Evaluation of psychometric properties of the eating restriction questionnaire and food involvement inventory in Iranian adults

The objectives of this study were to translate and validate the Persian version of the food involvement inventory (FII) and eating restriction questionnaire (ERQ) and to determine the measurement invariance based on gender, body mass index (BMI) status, and age. This cross-sectional study included 1100 Iranian adults. Exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) were used to evaluate the construct validity of FII and ERQ. Convergent and discriminant validity, measurement invariance in gender, BMI and age, reliability including internal consistency, and stability were investigated for FII and ERQ. The results showed that the four-factor construct of the FII and the one-factor construct of the ERQ were 44.27% and 55.12% of the total variance, respectively. The factor loadings of all items were > .3 in both scales and none of the items were deleted. Fitting indices indicated that the four-factor construct of the FII and the one-factor construct of the ERQ had a good and acceptable fit among the Iranian adults. The Persian versions of the FII and ERQ, translated into Persian and localized according to international standards, had high construct, convergent and discriminant validity as well as high reliability.

Instruments.Demographic characteristics questionnaire.This questionnaire includes age, gender, educational level, marital status, residence, occupational status, economic status, weight, height, and BMI.
Eating restriction questionnaire (ERQ).This ten-item questionnaire developed by Jezewska-Zychowicz et al. (2020), contains restrictions on the amount of food, cereals, bread, potatoes, sugar or sweets, fats, high-fat foods, meats, dairy products, fish, raw fruits and vegetables 6 .In this study, the ERQ was developed by reviewing texts and interviewing 10 adults who met the inclusion criteria, and three items were added to the questionnaire: "I have eating restriction on nuts (walnuts/peanuts/almonds/pistachios/hazelnuts), " "I have eating restriction on cereals", and "I have eating restriction on birds' eggs (chicken or other birds' eggs)".

Food involvement inventory (FII).
The FII designed and psychometrized by Lee et al. (2019) reflects an individual's current food lifestyle and includes 25 items, and dimensions of affective, cognitive, behavioral-purchase, and behavioral-cooking attitudes.This instrument was scored based on a 5-point Likert scale ranging from 1 to 9 (strongly disagree = 1, disagree = 3, neither disagree nor agree = 5, agree = 7, and strongly agree = 9) 10 .

Translation process.
In developing the Persian version of the FII and ERQ based on the WHO protocol (2015), the forward-backward translation technique was used 19 .In the present study, the translation process was as follows: (a) permission from the instrument designer, (b) forward translation from English (original language) into Persian (target language), (c) reconciliation and matching of forward translations, and (d) backward translation from Persian (target language) into English (original language).Then, instruments' psychometric properties (validity and reliability) were examined in adults.
Validity.The current study examined face, content, construct, convergent, and discriminant validity.
Face validity.Face validity was determined qualitatively and quantitatively.To quantitatively evaluate face validity, the instruments were given to ten members of the target group and they were asked to comment on the levels of difficulty, irrelevancy and ambiguity of each item through individual and face-to-face interviews.To quantitatively determine face validity, the impact item was calculated using the following formula: Impact item ≥ 1.5 remained in the study 20 .
www.nature.com/scientificreports/Content validity.Content validity was determined qualitatively and quantitatively.To quantitatively assess content validity, ten experts (experienced in qualitative research and instrument development) assessed grammar, wording, item allocation, and scaling of tools.
To quantitatively determine content validity, the content validity ratio (CVR) and content validity index (CVI) 21 were evaluated.The same experts gave their opinion on each item using the following formula: Ne = The number of experts who selected "necessary", N = The total number of experts.
The minimum acceptable CVR was determined based on the Lawashe table 22 .The number of experts was 10, so the acceptable value of CVR was ≥ 0.62.The method of Waltz and Bausell 23 was used to test the CVI.The same experts determined the relevance of each item and calculated the CVI using the following formula: Items with a CVI > 0.79 were accepted, whereas items with a CVI of 0.70-0.79were revised 23 .
Construct validity.Construct validity was assessed through exploratory factor analysis (EFA), confirmatory factor analysis (CFA) as well as convergent and discriminant validity.A cross-sectional study was performed to evaluate the construct validity.Samples were selected using the convenience sampling method.Inclusion criteria were 18-60-year-old persons having literacy, as well as exclusion criteria included not completing the questionnaires completely.After measuring BMI, the research instruments were completed by the questioners and with valid instruments by the sample population in the presence of the researcher and given to the researcher.The presence of the researchers during the distribution of the questionnaires, the explanation of the accuracy in the completion of the questionnaires, the explanation of the research objectives, the maintenance of the confidentiality of the participants' information, and the validity of the instrument used to measure height and weight assured us of the quality and accuracy of the information collected.To measure the validation of ERQ and FII constructs, all samples (N = 1100) were randomly divided into two subgroups of 550 individuals.The first subgroup included 302 women and 248 men (M age = 37.73, SD = 11.56;M BMI = 26.77,SD = 6.36), and the second subgroup consisted of 262 women and 282 men (M age = 37.53, SD = 10.62;M BMI = 27.35,SD = 4.76).Exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) were conducted to assess construct validity.

EFA.
The EFA was performed on the first group samples (N = 550) for the FII and ER.The Kaiser-Meyer-Olkin Measure (KMO) and Bartlett's sphericity tests were applied to assess sample adequacy and sphericity, respectively.Then, the latent factors of both instruments were extracted using the principal axis factoring (PAF) method, varimax rotation, and scree plot.The presence of a single item in the factor was approximately 0.3 based on the following formula: The CV is the number of extractable factors and n is the sample size of the study 24 .
CFA. CFA of the two instruments was conducted on the samples of the second group (N = 550).Model fitting was carried out using the goodness of fit indices (GFI), Satorra-Bentler scaled chi-square test (S-B χ 2 ), comparative fit index (CFI), Tucker-Lewis index (TLI), standardized root mean square residual (SRMR), root mean square error of approximation (RMSEA), and confidence interval (CI) of 90% 14 .
Convergent and discriminant validity.The correlation between the FII and ERQ, the correlation between the subscales, and the correlation between age, gender and BMI status with the scales and subscales were investigated.

Measurement invariance.
Measurement invariance means that a measurement instrument measures the same psychological structures in different societies.In the current study, measurement invariance in gender, BMI status and age was evaluated based on multigroup CFA using MPlus 6.1 software 25 .The first stage of configural invariance was investigated by assuming the equivalence of model factors among groups, and other models were compared with configural invariance.In the next step, metric invariance was evaluated by assuming equivalence of factor loadings among groups.In the third stage, scalar invariance was assessed by assuming the equivalence of factor loadings and item intercepts among groups.Moreover, the ΔCFI index (CFI change between two nested models) was used as a model equivalence index.

Reliability.
Internal consistency was evaluated based on the Cronbach's alpha coefficient, McDonald's omega (Ω) and average inter-item correlation (AIC).The intraclass correlation method was used to evaluate the stability.In this method, 40 individuals who were randomly selected from all initial participants completed the FII and ERQ for the second time in a two-week interval.
The number of the specialists who have checked option3and4 The total number of specialists .In the present study, to evaluate the EFA, the R4.5 software with the Psych and Polycor packages was used for the FII and ERQ, respectively.MPlus6.1 software was used to evaluate the CFA and invariance (configural, metric, and scalar) of the instruments.Convergent and discriminant validity was tested with SPSS 24 using the Pearson correlation test.In addition, the receiver operating characteristic (ROC) curve was utilized to determine the cut-off point.

Ethical aspects. The current study was approved by the Ethics Committee of Babol University of Medical
Sciences (IR.MUBABOL.HRI.REC.1400.046).All methods were carried out in accordance with the relevant guidelines and regulations.The objectives of the ongoing study were explained to all participants and informed written consent was obtained from them.Participant's right to withdraw from the research and maintain confidentiality was observed.Face and content validity.No items were deleted when studying face validity qualitatively but the statement "When I buy food, I check the information on the package" was deleted when evaluating face validity quantitatively because its value was less than 1.5.The appearance of the items was changed when studying face validity qualitatively.This scale was evaluated by experts for content validity.The CVR and CVI of all samples were acceptable.No item was deleted in this section.

Construct validity. EFA of FII.
The results of KMO (0.928) and Bartlett's sphericity test (5504.831)(P < 0.001) indicated that the data were suitable for EFA.Four factors with eigenvalues > 1 were identified for the FII construct and confirmed based on the screen plot.The data were rotated by varimax rotation, and the fourfactor construct of the FII accounted for 44.27% of the total variance.Totally, the first 8-item factor "behavioralcooking attitude", the second 6-item factor "affective attitude", the third 5-item "behavioral-purchase attitude" and the fourth 5-item "cognitive attitude" accounted for 13.01%(eigenvalue = 3.12), 12.75% (Eigenvalue = 3.05), 10.07%(eigenvalue = 2.41) and 8.44% (eigenvalue = 2.03) of the total variance, respectively.Additionally, there was a weak correlation (< 0.3) between the four factors.The findings revealed that factor loadings of all items were > 0.3 and no items were deleted.The correlation between all items and the total score was higher than the minimum acceptable value of 0.3 (Table 1).

EFA of ERQ.
The KMO (0.901) and Bartlett's sphericity test (2339.008)(P < 0.001) illustrated that the data were suitable for EFA.A factor with a eigenvalues of > 1 was identified for the ER construct and confirmed according to the screen plot.The data were rotated by varimax rotation, and in total, one-factor construct of ER allocated 55.12% (Eigenvalue = 7.33) of the total variance.Factor loadings of all items were > 0.4 and no items were deleted.The correlation between all items and the total score was higher than the minimum acceptable value of 0.3 (range = 0.47-0.72,mean = 0.530) (Table 2).

Reliability (internal consistency and stability) and convergent and discriminant validity.
The internal consistency values of all samples based on Cronbach's alpha were 0.814, 0.793, 0.767, and 0.754 for the FII subscales, including behavioral-cooking attitude, affective attitude, behavioral-purchase attitude and cognitive attitude as well as the total score of internal consistency of FII and ERQ was 0.786 and 0.865, respectively.The internal consistency values of all samples based on McDonald's omega were 0.811, 0.792, 0.765, and 0.753 for the FII subscales, including behavioral-cooking attitude, affective attitude, behavioral-purchase attitude and cognitive attitude as well as the total score of internal consistency of FII and ERQ was 0.780 and 0.854, respectively.The results exhibited that the intraclass correlation coefficients (ICC) were 0.95, 0.91, 0.94, and 0.93 for the FII subscales consisting of behavioral-cooking attitude, affective attitude, behavioral-purchase attitude, and cognitive attitude as well as the ICC total score of FII and ERQ was.92 and 0.96, respectively, which were significant at the level of 0.001.
Table 3 shows the correlation between the FII and ERQ and the variables of age, gender and BMI status.There was a moderate to weak correlation (0.205-0.477) between all FII factors.In addition, a positive and significant www.nature.com/scientificreports/correlation was found between BMI status, gender, FII and ERQ.Furthermore, a positive and significant correlation was observed between age and ER (r = 0.478) and a negative and significant correlation was seen between age, affective attitude (r = − 0.267), and cognitive attitude (r = − 0.301).There was a significant negative correlation between FII and ERQ.
Analysis of measurement invariance based on gender, BMI status, and age.Before analysis of measurement invariance, the fitting status of the four-factor model of the FII and one-factor model of the ERQ was evaluated based on gender, BMI status, and age of the samples.Then, configural, metric, and scalar invariances were assessed based on gender, BMI status, and age.The results revealed that by observing the values of factor loadings and intercept, the CFI change between configural and constrained models (∆CFI) was < 0.01.Thus, the configural, metric and scalar invariances in the FII and ER scales were confirmed based on gender, BMI status, and age (Tables 4 and 5).

Cut-off point of FII.
A cut-off point of FII was determined by considering the ERQ scale based on the ROC analysis.The area under the ROC curve was 0.851 with a standard error of 0.023 (CI 0.807-0.896)(Fig. 1).

Discussion
The aim of the present study was to translate the FII and ERQ scales, to evaluate their psychometric properties and to determine the measurement invariance in terms of age, gender and BMI status.The results of evaluating KMO and Bartlett's sphericity test indicated the suitability of the data for EFA.Since the KMO ranging from 0.7 to 0.8 and from 0.8 to 0.9 is good and excellent, the KMO values were excellent for both scales of the current study 26 .
The results of the current study showed that the four-factor construct of FII including behavioral-cooking attitude, affective attitude, behavioral-purchase attitude and cognitive attitude accounted for 44.27% of the total variance.This value was 63.32% in the study of Lee et al. (2019)  10 .The results of the present study suggested that there was a weak correlation between the four factors of FII, indicating that this scale had four independent factors.The findings of the ongoing study exhibited that in total, one-factor construct of ER allocated 55.12% of the total variance, and based on the screen plot diagram, it was confirmed that the ER scale was a single factor.
The results demonstrated that the correlation between all items and the total score in both scales was higher than the minimum acceptable value, representing that all items on both scales had inclusion criteria for performing EFA.Based on Cohen's recommendations, the correlations of 0.20, 0.40 and 0.60 were classified small, moderate and large, respectively 27 .
Since the cut-off points of CFI and TLI > 0.90 as well as SRMR and RMSEA ≤ 0.08 were considered as acceptable and good limit 28 the fitting indices of the present study showed that the four-factor construct of FII and one-factor construct of ERQ had good and acceptable fit in Iranian adult community.
To evaluate the reliability of the scales, internal consistency was investigated based on Cronbach's alpha.The results showed that the internal consistency values of all samples in the FII scale, its subscales and ERQ scale were higher than the recommended value of 0.7 29 .On the other hand, considering that in this study, the Ω and α coefficients of the FII and ERQ scales were greater than 0.7 and the AIC values were between 0.2 and 0.4; therefore, the findings represented that both scales had good internal consistency 30 .The findings of the present study suggested that the ICC for the scale and subscales of FII as well as the ERQ scale was higher than the recommended value of 0.8 29 , illustrating that the two scales had acceptable stability.The FII reliability in the study of Lee et al. (2019) was confirmed based on an overall Cronbach's alpha coefficient of 0.94 and item discrimination (with corrected item-total correlations greater than 0.40; ranging from 0.44 to 0.77) 10 .
Overall, the results manifested that both scales had appropriate and acceptable convergent and discriminant validity 29 .There was a significant positive correlation between BMI status and gender with FII and ERQ scales, representing that FI and ER were higher in overweight adults than in normal-weight ones and in women than men.
The findings revealed that ER in adults increased with age.Furthermore, there was a significant negative correlation between age with affective attitude and cognitive attitude, meaning that with increasing age, a person's cognitive and affective attitude decreases.It seems that as people get older, they talk less about food, food is less important for their lives or they spend less time and effort choosing food, which may increase the risk of malnutrition in these people.
Studies suggest that cognitive and affective components play an important role in weight gain or weight loss 31,32 .For example, people who are overweight or obese eat regardless of their physiological state and are more likely to react to internal situational or emotional factors such as negative mood, fatigue or boredom.These people are more prone to emotional eating, which is detrimental to their self-efficacy and motivation to maintain weight loss over time 33 .Therefore, screening for eating disorders and FI as risk factors for malnutrition in individuals > 35 years old, overweight women and overweight people can help prevent malnutrition in adults.
The results illustrated that there was a significant negative correlation between FII and ERQ.In this way, with increasing FI, ER was reduced.However, the available evidence shows that there is a direct and two-way relationship between ER and FI.For example, the ER on the three foods "fast foods and sweets" and "meat and meat products" was reported more in people with more FI.In addition, people with more FI than individuals with less FI reported a tendency to have more ER 6 .
The results displayed that the CFI change between configural and constrained models (∆CFI) was < 0.01.Hence, the configural, metric and scalar invariances in FII and ER scales were confirmed based on gender, BMI status and age.The ΔCFI ≥ 0.01 indicates a significant decrease in model fit and leads to the rejection of the constrained model 26 .In the ongoing study, the two FII and ERQ scales were invariant and unbiased between groups (gender, BMI status and age).Therefore, the differences between the groups in the scores of the two scales can be interpreted based on subgroups.This means that both FII and ERQ scales can be used with the structure presenting in both groups according to gender, BMI and age.The strengths of this study are (a) localizing two scales that measure the recent lifestyle of individuals, (b) screening the risk of malnutrition using these two scales, (c) having a large sample size for EFA and CFA, (d) evaluating measurement invariance in terms of gender, BMI status and age, (e) determining the FII cut-off point based on ROC curve and (f) using a weight estimator for CFA to increase the accuracy of the results and the lack of dependence on the assumption of normal data.The use of a convenient sampling method may limit the generalizability of the results.Physical, psychological, social and environmental variables may be effective in ER and FI, a few of which have been addressed in the present study.Both scales were self-reported, which may lead to reporting errors.

Conclusion
The FII and ERQ were translated into Persian according to international standards; their psychometric properties were studied in Iranian adults that had high construct, convergent, discriminant validity and reliability.Both scales had acceptable validity and reliability for men and women, individuals with normal BMI and overweight, and individuals aged < 35 and > 35 years.The use of these scales due to the small number and simplicity of their items is recommended to measure the recent diet of adults and diagnose the risk of malnutrition.

Table 2 .
Descriptive statistics, exploratory and confirmatory factor analyses of the ERQ.

Table 4 .
FII: goodness-of-fit indices of the four-factor model and measurement invariance based on gender, BMI status and age.S-B χ 2 Satorra-Bentler scaled chi-square test, df degree of freedoms, TLI Tucker-Lewis index, SRMR standardized root-mean square residual, RMSEA 90% CI root mean square error of approximation RMSEA and its confidence interval, CFI comparative fit index, D difference values, CI configural invariance, WI weak (or metric) invariance, SI strong (or scalar) invariance.