Размещено на http://www.allbest.ru/

Размещено на http://www.allbest.ru/

FEDERAL STATE EDUCATIONAL INSTITUTION

OF HIGHER EDUCATION

NATIONAL RESEARCH UNIVERSITY

HIGHER SCHOOL OF ECONOMICS

Saint Petersburg School of Economics and Management

Department of Management

Determinants of satisfaction with meals: linking recipe ratings to nutritional information

Bachelor's thesis

In the field 38.03.02 Management

Educational programme `Management'

Reviewer

Head of Information and

Analytical Department of

Center of Spatial Research LLC

Pustovalova E.A.

Supervisor PhD in Economics Antipov E.A.

2019



Abstract

Researching determinants of satisfaction with meals is relevant due to the fact that several researches showed how satisfaction with meal affects satisfaction with life, work (Haugaard, Stancu, Brockhoff, Thorsdottir, & Lahteenmaki, 2016). US army study how differs results of soldiers depending on their satisfaction with food and what are the determinants of feeling satisfied (Cole, et al., 2018). At the same time, increasing in index of world food consumption leads to several diseases like obesity (A. Andrews, Lowe, & Clair, 2011). However, we should research field of food consumption more to know how increase people's satisfaction with life, their productivity and heal from diseases. Aim of the research is to provide list with determinants of satisfaction with meals. Determinants will be found using literature analysis to found factors which influence overall satisfaction with meal that was found and analysis of database with information concerning recipe's nutritional information and rating given by visitors of web-site epicurious.com. Epicurious is a web-site which provides recipes of meals for amateur cookers. Purpose of database analysis is to find relationships between nutritional facts of the food and people's satisfaction after intake of it. Firstly, datasets were tested on normal distribution. Secondly, datasets were tested on significant difference of medians. Finally, dependence between variable rating and variables containing nutritional information was found using nonparametric regression because datasets weren't normally distributed, and they had significant statistical difference. In addition, scatterplots of distribution recipe's nutritional information by its' rating to approve several hypotheses. In conclusion, we made a list with determinants of satisfaction with meals and gave recommendation for workers of food business sector and provide information about possible future research in the field of food consumption and nutrition.

Key words: nutrition, satisfaction with meals, determinants, regression analysis.



Introduction

satisfaction meals index food

Consumption of food is a basic need of a person like drinking, sleeping, etc. Index of food consumption is raising each year and brings humanity several diseases like obesity, dyspnea, stroke or heart attack (A. Andrews, Lowe, & Clair, 2011). Diversity in choice of food brings negative effect for people's health and overall satisfaction with life. Although, it makes people create special ratings like Michelin stars for restaurants to know where they can food which give them the best satisfaction. Michelin star for restaurants means that they can make really big prices for meals and have all tables reserved for the next half a year. It evaluates in the rising of restaurant business sector (Vбsquez & Chik, 2015). US army studied how satisfaction with meals affects results of soldiers to know how to improve diet of soldiers (Cole, et al., 2018). Concerning these facts, we can think that studying field of food consumption and how it affects people's life is relevant.

The aim of the research is to specify determinants or factors of satisfaction with meals and give recommendations for people who need to know how they can improve their business using satisfaction with food across consumers or workers. Steps for the research are following:

1. Analyze previous researches to specify determinants of satisfaction with meals that was found.

2. Collect data from epicurious.

3. Make data sets' tests, analysis and visualization

4. Merge results from previous tasks

5. Give recommendations

The research question is following: “What are the determinants of satisfaction with meal of an ordinary person?”. The object is information about recipes' calories, protein, fat, sodium and rating collected from epicurious. The subject is determinants which influencing overall person's satisfaction with meal which evaluates as recipe's rating. We will test three hypotheses. They are following:

1. Majority of the recipes with number of calories more than 1000 will have rating more than 3;

2. Nutritional facts are not the most significant determinants of satisfaction with meal;

3. Majority of the recipes which have 100 grams of protein and more will have rating more than 3.

In terms of structure work will look as following:

1. Introduction;

2. Literature review, where we will find determinants of satisfaction with meals and gap for our research;

3. Methodology, where we give description of data collection and data analysis process;

4. Results;

5. Conclusion.

1. 

1. Literature review

1.1 Relevance of the research

The main goal of the paper is to specify determinants of satisfaction with meals using analysis of database, which is collected from epicurious web-site for amateur cookers. Database includes information about nutritional facts of the food (calories, protein, fat and sodium). Problem of the research rises from the gap. The problem is “How nutritional facts of recipe affects satisfaction with meals, which is measured as a variable rating in the database”. Results of the research could be used in the different spheres of nutritional business. Hospitality management can use our results for improving overall satisfaction with hotel across customers. Restaurants, cafes could improve their menu during lunch. In addition, US army can improve their research using results of ours because field of nutrition for US army is relevant because they have special program which research diet of soldiers and its' effect on productivity of soldiers. “The HPP performance dietitians were successful in making high-quality food choices accessible to soldiers, which was positively associated with nutrition-related behaviors. This program supported the USASOC HPP goals as well as the US Army's Performance Triad and the Chairman of the Joint Chiefs of Staff Total Force Fitness Framework” (Cole, et al., 2018). Our paper will help them with making list of determinants of satisfaction with meal, which in order will help to design diet of US soldiers and rise their productivity. Question of the paper is not in the specifying factors of satisfaction with food of specific person. The main reason for the paper is to find similar determinants of satisfaction with food across big amount of people. It is evidence from the fact that there is lack in researching people's factors of satisfaction across people not divided by geographical position, social status.

Studying food preferences of people always was relevant because people's basic needs are never ending with time. Even though need in food have bad impacts for people, for example obesity (A. Andrews, Lowe, & Clair, 2011). It also helps people to reduce stress (Osdoba, Mann, Redden, & Vickers, 2015). Katie E. Osdoba and others (2015) proved that meals help people to reduce their stress level. They manipulated stress across participants using Trier Social Stress Task. Results showed that not just eating reducing stress but also meal preparation. Participants were divided into four groups depending on control of menu and preparation: with control of a menu and preparation; without control of a menu and with control of preparation; with control of menu and without control of preparation; without control of menu and preparation. Group where people had no control for preparation of the meal and no control for menu showed the best results in reducing stress. It describes why people prefer eating outside their houses, why restaurant sphere of business is growing each year and why Michelin rating was introduced. In addition, restaurants where people have no control on what they are going to eat grow its popularity. Researching further food preferences can not only help restaurants increase their popularity and revenue but also ordinary people can improve their life, for example reduce stress with help of finding in Katie E. Osdoba's research. In addition, food is one of the major sources of human exposure to undesired chemicals (Flynn, et al., 2019). It proves the relevance of the research and that research in the field of interaction between human and food will lead to reducing illnesses and diseases.

Studying impact of satisfied meal is relevant. Agnes Giboreau (2018) claimed that positive emotions increased after the meal while negative emotions remains low before and after meal. Although, earlier research showed that negative emotions increased after meal. Difference explained as dissimilar dinner. Earlier research had design in context of low-quality food. One more research proved limitation that quality of food has one of the biggest impact on satisfaction with meal.

1.2 Determinants of satisfaction with meal

If people are not pleased with the quality of food, it will lead to decrease in number of customers in terms of hotel management. Abraham Pizam and others (2016) in the research called “Customer Satisfaction and its measurement in hospitality enterprise: a revist and update” are proved how important is satisfaction with food provided by the hotel's restaurant across customers in terms of most valuable factors of evaluating hotel's quality beyond customers. “The survey results suggested that guests react favorably to a clean neat restaurant, neat employees, ample portions, and responsiveness to complaints. The quality and quantity of service, food quality, helpfulness of the employees, and the prices of drinks, meals, and other services appear on the list of the most frequent compliments.” (Pizam, Shapoval, & Ellis, 2016) Authors suggests that the most valuable factor of satisfaction with meal is size of portion. At the same time, research has qualitative approach instead of quantitative. This factor can elaborate to the results as recipes with high number of calories will mostly have rating's value more than 3 and means that people are satisfied with these recipes. Qingqing Tan, Ade Oriade and Paul Fallon (2014) made a research across Chinese schools and pupils studying in it. They conclude that quality of food is a major determinant of satisfaction with food. People may have different satisfaction with food depends on what quality type of ingredients was used. Unequal ingredients could bring feeling of total dissatisfaction with food. “The food quality is the most significant dimension in service quality, as well as the most significant predictor of customer satisfaction” (Tan, Oriade, & Fallon, 2014)

“Serving in multi-use plates positively influences the satisfaction with the level of school food tastiness whereas the use of pitchers for water negatively influences it. The number of pupils eating at school is also a negative and highly significant determinant. A possible explanation is the adoption of the cook-and-chill catering system in the case of a large number of meals instead of the cook-and-hold one” (Ornella & Maria, 2016). This research stated list of factors which influence pupil's satisfaction with meals in Italian school. Such factors as overweight, gender, amount of canteen visitors, is pupil's meal free or not, relationship between school and food organization have positive or negative impact on satisfaction with meal.

Food satisfaction was best explained by sensory satisfaction (Andersen & Hyldig, 2015). Sensory satisfaction is when person likes tastes of the food and appearance of it both at the same time. Authors used questionnaire as a data collecting tool like most of other researchers. They consider that appearance and taste should be not two but one factor of satisfaction with meal. It means that further it should be checked that more than one variable can show dependence at the same time while they could not show results separated from each other. Such factor as time of day didn't influence final satisfaction with meal.

Same food could give different satisfaction to the particular person. Barbara Van Andersen and Grethe Hyldiq stated that “at least four major context effects that can alter the perception of food during consumption; its function as a meal component, social interaction during consumption, the environment and the freedom of food choice” (2015). It means that the same person able to reach different satisfaction from the same food depending on by whom was done the meal, quality of components, surroundings and control over the menu. Researchers suggest list of factors which influence food satisfaction. They group factors in three groups: before intake, during intake and after intake. Before intake factors consist of desire, expected liking, hunger; during intake consist of liking the appearance, taste, texture, variety, novelty, sueprisingness; after intake consist of satiation, satiety, energy level, desire for other foods; moreover researchers stated factors in context: physical eating context, occasion, social eating context, person related characteristics, product information (Andersen & Hyldig, 2015). Participants of Andersen and Hyldig (2015) research mentioned hunger as a major factor of satisfaction with meal, they could not believe staying hungry after the meal and been satisfied with food at the same time. “Hunger - a feeling of discomfort or weakness caused by lack of food, coupled with the desire to eat, a severe lack of food, a strong desire or craving” (Oxford dictionaries, 2019). If person unsatisfied with the food because he or she still feels hunger it means that person (Han, Lee, Chua, Lee, & Kim, 2019)did not get enough food or calories, proteins, fat, etc. of which food consists and what peoples' body needs to not feel hungry. Further, it will be tested that calories, fat and sodium have dependence on the rating gave by visitors of the website.

Heesup Hun and others (2019) conducted research how satisfaction with food of airline impacts on re-flying intention. They suggest that satisfaction of the meal taken during the flight increase passengers' satisfaction of the whole flight and possibility what passenger will choose again the same flying company. Researchers stated that tastiness, portion size, freshness, quality of ingredients, easy-to-digest, right temperature, presentation of the meal (color, variety), food and beverage delivery system are major factors of satisfaction with food (Han, Lee, Chua, Lee, & Kim, 2019). Authors proved that passengers who were satisfied with food are more willing to choose again the same flying company than passengers who didn't eat or were unsatisfied with provided food. In addition, food history (origin of the ingredients, condition of food storage, delivery process) have impact on final satisfaction. It brings limitation to the study due to impossibility to collect information from all people who left mark to the recipe on web-site. People may get different satisfaction because they all used different ingredients with different origin, storage and delivery process. Perhaps, people also could get used with ingredients from their local shops and feel the same satisfaction because they ordinary eat the same ingredients on their daily basis.

Satisfaction with food affect the satisfaction of life (Seconda, et al., 2017). In addition, study conducted by Seconda and others (2017) showed what people who has a lot organic food in their diet showed slightly more satisfaction with life than people who had a lot of unhealthy food in their diet. Researchers used semi-quantitative approach, high number of volunteers, questionnaire to evaluate satisfaction with diet and overall satisfaction with life. Although, better satisfaction with life described by authors not only by the fact that organic food leads to better health of people who eat it but also that people who buy organic food think that they help environment and save planet and feeling of it brings positive mood and increase satisfaction with life of a person. Although, significant difference between life satisfaction based on containing organic food in a diet was showed only in two fourth and fifth quartiles and it was slight significance. Moreover, researchers suggest that existence of organic food in the diet not the only factor by which overall life satisfaction is described because volunteers who didn't particularly eat organic food on daily basis also showed high satisfaction with life. Even though, it is still relevant to study people's ordinary life activities to know how to increase overall satisfaction with life. Study of the Seconda (2017) proves that field of food consumption should be researched more to characterize determinants of food satisfaction and how it increases life satisfaction. “Higher contribution of organic food to the diet may help to improve life satisfaction of people aged 45 and more through hedonist or eudemonic approaches but longitudinal studies are needed to better characterize the direction of causation” (Seconda, et al., 2017). Another study of Beate Andre and others (2017) proves that healthy diet increase overall life satisfaction. Their study were conducted among people ages 65 and higher among Norwegian citizens. Results also showed that seniors who has healthy diet also have less mental health problems.

It's important to mention that one of key drivers of choosing organic food for meal is to desire eat healthy (Jannssen, 2018). It also a determinant for satisfaction with food - desire to eat particular food. People want to eat specific meal and after in-take they feel more satisfied than after ordinary food in-take. Jannssen (2018) also conclude that income of a person and his or her education showed dependence to the purchasing organic food. Jannesen (2018) stated that “significant direct effects on organic food purchases were also recorded for three socio-demographic characteristics, namely income, education and young children, i.e. these characteristics had a greater influence on organic food purchases than on attitudes towards organic food”. It means that education and income have impact on preferences of a person and people with higher income and good education could feel more satisfied with food. Perhaps, education and income don't directly have dependence on satisfaction with specific meal. At the same time, person with higher education and higher income than ordinary person can feel more satisfaction with meal because he or she is more willing to try something new than person who has ordinary income and education because he or she get used to common food and won't like something to try with completely different taste from ordinary food. Hence, education and income won't be as a determinant of satisfaction with meals in our paper because these factors are determinants of purchasing or satisfaction with particular receipt but in our case, we want to know determinants of satisfaction with meals in common. Education and income won't be considered as a factor of satisfaction with meals in this paper.

If income of a person affect satisfaction with purchasing food price of the ingredients itself also could affect satisfaction with meals (Jannssen, 2018). Person could expect from more expensive food feeling more satisfaction than to his or her ordinary food. Hence, cost of the ingredients could affect the overall satisfaction with meal. At least, it isn't true that people expect bigger size of the food from more expensive food (Zuraikat, Roe, Smethers, Reihart, & Rolls, 2018). Zuraikat and others (2018) stated that “In this controlled study in a restaurant-style setting, changes to the cost of a meal did not affect the response to increasing portion size of the main dish”. In previous research Zuraikat and others (2018) found that “serving larger portions led subjects to consume more food, and we extended these findings by demonstrating that this effect was not exacerbated by doubling the price paid for the meal from $8 to $16”. As a result, meal cost did not affect overall satisfaction with expectations of a size of portion (Zuraikat, Roe, Smethers, Reihart, & Rolls, 2018). Hence, hypothesis of Zuraikat (2018) that people expect from more expensive meals larger portions was rejected.

Tino Bech-Larsen, George Tsails (2018) stated “the constructed knowledge and experience scales are cumulative, have high levels of reliability, and indicate the positive effects of such knowledge and experience on consumers' food-related life satisfaction”. It means that cooking skills of a person could have positive or negative impact on final satisfaction. Hence, it's impossible to measure cooking skills of web-site visitors. Further, cooking skills will be a limitation of the research. Perhaps, data collected from the web-site for amateur cookers. It will be considered that all people who left mark for the food have similar cooking skills.

Personal characteristics are another factor of satisfaction with meals (Hartwell, Shepherd, Edwards, & Johns, 2016). In addition, H.J. Hartwell and others (2016) stated that food service, quality of food, social environment are having impact on final satisfaction with food between patients in hospital. Researchers used questionnaire as a primary tool for the study. They are describing ambiguity as a major problem of questionnaire. They tried to avoid it with more direct work with patients for questionnaire. Although, word “service” is used in multiple senses. Hence, it's important to take these factors as a limitation of the research due to the fact that the goal is to describe dependence of people's choice on meals across the nutritional facts and ingredients.

Significant impact on understanding determinants of satisfaction with meals could bring papers that study satisfaction with meal across pupils or students. Pupils and students are eating the same food from canteen on daily basis through all years of education. Many researches used questionnaires to find factors which influence satisfaction with food among students. Hence, results of other scientists could bring limitations or hypotheses to our research. Although, majority of the papers didn't used quantitative approach, regression analysis and etc. to prove their results.

Pernille Haugaard and others (2016) in “Determinants of meal satisfaction in a workplace environment” made a research concerning factors of satisfaction with meals among workers in the same company. All 71 participant were working for a long time at the same company. Canteen was able to present 519 meals for the workforce during mealtime. Authors used longitudinal approach. It was significant for researchers to give respondents questionnaire before and after they eat their meal. Next step was analyzing the collected information. Scientists used mixed modelling approach for this step. Correlation matrix and LMM analysis was chosen as tools for analysis. In addition, they give demographic of participants and description of meal satisfaction and its determinants. Authors took all determinants from the studies of other researchers. “From this study we conclude that providing high quality meals in a cozy ambience increase meal satisfaction and perceived satiety without increasing reported energy intake. These findings support the role of meal satisfaction in promoting overall wellbeing without causing overeating” (Haugaard, Stancu, Brockhoff, Thorsdottir, & Lahteenmaki, 2016). This research gives limitation that findings which will be conducted further will be applicable only in terms of good quality food. Perhaps, low quality food could cause understatement of recipe's rating.

Cooking skills is a major factor for satisfaction with food because meal will not satisfy person if it is not cooked properly, even if ingredients had good quality and were fresh. Cooking skills also include skills of serving. Appearance of the meal give to the people expectation for the taste. Thus, research of the Mary Hannan-Jones and Sandra Capra (2018) conclude that appearance have not got significant difference for customer satisfaction with meals. “It is our opinion that the data from the both of the studies suggest that the form and colour of the plate is not a major issue to patients, with overall very few of the comments being made about the meals relating to the plates themselves” (Hannan-Jones & Capra, 2018). Due to the fact that author used statistical method. We conclude that people who used epicurious to find recipe for the meal cooked properly and serve dish in the way they want. Hence, bad cooking skills will not have impact on the results in our paper.

The diverse in food preference is significant limitation because different genders, people of different ages, pupils from different classes and schools have different preferences (Ali & Akbar, 2015). It is described by the fact that each person has its own meal preferences and personal factors by which decided whether they like specific food or not. Further, r-squared in regression could not show high significance because of lack in knowledge of evaluation personal preferences. Quantitative method isn't suitable for evaluation and finding dependence for specific persons' preferences. Other researchers used qualitative methods as primary tools for their research. Therefore, high number of factors will definitely have impact on results of regressions.

1.3 Gap

Literature analysis was made for finding the gap and determinants of satisfaction with meal that already were found. We conclude that there is a gap in the field of studying nutrition and satisfaction with meal. There is a lack in researchers that used quantitative approach and regression analysis as a tool for the research. In addition, there is a lack in researching how nutritional information impacts on overall satisfaction with meal.



2. Methodology

2.1 Research problem

Food consumption groves rapidly within time. People increased consumption of food by nearly 400 kcal per person per day between 1969-1971 and 1999-2001 (Kearney, 2010). It rises from 2411 kcal to 2789 kcal. At the same time countries as Somalia, Burundi, Rwanda, Kenya which located in sub-Saharan Africa showed a decline in food consumption from a very low level of food consumption (Kearney, 2010). Even though, we see a positive trend in food consumption. A lot of researchers were made about how important healthy diet is, but obesity with food still exist (A. Andrews, Lowe, & Clair, 2011). Even though, researching food consumption is able to clear how to improve all aspect of people life like work, overall mood, sleeping and etc. due to the fact that basic needs like eating directly affects day to day life of a person.

Need in researching of food consumption by people and how it affects their ordinary life is important due to the numerous variants of how satisfaction with meals is affecting human's work, habitats, ordinary life. In addition, a lot of researchers were conducted to make field of meal consumption clearer. It was found that even colors of food are affecting the overall satisfaction with meals and consequently satisfaction with work itself (Paakki, Aaltojдrvi, Sandell, & Hopia, 2018). Results of Paaki's (2018) research showed that people enjoyed food with diversity in colors, contrasts and bright colors, natural and simply colors. Most of the enjoyed colors were colors of vegetables. Moreover, results mean that more enjoyable colors person have in their diet healthier it gets thus most enjoyable food's colors are colors of vegetables.

Learning of the previous researchers which were described in literature review we found a problem that there is lack in researching how nutritional facts and food components affects the overall satisfaction with meals using quantitative approach. A lot of researchers were made about what are the determinants of satisfaction with meals but mostly they were founded using control groups and questionnaire. These types of research approach helped researchers found that colors are one of the determinants of satisfaction with meal (Paakki, Aaltojдrvi, Sandell, & Hopia, 2018) or hunger (Andersen & Hyldig, 2015) and etc. There is a lack in understanding how proteins and other parts of food which contains in the meal affects the overall satisfaction.

Steps for the research are following:

1. Collect data from a cooking website

2. Analyze data using Stata and Jupyter

3. Determine factors which would influence the rating of the recipe

4. Give recommendation to recipe's author based on the founded factors

2.2 Data collection and data description

The database was downloaded from kaggle.com. Kaggle.com is a free open source website where users can upload and download different databases, provide description for the data, update database, provide distribution for variables. Data was collected from the website www.epicurious.com. Website have more then 330 000 recipes, special overviews of specialists, distributed recipes, search, videos, columns with advices, free daily email subscription. People can leave their mark for each recipe. Mark which can be leaved is zero spoons, one spoon, two spoons, three spoons or four spoons. In database marks were transformed in 5 scale rating. Four spoons are 5, three spoons are 3,75, two spoons are 2,5, one spoon is 1,25 and zero spoons are 0. In addition, database from Kaggle was checked for the recipes which are don't have enough information. The food from which recipe consists was added as dummy variable. If recipe have specific ingredient one was putted in the variable and if don't zero was in the variable. Perhaps, to represent 330 000 recipes with confidence interval of 5% data with 384 recipes is enough. Database which was downloaded is consist of more than 20 000 recipes. Hence analysis' conclusion of database with 20 000 recipes will be representative for the majority of people. Moreover, it is including rating, nutritional facts (protein, fat, sodium and calories), was it or not in reviews of website's columnist (“#cakeweek”, “#wasteless”) and the products consist in the recipe. Epicurious is powered by website edamam.com. There nutritional facts and ingredients are ordered better than epicurious. Edamam doesn't have recipe step by step. Due to the fact that for making database step by step recipe isn't needed, edamam.com was used for taking information about nutritional facts, ingredients for increasing speed and sufficient of making database.

In addition, edamam was used to fill all missing values in the database. If there wasn't information about nutritional facts of a receipt variable with missing values of a receipt was deleted. As a result, from more than 20 000 observations approximately 4000 observations were deleted due to the inexistence of nutritional facts on edamam resource or epicurious web-site and approximetly 1000 of observations were filled with information about nutritional facts taken from edamam or epicurious sources. For some variable nutritional facts were located on epicurious web-site in the comments of author in the end of receipt.

Table 1. Descriptive statistic of variable rating

Rating
	Percentiles	Smallest
1%	0	0
5%	0	0
10%	2.5	0	Obs.	15233
25%	3.75	0	Sum of Wgt.	15233
50%	4.375		Mean	3.759396
		Largest	Std. Dev.	1.28094
75%	4.375	5
90%	5	5	Variance	1.640809
95%	5	5	Skewness	-1.99268
99%	5	5	Kurtosis	6.316895



Table 2. Descriptive statistic of variable calories

Calories
	Percentiles	Smallest
1%	28	18
5%	71	18
10%	114	18	Obs.	15,233
25%	198	19	Sum of Wgt.	15,233
50%	324		Mean	424.8443
		Largest	Std. Dev.	330.4139
75%	560	1982
90%	858	1985	Variance	109173.4
95%	1098	1990	Skewness	1.645197
99%	1620	1993	Kurtosis	6.189183

Table 3. Descriptive statistic of variable protein

Protein
	Percentiles	Smallest
1%	0	0
5%	1	0
10%	1	0	Obs.	15,233
25%	3	0	Sum of Wgt.	15,233
50%	8		Mean	17.93041
		Largest	Std. Dev.	24.37988
75%	25	242
90%	47	245	Variance	594.3786
95%	65	253	Skewness	2.958103
99%	111	264	Kurtosis	16.49829

Table 4. Descriptive statistic of variable fat

Fat
	Percentiles	Smallest
1%	0	0
5%	0	0
10%	1	0	Obs.	15,233
25%	7	0	Sum of Wgt.	15,233
50%	17		Mean	23.7254
		Largest	Std. Dev.	24.07055
75%	31	181
90%	54	192	Variance	579.3912
95%	73	194	Skewness	2.039956
99%	114	194	Kurtosis	8.706025

Table 5. Descriptive statistic of variable sodium

Sodium
	Percentiles	Smallest
1%	1	0
5%	5	0
10%	14	0	Obs.	15,233
25%	81	0	Sum of Wgt.	15,233
50%	285		Mean	496.3548
		Largest	Std. Dev.	612.9327
75%	675	4603
90%	1264	4606	Variance	375686.6
95%	1690	4626	Skewness	2.384462
99%	2941	4646	Kurtosis	10.59085

Figure 1 Distribution of variable fat

Table one, two, three, four and five shows descriptive statistics for variable rating, calories, protein, fat and sodium. These variables were chosen for descriptive statistics because they are the most important for this research. Further, correlation between each over will be found, dependence between rating and protein, fat and sodium will be found with regressions.

Figure 2 Distribution of variable calories

Figure 3 Distribution of variable protein



Figure 4 Distribution of variable sodium

Figure one, two, three and four shows distributions of variables calories, fat and sodium. Variable calories closest to the normal distribution because it contains a lot of unique variables - variables with once used values. Variables fat and sodium less similar to the normal distribution because receipts are taken from the web-site for households and majority of the recipes are made for households and contains instruction about making healthy food which contains low amount of fat and sodium. This is why variables fat and sodium have majority of values close to the zero.

Figure 5 shows distribution of variable rating. This variable normally distributed. Values 4,25 and 3,75 have highest density. Both of the values are actually 3 spoons out 4 in measure system of epicurious web-site but 4,25 means that people are more likely to cook this particular receipt than other. We conclude from the figure 5 that people are more likely to give zero to the recipe they didn't like than to give it low mark. In addition, if person liked the recipe and would like to cook it again it doesn't mean that he or she will give maximum mark. People are more willing to give almost the best mark.



Figure 5 Distribution of variable rating

2.3 Aim of the research

Aim of the research is to find specific factors based on the nutrition facts and other determinants of the recipe that influence people's opinion on the recipes' rating. To find relationship between variable will be used correlation. In addition, T-test for unpaired samples will show there is statistical difference between variables or not if database is normally distributed. If datasets will not have normal distribution then nonparametric tests will be used.

Therefore, analysis will be conducted using regression for finding the dependence between variable. Software which will be used: Excel, because the database was downloaded in .xlsx format; Stata 14.2 to do all correlations, regressions and tests; Jupyter notebook (notebook for python) with libraries of numpy, pandas, pylab, seaborn, matplotlib. pyplot, sklearn, sklearn. linear model, sklearn.preprocessing, collections, statsmodels. formula.api, statsmodels. api for making graphs and comparison with results of stata. Python has several benefits to stata. Python is free, python language similar to other programmer languages like R, C++, etc. In addition, python can read all format of files, python uses less computer hardware for making analysis, python more widespread around the world, python has more setting which can be changed, and it is more functional. Perhaps, Stata has easier command for regression, clustering, correlation. It's easier to read files and make output files and new user can start to use stata faster than python. There isn't significant difference in methods of how programs make analysis. Therefore, there won't be difference in the output of correlations and regression between two programs.

2.4 Limitations

Due to the fact that our database contains information only about nutritional facts and ingredients other factors that were mentioned by previous researchers will be limitations for our research because they do not exist in the database.

Hunger is a major determinant of satisfaction with meal (Andersen & Hyldig, 2015). Epicurious doesn't have any tools of collecting information about hunger of people who used web-site and left a comment before intake of meal and after it. It could be really important information and have significant impact on the results if we had access to the information about hunger of people. It will have significant impact because Andersen and Hyldig (2015) provided extended proof why hunger is so important. Participants of their research mentioned that they couldn't realize being satisfied with food after in-take and feel satisfied at the same time.

Appearance and taste won't be considered as limitation. Firstly, taste is a complex indicator of the food which contains such factors as how much proteins, sodium meal has and ingredients. All these factors are located in the database. Appearance won't affect the results because epicurious web-site is made for household and everyone who want to cook on their own. Hence, each person who left a comment serve a dish in the way they want and like. Also, people could change some ingredients to feel more satisfied with meal but didn't mentioned that in a comment section.

Personality is a really important limitation. Each person has unique preferences for food which are have significant impact on final satisfaction with meal (Hartwell, Shepherd, Edwards, & Johns, 2016). This fact also refers to the problem why some receipts have very high and very low marks left by the people of epicurious. Different people can leave marks that diverse from each other really significantly. It happens because taste of a person can significantly differe from other people even if they are studying in the same school and class (Hartwell, Shepherd, Edwards, & Johns, 2016).

Epicurious didn't provide any information about gender of people who left their marks. At the same time gender is one of factors which influence satisfaction with meals (Ornella & Maria, 2016). Not having gender as a variable in the database could affect the results but Ornella and Maria (2016) didn't provide us information about how significantly it affects overall satisfaction. Hence, we could suppose that gender didn't have significant impact that will ruin results if we will not include it. Although, there are numerous factors that influence satisfaction with meal besides gender.

The limitations of the research are quality of food that people used while cooking the recipe and cooking skills which also could affect the final satisfaction with meal. Using quality food to understand the true satisfaction with meal is very important (Tan, Oriade, & Fallon, 2014).

If people who tried recipe from epicurious were using the same good quality food rating could be different significantly. Using only OSL regression, correlation, etc. is other limitation because research could miss several results or other relationship and dependence could be found if using more methods for data analysis.

Overall, the most important factor of the satisfaction with meals was mentioned almost be every research that was used in this paper is quality of food (Tan, Oriade, & Fallon, 2014). We think that it won't have significant impact on the results in our research because people who tried to cook meals using recipes from epicurious used ingredients from the same supermarkets and shops that they used to buy food ordinary.

3. Results

3.1 Testing normal distribution

Table 6. Shapiro-Wilk W test for normal data

Variable	Obs	W	V	z	Prob>z
Rating	15,233	0.76974	1641.284	20.033	0.00000
Calories	15,233	0.85549	1030.082	18.773	0.00000
Protein	15,233	0.70404	2109.613	20.713	0.00000
Fat	15,233	0.82984	1211.432	19.211	0.00000
Sodium	15,233	0.74858	1792.179	20.271	0.00000

We perform Shapiro-Wilk test to test normal distribution for the database. We test variables rating, calories, protein and fat as they are the most important for this research. Prob<z for all five variables is zero. If we will reject null hypothesis about normal distribution of data, we will not make a mistake with probability of 0%. It means that our data isn't normally distributed according to Shapiro-Wilk test. Hence, we are not able to perform T test for testing null hypothesis that variables are statistically different.

3.2 Testing equitation of medians

Despite the fact that our data is not normally distributed we still can perform our analysis using non-parametric tests. For testing equitation of medians between datasets we are not able to use ANOVA test. We can use Kruskal-Wallis H-test because this test is non-parametric.

Results of Kruskal-Wallis H-test for variables rating and calories:

chi-squared = 1615.663 with 1514 d.f.

probability = 0.0344

chi-squared with ties = 1778.074 with 1514 d.f.

probability = 0.0001

In both variants' probability is less than 0.05. That means median difference between these two groups is significant and we can perform further analysis for this two groups.

Results of Kruskal-Wallis H-test for variables rating and protein:

chi-squared = 354.174 with 188 d.f.

probability = 0.0001

chi-squared with ties = 389.776 with 188 d.f.

probability = 0.0001

In both variants' probability is less than 0.05. That means median difference between these two groups is significant and we can perform further analysis for this two groups.

Results of Kruskal-Wallis H-test for variables rating and fat:

chi-squared = 348.087 with 168 d.f.

probability = 0.0001

chi-squared with ties = 383.193 with 168 d.f.

probability = 0.0001

In both variants' probability is less than 0.05. That means median difference between these two groups is significant and we can perform further analysis for this two groups.

Results of Kruskal-Wallis H-test for variables rating and sodium:

chi-squared = 2139.784 with 2146 d.f.

probability = 0.0001

chi-squared with ties = 2354.881 with 2146 d.f.

probability = 0.0001

In both variants' probability is less than 0.05. That means median difference between these two groups is significant and we can perform further analysis for this two groups.



3.3 Analysis

Ordinary OLS regression is not suitable for our database because it is parametric analysis and our datasets are not normally distributed. Only nonparametric analysis will fit our research. That is why we decided to choose local-linear regression for this research

Table 7. Bandwidth for local-linear regression

	Mean	Effect
calories	99.16469	132.8057
protein	7.318297	9.800987
fat	7.222977	9.673329
sodium	183.9728	246.3845

Table 8. Local-linear regression

Local-linear regression Kernel : epanechnikov Bandwidth: cross validation	Number of obs.	15,233
	E(Kernel obs)	15,233
	R-squared	0.1042
rating	Estimate
Mean
rating	3.721157
Effect
sodium	.0003239
fat	.0154343
protein	.0082677
calories	7.98e-06

Table 8 shows local-linear regression for datasets rating, protein, calories, fat and sodium. Dependent variable is rating and independent variables are protein, calories, fat and sodium. Regression shows that there is very small dependence between rating and protein, calories, fat, sodium. Regression model of variable rating based on variables sodium, fat, protein and calories can only describe 10,42% of variable rating's values.

Therefore, data visualization is very important because it can bring conclusion which is hard to make through analysis or data scanning.

Figure 6 Scatterplot of distribution of recipe's calories grouped by rating

Figure six shows that distribution of calories below 500 is approximately the same. That means that we cannot even suppose satisfaction with meal which is bellow 500 calories between people. Even though, we can see difference starting from 500 calories and more. If we suppose that rating with values more than 3 means that majority of people satisfied with recipes, we can see that people are satisfied with majority of recipes that have more than 500 calories. Perhaps, it can be seen that people are mostly satisfied with recipes that have calories between 500 and 1000 calories, but there are big number of recipes with such amount of calories that have rating's value zero. Hence, people were totally unsatisfied with recipe. Although, recipes with calories more than 1000 are getting higher rating than other recipes. It means possibility that people will be satisfied with meals which have more than 1000 calories more than with meals which contain less than 1000 calories.



Figure 7 Scatterplot of distribution of recipe's protein grouped by rating

Figure seven shows that distribution of protein with values below 50 is approximately the same. We can see difference starting from 50 and above. People were satisfied and give rating more than 3 to the majority of the recipes which contain more than 50 grams of protein. Bigger difference starts from 150 grams of protein and more. Only 4 recipes which contains more than 150 grams of protein get rating's value less than 3. All 4 of them get zero rating. At the same time, all other recipes get rating's value more than 3 which means that they were satisfied with meal and would like to repeat to cook it.

Figure eight shows distribution of recipe's fat by its rating. We can see that there is no statistical difference between rating's value and fat below fat's value of 50. At the same time, we see difference starting from 50 and higher. Majority of meals get rating's value more than 3 which means that they were satisfied with recipe and will re-do it again. Hence, the difference decreases after 100 grams of fat and majority of recipes have rating's value more than 3. After 150 grams of fat we see only one recipe with rating's value lower than 3 which gets zero out of five for rating.



Figure 8 Scatterplot of distribution of recipe's fat grouped by rating

Figure nine shows distribution of recipe's sodium by rating. We do not see difference of distribution for recipes which contains less than 2000 mg of sodium. It is necessary to point out that bigger number of recipes with more than 2000 mg of sodium get rating's value more than 3. The difference in number of recipes which get rating's value more than 3 compare to which get rating's value below 3 increased.

We see one major tendency from figure five, six, seven and eight. Bigger number of calories, protein, fat or sodium recipe has more chance of it get rating with value more than 3. We consider that recipes with high amount of protein, fat, sodium or calories have more chances to satisfy

Table 9 shows correlation between variables rating, calories, protein, fat and sodium. One gram of protein means 4 calories. Hence, we see that protein and calories are highly correlated between each other. Also, calories have high correlation with fat, because a lot of calories comes from fat. Rating is not correlated with variables protein, fat, calories or sodium. Sodium have not high correlation with calories as protein or fat.

Figure 9 Scatterplot of distribution of recipe's sodium grouped by rating

Table 9. Correlation matrix

	Rating	Calories	Protein	Fat	Sodium
Rating	1
Calories	0.1309	1
Protein	0.1201	0.7340	1
Fat	0.1340	0.8897	0.6144	1
Sodium	0.0826	0.4439	0.5094	0.3590	1

Majority of recipes which get rating's value of 5 have low amount of fat, sodium and calories and have high amount of protein. It means that people prefer healthy food, because high amount of calories, fat and sodium in a diet leads to several diseases.

3.4 Results of analysis

In the beginning, we had three hypotheses. They are:

1. Majority of the recipes with number of calories more than 1000 will have rating more than 3;

2. Nutritional facts are not the most significant determinants of satisfaction with meal;

3. Majority of the recipes which have 100 grams of protein and more will have rating more than 3.

Through analyzing database, we consider that majority of the recipes with high amount of calories (more than 1000) get rating's value more than 3 which means that people were satisfied with meal and would like to repeat to cook it again. Hence, first hypothesis is approved. Nonparametric regression showed that using nutritional facts (calories, protein, fat, sodium) we are not able to predict rating of the meal. In addition, many earlier researchers found a lot of determinants of satisfaction with meals besides nutritional information about meal. Second hypothesis is approved. Finally, we made scatterplots which showed that majority of the recipes with high amount of protein are top-rated



Conclusion

In conclusion, we will describe protentional researchers which could be made, evaluate work which has been done. We made analysis of database which include nutritional information and rating given by the people based on their opinion about recipes from Epicurious. Analysis showed that nutritional facts did not affect the rating. Hence, nutritional facts have not got dependence on the satisfaction with meal and hypothesis was approved. In addition, two more hypotheses about number of calories, amount of protein and how they affects the rating were approved. Through analyzing literature review we found out the limitations which were described earlier and determinants of satisfaction with meal across different groups of people. Most important determinants were sensory satisfaction (Andersen & Hyldig, 2015). Sensory satisfaction is when person likes tastes of the food and appearance of it both at the same time. Authors used questionnaire as a data collecting tool like most of other researchers. They consider that appearance and taste should be not two but one factor of satisfaction with meal. In addition, determinants like hunger, texture of the meal, colors, surroundings of the person, quality of food, controls under cooking and menu are significant factors. Group where people had no control for preparation of the meal and no control for menu showed the best results in overall satisfaction with meal and reducing stress (Osdoba, Mann, Redden, & Vickers, 2015). It should be mentioned that the most important limitation of the research was that we could not collect information about hunger and transfer it to numeric variable and previous research showed that taste of a person is unique (Ali & Akbar, 2015).