Elham Makiabadi; MSc ¹, Mohammad Hossein Kaveh; PhD ^*2,Abdolrahim Asadollahi; PhD ³ & Jeyran Ostovarfar; MSc ^2
1 Department of Community Nutrition, School of Nutrition and Food Sciences, Shiraz University of Medical Sciences, Shiraz, Iran.
² Department of Health Education and Promotion, School of Health, Shiraz University of Medical Sciences, Shiraz, Iran.
³ Department of Healthy Aging, School of Health, Shiraz University of Medical Science, Shiraz, Iran.

 
Introduction

Health literacy is considered as one of the basic skills required for challenging health-related decisions. However, evidence indicates deficiencies in individuals' knowledge and ability for self-management and particularly health literacy related to the nutrition (Gibbs et al., 2015). Nutrition literacy can be defined as a degree in which individuals have the capacity to obtain, process, and understand basic nutrition information (Zoellner et al., 2009). Studies revealed some of the causes or factors involved in nutritional literacy such as following nutritional standards, interpretation of food labels, and appropriate decisions about diet (Gibbs and Chapman-Novakofski, 2013, Zoellner and Carr, 2010). Nutritional literacy as a skill-based process causes people to identify and transform nutrition messages into knowledge. In general, food choices in people having sufficient nutritional knowledge are healthier (Parmenter et al., 2000, Yajima et al., 2001).
Although health literacy plays a vital role in decision-making related to health and nutrition, it still has not an ideal situation in the communities. According to surveys, approximately 80 million of American adults (36%) (Berkman et al., 2011) and 56.6% of the participants in five Iranian states had inadequate health literacy (Tehrani Banihashemi et al., 2007). In Zollner's study, 24% of the participants reported very low, 28% reported low, and 48% reported adequate nutrition literacy (Zoellner et al., 2009). Another study also indicated that adults in Jiangxi Province in Eastern China had low level of nutrition literacy (Du et al., 2010).
Theories of health behavior help us to identify behavior determinants and design targeted educational intervention (Sharifirad et al., 2008). The predictive power of the theory of planned behavior (TPB) was proved in many social and health behavior studies, such as food behavior, physical activity, self-care, and screening tests (Kassem et al., 2003). It is also one of the most important models in the field of food choices (Shepherd and Towler, 1992). The TPB assumes individuals as logical actors, so that they process the information before performing a behavior. During this process, basic individual beliefs and consequently, individual behavior may change (Rashidian et al., 2006). Accordingly, the most important predictor of adopting a behavior is individuals’ intention to perform that behavior. The intention is determined by the attitude toward behavior (ATB), subjective norm (SN), and perceived behavioral control (PBC) (Sharma, 2016).
Based on the literature, it is necessary to develop and improve tools that can accurately assess individuals’ nutrition literacy in public health systems (Ndahura, 2012). Therefore, the present study aimed to develop and validate the Nutrition Literacy Promoting Behavior based on TPB (TPB-NLPB) questionnaire in the youth in southern Iran in 2017.
Materials and Methods
This cross-sectional study aimed at developing and validating the TPB-NLPB. In this study, the cluster random sampling method was used to select 203 students (103 men and 100 women), who were in four dormitories of Shiraz University of Medical Sciences in southern Iran.
Sample size determination: The sample size was calculated using 1:5 N/p ratio, i.e., the ratio of the number of item to participants. This indicated that five responders were required for each question on the scale (De Vet et al., 2005). Therefore, the 37-item questionnaire required a sample size of 185 participants. The inclusion criteria were active students of the undergraduate levels, who were willing to participate in the study. Exclusion criteria included having an incomplete questionnaire.
Development the instrument: The assessment tool for nutrition literacy was a self-administered inventory. This questionnaire was prepared by the research team through literature review to achieve the research goals (Francis et al., 2004, Ndahura, 2012, Pettersen et al., 2009, Song, 2014, Vanderlee and Hammond, 2014, Zoellner et al., 2009). Then, the face and content validity of the initial version of the questionnaire were evaluated and confirmed by a panel of experts (n = 6) after some revisions. The content validity ratio (CVR) and content validity index (CVI) values were also calculated. In the next step, face validity was assessed using a pilot test on 10 participants from the study target group; these participants were not interviewed in the next step. The necessary revisions were made to the final application test. Finally, for the main stage of the research, the tool was used on a sample of 203 people from the research population.
The TPB-NLPB included four subscales: attitude toward behavior (12 items, e.g., information on the nutrition labels are not related to the food quality), subjective norm (5 items, e.g., if I do not use nutrition labels for food choices, I feel I am under social pressure), perceived behavioral control (13 items, e.g., I do not understand the meaning of nutritional information on the food products), and behavioral intention (7 items, e.g., I intend to become acquainted with recommendations of the WHO for fruits and vegetables until the next month). Each item should be responded on a five-point Likert scale. For positive questions, scores of five, four, three, two, and one were assigned to “Strongly agree”, “Agree”, “Neither agree nor disagree”, “Disagree”, and “Strongly disagree” options, respectively. The negative items were scored reversely. The attainable scores could range from 37 to 185.
Data collection: When collecting the data, the researcher visited dormitories of Shiraz University of Medical Sciences, introduced herself to the students, and explained the study objectives. The Persian TPB-NLPB was then distributed among the participants and the questionnaires were answered anonymously with a response arte of 98%.
Data analysis: All statistical analyses were conducted using the SPSS-23. To assess the questionnaire reliability, its internal consistency was determined. In this approach, a Cronbach’s alpha coefficient of > 0.7 was achieved that represented acceptable reliability for the instrument (Jones et al., 2004). The reliability of the TPB-NLPB was also assessed using the split-half method, and the Spearman-Brown coefficient was calculated for the whole scale and its halves. This index is normally used in questionnaires with a large number of questions (Seif, 2004).
The CVR was used to quantify the experts’ agreement. In general, a CVR score of 0.80 or higher indicates good content validity (Lawshe, 1975). In this study, the CVR mean for essentiality criterion and the CVI means for simplicity, clarity, and relevance criteria were calculated as 0.91, 0.91, 0.95, and 0.98, respectively. As a result, questions that had problems were reviewed and revised after consulting with the experts in health education and nutrition sciences. Next, the students answered the modified questionnaire. In addition, confirmatory factor analysis and fitness indicators were used to confirm the validity of the questionnaire.
Ethical considerations: Regarding the ethical considerations, the participants were assured that their information would be kept confidential. Later, informed consent forms to enter the study were obtained from all participants. All participants received verbal explanation about the study objectives and procedures and then signed written informed consents for taking part in the study. The participants were also reassured about the anonymity and confidentiality of their information. All procedures of the study were in accordance with the ethical standards of the institutional or the national research committee and with the 1964 Helsinki declaration. Its later amendments including informed consent and confidentiality of all personal information were also in accordance with the ethical standards.
Results
A total of 203 students were investigated; 103 men (51.0%) and 100 women (49.0%) with a mean age of 22.32 years (SD = 2.18). The mean weight of the male and female participants were 70.33 kg (SD = 12.41) and 57.13 kg (SD = 8.20), respectively. The mean height of the male and female participants were also 176.82 cm (SD = 0.70) and 161.85 cm (SD = 0.05), respectively. All the students replied to TPB-NLPB. Table 1 shows the demographic characteristics, comparing total score of TPB-NLPB and its classified scores made between males and females regarding their gender and the study field; it did not reveal any statistically significant difference between the samples except in the study field (See Table 1).
Internal consistency: The coefficients of Cronbach’s alpha (α = 0.87), Guttmann method (λ₁ = 0.84 to λ₆ = 0.91), and convergent validity (0.74) were estimated, which were significant at P < 0.01. The discriminative power in the TPB-NLPB of sub-scales with overall score using Kolmogorov–Smirnov and Shapiro–Wilk tests of the normality demonstrated a normal distribution of data (K-S = 0.170, S-W = 0.085, P > 0.05). Mean overall score of TPB-NLPB was 129.08 (CI = 127.11-131.05), SD = 14.24, Skewness = 0.022, Kurtosis = 0.953, with a minimum and maximum of 86 and 173, respectively. Discriminative power testing showed that the domains had a normal distribution.
Regarding the criterion validity, Pearson’s correlation coefficients were significant and appropriate for all the sub-domains of TPB-NLPB. This finding could suggest some specificity of these domains.
Contrast validity: According to Table 2, the Kaiser-Meyer-Olkin (KMO) test and Bartlett’s test of sphericity were conducted to evaluate the factorability. The KMO was 0.788 and the Bartlett’s test of sphericity was less than 0.001, meaning that Exploratory Factor Analysis (EFA) could be applied to the obtained factors (See Table 2). The diagonals of the anti-image correlation matrix were all over 0.54, confirming inclusion of each item in the factor analysis.  Finally, the communalities were all above 0.4.
The exploratory factor analysis demonstrated that 37 items of TPB-NLPB were organized into five factors, explaining 64.91% of the scale variance (Initial eigenvalue = 1.090). This scree plot shows that five factors explain the most variability since the line starts to straighten almost after factor five (See Figure 1).
Second-order confirmatory factor analysis indicated that the factors were designed well upon a principal factor. According to Table 3, the rotated factor matrix pattern of Varimax was considered for the TPB-NLPB's subscale questions. In this regard, questions with factor loadings above 0.35 were selected (See Table 3).
Covariates exist between some items, i.e., item ATB5 between factors No. 1 and 2; item ATB10 between factors No. 1 and 2; and item PBC7 between factors No. 1 and 5 in Persian version of TPB-NLPB. This may indicate that the covariate item of these factors can be reconstructed. Additionally, no items were removed. Table 4 presents the study descriptive statistics. The skewness and kurtosis results were within a tolerable range for assuming a normal distribution. Furthermore, Varimax rotation was applied and the least correlations between each of the composite domains were reported.
Confirmatory factor analysis: The maximum likelihood with robust standard errors and chi-square (ML) were used for estimation of the method available in AMOS-22. The aim was to confirm fitness of the five-factor model that emerged from EFA. The adequacy of the model was examined using the comparative fit index log likelihood, Akaike's information criterion (AIC), finite sample corrected AIC (AICC), Bayesian information criterion (BIC), consistent AIC (CAIC), root mean square error of approximation (RMSEA), goodness-of-fit statistic (GFI), adjusted GFI, incremental fit index (IFI), and comparative fit index. These values for the five-factor 37-item model suggest that the model provides a moderately good fit. Consequently, the five-factor model was appropriate for the data and the fit index techniques were appropriate for adjusting the scale. According to Table 5, indexes of the model's goodness of fit refer to integrity of the five-factor model with data. The χ² degree of freedom was less than two in efficient models; it is better when it is closer to zero. The root mean square error of approximation (RMSEA) must be less than 0.05 to indicate a good model (Asadollahi et al., 2013, Asadollahi et al., 2016). The model pointed out goodness of fit of TPB-NLPB. Additionally, AIC, AICC, CAIC, and BIC (Schwarz criterion) values are shown in bold face to indicate that the corresponding model is favored by the criterion.
Measures closer to 1 in the comparative fit index, GFI, the adjusted GFI, and the incremental fit index, refer to the model's goodness and fit (Baumgartner and Homburg, 1996, Doll et al., 1994). In our study, these measures were more than 0.90 (see Table 5). Results from the factor analysis indicated that all the item loadings were significant at the P < 0.05 level and all were above 0.7. This indicates further improvement for generalizability of the revised model to samples in academic settings.