Participants and procedures
Data for this study came from two Australian research projects both of which had the development and validation of the Feeding Practices and Structure Questionnaire for infants and toddlers as primary (Sample 1) or secondary aim (Sample 2) (see Figure 1).
Sample 1 (“the FPSQ for Infants and Toddlers Study”) was specifically designed to develop, test and validate the Feeding Practices and Structure Questionnaire for infants and toddlers. Sample 1 was recruited as a supplement to Sample 2, running concurrently with the follow-up assessments of Sample 2, to specifically cover the child age range 12-24 months and expand some of the feeding practices concepts that were not necessarily age appropriate in under 12-month-olds and therefore not examined in Sample 2 (e.g. using rewards). Parents were recruited through university staff and student email distribution lists, parenting forums and social media websites that were exclusively targeted towards Australian parents. For instance, social media websites included Australian Facebook groups (e.g. mums and bubs groups; location specific pregnancy, baby and child buy/swap/sell groups). Eligible participants were 18 years or above, had internet access to the online survey; their child was between 6-24 months old and had not been diagnosed with any feeding disorder. In total, 530 participants commenced the online survey. Of those, 491 provided relevant data for the development and validation of the FPSQ for infants. Notably, 19 responses from parents with children younger than 6 months (minimum age 1 month) were accepted due to overlap with the same age of Sample 2.
Respondents completed the self-administered online questionnaire and reported on the following demographic variables: child age, child gender, parent age, parent gender, relationship with child, feeding responsibility, feeding mode and education level. Additionally, participants indicated whether or not they were of Aboriginal or Torres Strait Islander origin (0.8% yes). Participants could go into the draw to win one of four AUD25 gift vouchers. Approval was obtained from the Queensland University of Technology Human Research Ethics Committee (REF NO. 1400000033).
Sample 2 (the Tresillian Feeding Study) was designed to examine the feeding practices of Australian parents of infants and how practices related to the development of infant eating and weight longitudinally. Participants were recruited at the Tresillian Family Care Centres (https://www.tresillian.org.au/), an early parenting support service, in the state of New South Wales, Australia and via advertisements posted to the Tresillian Facebook group. At the Tresillian Family Care Centres (residential and day stay centres), flyers and posters were displayed around the centres and handed to parents or caregivers by Tresillian nurses. Interested parents and caregivers were provided with a plain language information sheet prior to consenting to participate. They returned their completed paper-and-pencil questionnaire to a sealed box for subsequent collation and data entry by research staff. Parents and caregivers who responded to the flyer posted on the Tresillian Facebook group were directly linked to an online version of the plain language information sheet and survey, hosted on SurveyGizmo. To be eligible for participating in the survey, parents/caregivers needed to be 18 years or older, have an infant less than six months of age, and be able to read and write in English. Participants could enter into a draw to win one of two iPads. In total 496 participants provided some data. Participants were excluded if their baby was older than 6 months of age at baseline, less than 35 weeks gestation, <2500g birthweight, living outside Australia, had a health condition that affected feeding, or if data on feeding practices were missing. In total, 59 participants were excluded, leaving 437 participants with relevant data for the development and validation of the FPSQ for infants at baseline assessment (child ages 0-6 months) and 148 participants at follow-up assessment (child ages 6-15 months).
Respondents completed the self-administered (online) questionnaire which included the following demographic variables: child age, gender, multiple birth, parent age, gender, relationship with child, feeding responsibility, feeding mode and education level. Participants of Sample 2 also indicated their country of birth (86% were born in Australia) and whether or not their child is of Aboriginal or Torres Strait Islander origin (4.3% yes). Ethical approval was granted by the Sydney Local Health District Human Research Ethics Committee (Protocol No X15-0233) and the University of Technology Sydney Human Research Ethics Committee (REF NO. 2015000528).
Generation of constructs and items
Generation of constructs and items was based initially on the original FPSQ and its underlying theory of authoritative feeding (22), and were adapted for use with children under the age of 2 years. Thus, item construction and selection were mainly conducted a priori. Due to the differences in developmental stages of infants and the feeding mode (milk, or [semi-] solid foods), it was necessary to devise two versions: one for milk feeding interactions (FPSQ-M) and one for (semi-)solid feeding interactions (FPSQ-S). The feeding practices constructs and items to be included in the infant and toddler version of the questionnaire were derived in three ways (see Figure 2). Proposed constructs are shown in Table 1.
In the next phase, proposed constructs for the milk and solid feeding versions respectively were independently reviewed by 12 international experts, chosen based on their clinical and/or research experience in infant feeding. The experts’ feedback confirmed the decision to develop two versions of the feeding practices questionnaire based on the feeding mode (i.e. infant still having milk feeds vs. infant having (semi-)solid feeds, depending on how the parent would currently mainly feed their child), rather than splitting the questionnaire by age (e.g. <6 months, 6-12 months and >12 months). Based on feedback to increase the distinction between both questionnaire versions, terminology was adjusted so that the milk feeding version refers to ‘baby’ and ‘feeding’ (e.g. “I let my baby decide when he would like to have a feed”), while the solid feeding version refers to ‘child’ and ‘eating’ (e.g. “I let my child decide when he would like to eat”). Finally, the revised questionnaires were piloted with 4 participants. While most comments related to the length and ease of online completion, another suggestion included providing a time frame (e.g. think about feeding your baby within the last two/few weeks) as feeding interactions at this early age change quickly. Inclusion of a ‘not applicable’ option was also suggested due to variability in developmental stages. All items were scored on a 5-point Likert scale with responses from 1=never to 5=always. A ‘not applicable’ response option was available for all items.
Item consolidation, factor identification, specification and validation
Once data were collected for both questionnaire versions in the Samples 1 and 2, items were closely screened. Issues with response distribution and normality (i.e. kurtosis values >3 and high skewness as visually inspected via histograms) were noted down for each item. Similar to Llewellyn et al.(25), items were discarded if >80% of respondents ticked the same response option (e.g. ‘never’ or ‘always’) or if a large number of respondents (>5%) selected the ‘not applicable’ response option. Next, Samples 1 and 2 were combined to conduct Confirmatory Factor Analysis for both versions of the questionnaire (see Figure 1).
Confirmatory Factor Analysis
Confirmatory Factor Analysis (CFA) was conducted for the statistical construct specification of the full model. The aim was not only to verify the proposed model, but also to identify the strongest set of items for each feeding construct, confirm the factorial validity, flag issues with cross-loading items, and examine factor-factor correlations. Furthermore, CFA was chosen in order to follow the development procedure for the original FPSQ questionnaire. Given the large sample size and items being developed for particular constructs, thus providing a specific model to be tested against the observations, CFA was selected as preferred method. CFAs were conducted in Mplus Version 7.3 (26) with the weighted least squares estimator (WLSMV; for ordinal categorical indicators)(27). Model specifications included fixing one regression weight per factor to 1 and correlating all factors with one another. The following indices and acceptable cut-offs were used to evaluate overall model fit: the normed chi-square (χ2/df) with values between 1.0–2.0, Comparative Fit Index (CFI) and Tucker-Lewis Index (TLI) >0.90, and Root Mean-Square Error of Approximation (RMSEA) <0.08 (28-30). Post hoc modifications were undertaken to improve model fit if acceptable levels were not achieved. Model re-specifications, such as loading an item onto another subscale than initially designed, were guided by modification indices and conceptual justifications. Items identified as having poor measurement properties (i.e. non-significance with p ≥0.001, item-factor loading <0.4, squared multiple correlation <0.2) were removed. Internal consistency was determined using Cronbach’s alpha and Coefficient H. As for the original development of the FPSQ, subscales with values <0.6 were deemed to have poor reliability and consequently excluded (22).