Are health oriented consumers as aware as they believe they are?

(1)

Master Thesis

Are health oriented

consumers as aware as

they believe they are?

Does the health halo around protein, which has

a beneficial nutritional value, make the health

oriented consumers neglect the sugar content of

the same product?

Author: Supervisor:

Anneloes Beverwijk Prof. Dr. Ir. Koert van Ittersum

anneloesbeverwijk@hotmail.com k.van.ittersum@rug.nl

This thesis is submitted in the fulfilment of the requirements by the University of Groningen for the degree of Msc. Marketing Management

Faculty of Economics and Business

(2)

Abstract

Quantitative research was conducted to investigate whether the health halo around protein, results in consumers with a high interest in health-orientation rating protein bars high in protein as more healthily than low protein alternatives, regardless of the sugar content. Secondly, the moderating effect of a traffic light label on the relationship between health halo’s and the rating of the healthiness of a protein bar is investigated. The research is divided in two studies, the first study investigates whether there is a statistically significant difference for all conditions: protein high or low, protein displayed or not and consumers with a high or low interest in health on their estimations of the sugar content of a protein bar. The second study investigates whether the implementation of a traffic light label will influence the participants choice in picking a healthier option. Both studies are not statistically significant, but there is significant proof that consumers with a high interest in health will pick high protein/low sugar over other alternatives lower in protein and/or higher in sugar. Suggestion for further research are provided as a result of insignificant hypotheses.

Keywords

(3)

Introduction

We live in a world where terms like health nut and health freak go hand in hand with sugar addict and sweet tooth. Where the world population craves artificial sugars more and more, we simultaneously crave knowledge about nutrition, health and physical activity. This results in ever growing numbers for both obesity and the wellness industry, respectively 27.5% (between 1980 and 2013) and 10.6% (between 2013 and 2015) (Medicalnewstoday; Global Wellness Institute).

Lately, taking care of one’s health is an ongoing trend and where 49% believes they are overweight a slightly larger percentage is actually admitting to take care of the excess weight, 50% (Nielsen, 2015). In a survey among 30.000 respondents in 60 countries, who have Internet accessibility, acquired by the Nielsen Global Health & Wellness (2015), a staggering 72% of respondents are trying to shed the excess pounds by exercising and an extra 11% reduces their weight with supplements like diet bars, pills and shakes (Nielsen, 2015). In addition there is a 55% increase over the last three years of respondents admitting they eat more natural and fresh foods. Furthermore, slightly over one-third (32%) seek for foods high in protein and the exact same amount of respondents plead for low sugar foods (Nielsen, 2015). Marketers can use the healthy attributes of products that are already perceived healthy by the consumers, to increase their sales (Nielsen, 2015). The growing interest in fitness and healthy foods results in a growing interest in healthy products for marketers. As Deloitte presented in their health and wellness report of 2017, the number of products that were formulated or reformulated to meet the companies policies on health and nutrition grew from 22.500 in 2014 to 179.600 in 2016. Food manufacturers and retailers provide consumers with a healthy lifestyle focus with products which fit in their diets, or accompany their extensive exercise regime, providing them products high in desirable attributes (Nielsen, 2015). One of these desirable attributes is protein, which is necessary for the recovery of muscle damage after exercise (Børsheim, Tipton, Wolf & Wolfe, 2002). But protein has more to it, it increases satiety, which results in weight loss, and protein increases the overall body composition (Ridge, Devine, Lyons-Wall, Conlon & Lo, 2018). However, the demand for protein grows faster than the actual necessity, because usually individuals eat enough protein with a regular diet, except for vegans, vegetarians, dieters or people with an extreme exercise regime and physical jobs (Molyneaux, 2015). The lack of knowledge about the actual effects of protein is an opportunity manufacturers enthusiastically seize, they produce items enriched with protein, but the increase in protein often goes hand in hand with the increase of

undesirable ingredients such as sugar, sodium and cholesterol (Nielsen, 2015). The result of the fitness lifestyle becoming mainstream, instead of a market for bodybuilders only, is that consumers with this lifestyle continuously feel a lack in their protein consumption, resulting in 75% of these consumers buying some form of protein enriched products. (Molyeaux, 2015). The interest in protein creates a health halo around this nutrient, resulting in consumers believing a protein enriched product is healthy, even though the sugar content can be as high as a candy bar (Painter & Prisecaru, 2002).

(4)

potential solution to this problem will be investigated, to increase the consumer awareness regarding the sugar content of protein enriched products.

Literature review

Weight Loss, Resistance Exercise and Protein

As aforementioned, 50% of the respondents did claim they were seeking weight loss, whereas 72% tried to accomplish this goal through exercise (Nielsen, 2015). From all forms of

exercise, resistance training is most effective in terms of weight loss, due to the fact that resistance training in combination with a diet results in a higher Fat Mass loss and has a more sparing effect on Fat Free Mass loss, compared to aerobic training or no exercise training (Hunter, Byrne, Sirikul, Fernández, Zuckermann, Darnell & Gower, 2008). Resistance training results in a higher percentage of lean muscle tissue and therefore there is an increase in Resting Energy Expenditure (REE), which is the calorific expenditure of one’s body in utter rest (Hunter et. al., 2008). In the study conducted by Hunter et. al. (2008), the Resting Energy Expenditure significantly decreased with weight loss for aerobic trainers and non-trainers, while the Resting Energy Expenditure did not significantly decrease in the group of resistance trainers. In addition, this maintenance of the Resting Energy Expenditure results in the individual burning a higher amount of kcal compared to their aerobics and non-training counterparts from dawn-to-dark, thus helping in their weight maintenance after weight loss (Hunter et. al., 2008).

After resistance training, both muscle protein breakdown and muscle recovery are common. Resistance training will leave the body with a negative net muscle protein balance, for which it is solely possible to get to a positive by the intake of nutrients (Børsheim et. al., 2002). The intake of Essential Amino Acids (EAA, found in protein rich foods) immediately after exercise, results in a stimulation of muscle protein synthesis (Børsheim et. al., 2002). The muscle protein synthesis reduces muscle damage and will accelerate the recovery of the muscle which benefits the muscle function (Pasiakos, Lieberman & McLellan, 2014). Furthermore, immediate ingestion of a protein supplement before and after an resistance exercise session results in a more rapid muscle recovery, which in turn results in significantly more repetitions in the next exercise session among the supplement takers compared to the placebo takers (Hoffman, Ratamess, Tranchina, Rashti, Kang & Faigenbaum, 2010). In addition, greater recovery results in greater strength over time (Hoffman et. al., 2010). Lastly, an individual who is invested in the sport of bodybuilding, requires a 100% higher protein intake compared to a sedentary individual (Lemon, Tarnopolsky, MacDougall & Atkinson, 1992). These higher levels of protein intake result in a 7.9% increase in strength and 8.8% midarm flexor area gains after a month of training for amateur bodybuilders, thus individuals new to the bodybuilding sport (Lemon et. al., 1992).

(5)

the age of 65) in Western populations are prone to a rise of chronic diseases and age related conditions, whey protein has proven to improve age related changes in the body such as strength, functional ability and the loss of lean tissue (Ridge et. al., 2018).

The beneficial effects of proteins for both weight loss and muscle recovery gain worldwide interest, which subsequently results in a drastically growing demand for protein (Molyneaux, 2015). The knowledge of protein, however, beyond its muscle building benefits, is limited, thus leaving room for protein food/beverage manufacturers to seize their

opportunities (Molyneaux, 2015). In spite of the fact that people usually consume enough protein in their diet, this may be different for some groups of consumers (such as: vegans and vegetarians, people with strenuous exercise levels, elderly and dieters), all consumers strongly seek for high protein products (Molyneaux, 2015). Moreover, due to the rise in exercise trends and diet trends lacking animal protein, an increasing percentage of the consumers feels a severe lack of protein in their diet, resulting in 75% of consumers acknowledging to consume some form of high-protein foods or beverages in the past year (Molyeaux, 2015). Where high levels of protein were associated with bodybuilders, it has now become a ‘must have’

ingredient for the whole population (mainly Millennials), thus pushing food/beverage manufacturing towards producing protein enriched products (Molyneaux, 2015). In order to take advantage of this trend, food manufacturers alter their products or invent products that are high in protein, but on the other hand high in undesirable attributes like sugar, sodium, cholesterol, trans- and saturated fat (Nielsen, 2015).

As Etzel (2004) found in his article, these processed, palatable high-protein foods are manufactured to provide consumers convenient options for consuming their protein, for example: high-protein beverages. However, in order to make these product palatable, there has to be a balance between the pH level and sugar content, drinks that are very acidic use sugar to balance the tartness (Etzel, 2004). Whenever the pH value is low (pH=3 to 7), the acidity has to be compensated with sugar (6g of protein and 24g of sugar per 240ml.), resulting in the protein beverage being equally high in calories and only slightly lower in sugar as a soft drink, but will carry the claim ‘good source of protein’(Etzel, 2004). In

addition, products from the ultra-processed food products group 3 (protein bars belong to this group) contain on average 33% more sugar than unprocessed or minimally processed foods (such as fresh or frozen vegetables, dried fruits etc.) (Monteiro, Levy, Claro, de Castro & Cannon, 2010). These sugars, preservatives and cosmetic additives are designed to create products that are convenient, very palatable, ready-to-eat and have a very long shelf-life (Monteiro et. al., 2010).

(6)

also related to an array of diseases like cardiovascular disease, the metabolic syndrome, diabetes and obesity (Bray & Popkin, 2014).

Health Branding

Products with a health-related attribute are often perceived differently and thus more preferably compared to those who don’t (Wansink, Ittersum & Painter, 2004). In addition, whenever these attributes are exposed on the labels of products, thus being descriptive, the sales increase by 27% (Wansink, Painter & Ittersum, 2001). These descriptive attributes, when health or diet-related can give a positive halo to a certain food (Wansink et al., 2004). The consumer can create a positive or negative bias on expectations, when a certain ingredient or attribute is presented (Wansink et. al., 2004). For example, energy bars (thus protein bars, meal replacement bars, nutrition bars etc.) are commonly seen as healthy, even though they can be made of ingredients high in sugar and thus have a high GI (glycemic index), which is commonly viewed as unhealthy (Painter & Prisecaru, 2002). The contrast in evaluations is a result of the positive claims provided by the manufacturers, they claim their bars provide ‘quick energy’, ‘greater post-exercise recovery rates’ or ‘reduce hunger between meals’, which may trick the consumer (Painter & Prisecaru, 2002).

Chandon & Wansink (2007), present an interest addition about health halo’s. They studied whether health claims make consumers underestimate the calorie content of meals in fastfood restaurants at both sides of the spectrum, thus being perceived healthy and unhealthy. Shockingly, the healthy appearance of a restaurant results in a 35% lower calorie estimation compared to a meal with the same calorie content from a restaurant which is perceived as being unhealthy (Chandon & Wansink, 2007). In addition, this health halo bias is as strong for consumers that have a high interest in nutrition as it is for consumers with a low interest in nutrition (Chandon & Wansink, 2007).

Therefore, health branding can be quite dubious, since on one hand the consumer recognizes the brand and will acknowledge it’s products as being healthy and thus won’t feel the necessity to read the nutritional information on the back, while on the other hand this lack of reading can cause consumers to believe in potential misleading information about healthy products and they won’t develop autonomous skills concerning reading the nutritional labelling (Anker, Sandøe, Kamin & Kappel, 2011). Health branding can be separated into different elements, whereas the focus of this paper is on functional claims, which imply that an attribute of a product has beneficial nutritional properties (Anker et. al., 2011). As a matter of fact, reported by Cowburn & Stockley in their 2005 article, consumers find great

difficulties in understanding the nutritional information labels. Where consumers did report they understand the basic terms such as calories, sugar, salt, vitamins and fat, the least understood relationships where those between sugar and carbohydrates, calories and energy, salt and sodium and cholesterol and fatty acids (Cowburn & Stockley, 2005). Cowburn & Stockley also reported that people with a high interest in health commonly show higher levels of label reading. However, consumers often find it hard to understand how a product fits in their everyday diet, even though they were using the nutritional information labels (Cowburn & Stockley, 2005). Health branding mostly aims to create higher purchase intentions,

(7)

the consumer (Anker et. al., 2011). Lastly, healthy branding can distort a consumers’ everyday knowledge of healthy eating, meaning that healthy branding can result in people believing a certain product is healthy, while this is at odds with the public health culture which professionals try to establish (Cowburn & Stockley, 2011).

In this paper the focus is on protein bars (this study is narrowed down to the consumption of protein bars, in order to exclude differences in opinions for different protein enriched

products), which are often used as a healthy post workout snacks, or as meal replacements for weight loss, but there are numerous protein bars which are packed with a higher percentage of sugar than protein (Appendix 1). The sugar content of these bars is often overlooked due to the positive attribute, which is the amount of protein packed for a snack. It can be reasoned that nutritional labelling can be a barrier for many consumers, where you should have literacy and numeracy skills in the first place and a slight understanding of nutritional information in the second place (Sonnenberg, Gelsomin, Levy, Riis, Barraclough & Thorndike, 2013). Furthermore, even though nutritional labels provide a guideline in the nutritional value of a product or a percentage of the daily recommended intake, it is demonstrated that consumers are more likely to identify healthy options with traffic light labels, as these are fairly easily understood (Sonnenberg et. al., 2013). In the article by Ollberding, Wolf & Contento (2010) is stated that individuals who use the nutrition labels on food, tend to have a healthier

consumption than those who don’t. However, the less health oriented consumer benefits from the front-of-package traffic light labelling, since a quick glance at the packaging will already inform them with the nutritional values of the product (Brownell & Koplan, 2011). Moreover, the sales of healthy items increased, while simultaneously the sales of the unhealthy items dropped whenever the traffic light was introduced in a large hospital cafeteria (Thorndike, Sonnenberg, Riis, Barraclough & Levy, 2012). All in all, when consumers are confronted with a traffic light food label, they are more likely to consider their health and pick the healthier option at point-of-purchase (Sonnenberg, Gelsomin, Levy, Riis, Barraclough & Thorndike, 2013).

Conceptual framework

When a consumer is already health oriented and has an appropriate understanding of nutritional information labels, they won’t feel the necessity to read the labels on the back whenever they acknowledge the healthiness of a brand (Anker et. al., 2011). The amount of protein in a protein bar is seen as a healthy attribute, which can create a positive halo around the consumed product (Wansink et. al., 2004). This health halo results in consumers

underestimate the calorie content of a product and thus believe the product they are

consuming is healthier than it actually is (Chandon & Wansink, 2007). The reason for healthy eaters underestimating the calorie content of products has a lot to do with someone’s

(8)

Wansink, 2007). As Okada (2005) found in her study, consumers will give in to their hedonic cravings more often when there is a chance that they can justify the consumption, therefore, the protein content may put the attention of the sugar content of the same product. For example, there are 10.3 grams of sugar in every 100ml of cola, whilst there is twice or even triple the amount of sugar per 100 grams in certain protein bars (Appendix 1). Thus,

percentage wise, there is twice or three times as much sugar in a protein bar compared to cola, however the protein bar is categorized as being a healthy snack, while cola is definitely categorized as extremely unhealthy. In addition, the World Health Organization suggest that the daily sugar content of free sugars (which are added sugars and sugars present in syrups, honey and fruit juices and concentrates) should be at least below 10% of your daily energy intake, but ideally below 5% (World Health Organization, 2015). Therefore, the daily intake should be below 50 grams for woman, and 62 grams for man, but ideally these amounts should be 25 grams and 31 grams respectively (World Health Organization, 2015)(Appendix 2). Some of the protein bars listed in Appendix 1 have a sugar content nearly as high as the daily recommended intake, or at least have a sugar content which exceeds 50% of the daily recommended intake. Is there a possibility to inform consumers more extensively when we add a traffic light label to these products, to get an understanding whether the sugar content of the given protein bar is low or high. This brings up the following research question:

Does the high protein content of a protein bar, results in a health oriented consumers being more likely to neglect the sugar content and can traffic light labelling interfere?

As stated in the article by Painter & Prisecaru (2002) and Wansink and others (2004), consumers often neglect the negative attributes when a positive attribute is brought more to daylight by the manufacturers. Usually health oriented consumers are invested in a form of sport and according to the survey by Nielsen Health & Wellness (2005), 72% try to shed excess weight by exercising and an extra 11% by the use of diet bars, shakes and pills. This total of 83% is the target group for protein/energy bars. Whenever an energy bar is promoted for its high protein level and thus quick post-workout recovery or quick energy on the

packaging, often the consumers forget to glance at the nutritional label of the product and thus neglect the sugar content (high Glycemic Index) (Painter & Prisecaru, 2002). This is striking, because most of the health-oriented consumers are very reluctant to the consumption of processed sugar, whilst this is found to be toxic, bad to one’s health and equally addictive as

Figure 1: Conceptual Model

Hypothesis 1

(9)

alcohol (Lustig et. al., 2012). Therefore, a consumer can be oblivious to the sugar content when a healthy attribute is so prominently present (Wansink et. al., 2004). Moreover, health-oriented consumers are more likely to react to health primes due to guilt and self-presentation (Chandon & Wansink, 2007), because they will use this health prime to justify a hedonic purchase, such as a protein bar (Okada, 2005). The need for justification (Okada, 2005) and the health halo around brands, products or nutrients with a healthy image combined(Painter & Prisecaru, 2002), will make the health-oriented consumer more prone to the health primes (Okada, 2005).

A health-oriented consumer can be categorized as being generally interested in health (Roininen, Lähteenmäki & Tuorila, 1999). Having a positive attitude toward general health interest will make a consumer rate specific foods differently from those with a negative attitude toward general health interest (Roininen et. al., 1999). The consumers with a positive attitude rated products with a healthy attribute (such as artificially sweetened soft drinks) higher on healthiness than the consumers with a negative attitude (Roininen et. al., 1999). Consumers who score high on general health interest, compared to consumers scoring low or moderate, tend to make more healthful-not-pleasant food choices, are more health-oriented and are willing to eat items which are less pleasantly if they are more healthful (Roininen et. al., 1999). In addition, Hayes & Ross (1987) found that a consumer’s health and appearance are their biggest motivation for their healthy eating habits. Thus, the appearance is one of the biggest motivations to eat healthily (Hayes & Rosss, 1987), and the benefits of protein for weight loss and muscle recovery (improvements in physical appearance) have gain worldwide interest and have become a trend (Molyneaux, 2015).

All in all, consumers who are health-oriented are more sensitive to the health primes marketers use to sell their protein enhanced products. Therefore the first Hypothesis will be: H1: It is more likely that the consumer with a high interest in health-orientation will rate the high protein bar as more healthily, and will pick the high protein bar over a low protein alternative, regardless of the sugar content, compared to a consumer with a low interest in health orientation.

It is found that consumers with an interest in diets and health are keener to reading the nutritional information labels (Cowburn & Stockley, 2005) and consumers who read the nutritional information labels make healthier food choices at point-of-purchase (Ollberding et. al., 2010). However, when consumers recognize a brand as being healthy, they won’t feel the necessity to read the nutritional label and assume the product as being entirely healthy, even if it’s not (Anker, Sandøe, Kamin & Kappel, 2011). An interesting solution is a traffic light label, this type of labelling will reflect the sugar content by colour, where yellow stands for a low amount of sugar (<5%), orange for a medium amount of sugar (5%≤15%) and burgundy red for a high amount of sugar(>15%). Whenever a traffic light label is found on the front packaging of a product, a consumer is more conscious about the nutritional values of the given product (Sonnenberg et. al., 2013). Traffic light labels are easily understood and

(10)

effect between the dependent and independent variable as can be seen in the figure above. Therefore, the second hypothesis will be:

H2: Traffic light labelling will moderate the effect (consumers with a high interest in health-orientation are more likely to pick high protein bar over a low protein bar, regardless of the sugar content) and will make consumers aware of the sugar content of a protein bar.

Study 1: The Health Halo around protein.

The first study will test our first hypothesis, whether health-oriented consumers are more sensitive to the health halo around protein, because they are more interested in the effects of protein on either their weight loss or muscle recovery (Molyneaux, 2015).

Interestingly, these health-oriented customers (those who have a high interest in general health) tend to rate products with a positive attribute (for example, low fat cheese) as being more healthily compared to consumer with a low interest in general health (Roininen et. al., 1999). These health oriented consumers thank their healthy eating habits to their

motivation for maintaining a good health and a good physical appearance (Hayes & Ross, 1987).

Therefore, consumers with a positive attribute toward general interest in health and a general interest in fitness, will be more prone to health primes due to guilt and

self-presentation (Chandon & Wansink, 2007). The reason for this, is the justification effect of the healthy attribute (protein), which will reduce their feeling of guilt and will make the protein bar feel more as a utilitarian product than a hedonic one (Okada, 2005).

In this study we are going to test if consumers with a high interest in health-orientation, will estimate protein bars lower in sugar compared to consumers with a low interest in health-orientation. Furthermore, we investigate whether consumers with a high interest in health will rate protein bars with a high amount of protein lower in sugar than consumers with a low interest in health. The results will provide us with the answer if a consumer with a high interest in health will categorize the high protein option as more healthily compared to the low protein alternatives, regardless of the sugar.

Method

(11)

approach these sportive and health-oriented individuals to reach the 50% mark of respondents with a high interest in health. The participants were randomly assigned to one of the available four control groups, in order to keep the results as generalizable as possible, presented below:

(IV) Protein displayed

(IV) Amount of Protein Protein displayed Protein not displayed High Amount Protein High Protein / displayed High Protein / not displayed Low Amount Protein Low Protein / displayed Low Protein / not displayed

The study used a 2x2x2 between subject design (three-way ANOVA), the respondents were randomly assigned to one of the 8 conditions, Figure 2 present the 4 condition for either consumers with a low interest in health and consumers with a high interest in health. The independent variables are amount of protein, protein displayed and interest in health to test whether the dependent variable (amount of sugar guessed) can be explained, and whether a high level of protein displayed on the packaging makes consumers with a high interest in health-orientation more likely to estimate the sugar content lower compared to consumers with a low interest in health-orientation. The results of the survey will be analysed in SPSS, to get detailed insights in the differences among the conditions.

To test whether the participants had a low or a high interest in health orientation, they had to answer two sets of statements at the end of the survey. These were asked at the end of the survey, to decrease the chance that the participant would understand what the true

meaning of the study was investigating, in order to get more reliable results. The interest in general health among the participants was tested with the statements presented by Roininen et. al. (1999) (Appendix 3). These eight statements were scored from ‘strongly disagree’ to ‘strongly agree’ on a 7-point Likert scale. The interest in exercise and protein was tested with six statements asking the participants interest in either exercise or protein consumption (Appendix 4), these statements were also scored on a 7-point Likert scale ranging from ‘strongly disagree’ to ‘strongly agree’. Some of the statements in both general health interest and interest in exercise and protein were reversed (the statements 1, 3, 7 & 8 are reversed for general interest in health and the statements 2 & 5 were reversed for interest in exercise and protein), to make sure the participant was paying attention to the questions and is answering it truthfully.

The participants were also tested on their understanding of protein. Three statements were formed to test if the participants are indeed prone to the health halo that protein is healthier. Three statements were formed (Appendix 5), which were scored on a 7-point Likert scale ranging from ´strongly disagree´ to ´strongly agree´, where one statement was reversed (The more protein a product has, the more sugar it has) to check whether the participant was paying attention. Furthermore, the participant was asked a multiple choice question why they consume protein bars with six options to choose from (‘I never eat protein bars’, ‘I eat them as a pre-workout snack’, ‘I eat them during my workout’, I eat them as post workout for recovery’, ‘I eat them as a healthy protein enriched snack’ and ‘I eat them as meal replacements for my diet’), in order to understand the need for protein bars among the respondents.

(12)

All participants received a Qualtrics survey which would take approximately 5-10 minutes to complete. The participants were asked to estimate the grams of sugar in each protein bar, the actual weight of the bars was given in all four conditions, in order to make more realistic estimations. Afterwards the actual grams of sugar are subtracted from the guessed grams of sugar, to see how many grams of sugar the participants under- or

overestimated. It is expected that the estimated grams of sugar will be lower whenever the grams of protein are displayed compared to the condition where the protein content is not displayed. It is also expected that the respondents with a high interest in health will estimate the grams of sugar lower than the respondents with a low interest in health. In addition, it is expected that the health halo will have a stronger effect on the protein bar high in protein compared to the bars low in protein.

Interest in Health Orientation

A Cronbach’s alpha was conducted for the 8 questions of the first set of Likert type questions: whether there is a general interest in health. The internal consistency of the questions is tested to check if they are all reliable in measuring the scale general health interest. In the Reliability Statistics the Cronbach’s Alpha for the questions of the scale general health interest is .722 (Appendix 6), a reliability coefficient of at least 0.7 can be considered as having a high

internal consistency. Even though deleting the item reverse Q26_1 could lead to a Cronbach’s Alpha of .741, and deleting the item: ‘it is important to me that my diet is low in fat’ could lead to a Cronbach’s Alpha of .748, these improvements are minor and the Cronbach’s Alpha is already above the needed threshold (Appendix 7), therefore deleting the items is a loss of information.

The second set of Likert type questions are concerning whether there is an interest in sports and thus protein. We want to check whether there is internal consistency between these questions and if these are suitable to transform into a scale for interest in exercise and protein consumption. The Cronbach’s Alpha for these questions is .702 (Appendix 8), which is just slightly above the threshold of 0.7 and therefore there is a high internal consistency between the items. Furthermore, if one of the questions was removed, the Cronbach’s Alpha would be below 0.7, therefore none of the questions should be removed (Appendix 9).

(13)

Sugar estimations:

The respondents had to guess the sugar content of four different protein bars. For the groups who had bars low in protein the following bars were chosen: NatureValley (high in sugar), Atkins (low in sugar), Eatnatural (medium in sugar) and Trek (high in sugar). For the groups who has bars high in protein the following bars were chosen: Vifit (high in sugar), Barebells (low in sugar), QNT (medium in sugar) and Mars (high in sugar). The bars are combined so that the sugar content is either low, medium or high for the low protein and high protein options. So this won’t influence the results. Therefore, NatureValley & Vifit, Atkins & Barebells, Eatnatural & QNT and Trek & Mars are combined as a dependent variable. On every couple of protein bars (NatureValley & Vifit etc.), we are going to perform the three-way ANOVA, to understand whether the independent variables influence the dependent variable, and if the effect is similar for all four sets of protein bars.

Analysis three-way ANOVA:

Before we can perform the three-way ANOVA for the three independent variables (amount of protein; high or low, and protein displayed; yes or no, interest in health; yes or no) on our dependent variable (how much the consumer under- or overestimates the sugar content), we should check if our data is suitable for carrying out this analysis. The raw data collected in the Qualtrics survey should meet certain criteria before we can expose it to a three-way ANOVA. There are six assumptions which should be met: the dependent variable should be continuous, the levels of each independent variable should be categorical, there should be independence of observations, there should not be any significant outliers, the dependent variable should be normally distributed and there needs to be homogeneity of variance.

The first three assumptions are already met. Firstly, our dependent variable is continuous, because it’s a ratio variable, whenever a consumer overestimates the sugar content with 12 grams, we know it’s 10 grams more than a overestimation of 2 grams, and an overestimation of 10 grams is twice as much as an overestimation of 5 grams, also the variable has an absolute zero, because zero tells us that there is no over- or underestimation. Secondly, our independent variables are dichotomous, they all have two categories; high and low for amount of sugar, yes or no for protein displayed and yes or no for interest in health. Thirdly, our observations are independent, since all respondents were completely at random assigned to one of the eight conditions, therefore, none of the respondents can be on more than one condition. The last three assumptions should be tested for, before we are able tell if they are met, which will be discussed thoroughly below.

Assumption 4: outliers:

We will use the outlier labelling rule to identify if there are any outliers in our dataset. We will use the mathematical construct presented in the article of Hoaglin & Iglewicz (1987). The quartiles are used together with a factor (k) of 2,2, since this number is more suitable for the number of observations in our study (Hoaglin & Iglewicz, 1987). The formula for the Lower (Q1) and Upper(Q3) bound will be:

(14)

Since we have tested our first study on four different protein bars (respondents had to guess the amount of sugar for four different sets of protein bars), we have to check the possible outliers for all four questions. Q1 is the 25th percentile and Q3 is the 75th percentile (Appendix 10).

Question 1, NatureValley & Vifit:

Lower bound: -3,95 – 2,2(5,8-(-3,95)) = -25,4, Upper bound: 5,8 + 2,2(5,8-(-3,95)) = 27,25. As we can see in appendix 10, there is no value lower than -25,4, but there are two values higher than 27,25, case 116 has a value of 28,7 and case 36 has a value of 63,8. If we look at the original data of case 36, we can see that this respondent guessed that there were 70 grams of sugar in the protein bar, while the weight of the protein bar is 40 grams, the respondent guessed for three out of four question a number of grams higher than the actual weight of the bar, therefore the responses are impossible and useless and the best solution to deal with this respondent is to delete the response. The same accounts for case 116, this respondent had the condition of displayed protein, and guessed the sugar content nearly as high as the weight of the bar, while there was stated that the bar also contained a high amount of protein (between 18 and 20 grams), therefore this answers are impossible and can therefore be treated as an outlier and deleted so they will not influence the results of the three-way ANOVA.

Question 2, Atkins & Barebells:

Lower bound: 4,3 – 2,2(17,7-4,3) = -25,2, Upper bound: 17,7 + 2,2(17,7-4,3) = 47.2. As we can see in Appendix 10, none of the higher or lower values fall out of these boundaries, however case 76 seems to have a pretty large overestimation, taking a look at the original data the respondent only guessed the sugar amount this high for this particular protein bar, and we should leave the case at it is.

Question 3, Eatnatural & QNT:

Lower bound: 1,8 – 2,2(15,3-1,8) = -27,9, Upper bound: 15,3 + 2,2(15,3-1,8) = 45. As we can see in Appendix 10, none of the values violates the boundaries, however case 95

overestimated the sugars in a protein bar with 43,3, which is extremely high. Taking a closer look at the original data, this respondent guessed the sugar content of two bars nearly as high as the weight of the bar and for one bar higher than the weight of the bar, therefore these responses are impossible and the best solution to deal with this case is to delete it from the dataset.

Question 4, Trek & Mars:

(15)

Assumption 5: normality:

The sugar estimations of most of the groups were normally distributed (p>.05), except for three groups: (1) Protein bars with a low amount of protein, protein not displayed and evaluated by a consumer with a low interest in health orientation (p=.009), (2) Protein bars with a low amount of protein, protein not displayed and evaluated by a consumer with a high interest in health orientation (p=.038), (3) Protein bars with a high amount of sugar, protein displayed and evaluated by a consumer with a low interest in health orientation (p=.013), assessed by Shapiro-Wilk’s test of normality (Appendix 11).

We will try to transform the data in order to have all group normally distributed. The three groups which are not normally distributed show a positive skewness (Appendix 12), we can transform the data with a log10 or square root, however our dependent variable has negative values, therefore we should add the maximum negative value (this is -12.3 in our data) to all the data plus 1, so that the minimum number is 1 and we are able to perform a log10 or square root. With the square root we still had one group with a non-normal distribution and with the log10 we have two groups left with a non-normal distribution (Appendix 13), therefore we use the square root for our transformation on which we are going to run test comparisons and investigate whether this gives us more reliable results compared to the original data with more non-normal distributed groups.

The sugar estimations of six out of eight groups were normally distributed (p>.05). Protein bars with a low amount of protein, protein not displayed and evaluated by a consumer with a low interest in health orientation (p=.029) & protein bars with a low amount of protein, protein displayed and evaluated by a consumer with a high interest in health orientation (p=.038) are not normally distributed, assessed by Shapiro-Wilk’s test of normality (Appendix 11).

We will try to transform the data of the above mentioned groups who violate the assumption of normality. The group protein bars with low amount of protein, protein not displayed and evaluated by consumer with a low interest in health orientation shows a high level of positive skewness (skewness = 1.004), but the variable protein bars with low amount of protein,

protein displayed and evaluated by a consumer with a high interest in health orientation shows a strong kurtosis which is light tailed (kurtosis = -1.468). Firstly, we have to get rid of the negative values of the dependent variable, before we can apply any transformation, therefore we add 3,3 to every value (the lowest value is -2,3 +1 results in a positive value of 1). The first transformation (log10), resulted in 3 groups violating normality, assessed by Shapiro-Wilk’s test of normality, therefore this transformation is not worthwhile. When we transform the data by using the square root, we have normality for all group according to the Shapiro-Wilk’s test, with 1 group having a p=0.05, thus having the exact value of our significance level.

(16)

with a low amount of protein, protein not displayed and evaluated by a consumer with a high interest in health violates normality (p=0.005), assessed by Shapiro-Wilk’s test of normality (Appendix 11).

We will try a transformation of the data in order to receive normality for all groups. This group shows a very large level of positive skewness (=1,897) and kurtosis (=4,096). In order to be able to transform the data, we have to get rid of the negative data (the underestimations), therefore we add the lowest value + 1 to every value (lowest value = -7), so that the lowest value becomes 1 and we are able to perform a square root. After performing a square root transformation, all group are normally distributed.

The sugar content of six out of eight groups were normally distributed (p>.05). Protein bars with a low amount of protein, protein not displayed and consumer with a low interest in health (p=0,048) and protein bars with a low amount of protein, protein displayed and consumer with a low interest in health (p=0,019) are both not normally distributed as assessed by Shapiro-Wilk’s test of normality (Appendix 11).

Both of the groups which are not normally distributed show a moderate of high level of positive skewness (=0.720 & =1,185 respectively). In order to be able to perform a

transformation, the variable should consist of values ≥ 1, therefore we have to add the largest minimum value + 1 (lowest minimum value = -13). After performing a square root

transformation, all groups are normally distributed.

Assumption 6: homogeneity of variances

Whenever we use the untransformed original data, there is a homogeneity of variance, according to the Levene’s test for equality of variances, p=0.546, this indicates that the Levene’s test is not significant (p>.05). However, when we use the transformed (square root) dependent variable, there is no homogeneity of variance, since p <.05, because p=.031. Since the transformation of the data results in a violation of homogeneity of variance and since ANOVA is considered to be robust to non-normal distributions, especially when the groups violating normality are similarly skewed (they are all positively skewed, appendix 12), we will run the test regardless of the violation of normality for 3 out of the 8 groups, to also retain the clarity of the original data.

For both the transformed variable as the original variable, there was homogeneity of variance, according to the Levene’s test for equality of variances, p=0.472 (the original variable) and p=0.991 (the transformed variable sqrt).

(17)

There is homogeneity of variance, assessed by Levene’s test for equality of variances, p=0,076 (the original variable) and p=0,152 (the transformed variable sqrt).

Results: Question 1, NatureValley & Vifit:

There is no statistically significant three-way interaction between our independent variables amount of protein, protein displayed and interest in health, F(1,126)=0.988, p=0.322. Furthermore, there are no statistically significant two-way interaction either, for amount_of_protein*protein_displayed: F(1,126)=0.361, p=0.549, for

amount_of_protein*interest_in_health_combined: F(1,126)=0.004, p=0.953, and for protein_displayed*interest_in_health_combined: F(1,126)=0.011, p=0.915 (Appendix 14). Question 2, Atkins & Barebells:

There is no statistically significant three-way interaction between our independent variables amount of protein, protein displayed and interest in health, F(1,125)=0.911, p=0,342. In addition, there are no statistically significant two-way interaction among the independent variables: amount_of_protein *protein_displayed: F(1,125)=0,518, p=0,473,

amount_of_protein*interest_in_health_combined: F(1,125)=0,161, p=0,689 and

protein_displayed*interest_in_health_combined: F(1,125)=1,118, p=0,292 (Appendix 14). Question 3, Eatnatural & QNT:

There is no statistically significant three-way interaction F(1,99)=0,587, p=0,445. Furthermore, there are no statistically significant two-way interactions:

amount_of_protein*protein_displayed: F(1,99)=0,023, p=0,879,

amount_of_protein*interest_in_health_combined: F(1,99)=0,639, p=0,426,

protein_displayed*interest_in_health_combined: F(1,99)=2,075, p=0,153 (Appendix 14). Question 4, Trek & Mars:

There is no statistically significant three way interaction F(1,125)=0,459, p=0,499. Furthermore, there are no statistically significant two-way interactions:

amount_of_protein*protein_displayed: F(1,125)=0,23, p=0,632,

amount_of_protein*interest_in_health_combined: F(1,125)=0,721, p=0,397,

protein_displayed*interest_in_health_combined: F(1,125)=0,034, p=0,853 (Appendix 14).

Overall result study 1:

Since none of the three-way interactions for all combinations of protein bars were statistically significant, we have to reject our first hypothesis. There is no statistically significant

difference between consumers with a high interest in health and a consumer with a low interest in health on their estimations of the sugar content and there is no statistically

significant difference between the protein being displayed or not and whether the protein bar had a high or low amount of sugar.

(18)

disagreed upon. For the statement ‘if a product contains more protein it’s healthier’, only 16.5% agreed, 6.1% agreed to the statement ‘by eating protein I automatically gain muscle’ and 6.1% agreed to the statement ‘the more protein a product has, the more sugar it has’. This may indicate that the respondents, regardless of their interest in health-orientation (and in spite of the fact that 58.6% of respondents replied they never eat a protein bar), do have a good understanding of the effects of protein and the different condition (high or low interest in health) may differ too little from one another.

Study 2: The moderating effect of traffic light labelling.

In this study we will test whether traffic light labelling will moderate the effect of health halo around protein and provide answers to our second hypothesis. Ollberding et. al. (2010), found that consumers who do read the nutritional information label on food and beverage items tend to generally have a healthier consumption than those who don’t. However, reading the

nutritional information label sounds easier than it is in fact, with many consumers not being able to understand the information provided (Cowburn & Stockley, 2005). A high interest in, and knowledge of, healthy foods is needed in order to process this information and understand whether a food item fits in an everyday diet (Cowburn & Stockley, 2005). The traffic light labelling is invented to make the nutritional information more easily interpretable, since a quick glance at the packaging will inform the consumer is a second (Brownell & Koplan, 2011). Thorndike et. al. (2012), proved the effect of traffic light labelling in their study, after implementing the coloured labels, the purchases of healthy items increased and the purchases of unhealthy items decreased. In addition, traffic light labelling will result in a healthier decisions at point-of-purchase (Sonnenberg et. al., 2013). All in all, the results of previous studies have provide very promising content to build this study on.

Method

This study is an addition to the previous study presented to the respondents in Qualtrics, therefore the same pool of participants is used, together with the same statements measuring the general interest in health, the interest in exercise and protein and their understanding of proteins and the reason why they consume protein bars.

We will test whether the implementation of a traffic light label can intervene between the protein presented on the packaging and the consumer with a high interest in

health-orientation being more likely to rate a high protein bar as more healthful, thus the traffic light label works as a moderator. Therefore, will a traffic light label make a consumer aware of the sugar content of that protein bar. This study uses a Multinomial Linear Regresion between subjects design and the results will be analysed using SPSS.

In the Qualtrics survey, the respondents were randomly assigned to either one of the following conditions: the traffic light being displayed or it not being displayed. The

(19)

normally done in store or online. The participants were given the weight of the bars, the grams of protein and the grams of sugar for each protein bar. In order to solely investigate the effect of the traffic light label being present or not. The participants had to pick their favourite options out of three individual sets of six options (Figure 3).

Nutritional value Question 1 Question 2 Question 3 High protein/low sugar Fulfil Carbrite Mission High protein/medium sugar Oh yeah Cake bites Combat High protein/high sugar So good Xxldelicious Mars

Low protein/low sugar Atkins Smartbar Mealpro

Low protein/medium sugar Supreme Nine Eatnatural

Low protein/high sugar Primal Bounce Isostar

The respondents were asked after each set of six options to tell if they had tasted the bars, in order to investigate whether the picked option is a result of past purchases of the brand or entirely because of the protein and or sugar content.

It is expected that whenever the traffic light is presented, merely all respondents will pick the high protein/low sugar, because of the health halo of protein together with the least amount of sugar.

All the respondents in both condition (traffic light displayed and not displayed) knew the protein content and the accompanying sugar content of each protein bar they had to choose from. This was presented to keep ‘all other things equal’, thus to be able to examine the true effect of the traffic light labelling.

Analysis Multinominal Logistic Regression

Since our dependent is categorical instead of continuous, we are not able to perform a ANOVA for this study. The dependent is Nominal, because the respondents have to choose their favourite protein bar out of six options: high protein/high sugar, high protein/medium sugar, high protein/low sugar, low protein/high sugar, low protein/medium sugar and low protein/low sugar (Figure 3). We want to investigate whether the implementation of a traffic light label influences the choices being made and whether that differs for consumers with a low or high interest in health orientation.

To be able to understand the different effects of our independent variables (interest in health orientation & traffic light displayed) on our categorical dependent variable (protein bar selected), we should perform a Multinominal Logestic Regression.

Prior to the Multinominal Logestic Regression, our dataset should meet six assumptions, to investigate whether it is appropriate for this dataset to use this analysis and provide us with valid results. The six assumptions are the following: the dependent variable should be

nominal, there are one or multiple independent variables that are either continuous, ordinal or nominal (dichotomous independent variables are also allowed in this analysis), there should be independence of observations, there should be no multicollinearity, any continuous independent variable should have a linear relationship with the logit transformation of the

(20)

dependent variable, high leverage values & highly influential point and outliers are not allowed.

All assumptions are met, except for the fourth assumption for which we have to run an analysis, since our dependent variable is nominal (as previously explained), our independent variables are both dichotomous (interest in health: yes or no, and traffic light displayed: yes or no), there is independence of observations, since the respondents are completely at random assigned to either the condition where traffic light was displayed or to the condition where it was not displayed, we don’t have any continuous independent variables (just dichotomous independent variables representing yes or no) and there are no outliers for these dichotomous variables (± 50% of the respondents belong the either of the two possibilities for both the independent variables).

Assumption 4: multicollinearity

In order to test whether there is multicollinearity between de independent variables, we are going to run a linear regression, and test it two times. Firstly, we insert one of the independent variables in the dependent variable box and the other independent variable in the independent box, when we check the collinearity diagnostics and run the analysis we get a VIF of 1 (Appendix 15), meaning that there is no multicollinearity since it is below the threshold of 3. Secondly, we insert the dependent variable in the dependent variable box and both of the independent variables in the independent box and run our linear regression, when we run the analysis we have a VIF of 1,012, which indicates that there is no multicollinearity (Appendix 15).

Results

The respondents had to pick their favourite protein bar out of the six possible options, for three questions in a row. Therefore, we are able to check whether the independent variables do have an effect on the protein bar picked, for different sets of protein bars. In this second study, we want to investigate whether the protein bar a consumer decides upon, can be predicted by the presence of traffic light labelling and the interest in health orientation. When performing our Multinominal Logistic Regression, we are going to test it for all three

questions. Question 1:

In the first questions, the respondents had to pick their favourite out of six protein bars: oh yeah (high protein/medium sugar), fulfil (high protein/low sugar), so good (high protein/high sugar), supreme (low protein/medium sugar), atkins (low protein/low sugar) and primal (low protein/high sugar). In figure 4, we collected the responses of all participants regardless of the condition they were placed in.

Protein bar Frequency Percentage Tried the bar

Oh yeah (high protein/medium sugar)

29 21.8% 12%

(21)

So good (high protein/high sugar) 8 6% 6.8% Supreme (low protein/medium

sugar)

3 2.3% 3%

Atkins (low protein/low sugar) 11 8.3% 17.3% Primal (low protein/high sugar) 5 3.8% 4.5%

Figure 4 shows that most of the participants have picked the high protein alternatives over the low protein alternatives (85,7%) and only 10% of respondents picked a high sugar option and most of the respondent picked the Fulfil bar, even though less than 10% has tried this bar, which indicates that this is probably not a result of brand recognition. However, were the independent variables able to predict this outcome?

With p=0.170, the model is not statistically significant and therefore the full model does not predict the dependent variable. Moreover, a statistically significant result (p<0.05) for the Pearson chi-square statistic indicates that there is a poor fit for the model and our p=0.031. In addition, the variance explained by the model is also very poor, the Pseudo R-Square for Nagelkerke=0.110. Lastly, both of our independent variables were not statistically significant: interest in health, p=0.069 and traffic light displayed, p=0.539.

When running the analysis, a warning occurred that not all dependent cells had frequencies for all the options, because the low frequencies on mainly the low protein bars. Therefore, the protein bars were added as one category on the dependent variable, to test whether we get more information if we decrease the dependent variable to 4 categories: oh yeah (high protein/medium sugar), fulfil (high protein/low sugar), so good (high protein/high sugar) and low protein (combination of low, medium and high sugar = 19 responses).

This alteration made the model statistically significant: p=0.044, however, the Pearson chi-square statistic still indicates a poor fit (p=0.011) and the variance explained by Nagelkerke is still low (=0.104). The independent variable interest in health is statistically significant

(p=0.019) and traffic displayed is still not statistically significant (p=0.375). The only coefficient (Appendix 16) which is statistically significant indicates that it is less likely that consumer with a low interest in health compared to a consumer with a high interest in health will pick fulfil (high protein/low sugar) over low protein alternatives (p=0.026).

Question 2:

In the second question the options were xxldelicious (high protein/high sugar), Carbrite (high protein/low sugar), Cakebites (high protein/medium sugar), Bounce (low protein/high sugar), Smartbar (low protein/low sugar) and Nine (low protein/medium sugar). The frequencies are reported in Figure 5 below.

Xxldelicious (high protein/high sugar)

18 13.5% 14.3%

(22)

sugar)

Bounce (low protein/high sugar) 5 3.8% 0% Smartbar (low protein/low sugar) 20 15% 44.4% Nine (low protein/medium sugar) 5 3.8% 0.8%

Figure 5 gives insight that most respondents picked the high protein alternatives (77.5%) and only 17.3% of respondents picked a high sugar option regardless of the conditions. Equally to the first question, most of the respondents picked the Carbrite bar, which less than 5% of respondents are familiar with, therefore picking this bar is probably not a result of brand recognition.

The full model, again, is not improved by the independent variables compared to the intercept alone, with p=0.286 it is not statistically significant. In this model the data fits well (p>0.05), assessed by the Pearson chi-square statistic: p=0.338, but there is a very small proportion of the variance explained by the model (Nagelkerke=0.097). Both of the independent variables are not statistically significant: interest in health (p=0.347) and traffic light displayed

(p=0.271), resulting in none of the coefficients being statistically significant.

The same warning occurred and therefore we tried to decrease in categories for the dependent variable again, by adding all low protein alternatives together to one category: low protein, consisting of 30 responses. Unfortunately the model does not become statistically significant (p=0.352), both of the independent variables were not statistically significant: interest in health (p=0.358) and traffic light displayed (p=0.398).

Question 3:

In the last question the options consisted of Mission (high protein/low sugar), Mars (high protein/high sugar), Combat (high protein/medium sugar), Mealpro (low protein/low sugar), Isostar (low protein/high sugar) and Eatnatural (low protein/medium sugar). The frequencies are presented in Figure 6 below.

Mission (high protein/low sugar) 56 42.1% 4.5% Mars (high protein/high sugar) 26 19.5% 19.5% Combat (high protein/medium

sugar)

15 11.3% 6%

Mealpro (low protein/low sugar) 15 11.3% 3%

Isostar (low protein/high sugar) 4 3% 9.8%

Eatnatural (low protein/medium sugar)

17 12.8% 43.6%

Figure 6 gives insight that most respondents picked the high protein alternatives (72.9%) and 27.1% of respondents picked a high sugar option regardless of the conditions. The same accounts for this question as for the previous 2, we assume that the option the respondents picked is not a result of brand recognition, since the bar with the most frequencies has one of Figure 5: respondents picks for question 2

(23)

The model for the last question is statististically significant (p=0.01), however the fit of the model is poor (p=0.002) assessed by Pearson chi-square statistic and the variance explained is also very low (Nagelkerke=0.168). Furthermore, the independent variable interest in health is statistically significant (p=0.002), but traffic light displayed is not (p=0.567).

The following coefficients are statistically significant: it is less likely that a consumer with a low interest in health compared to a consumer with a high interest in health will pick a mission (high protein/low sugar) or a mars (high protein/high sugar) or a combat (high protein/medium sugar) or a mealpro (low protein/low sugar) over a Isostar (low protein/high sugar) bar (for all options p=0.000), it is more likely that a consumer with a low interest in health compared to a consumer with a high interest in health will pick a Mars (high

protein/high sugar) over a Combat (high protein/medium sugar) bar (p=0.017), it is less likely that a consumer with a low interest in health compared to a consumer with a high interest in health will pick a Mission (high protein/low sugar, p=0.004) or a Combat (high

protein/medium sugar, p=0.017) over a Mars bar.

Even though we had a statistically significant result, we decided to also reduce the dependent to 4 categories as explained in the previous questions, to reduce the warnings of some levels of the dependent variable not having any frequencies. All the low protein categories were put together, consisting of 36 responses.

After merging the category the model remains statistically significant (p=0.007), with the independent variable interest in health being statistically significant (p=0.002) and traffic light displayed not being statistically significant (p=0.720). The following coefficients are

statistically significant: It is more likely that a consumer with a low interest in health compared to a consumer with a high interest in health will pick a Mars (high protein/high sugar, p=0.004) or a Low Protein bar (p=0.005) over a Mission bar (high protein/low sugar), it is less likely that a consumer with a low interest in health compared to a consumer with a high interest in health will pick a Combat (high protein/medium sugar, p=0.017) over a Mars (high protein/high sugar) bar and it is more likely that a consumer with a low interest in health compared to a consumer with a high interest in health will pick a low protein bar (p=0.024) over a Combat (high protein/medium sugar) bar.

Overal result:

We have to reject our second hypothesis, since there is no statistically significant result for any of the 3 question regarding traffic light labelling. There are only some statistically

(24)

Discussion

Our findings suggest that there is no statistically significant difference between consumers with a high interest in health-orientation compared to those with a low interest in health orientation in their ability to rate the healthiness of a protein bar when the protein is present on the packaging or not and there is no statistically significant difference in estimating the healthiness of a protein bar when it is high or low in protein. In addition, the study doesn’t find any statistically significant proof that traffic light labelling can moderate the effect of the health primes. Simplified, our findings report that the health halo around protein does not make the consumer with a high interest in health-orientation more likely to believe that a high protein bar is healthier than a low protein bar and that these consumers with a high interest in health estimate the sugar content equally as high as consumers with a low interest in health. Furthermore, our study does not find a difference in results when the traffic light labelling is present for either consumers with a low and high interest in health-orientation.

Study 1 of this research provides findings that are in contrast of what we assumed, consumers with a high interest in health are equally aware of the sugar content of the protein bars as consumers with a low interest in health and they do not believe that the protein bar higher in protein is in fact healthier. In the second study our findings were also in contrast with our assumptions, we assumed that the traffic light label will drastically affect the choice of the consumer, but there was no statistically significant proof for this, because either consumers with a low and high interest in health already picked the ‘healthier’(higher in protein and/or lower in sugar) options. The only finding that has some statistical significance, is that consumer with a high interest in health-orientation were more likely to pick the high protein/low sugar option over options higher in sugar and/or lower in protein, but they were more likely to pick these high protein/low sugar options in either condition of the traffic light interference, thus consumer with a high interest in health-orientation are more knowledgeable about the protein/sugar ratio than expected.

Our results are in contrast with past findings, where Chandon & Wansink (2007) found that the health claim of a restaurant resulted in a 35% lower calorie estimations, compared to a meal with the same calorific content from a restaurant which is perceived as unhealthy, in our study, the health halo around protein does not make the respondents

estimate the sugar content lower for a high protein bar compared to a low protein bar. But our findings are also equal to Chandon & Wansink (2007), they found that the health halo bias is equally as strong for consumers with a low or a high interest in nutrition, which we also found in this study. Sonnenberg et. al (2010), provided findings that traffic light labelling leads to healthier options, however our findings did not find any statistically significant proof for this. In addition, Ollberding et. al., provide findings that when the nutritional labelling of a product is used, people tend to have a healthier consumption, our findings support this, because all respondents were given the sugar and protein content of each bar in study 2 and a very low percentage of respondents picked the high sugar options.

(25)

insignificant results. However, our study does provide evidence that consumer with a high interest in health-orientation are more likely to pick the healthier option (high protein/low sugar) over other options compared to consumer with a low interest in health-orientation.

Limitations and suggestions for further research

Our study’s respondents consisted of young people (average age of 29) and all were believed to have had some form of education, literacy is needed to read nutritional labels and this will probably have caused that the respondents have a better understanding of the nutritional value of the protein bars. For further research it is suggested to have a more diversified pool of participants, with respondents being older and/or less educated, to understand whether there is a lower understanding of nutritional labels among those respondents.

In addition to the previous limitation, nothing was asked about the respondents’ understanding of nutritional labels. Do they use these on an everyday basis, do they refuse to buy products when they believe the nutrients are not optimal, or do they relentlessly buy and eat everything they crave. It is suggested to add some questions to the questionnaire about the ability to read nutritional labels, because it might have been the case that all of the

respondents in this study were very keen on reading the nutritional label and therefore, were more informed beforehand, resulting in these insignificant results. Furthermore, maybe our respondents with a high interest in health were to knowledgeable when it comes to protein bars, so that they are aware of the sugar content of the bars and are thus not subjected to the health halo. It is probably a good suggestion to have more respondents who are have a high interest in health, but a low knowledge about health. For example, respondents who just started their weight loss journey, respondents who just started their muscle gain program, all in all respondents new to the healthy lifestyle, but with a very high interest in a healthy lifestyle.

The number of subjects should have been larger, when analysing the results in the multinomial logistic regression, some of the levels of the dependent variable had too little or even zero frequencies, resulting in us not being able to investigate those relationships. With a much larger respondents pool, the likelihood of having enough frequencies for every level of your variables increases drastically.

In the study two, there should have been 4 condition, to investigate whenever the sugar content and protein content of each bar was not given, the responses would have been

different. It could have been a possibility, that if these numbers were not given, the traffic light label would have made a bigger impact, because the sugar and protein content was given, it was more likely to pick the high protein/low sugar option for both consumers with a low and high interest in health-orientation. Having four conditions, results in investigating whether the traffic light labelling would have had more effect if less nutritional information was available.

In our study we only asked the respondents to estimate the sugar content of protein bars, resulting in all respondents being subjected to the health halo around protein, which may have been the cause of our results being insignificant. It would be suggested for future

(26)

Our study was conducted via a Qualtrics survey, it is probably more reliable to do a field study when investigating the usefulness of a traffic light label. In a field study in a sport supplement shop you are able to observe the behaviour of the respondent: whether they read the nutritional labels and if healthier options are made at point-of-purchase when a traffic light label is present. Because the nutritional information was written down in the question, the respondents are more likely to read this information compared to the nutritional label on the back of a protein bar.

Conclusion

The obesity paradox is a very interesting occurrence, there is a rising global interest in fitness and health, but simultaneously there is a growing obesity epidemic and other nutrition related diseases. The downside is that marketers become more knowledgeably in responding to the upcoming trend of a healthy lifestyle and are thus are increasing their sales and profits with their protein enriched products, while on the other hand it becomes harder for consumers to know the true nutritional value of a product with all the, sometimes misleading, health claims around. Due to these health claims, certain products or ingredients generate a health halo, which results in consumers neglecting the unhealthy attributes, such as high amount of sugar, salt and unhealthy fats.

The health halo around protein is believed to have a bigger effect on consumers with a high interest in health-orientation, however the findings in our study are not statistically significant for these assumptions. Furthermore, a traffic light label should moderate the effect of the health halo around protein, particularly for consumers with a high interest in health-orientation, but the findings for this statement were also not statistically significant. This study does provide proof that consumers who have a high interest in health-orientation are more likely to pick a healthier option (high protein/low sugar) over some options lower in protein and/or higher in sugar compared to a consumer with a low interest in health-orientation.