Adaptation of the Food Choice Questionnaire: the case of Hungary

Purpose The purpose of this paper is to examine the applicability of the original 36-item Food Choice Questionnaire (FCQ) model developed by Steptoe et al. (1995) in Hungary. Design/methodology/approach The national representative questionnaire involved 1,050 individuals in Hungary in 2015. Several multivariable statistical techniques were applied for the analysis of the data: confirmatory factor analysis, principal component analysis, and cluster and Log-linear analysis. Findings The results indicate that the original nine-factor model is only partially applicable to Hungary. This study successfully managed to distinguish the following factors: health and natural content, mood, preparation convenience, price and purchase convenience, sensory appeal, familiarity, and ethical concern. The FCQ scales proved to be suitable for the description of clusters based on specific food choices and demographic characteristics. By using the factors, the following five clusters were identified: modern food enthusiast, tradition-oriented, optimizer, easy-choice and un-concerned, all of which could be addressed by public health policy with individually tailored messages. Originality/value The Hungarian testing process of the FCQ model contributes to an examination of its usability and provides the possibility of fitting the model to different cultures.


Introduction
Food choices can be described as a complex process, whose influencing factors can be divided into two categories: internal (food) effects (e.g. sensory aspects) and external (non-food) effects (e.g. psychological, social and cultural factors) (Eertmans et al., 2005). Attempts have been made to describe and approach these factors from different perspectives.
The literature draws attention to the fact that consumer behavior is changing, and a kind of nutritional revolution is occurring. The "new" consumer builds on trust and takes on role Roos et al. (2012) also used the modified FCQ to examine parental family food choice motives in Finland. Furthermore, the adapted FCQ model has been implemented by many other researchers (Lockie et al., 2002;Ares and Gámbaro, 2007;Scheibehenne et al., 2007;Pieniak et al., 2009Pieniak et al., , 2013Honkanen and Frewer, 2009;Share and Stewart-Knox, 2012;Crossley and Nazir, 2002;Dowd and Burke, 2013;Pula et al., 2014;Gagić et al., 2014;Ooi et al., 2015;Dikmen et al., 2016). The structure of factors has also varied from study to study. For example, in Ooi et al.'s (2015) Malay adolescent sample, the new factors were religion, parents, peers and media, and in Pula et al.'s (2014) UK study, a new impression management factor was included.
As can be seen above, the FCQ has been applied at the national level, but there is also much interest in testing the cross-cultural validity of the FCQ model (Prescott et al., 2002;Eertmans et al., 2006;Pieniak et al., 2009;Milošević et al., 2012;Januszewska et al., 2011;Markovina et al., 2015). Prescott et al. (2002) was one of the first to apply the FCQ in Asia, and their results indicated a strong relationship between food choice motives and nationality. More recently, the measurement invariance has been demonstrated by Pieniak et al. (2009) across six countries, by Januszewska et al. (2011) across four countries and by Markovina et al. (2015) across nine European countries. In contrast to these findings, Milošević et al.'s (2012) analysis indicated that the original nine-factor design displayed a suboptimal fit for the Western Balkan Countries. Furthermore, Eertmans et al.'s (2006) results of the test across three countries (Canada, Belgium and Italy) did not support the generalizability of the FCQ's factor structure. They suggested that its items and construct may have different connotations across western urban populations, whether in English-speaking or non-English-speaking countries. Fotopoulos et al. (2009) also did not reinforce the robustness of the original FCQ. There have been some psychometric problems, but in their opinion, at the subpopulation level the original FCQ can form the basis for a new typology.
When analyzing the data, answers to three main questions were sought: (1) Can the original FCQ be applied to this sample?
(2) If the original model cannot be applied, how are the factors modified?
(3) Can the FCQ model be used to create individual consumer segments based on consumers' food choice habits?

Sampling
The data collection was carried out in November 2015 with the help of a market research firm. The sample consisted of 1,050 individuals. For sample size calculation, we used the following formulas: where Z is the Z-value (e.g. 1.96 for 95% confidence level), p is the percentage picking a choice in a question and err is the margin of error or confidence level. In this formula, we used 8,000,000 for the population, Z ¼ 1.96 for 95% confidence level, the error rate was 3 percent, for the worst case scenario, p ¼ 0.5 SS was 1,067 and for p ¼ 0.4 SS was 1,024, so we set the sample size to 1,050. Representativeness for regions and for types of settlement was ensured by the applied quoted sampling method. The sample pattern met the quotas previously defined by the Hungarian Central Statistical Office. In the assigned settlements a random walking method was used to ensure total randomness in selection. In the second step, the interviewed individual within a household/family was selected by using the so-called birthday-key. With this method, randomness was ensured in the second step as well. (The reason for the involvement of young people under the age of 18 was that young people, in the majority of cases, make independent purchasing decisions and their influence on the family's purchases of food is extremely high).
Since random walking does not ensure the sample is a reflection of the entirety of the population (the number of the female and elderly respondents was higher than the national average), the sample of the people was corrected by multi-dimensional weighting factors (gender and age). After these methods were applied, the sample was representative of the structure of the Hungarian population in all the three aspects (region of residence, gender and age).
We took care that the personal data of the interviewed individuals enjoyed complete protection and anonymity. The data were collected through a standard questionnaire. The questionnaire used in this research was completed after asking several individuals from the target group (pre-testing phase).

Questionnaire
In the analysis, the 36-item questions of the original FCQ with a five-point Likert scale were used. An answer of 1 represented "I do not take this into account at all," while a 5 represented "I fully take this into account." The distribution of the sample is shown in Table I.

Data analysis
Several multivariable statistical techniques were applied for the analysis of the data: confirmatory factor analysis (CFA), principal component analysis (PCA) and cluster analysis. As a first step, a CFA was applied and a fit estimated for Steptoe et al.'s (1995) nine independent factor FCQ model for Hungary. Goodness of fit was assessed using a variety of measures.
A highly significant difference was found between the original and the expected covariance structure according to the χ 2 indices, which indicated there was a large unexplained part in the original model ( χ 2 (630) ¼ 16,798.2; p o0.001). The Tucker Lewis index also lay outside the conventional acceptance limits of 0.95. Although the RMSEA value of 0.065 indicated a good fit, as it was below the conventional value of 0.08, CFA goodness-of-fit measures suggested that the original factor structure cannot be applied directly to the Hungarian data, and therefore explanatory factor analysis should be followed (Linting et al., 2007;Fotopoulos et al., 2009). PCA was employed with varimax rotation to correctly identify the independent principal components (PCs). In order to detect the internal consistency of the PCs and measure scale and subscale reliability, Cronbach's α was calculated for each subscale and for the whole scale. The identified dimensions have a reliability above the recommended 0.6 -0.7, and all the component loadings were greater than or equal to 0.4 -0.5, while the extracted variance was around 0.7 (Hair et al., 1998). The assessment of statistically significant differences across the PCs was established by analysis of variance (ANOVA) and an independent sample t-test. A significance level of 5 percent was used for evaluating the results. The Levene's test was also applied to check homogeneity and found the studied groups to be homogenous in terms of their standard deviation. In order to measure the strength of the relationships, effect sizes and the 95% confidence intervals for the difference between subsets were reported.

1477
Adaptation of the Food Choice Questionnaire Finally, cluster analysis was performed using the standardized factor scores from the PCA. The following clustering methods were used and compared: hierarchical clustering with Euclidean distance (using simple, complete and average linkage, centroid, median and Ward's method) and K-means clustering (using MacQueen's method). The above-mentioned clustering algorithms were performed with different numbers of clusters 3 -6. The following clustering quality indices were used to determine the proper clustering algorithm and to establish the suitable number of clusters: the Silhouette index (Chen et al., 2002), the Calinski-Harabasz index (Zhao and Karypis, 2005;Shu et al., 2003) and the Dunn index (Bolshakova and Azuaje, 2003). For all indices, the highest value indicates the most appropriate cluster configuration. Based on the applied indices, K-means clustering with five clusters proved to be the proper configuration as it resulted in the highest quality indices. In order to validate the adaptive FCQ clusters, a leave-one-out (LOO) validation technique was used. During the LOO validation, each FCQ factor was omitted one by one and the stability of the cluster structure was examined. The established FCQ clusters proved to be stable during the validation. In the case of scale profile variables, an ANOVA test was applied to compare average cluster scores.
Furthermore, categorical data were analyzed by log-linear analysis. The relationship between four factors (FCQ clusters, age, gender, education level) was investigated. If two factors are not independent of each other, then an interaction effect exists between them.

Sample distribution
Census proportions a Name Individuals % % The saturated model contains all possible interactions between the studied factors while the restricted model contains only some of the interaction effects. Log-linear analysis uses Pearson χ 2 statistics in which the restricted model is compared to the saturated model. The null hypothesis is that there is no difference between the two models; therefore, the lack of difference ( p W0.05) indicates a good model fit. Among all possible interactions, the following conditional independence model could be fitted to our data: where U denotes the logarithm of the observed frequencies and × denotes interaction effect.
The R 3.0.3 statistical software package was used for all analyses, the clusterCrit package was used for measuring the cluster quality and the CFA was calculated by the LAVAAN package (LAtent VAriable ANalysis), while log-linear analysis was performed using loglm function in the MASS package.

Food choice
Since the nine-factor model which derives from the original FCQ statements is not appropriate for the sample, PCA analysis was used to discover what modifications occurred in the original model's factors (Table II).
As can be seen from the results, the original model can only be partially applied to this sample. Three other factors were listed among the statements. Of the original factors only sensory appeal, mood, familiarity and ethical concern remained in their original form. From the convenience factor, the items referring to ease of obtaining the food moved to the price factor, and together formed the price and purchase convenience factor. Health and natural content, and weight control, which were originally separate factors, entered the new model as a single factor. The new factor structure is shown in Table III. According to our results, seven factors could be identified. These were health and natural content, mood, preparation convenience, price, sensory appeal, familiarity and ethical concerns.
To conclude this stage of the investigation, clusters were developed from the seven factors which had emerged previously. The results are shown in Table IV.
From the seven factors, five clusters were created. With the first cluster (modern food enthusiast), factors 2, 5 and 3 were decisive. With the second cluster (tradition-oriented) it was factor 6, with the third cluster (optimizer) factors 4 and 5 and with cluster 4 (easy-choice) factors 6 and 3. With the fifth cluster (un-concerned) negatively valued factors appeared.
In order to better understand the internal structure of clusters, log-linear analysis was performed on the demographic background variables. The results are shown in Table V regarding age and education levels. The Pearson χ 2 value ( χ 2 (df ¼ 144) ¼ 167.05) indicated an acceptable model fit ( p ¼ 0.092).

FCQ structure
The first two questions of our research attempted to discover whether the original FCQ construction could be applied to our sample and, if not, how the structure of the factors could be modified. The nine-factor FCQ construction developed by Steptoe et al. (1995) has been widely used to investigate motivations behind the choice of foods. Later on, Lindeman and Väänänen (2000) extended the range of aspects with three ethical factors. During the widespread application of this measuring methodology several authors, such as Prescott et al. (2002), and Januszewska et al. (2011) demonstrated measurement invariance cross-culturally, while other authors' results did not support the generalizability of the FCQ's factor structure (these included Eertmans et al., 2006;Fotopoulos et al., 2009;Pula et al., 2014).
The factor structures which were developed during the examinations ranged from 5 to 13 factors, of varying content. As a result of our representative survey, it is established that the original nine-factor model is only partially applicable to the Hungarian sample. On the basis of our results, seven factors were distinguished, namely health and natural content (+ weight control), mood, preparation convenience, price and purchase convenience, sensory appeal, familiarity and ethical concerns. Comparing these to the results of other researchers, there is a clear tendency for the health and natural content (as well as, in some cases, weight control) factors to merge together (see Eertmans et al., 2006;Milošević et al., 2012;Ooi et al., 2015;among others). At the same time, another characteristic can also be observed, in which the convenience factor becomes divided into purchase and preparation convenience. Our results also support this, since the purchase convenience, when added to price, creates one factor, while preparation convenience is its own independent factor. This is also apparent in the work of Eertmans et al. (2006), Ares and Gámbaro (2007), Milošević et al. (2012), Gagić et al. (2014) and Ooi et al. (2015).
Regarding the order of importance of the various factors, the Hungarian sample order agrees with research studies that place sensory appeal, price and purchase convenience, and preparation convenience factors at the top of the list. (The average values of the individual factors are shown in Table III.) Similar preferences are reported with British (Steptoe et al., 1995) and Russian (Honkanen and Frewer, 2009) consumers as regards motivating factors in the choice of food.
Moreover, the partial results of Januszewska et al. (2011), according to which for Hungarian consumers the health factor is less important, confirm this, unlike, for example, the full samples on the same research, and unlike the evaluations of Steptoe et al.'s (1995) British sample, or Gagić et al.'s (2014) Serbian sample, where consumers consider health to be the second most essential aspect. At the same time, in our research Familiarity occupies a high, fourth place, while in other cultures it is considered to be among the least important. It seems that for Hungarian consumers (and, as expected, most particularly men) the fact that food is familiar is much more important, and they are much less open to novelty. However, for Hungarians ethical questions are also less important.
Analysis on the basis of gender revealed that three of the FCQ seven factors showed a significant difference. The three factors are health and natural content, price and purchase convenience, and familiarity. For women, health and natural content (weight control) was relatively more important, as well as price and purchase convenience, while for men familiarity proved to be the most essential. Compared to other FCQ measures, differences

1483
Adaptation of the Food Choice Questionnaire between the genders in the evaluation of the factors listed above also appear in the work of other researchers. The importance of the health factor among women respondents was noted in the original research of Steptoe et al. (1995). At the same time, Crossley and Nazir's (2002) results showed that among dental students, women were significantly more aware ofamong other thingsthe health and convenience aspects when they were choosing food. The differences are also confirmed in the research carried out by Januszewska et al. (2011), in which women over the whole sample gave more importance to natural content and weight control factors.
At the same time, there are research studies which have not found gender differences in the factors (e.g. Milošević et al., 2012). Although Gagić et al.'s (2014) study also showed that there is almost no gender difference between the importance of the factors, womenjust as in our studyevaluate convenience higher than men (although this refers to the original convenience factor). In their work, familiarity shows no significant difference, unlike in our research.
If we examine the order of the factors in terms of another background variable, age group, it becomes clear that the mood, price and purchase convenience, preparation convenience and ethical concerns factors show differences. Among these factors, it is ethical concerns which shows agreement with the age group-based results obtained by Steptoe et al. (1995). Similarly to the present research results, Prescott et al. (2002) found that it was the older consumers who evaluated the ethical concern and familiarity factors more highly than the young. However, the mood factor in our study was decisive in the under 30 age group, while in Prescott et al. (2002) it was more significant for older consumers. Rahman et al. (2013) similarly to our resultsfound age to have a significant negative correlation with mood (i.e. the older a person becomes, the less important this factor is for them).
On the basis of our results, it can be stated that in almost all factors there was a difference in terms of level of education, and compared to other background variables, it had a greater effect. Steptoe and Wardle (1999) showed differences based on educational status, but only in the case of price, familiarity, mood and sensory appeal factors. The evaluations made by those with lower educational qualifications, however, agree with those in our research, given that they gave great emphasis to price and familiarity factors. In this research, however, the greatest effect was observed in the cases of ethical concerns ( for those with high-level qualifications), and the preparation convenience ( for those with lower-level qualifications).

FCQ clusters
During the analysis, five clusters were created from the seven FCQ factors on the basis of differences in motivational preferences.
In the case of Cluster 1, called the "Modern food enthusiast" group, the mood, and the sensory characteristics were the most important factors, but the ease of preparation factor also had an effect. Based on the analysis, men and young people ( from 14 to 29 years of age), and people with higher and secondary education are more likely to belong to Cluster 1 than to other clusters. In fact, in terms of age group they represent the y and z generations. First, this age group cannot yet earn a living for themselves, so the primary criterion in the decision-making process is the current mood, or sensory appeal; second, this age group represents young adults (men) for whom long meal preparations are not attractive, but who expect a consistent sensory quality. In addition, this age group is most affected by the daily stress and challenges of life, and the food and meals consumed may actually help to combat stress and relieve tension.
In Cluster 2 or the "Tradition-oriented" group, as is clear from the name, consumers prefer to choose foods which they are familiar with. It is mainly men and middle-aged people ( from 30 to 59 years of age) with higher and secondary education who are more likely to belong to Cluster 2. The analysis of the FCQ factors also showed that familiarity is a particularly important factor for Hungarian customers, especially for men. It probably gives a sense of stability, and perhaps also assists them in making speedy choices, and therefore means not much cognitive capacity is required when purchasing food. It is noteworthy that this is the cluster where consumersif only slightlyalso take ethical aspects into account. Familiarity perhaps also represents a kind of guarantee of the place of origin.
Naturally, a price-oriented group emerged as well, the Cluster 3 group, referred to as "Optimizer," although this group cannot be termed particularly price-sensitive, since it was also important for them that the product had the right taste, aroma, texture and appearance. It is mainly middle-aged, and older women who belong to Cluster 3 (the proportion of consumers over 60 is significant). With regard to educational qualifications, there is no outstanding category. The cluster features reinforce the fact that the price sensitivity and purchase convenience is the most important motive for women, for whom the sensory/ gastronomic appeal of the food is just as important.
Cluster 4 included those who experienced a feeling of pleasure when purchasing food, the "Easy-choice" group. For them the familiarity of the food and its ease of preparation were important, but sometimes they were influenced by mood when making a choice. The likelihood of belonging to this cluster is almost equal in terms of gender distribution. Young people with lower qualifications are more likely to be included in this cluster.
The Cluster 5 group was termed the "Un-concerned," since we could not identify any decisive factor in their case; they approach the task of choosing and buying food with indifference, and do it as a routine. Older women (over the age of 60) are more likely to be members of this cluster as well as those with loweri.e. vocational and primary schooleducational qualifications.

Conclusions
During the examination, we sought answers to the three questions that were presented in the Introduction.
Regarding the first and second questions (can the original FCQ be applied to the sample?if the original model is not applicable, how can the factors be modified?), our conclusion was that the original FCQ model, with its 36 statements, cannot in its entirety be applied to the sample, although the differences experienced were minimal. Instead of the original nine factors, seven factors were identified in the sample. The study considered the reason for these differences to lie in cultural differences.
Another conclusion of the study is that the FCQ scalesalthough on a different factor structurehave also proved in the Hungarian sample that food choices and demographic characteristics can be described as clusters. These clusters provide an opportunity for public health policy to influenceon appropriate platforms and with target group-specific messagesthe food choice and consumption habits of the Hungarian consumer. Taking into account the fact that factors related to health claims, natural ingredients and ethical considerations did not show an outstanding value in a single cluster, a clear strategic direction for public health and nutrition policy is evident.
The results show that there are seven factors simultaneously present in the mindset of the Hungarian population. Accordingly, companies and public health organizations can tailor their media messages in a customized way, according to the way consumers think. The main messages are best formulated along the lines of health and naturalness, a good state of mind, food preparation comfort, price sensitivity and convenient availability.
The research also identified those target groups who could be targeted with messages in a differentiated way. For the modern food enthusiast group, the media message can be formulated along lines of good mood and enjoyment, while for the tradition-oriented cluster it can focus on intimacy and ethical considerations. For the optimizer, price, comfort and enjoyment can be an effective argument, while in the Easy-choice cluster all aspects that 1485 Adaptation of the Food Choice Questionnaire make it easily available are important. Last but not least, targeting and reaching the Un-concerned is doubtful as this group has no single value, except for the acceptance of favorable prices. Based on the results of the research, public health organizations have the opportunity to influence the health of the Hungarian population in a positive direction, an important element of which is the emphasis on conscious food choice.