1. Introduction
Traditionally, food safety knowledge has been seen as an important factor in improving food safety behaviour, and a substantial amount of information regarding consumer knowledge and self-reported practices has been reviewed [
1,
2,
3,
4,
5]. However, there is no doubt that a complex relationship exists between knowledge and behaviour [
6,
7].
Mullan et al. [
8] criticised interventions targeting food safety based on knowledge. Applying the Theory of Planned Behaviour (TPB) model, knowledge was demonstrated to be a necessary but not a sufficient condition for particular behaviours, and when other factors, such as norms and perceived control, were taken into accounts in the analysis, the knowledge factor was not the sole predictor of behaviour [
8]. Food safety practices refers to handling in terms of cleaning, cold food storage, cooking, chilling, hand-washing, and avoidance of cross-contamination. The combination of Knowledge, Attitude, and Practice has been termed KAP [
6,
9,
10]. Da Cunha [
11] highlighted several caveats of the KAP approach related to difficulties translating knowledge into practices. Human beings are complex and factors such as optimistic bias, lack of motivation or inadequate infrastructure may hamper optimal behaviour [
8,
11]. Furthermore, risk perception is a complicated concept; people are generally more likely to believe that they will win the highest profit rather than that they will suffer from something negative, such as, for instance, food poisoning. This “optimistic bias” or “risk related optimism”, defined by Weinstein [
12], may prevent consumers from absorbing information and following rules.
Actually, attitudes have been studied in relation to behaviour for decades. Ajzen [
13] proposed that an individual’s attitude towards any object is a function of the strength of his or her personal beliefs about the object and the evaluative aspect of those beliefs. If people believe that a certain behaviour will lead to a desirable outcome, then they are more likely to have a positive attitude towards the behaviour. Alternatively, if individuals believe that a certain behaviour will lead to an undesirable or unfavorable outcome, then they are more likely to have a negative attitude towards the behaviour [
13].
Illustrating the relationship between attitudes and food safety behaviour, Unklesbay et al. [
14] developed an early survey instrument to assess the attitudes, practices, and knowledge of more than 800 college students. It was shown that women who had enrolled in a course including food safety information had significantly higher scores for both attitude and practice.
Relationships between knowledge, attitudes, and behaviour are more nuanced in this case because food safety attitudes might be crucial when it comes to food safety behaviour; international investigations using Structural Equation Model (SEM) may confirm this. Ko [
15] used SEM to investigate the relationship between food safety knowledge, attitude, and Hazard Analysis Critical Control Points (HACCP) practices among 421 restaurant employees. It was similarly found that attitude mediated the relationship between knowledge and HACCP practices. Baser et al. [
16] used SEM on food safety data regarding knowledge, attitude, and behaviour for 498 hotel staff and did not find any relationship between food safety knowledge and behaviour. However, the results from their analysis showed that there was a positive relationship between knowledge and attitude, as well as between attitude and behaviour. Furthermore, Sanlier and Baser [
17] examined the relationships between food safety knowledge, attitudes, and self-reported behaviours using SEM. In a survey targeting 1219 young women, the mediating role of attitude was demonstrated. Results from that path analysis did not show any strong causal relationship between food safety knowledge and behaviour, while the results indicated a full mediation of the effect of knowledge on behaviour by attitude [
17].
Limited data have been collected regarding the influence of different factors on food safety behaviours in Sweden, although these findings provide support for the hypothesis that the relationship between food safety knowledge and behaviour is mediated by attitude. These interactions are further investigated here. A Swedish survey performed on 606 university students has demonstrated that food safety education made a difference when it comes to food safety knowledge [
5]. There were a significantly higher number of correct answers on a food safety knowledge questionnaire and this was correlated with more optimal self-reported food safety behavior. Thus, the investigation indicated a positive correlation between food safety education, knowledge and optimal self-reported food safety behaviours [
5]. The present SEM study proposes to include the Background factor in order to possibly demonstrate a relationship with the factors of knowledge, attitude, and behaviour and to further evaluate the mediating role of attitude. This may have implications for the development of food safety education. International investigations using SEM have investigated the relationship between food safety knowledge, attitude, and behavior [
15,
16,
17]. However, the novelty of this current investigation, as far as the authors are aware, is that there has not been an SEM analysis of food safety data which included the Background factor. In the present study, the Background factor includes the following issues: genus; experience cooking and handling different food items; experience taking a food safety course at different levels; and foremost source of food safety knowledge (
Table 1). Specifically, the present study aims to investigate self-reported food safety background, knowledge, attitudes, and behaviour among university students in Sweden using SEM to examine factors influence on food safety behaviour.
3. Structural Equation Model
A structural equation model (SEM) is a tool for analysis of the interrelationships among latent variables measured using multiple correlated observable indications. The SEM in the present study was performed on the dataset in two steps. In the first step, a selection was made based on the relevance of the type of question to the objective of the analysis. Out of 606 respondents in the study by Marklinder et. al. [
5], 408 were deemed usable for the SEM based on the respondent’s choice to opt out on certain questions in the questionnaire. The opt-outs were taken to include all observations from respondents who answered questions indecisively, such as “I don’t eat meat” or “I avoid this food for [a different reason]” or “I do not know” on any of the questions in the questionnaire. The original data contained many variables. Therefore, a selection of variables was undertaken according to determined criteria that were relevant for the analysis. To be sure of selecting them correctly it was performed in consultation with a food safety expert.
In the second step, factor loadings were tested to establish which variables could be used in the analysis. A factor loading is a standardized measure of the relationship between variables and the underlying structure, ranging from −1 to 1. A loading that is closer to 1 means that there is a strong effect between the variables and its factor or between factors. When forming factors in SEM, a loading from a variable to the latent variable is deemed acceptable at or above 0.5 [
19]. However, according to Hair et al. [
19], a loading of 0.3 is seen as sufficient to form the structure of a factor. Further, Matsunaga [
22] pointed out that, as applied to social studies, loadings can be as low as 0.2. This cut-off point for factor loadings was applied to this study, along with the relevance of the variables themselves. More than half of the variables had loadings sufficient to their respective factors to be included, and all were relevant to the study. The theoretical underpinnings of the structure used for this analysis have been established by Ko [
15], Baser et. al. [
16] and Sanlier and Baser [
17]. In addition, the Background factor was added to the present study.
Figure 1 illustrates the four factors used for the model: Background; Knowledge; Attitude; and Behaviour.
3.1. The Factors
The different variables (B
1–B
9; K
1–K
8; A
1–A
7; H
1–H
6) forming the factors are explained in the
Table 1,
Table 2,
Table 3 and
Table 4. These are the variables that had the appropriate factor loadings to be included in the model. B
1–B
9 are variables for Background; K
1–K
8 are variables for Knowledge; A
1–A
7 are variables for Attitude; and H
1–H
6 are variables for Behaviour. In this study, the model path analysis has the novel addition of testing the causality of how a respondent’s Background affects Knowledge.
The questions used for Background were mainly concerned with whether respondents were women or men, had experience of cooking or handling different food items, the foremost source of their food safety knowledge, and whether it was informal (family and friends) or formal (food safety education) (
Table 1). All variables were treated as binary, except for B
2 and B
3 which were on an ordinal scale.
The variables for Knowledge in the questionnaire, formed true/false questions with one or multiple answers, have been analyzed as binary data. These are dummy variables where the true answer to the question is valued as 1 and the rest are valued as 0 (
Table 2).
The variables for Attitude included dealing with the importance of washing hands before food handling, after handling of raw minced meat, raw chicken, or raw eggs, and after toilet visits, as well as with cold food storage. One variable was an evaluation of the respondent’s level of food safety knowledge. The variables for Attitude were treated as ordinal. Response options were assessed as the means of six semantically different scales: “Very important”; “Rather important”; Neither important nor unimportant”; Not especially important”; Not at all important”; I have never been in this situation”/I never use leftovers” (
Table 3).
The variables for Behaviour (H1–H6) were treated as ordinal variables where the least correct behaviour has the lowest score and the best behaviour has the highest value (
Table 4).
3.2. Data Analysis
This structural equation model (
Figure 1) is based upon ordinal data based on the nature of the questions in the questionnaire. As an example, in terms of attitude variables (
Table 3), this ranges from 1, meaning not at all important, to 5, meaning very important. Behaviour variables (
Table 4) are judged by what researchers deem to be the least desirable behaviour to the most desirable behaviour. As we use ordinal data, we used the diagonally weighted least squares (DWLS) estimator and polychoric correlations, in accordance with Yang-Wallentin et al. [
23]. Binary variables such as dummy variables can be seen as a special case of ordinal variables. Because there is an intrinsic ordering in our study between how the respondents answered on binary questions DWLS can be used.
The collected data were processed and analysed in RStudio using the Lavaan package in order to perform the structural equation modelling. The Structural Equation Model’s goodness of fit was judged from four goodness of fit (GFI) sets; their indices are presented in
Table 1. They consist of the comparative fit index (CFI), from comparison of the observed model to a null model; the absolute indices Root Mean Square Error of Approximation (RMSEA); the chi-square; and the normed chi-square (
χ2 &
χ2-normed). The chi-square test is usually used for testing the significant differences between observed data and estimated data. However, in SEM it is desirable to look for no differences between the data, i.e., a low observed
χ2-statistic. Nevertheless, having performed a chi-square test, it is recommended to use
χ2-normed as well as RMSEA, as both correct for chi-square inflation, along with CFI, which is the most widely used indice.
For this analysis, Hair et al. [
19] recommend Goodness of Fit Index, as shown in
Table 5, i.e., a
p-value that is lower than 0.05, RMSEA lower than 0.08, and CFI larger than 0.90. These cut-off values are similar to those used by Ko [
15], Baser et al. [
16], and Sanlier and Baser [
17].
6. Conclusions
The structural equation model for this study confirmed that the Background factor, involving topics such as having experience of food safety education or declaring that the foremost source of food safety knowledge was a formal university/college course, strongly influenced Knowledge. The Knowledge factor, in turn, strongly affected Attitude and did not directly affect Behaviour in the same way as Attitude. Thus, the Attitude factor seemed to have a mediating role between Knowledge and Behaviour. The hypothesis that the relationship between food safety knowledge and behavior is mediated by attitude was confirmed.
This study indicated that attitude has a stronger impact on behaviour than knowledge, which may have an impact on food safety behaviour. This has implications for the development of food safety education, and warrants further investigation and practical development.