be measures of food pickiness. Food pickiness can also be measured by direct self-report; Raudenbush et al. (1995) developed a pickiness scale (with 4 items; e.g.I consider myself a picky eater) from a preliminary set of questions in Frank & van der Klaauw (1994). In a group of 101 undergraduates, they found significant correlations (0.61 >= |r|s >= 0.44) with all three FAS measures.
Walsh’s (1993) don’t like—never tried and the FAS won’t try sums appear to index food neophobia. Again, food neophobia can also be measured by direct self-report. Raudenbush et al. (1995) also developed a neophobia scale (7 items; e.g. I enjoy trying new foods), which again showed significant correlations (0.64 >= |r|s >= 0.34) with all three FAS measures (see 1.4.2 below). Raudenbush et al.’s (1995) pickiness and neophobia scales turned out to be highly correlated with each other (r = 0.66) and, as all three of the FAS measures correlated with both scales, this suggests that the two are closely related.
There are many procedural details in these studies to consider. Methodological shortcomings may compromise the validity of the measures, while the variation in equally valid designs may make it hard to compare studies. The studies reported above have used various questions about the foods on the list. The relationship between these different scores has not been investigated.
The number of foods on a list varies widely, from 20 (Wallen, 1945; 1948; Gough, 1946; Altus, 1949) to 455 (Frank & van der Klaauw, 1994). By using a longer list, quirks about individual foods from idiosyncratic aversions or allergies can be avoided (though how long is necessary has not been systematically studied). However, certain dietary regimes are category-restrictive, like vegetarianism or some religious taboos. For category-restrictive diets (CRDs), increasing the size of the list will make no difference if the proportion of the interdicted food type remains constant. Frank & van der Klaauw (1994) are wrong to say that the length of the FAS will "minimize the impact of... food practices (e.g. vegetarianism)" (p. 102). Vegetarians will always score more highly if there are meats on a list. All previous studies have failed to take this into account; thus their findings may be confounded by differences between, say, vegetarians and non-vegetarians. However, the proportion of vegetarians will often be so low as to negate this problem; this is particularly true for those studies carried out in pre-1960 USA. Either we have to use a food list without, for example, any meats, which would be unrepresentative for most people, or we must exclude individuals with CRDs from any analyses. Equally, CRDs may themselves be manifestations of food pickiness or neophobia.
The proportion of different types of foods on the list has wider implications as well. For example, a predominance of bitter-tasting foods in the list may produce a score that is closely related to bitter sensitivity. Frank & van der Klaauw (1994) and Wallen (1945) both report high internal reliabilities for their lists using split-half correlations. However, as the various types of foods were randomised throughout their lists, these high split-half correlations merely show that, with a constant proportion of different types of foods, different lists give highly related scores. What happens if the proportion of different types of foods varies is unclear. A priori, I suggest particular differences that may be important. Pliner & Pelchat (1991) present evidence that neophobic responses to meats and non-meats are very different, with novel meats evoking a "disgust" reaction, while Terasaki & Imada (1988) found preference ratings for meats showed a different pattern of relationships than vegetables or fruit. Also, the more novel foods there are on a list, the more a simple rejection sum will become a measure of food neophobia. Given the close relationship between neophobia and pickiness, how important this may be is uncertain.
In summary, list heuristic sum measures may be sensitive to the exact procedure used. One aim of this Section is to determine to what extent procedural variations influence the results. As it seems possible that the content of the list has an important effect, such problems will be addressed through the use of subset measures, i.e. measures calculated from a subset of the foods from a list. Thus, a score can be calculated for just the non-meat foods on a list. (Such a measure can also be used for vegetarians.) With a priori beliefs as to important differences in food type, by comparing the scores from, say, a meat and non-meat list, we can test the internal reliability of a list heuristic score more thoroughly than through ‘atheoretical’ analytic techniques.
The populations used in list heuristic studies have also varied. A number of the early papers used US military personnel (Wallen, 1945, 1948; Gough, 1946; Altus, 1949). Tussing (1939) used US college students and students have (unsurprisingly) dominated later studies’ populations.
1.4.2 Psychometric Measures
As mentioned above, a complementary way to study food-related responses is to devise questionnaires in the usual mould of personality measures. Pliner & Hobden (1992) designed a Food Neophobia Scale (FNS), which, through convergent evolution, is very similar to a combination of Raudenbush et al.’s (1995) neophobia and pickiness scales. The FNS was designed as a measure of "reluctance to eat and/or avoidance of novel foods" (p. 105, Pliner & Hobden, 1992). They report very high alpha values (0.88 in two samples). The scale was validated in a series of behavioural experiments and very good test-retest reliability was found (e.g., r = 0.82, p < 0.01, with a 15 week interval). Meiselman, Mastroianni, MacDonald, Buller & Edwards (1995) further showed that, in US undergraduates over three years at college, FNS still showed good stability.
Pliner & Hobden (1992) present data from numerous unpublished studies
concerning correlates of FNS. These results are shown in Table 1.1 (based
on Table 4, p. 114, Pliner & Hobden, 1992).
|Trait STAI*||r = 0.39, p < 0.05, n
r = 0.46, p < 0.01, n = 40;
r = 0.30, 0.05 < p < 0.1, n = 36
|State STAI*||r = 0.20, n.s., n = 30;
r = 0.45, p < 0.01, n = 40
|Ad hoc measure of anxious mood||n.s., 3 samples (ns = 30, 24, 25)|
|Sensation seeking and similar constructs|
|GNS**||r = 0.54, p < 0.01, n
r = 0.62, p < 0.01, n = 73
|Experience Seeking subscale of SSS***||r = -0.46, p < 0.01, n
correlations with other SSS subscales not significant
|Age||r = -0.27, p < 0.05, n
r = -0.19, 0.05 < p < 0.1, n = 71;
r = -0.27, n.s., n = 33;
r = -0.21, p < 0.05, n = 112
|Sex||No significant relationships ever found|
* State-Trait Anxiety Inventory (Spielberger, Gorsuch & Lushene, 1970)
** Generalised Neophobia Scale (Pliner & Hobden, 1992)
*** Sensation Seeking Scale (Zuckerman, 1979)
The authors conclude, "it appears that trait food neophobia is significantly and positively related to other trait-like measures of anxiety? but may be less strongly related to such temporary anxiety states as state anxiety or anxious mood" (p. 115). These results are discussed in 1.4.3.
A similar, though reversed, questionnaire—the Variety-Seeking Questionnaire (VSQ)—was devised by Van Trijp & Steenkamp (1992; see also Van Trijp, 1994), which Meiselman et al. (1995) showed to be highly correlated with FNS (-0.58 > rs > -0.71, ns >= 95).
At face value, there are, however, a number of problems apparent with such scales: individuals with CRDs would score artificially high (e.g., FNS question 8 I am very particular about the foods I will eat). (This may also be a problem for many measures of restraint.) More generally, some of the questions seem to relate more to food pickiness generally than to neophobia in particular (FNS questions 3, 8 & 9).
1.4.3 Relationships with Other Variables
Notwithstanding the problems discussed above, we thus have a body of list heuristic measures, measuring related traits of food pickiness and neophobia. This section concentrates on their relationship with certain personality traits. Two key traits have emerged in the literature as possible predictors: sensation-seeking (or related constructs) and anxiety. It should be noted that these two are frequently found to be negatively correlated, so any relationship to one may produce a relationship to the other.
Food neophobia and pickiness can be expected to be related to similar patterns of responses to external stimuli, such as sensation-seeking. Zuckerman (1979) argued that food has a strong stimulating value that would serve high sensation-seekers. Demonstrating an association, Walsh (1993) found significant negative correlations between Zuckerman’s Sensation Seeking scale (SSS) and the don’t like—never tried and don’t know scores (n = 80). In two studies, Raudenbush et al. (1995) found some significant correlations with their like (r = 0.22 in Study II; n.s. in Study IV) and won’t try (r = -0.35 in Study II; r = -0.29 in Study IV) scores, but not with the dislike score. They also found a significant relationship between their neophobia scale and SSS (r = -0.27 in Study IV).
Pliner & Hobden’s (1992) scale also showed a significant negative correlation with the Experience Seeking subscale of the SSS (r = -0.46), but not with any of the other subscales. FNS also correlated with their own General Neophobia Scale (rs >= 0.54)—"a preference for familiar situations and people vs. a willingness to experience new situations and people" (p. 107).
That someone is neophobic implies that they are making negative assumptions about the food, a trait associated with anxiety. Pickiness is also seen by "folk-psychology" as a neurotic trait. Thus, we might predict relationships with anxiety measures. Early studies compared various ‘neurotic’ groups with ‘normals’, the former being shown to have more food aversions (Wallen, 1945; Gough, 1946; Wallen, 1948; Altus, 1949). Of course, clinically anxious individuals may be more picky without anxiety being a predictor in a ‘normal’ population. All of these studies (Wallen, 1945; Gough, 1946; Wallen, 1948; Altus, 1949) used male, military personnel. Translating into modern psychiatric terminology (see 5.1.3), Wallen’s (1945) and Gough’s (1946) "neurotic" groups appear to have mainly had Generalized Anxiety Disorder. Wallen (1948) further showed an increased number of aversions in a range of psychopathologies, including epilepsy, intra-cranial injury and personality disorders and that the degree of neurosis was correlated with the number of aversions, the first evidence that anxiety as a continuous trait may be related as opposed to simply the clinical/non-clinical distinction. Altus (1949) found more aversions in illiterate Army recruits, which he explains by arguing that illiterate Army personnel are less intelligent and more neurotic (Altus, 1946). He also found that the number of aversions was correlated with a measure of adjustment within the illiterate group.
Smith et al. (1955a) took the next step, demonstrating a link in a non-clinical population: they compared the number of aversions with Taylor Manifest Anxiety Scale (Taylor, 1953) in two non-clinical populations and found a significant, though small, correlation (r = 0.17). They replicated this finding using a number of other measures in a second study (Smith et al., 1955b).
Although such findings were consistent, views on why food pickiness should be related to anxiety were less so. Wallen (1945) talks of "inadequate emotional control" in childhood leading to more food rejections. Smith et al. (1955a) also asked their subjects to give their reasons for their rejections. These results proved difficult to interpret, but suggested a common, low-level of rejections based on distaste plus variable, anxiety-related levels of aversions.
Pliner & Hobden (1992) found the FNS significantly correlated with trait anxiety (Spielberger State-Trait Anxiety Inventory, STAI) in three samples (0.46 >= rs >= 0.30). In contrast, Raudenbush et al. (1995) failed to find any significant correlations between FAS responses and the Fear Survey Schedule (Geer, 1965), a measure of levels of fear and phobic tendencies, despite a large sample of 143 subjects. Also of relevance is Pelchat & Pliner’s (1986) finding that 2-6 year old children who are poorly adjusted (based on reports by their mothers) were more likely to refuse a novel food.
Various animal studies also support a link between food neophobia and anxiety and sensation seeking (see Pliner & Hobden, 1992, for review). However, precisely how appropriate such animal studies are as models for the human work considered here is debatable. Many non-human studies have assumed that food neophobia makes for a valid animal model of human anxiety (e.g., Miyamoto, Kiyota, Nishiyama & Nagaoka, 1992). Many more have assumed that food neophobia is a valid measure of the animal’s general neophobia. Such views have been criticised (e.g., Cole, Robbins & Everitt, 1988). As non-human work has implications for the human studies, so to do those human studies have implications for the non-human work. If food neophobia were not to be related to anxiety and sensation seeking in human studies, we would have to re-consider what food neophobia in non-humans actually measures.
Gender differences are an obvious factor to consider, particularly as they are seen in both anxiety and sensation-seeking. The very earliest studies, both considering gender differences (Tussing, 1939; Wallen, 1943), used an alternate statistic based on a list heuristic: each food on the list was classified by whether more men or women rejected it. Both found that there were considerably more foods which more women than men rejected. Wallen (1943) theorised that his result related to women’s lower levels of (what is now called) sensation-seeking. Smith et al. (1955a) found women to have higher anxiety levels and to show more food rejections, results that they link. Smith et al. (1955b) found that women are specifically higher on neophobia. Pliner & Hobden (1992) reported no gender effects. Frank & van der Klaauw (1994) reported that women scored higher on the FAS dislike and won’t try scores. Neither Pliner & Hobden (1992) nor Meiselman et al. (1995) found any difference by gender on FNS; the latter also found no difference on the VSQ.
Age has been somewhat neglected as a variable; but even when considered, the limited age distributions of the sample populations makes it hard to draw any definitive conclusions. Pliner & Hobden (1992) found a marginal negative relationship to age: over five samples, -0.27 < rs < -0.19, significance varying with sample size.
The most extensively studied area of food-related traits is that relating to restraint. Frank & van der Klaauw (1994) suggest the importance of any link: "Perhaps the observed tendency toward food neophobia in females contributes to the increased vulnerability of women to eating disorders" (p. 104). Raudenbush et al. (1994) also compared FAS scores to several measures of eating restraint. There were no significant correlations between FAS measures and the Revised Restraint Scale (Herman, 1978) or the 8 subscales of the Eating Disorders Inventory (Garner, Olmstead & Polivy, 1983). The number of won’t try responses did show a significant, if weak, correlation with the Eating Attitudes Test (Garner & Garfinkel, 1979), but the number of like and dislike responses did not show a significant correlation. Westcombe & Wardle (1997), meanwhile, reported no significant correlation between the restrained eating scale of the Dutch Eating Behaviour Questionnaire (Van Strien, Frijters, Bergers & Defares, 1986) and FNS. It should be noted that restraint is an ill-defined area and different questionnaire measures of restraint may be measuring different psychological traits.
The FAS has also been compared to olfactory responsiveness (Frank & van der Klaauw, 1994; van der Klaauw et al., 1994), but such findings are not reviewed here.
1.5 Two New Studies
This section describe the use of the list heuristic in two studies with the aim of further exploring the relationship between sum variables and various personality traits and to determine the comparability of variations in the methodology used. 1.5.1 considers the importance of what question is asked about the foods and what foods are on the list, while 1.5.2 allows findings to be replicated on a different subject population using a different food list.
1.5.1 South Bank Sample
Subjects came from two classes of undergraduate engineering students at the University of the South Bank, London: one, first-year (n = 43) and one, second-year (n = 34). The nature of the study was explained afterwards. This study was accepted by the Research Ethics Committee of the Joint Institutes: Maudsley Hospital and Institute of Psychiatry.
The food list consisted of 58 items in three blocks (detailed in 5.1.1). The first block consisted of 43 real foods and 5 invented, novel foods, selected to cover the range of basic food types. The names and descriptions of the novel foods were based on Pliner & Pelchat (1991). Invented foods were used to guarantee their novelty to all subjects. Short phrases were used to describe the invented foods; some of the real foods also had accompanying short phrases to avoid drawing attention to the invented foods. The second block consisted of 6 plant items not normally considered food and the third, 7 meat items not normally considered food. These items were all non-toxic. The first question asked about each of these foods was Would you be happy to eat it?—for the first block, subjects had to answer Yes or No, while for the latter two blocks, subjects could also answer Maybe. The total number of Nos was summed; for the last two blocks, Maybe counted for half. For the 43 real foods, this sum—the total number of rejections—gave R[Real]. The five invented, novel foods from the first block similarly produced a score, R[Novel]. The last two blocks yielded scores out of 6 or 7 respectively: the ‘animal non-foods’ score, R[ANF], and the ‘plant non-foods’ score, R[PNF]. Subjects were also asked on the first block whether they had ever tried each item. Table 1.2 details the list-derived variables used here.
The 43 real items on the first block can be split by type and R scores calculated for specific subsets of the list. This was done for two categorisations. First, the foods were split into food types: meats versus plant products (see 5.1.1). R[Meat] is the number of rejections from the 10 meats; and R[Veg], from the 12 fruits, vegetables, grains, nuts and seeds. Secondly, the food items were ranked by how many subjects reported having tried them. The least tried tertile (tripe, eel, tofu, tongue, goat’s milk cheese, sunflower seeds, duck, cottage cheese, mackerel, black olives, mozzarella, semolina and vegetable curry) produced an R[Rare] score; and the most tried tertile (butter, double cream, mushrooms, prawns, walnuts, boiled rice, peppermints, soft-boiled egg, spaghetti hoops, cabbage, custard and mayonnaise) produced R[Common].
The third question asked in the first block was Do you like, or think
you'd like, the overall taste of it?, answered Like (+1), Neither
Like nor Dislike (0) or Dislike (-1). U[Real] was calculated
as the number of (real) foods rated as unpalatable (-1 on this question).
Note that U is a negative palatability measure.
|Number of real foods subject would not be happy to eat.|
|Number of real foods that subject rated as -1 [range -1 to +1] for overall taste. (Same items as R[Real]).|
|Number of invented, novel foods subject would not be happy to eat.|
|Number of ‘animal non-foods’ subject would not be happy to eat.|
|Number of ‘plant non-foods’ subject would not be happy to eat.|
|Subset of R[Real]: number of meats subject would not be happy to eat.|
|Subset of R[Real]: number of vegetables, fruits etc. subject would not be happy to eat.|
|Subset of R[Real]: number of common foods subject would not be happy to eat. (‘Common’ being defined as the upper tertile of what foods subjects have tasted.)|
|Subset of R[Real]: number of rare foods subject would not be happy to eat. (‘Rare’ being defined as the lower tertile of what foods subjects have tasted.)|
All subjects completed the Spielberger State Trait Anxiety Inventory (Spielberger, Gorsuch, Lushene, Vagg & Jacobs, 1983); "N" and "E" subscales of the Eysenck Personality Questionnaire (EPQ; Eysenck & Eysenck, 1975); a short-form of the Marlowe-Crowne Social Desirability scale (Reynolds, 1982) and an Anglicized version of the Food Neophobia Scale (see 5.1.2). Two different anxiety measures were used (trait STAI and EPQ "N") to increase reliability. Subjects also completed several demographic questions.
Subjects were given an A4 landscape questionnaire booklet at the beginning of a class, latecomers being handed theirs on arrival. Booklets were filled out individually. The booklet began with several background questions and demographic details, then the food lists, then the psychometric measures. Everyone attending completed a questionnaire. The study was carried out in March 1993.
There were only 5 women to 72 men, so only the results for men are presented here (although the inclusion of the women makes no difference to the pattern of results). With occasional omissions in completing the questionnaire, the exact population varies from analysis to analysis though typically no more than a handful is lost (see below for further discussion). Ages ranged from 18 to 45 (mean ± standard deviation was 22 ± 4.8) and were highly skewed (skew = 2.7). Age showed no significant relationships with the other background variables (ps > 0.1).
Scores on the personality measures agreed with known norms and produced the expected intercorrelations. Trait STAI and EPQ "N" showed a large, positive correlation (r = 0.72, p < 0.001, one-tailed) and both showed negative correlations with EPQ "E" (r = -0.50, p < 0.001, and r = -0.28, p = 0.008, respectively, both one-tailed).
Subjects were asked Do you have any dietary restrictions? among the demographic questions. Many reported problems with specific foods (e.g., allergies) and 7 reported having category restrictive diets (CRDs): either being vegetarian or following kosher food laws. These subjects had higher FNS scores, although the difference was not significant. They also showed higher scores on most of the food rejection measures: e.g., for R[Real], mean difference of 5.6, t60 = 2.1, p = 0.041. For analyses involving FNS or most R measures, these 7 subjects were excluded (although their inclusion has little effect on the pattern of results). (Those R measures for which these subjects were not excluded are detailed below.) Although subjects were asked whether they had any category restrictive diets, borderline cases may have been missed, e.g. people sympathetic to vegetarianism.
Food rejection measures:
A few subjects reported having tried some of the invented, novel foods: eight reported having tried naseberry, down to no-one reporting having tried pance. Compare Frank & van der Klaauw (1994) who reported a figure of 7% for responses implying a subject had experience of a fictitious food.
Cronbach’s alpha was calculated for the
various list measures, R[Real], R[Novel], R[ANF] and R[PNF]: 0.68 <=
alphas <= 0.90, ns >= 47 (see Table 1.3). Such results
are satisfactory; if anything, they are too high, implying a redundancy
in the length of lists used.
Table 1.4 reports descriptive statistics for these variables. The mean
standard deviation for FNS was 31 ± 12, n = 65.
|Variable||Mean number (standard deviation) rejected||
Completion rates for these variables were lower than for the background variables, particularly for R[Novel]. Those subjects for whom there were missing data for any of R[Real], R[Novel], R[ANF] or R[PNF] were compared to the rest on their scores on the background variables with t-tests. There were no significant differences.
R[Real], R[Novel], R[ANF] and R[PNF] all correlated with each other (0.28 <= rs <= 0.65, ps <= 0.034). FNS showed significant correlations (rs >= 0.4, ps <= 0.001) with all of them.
Subjects often omitted answering the palatability question or wrote "don’t know" for foods that they were unfamiliar with; only 24 subjects gave palatability ratings for all of the real foods, though only 9 missed more than 5 ratings. The high failure rate for the palatability question for unfamiliar foods is understandable and, perhaps, was encouraged by the inclusion of multiple questions about each food. U[Real] can be scaled appropriately to compensate for missing values, though this may produce a biased measure by excluding those foods that the subject is least familiar with. U[Real] shows a large correlation with R[Real]: for just those subjects who gave palatability ratings for all the foods, r = 0.76, p < 0.001, n = 24; using a scaled U[Real] for all the subjects, r = 0.86, p < 0.001, n = 58. The pattern of correlations produced with scaled U[Real] is the same as with R[Real], i.e. there are significant correlations with R[ANF], R[PNF], R[Novel] and FNS (rs > 0.2, ps < 0.03, all one-tailed).
Although those with CRDs have been excluded for analyses involving R[Meat], R[Rare] or R[Common], they have been included for analyses involving just R[Veg] or R[PNF] as neither lacto-ovo-vegetarianism nor kosher laws restrict consumption of such foods. Cronbach’s alphas for the five subset measures, R[Meat], R[Veg], R[Common] and R[Rare] ranged from 0.43 to 0.76. Both FNS and U[Real] significantly correlated with all of the subset measures (rs > 0.4, ps <= 0.001).
There was a significant correlation between R[Meat] and R[Veg] (r = 0.33, p = 0.011); both also showed positive correlations with R[Novel] (rs > 0.4, ps <= 0.001). R[Meat] was correlated with R[ANF] (r = 0.44, p = 0.001), but not significantly so with R[PNF]. R[Veg] showed the reverse pattern, being significantly correlated with R[PNF] (r = 0.29, p = 0.018), but not so with R[ANF].
R[Rare] and R[Common] were significantly correlated (r = 0.56, p < 0.001). R[Rare] showed significant correlations with R[Novel], R[ANF] and R[PNF] (rs > 0.3, ps <= 0.010). R[Common] did not significantly correlate with R[PNF] or R[ANF]; yet it did significantly correlate with R[Novel] (r = 0.43, p = 0.003).
FNS significantly correlated with all of the subset measures (0.41<= rs <= 0.59, ps <= 0.001). U[Real] also showed large correlations with the various subset measures (rs > 0.6, ps < 0.001).
Relationships with background variables:
None of the food rejection measures (R measures, subset R measures, U[Real] or FNS) showed any significant relationships with any of the background variables. For example, between FNS and trait STAI, there was a correlation of r = 0.04, n.s. As Type II error (failing to correctly reject a null hypothesis) is dependent on experimental power, I also report the 95% confidence interval for this result: -0.21 < rho < 0.28. Between FNS and EPQ "E", r = -0.18, n.s.
An insufficient n and the high correlations between dependent variables preclude meaningful use of multiple regression analyses.
Descriptive statistics for FNS agree very well with Pliner & Hobden (1992). They report a mean ± standard deviation for FNS of 34.5 ± 11.9 for their sample of psychology undergraduates, while the present population gives 32 ± 13 (with the inclusion of subjects with CRDs to match). However, the pattern of correlations found between FNS and other measures is radically different. With a larger sample size than any of Pliner & Hobden’s (1992) groups, the present study failed to find a significant correlation between FNS and trait STAI.
As both Study 1 and Pliner & Hobden (1992) correlate FNS and trait STAI, a direct statistical comparison between the results can be made. Confidence intervals for rho from Pliner & Hobden’s (1992) results were calculated: for n = 30, 0.66 > rho > 0.03; for n = 40, 0.67 > rho > 0.17; for n = 36, 0.57 > rho > -0.03. Alternatively, using an arctanh transformation of r values (Fisher, 1921), normal tests can be used to see whether the present result differs from Pliner & Hobden’s (1992) three results. Comparing the present value for r with Pliner & Hobden’s (1992) two lower figures gives zs < 1.63, ps > 0.1, but against their highest figure for r, z = 2.22, p < 0.05. On these results alone, the inconsistency between the results reported here and Pliner & Hobden (1992) appear to be only marginally significant. On the other hand, that there were no relationships between other food rejection measures and anxiety, suggests that there is a real difference. FNS also showed high correlations with measures of food pickiness, as opposed to neophobia: for example, with R[Common], r = 0.52, p < 0.001. This shows a worrying lack of specificity; a result echoed by Raudenbush et al. (1995).
Constraining how far these results can be generalized is the distinct nature of the population used here: engineering undergraduates from one university; the 100% return rate achieved is, however, notable. Previous studies have also used distinct populations, typically students, or they have not given many details of what populations were used.
Generally, the various food rejection measures (the R measures, U[Real] etc.) all correlated significantly with each other. It is difficult to interpret the correlation magnitudes seen with the subset comparisons. These comparisons were of an extreme nature (only meats versus only vegetables etc.) as compared to more usual reliability statistics. We can conclude that some of the methodological details raised in 1.4.1 are not barriers to comparing between studies. The main exceptions to this pattern were along the dimension of novelty. R[Common] did not correlate significantly with some more neophobic measures. Interestingly, there were also no significant relationships between R[ANF] and R[Veg] or R[PNF] and R[Meat], that is the animal and plant distinction was important for novel foods. This supports the findings of Pliner & Pelchat (1991).
1.5.2 Family Diet Study Sample
The main subject population consisted of mothers of 9-11 year old children, recruited through primary care physicians in south London. This, the general sample, consisted of 92 mother/child pairs. A group of Gujarati mothers and their children from the same region were also interviewed (n = 24 pairs). Only the mothers’ results are discussed.
The psychometric measures used and discussed here were: EPQ "E" and "N" subscales; FNS and the restrained eating scale of the Dutch Eating Behaviour Questionnaire (DEBQ; Van Strien, Frijters, Bergers & Defares, 1986). The Anglicized version of the FNS was again used. Only eleven out of twenty-four of the Gujaratis received the FNS and DEBQ: these subjects received the measures because of their better English skills, for other subjects, the measures were deemed inappropriate. The EPQ was not given to any of the Gujaratis. The only demographic variable I consider is age.
Subjects were given a list of 30 foods (5.1.1) and were asked How much do you like each of the foods below?, answers being given on a five point scale (-2 to +2). For each subject, the total number of items with a score of -2 or -1 was calculated, to give U. U is calculated on the basis of whatever items on the list were completed, with the value scaled appropriately to be out of 30. In the General Sample, there were no more than occasional missing values. The comparability of this use of a palatability question is validated by the high correlation between R[Real] and U[Real] seen in 1.5.1. Subset measures U[Meat] and U[Veg] were also calculated, as before. Table 1.5 details the list-derived variables used here.
For the Gujaratis, certain foods (chicken, sardines, sausages, eggs,
beefburgers and liver) were excluded from the list with respect to calculating
a U value as these foods are not components of the normal Gujarati diet,
|Number of foods that the subject rated as -2 or -1 [range -2 to +2] for overall taste.|
|Subset of U: number of meats that the subject rated as -2 or -1 for overall taste.|
|Subset of U: number of vegetables, fruits etc. that the subject rated as -2 or -1 for overall taste.|
|As U, but with items not common to the typical Gujarati diet excluded.|
The data were taken from a larger study on healthy eating practices and attitudes (the Family Diet Study: Gibson, Watts, Wrightson & Wardle, 1995; Gibson, Wardle & Watts, 1998). The questionnaires considered here were usually completed in the presence of an experimenter, with some being given to the subject to take away and return by post. The study was carried out Spring to Autumn 1993.
Rates of completion for the General Sample were much higher than for the similar question in the South Bank sample. This may be due to the more common foods used in this study and due to the different questioning techniques used.
The General sample:
The mean ± standard deviation age was 39 ± 4.9 years. There were four vegetarians and twelve on various other diets (the precise nature of these other diets was not available). Again, with occasional omissions in completing the questionnaire, the exact population varies from analysis to analysis, although typically no more than a handful are lost. The intercorrelations between the various background psychometric measures were as expected.
For analyses of FNS, U and U[Meat], subjects on diets were excluded, as before. The mean ± standard deviation for FNS was 30 ± 11, n = 74 (or 31 ± 12 including those on diets). FNS showed no significant correlations with any background variables. U was related to FNS (r = 0.26, p = 0.017) and EPQ "E" (r = -0.30, p = 0.010). Other correlations were not significant.
Cronbach’s alpha was calculated for U (alpha = 0.76, n = 73) and for U[Meat] and U[Veg] (0.43 <= alphas <= 0.55). U[Meat] and U[Veg] were significantly related to each other (r = 0.43, p < 0.001). U[Veg] showed no other significant relationships. U[Meat] showed a significant correlation with EPQ "E" (r = -0.29, p = 0.012). There were no significant results with any of the other background variables.
The Gujarati sample:
The mean ± standard deviation age was 38 ± 3.9 years. The U[Guj] items for the Gujarati subsample are all vegetarian. Moral/religious restrictions on the Gujarati subjects’ diets varied, however all the U[Guj] items could theoretically be eaten by all the Gujarati subjects. A significant correlation with FNS only was found (rS = 0.70, p = 0.024); a Spearman’s correlation was used given the small n of 10. There were no significant relationships between age, FNS and DEBQ.
Rates of completion for the General Sample were much higher than for the similar question in the South Bank sample. This may be due to the more common foods used in this study and due to the different questioning techniques used.
The subject population is radically different from either 1.5.1, Pliner & Hobden (1992) or the FAS studies, most notably in being considerably older, although it is still fairly specific in itself. The data from the Gujarati subsample, although limited, do confirm the relationship between FNS and U in a very different cultural group.
The main difference in results with 1.5.1 is in the significant relationship between U and EPQ "E". This result is puzzling given that the expected intermediate variable, FNS, fails to show a significant relationship with either variable and that the food list consists of more familiar foods than in the South Bank sample. Compare also Walsh (1993) and Raudenbush et al. (1995), both of whom found relationships between SSS and neophobia, but not pickiness measures. Also, FNS failed to correlate significantly with any of the three subset U measures. This is in contradiction to 1.5.1, but in agreement with the initial design of FNS as a measure of neophobia specifically. Perhaps this is simply because the subset measures are less reliable given the small number of foods in each.
1.6 A Synthesis?
The various studies reviewed and the results of the present studies agree on a number of results: neophobia and pickiness are very closely related to each other, both are somewhat related to sensation-seeking but not to measures of restrained eating. There is some inconsistency between whether food pickiness or neophobia measures are more closely related to sensation-seeking. Such differences and the overall weaker relationships found here may be due to the use of the EPQ "E" scale as opposed to Zuckerman SSS. Any relationship with anxiety is even less clear, with the early US studies and Pliner & Hobden (1992) supporting a link, whereas the present studies and Raudenbush et al. (1995) found no relationship. Pliner & Hobden (1992) report a number of relationships involving FNS, none of which were replicated here. However, I cannot rule out the possibility of Type II errors in our data. It is interesting to compare these results with those obtained by Otis (1984) who measured food neophobia in a realistic taste testing situation. She found that state anxiety and age were significantly related to neophobia responses, but not sensation-seeking. A metanalysis would be an appropriate next step in comparing between these apparently disparate results.
The various subset and alternative measures largely showed respectable correlations with each other. This suggests that the make-up of the food list (at least with respect to the meat/non-meat distinction or novelty) is not of great importance, nor is the exact question used. The weakness of any novelty effect is clearly tied to the close relationship found between measures of pickiness and of neophobia. Such procedural differences probably do not, therefore, explain the divergent results found in different studies. Some of the subset measures used did produce low alpha reliability scores, perhaps because of the small number of items in them. This suggests a lower limit for the number of items on a food list that should be used.
Although the exclusion of individuals with CRDs is an improvement over previous studies, it remains an unsatisfactory solution in populations where high proportions of individuals have CRDs, as is increasingly the case with vegetarianism in student populations. This and the limited cross-cultural application seen in Study 2 demonstrates some limitations of this heuristic.
A possible explanation for the divergence of results lies in the different populations used: however, it is difficult to identify what differences may be of importance. Although we should not be prejudiced against the early papers simply because they are old, clearly their social milieu may have been very different. Wallen (1943) shows admirable prescience in commenting on social changes that might reduce the sex differences he found. The gap between the midcentury US and today may also explain the differences in results concerning anxiety. If secular changes can explain these differences, this in itself implies that social factors—i.e., those factors that are different between populations—have an important influence on our food related behaviours. Suggestions can be made as to other factors that could be of influence: for instance, one variable that would differ between the subject populations used here is the subjects’ involvement in food preparation.
Our failure to replicate Pliner & Hobden’s (1992) results concerning the relationship of FNS to anxiety or sensation-seeking is more difficult to interpret, particularly with the distribution of FNS scores matching so well. The population similarities between Pliner & Hobden (1992)—Canadian psychology undergraduates—and Study 1—UK engineering undergraduates—makes the failure to replicate surprising. Pliner & Hobden (1992) do have a mixed sex group compared to the two single sex populations here; however, with no observed relationship between sex and FNS, this cannot explain the correlations. I also found high correlations between FNS and food pickiness measures, in contradiction to further work by Patricia Pliner (pers. comm.).
Notwithstanding that I made the theoretical distinction, the rejection of novel foods and of familiar foods comes out as very closely related (for instance, the high correlation between FNS or R[Rare] with R[Common] in 220.127.116.11). This may be due to methodological inadequacies in distinguishing between them. For example, questionnaire measures used to index neophobia confound the issue with questions of a more general nature, like I am very particular about the foods I will eat in the FNS. Raudenbush et al. (1995) argue along similar lines. The question used here to index familiarity, namely Have you ever tasted it? (compare also Wallen, 1943; 1945; Smith et al., 1955b), provides but a coarse estimate. A food that has been tasted once is not necessarily very familiar. Also, there may be a response bias: subjects may see it as undesirable to admit that they have not tasted a food for which they proclaim a dislike, although the R[Rare]/R[Common] distinction should still hold.
Assuming for the moment that this close relationship between the rejection of novel and familiar foods does hold, does this imply that the distinction is of no use or are these two distinct traits that happen to be closely related? The very nature of food rejection behaviour precludes most disliked foods from being familiar to an individual. The more unpleasant a food is on the first tasting, the less likely the individual is to taste it again. (Cases of familiar foods becoming unpleasant though initially liked—i.e., learned food aversions—occur infrequently.) This would imply that most rejected foods are somewhat novel to the individual, making the measurement of any trait related to the rejection of familiar foods difficult. Compounding this, any food one encounters will be edible and nutritious (to some degree) for that is how society defines "foods". Therefore, everything that society deems to be a food should be liked if we assume that we have evolved to learn to like anything edible and nutritious. Thus, the main reason left not to like anything is neophobia. One obvious exception to this line of reasoning concerns those foods that are inherently unpalatable—e.g., being too bitter, sour or ‘hot’—but which we learn to like—e.g. coffee or chili. However, these make up only a small proportion of foods; for example, in Study 1, the items vegetable curry, black coffee, peppermints and pickled onions. Other idiosyncratic exceptions may also spring to mind, yet overall there is a troublesome confound between liking and familiarity.
The alternative is that there is no valid distinction between the rejection of novel and familiar foods. Given the other possibilities discussed above, this would be a rather assertive claim to make from these studies. Nevertheless, perhaps we should be more careful in assuming such a distinction.
Given the importance of food choice behaviour, it seems likely that food choice cognitions are, at least in part, of ancient evolutionary heritage. Learned food aversions (Section 3), for instance, have been observed across an extremely wide taxonomic range (from humans to slugs; Garcia, Lasiter, Bermudez-Rattoni & Deems, 1985). Perhaps food cognitions are fairly distinct from other brain operations, which would explain a lack of a relationship between food rejection and anxiety measures. If there is a connection, any research of this nature still leaves open the question of how, say, a general level of anxiety leads to food pickiness or food neophobia. I tackle such issues in Section 2 and other research has already begun to do so (e.g., Pelchat & Pliner, 1995).
That the list heuristic and the FNS produce inconsistent results in different studies casts some doubt on their utility (although Type II errors may explain the differences). Study 1 demonstrated that the list heuristic is fairly robust when it comes to changes in the methodology used. It is, thus, unclear what underlies the inconsistency of our results with respect to earlier studies, though social factors are a likely possibility. Over time, region and occupation, these studies have used a wide variety of populations. Studies taking a more ‘epidemiologically’ rigorous approach to generalizability may be called for. Until these questions are resolved, perhaps all such studies’ findings should be taken with a pinch of salt.