The feeding siblings questionnaire (FSQ): Development of a self-report tool for parents with children aged 2 – 5 years

Over the last decade, there have been repeated calls to expand the operationalisation of food parenting practices. The conceptualisation and measurement of these practices has been based primarily on research with parent-child dyads. One unexplored dimension of food parenting pertains to the evaluation of practices specific to feeding siblings. This study describes the development and validation of the Feeding Siblings Questionnaire (FSQ) – a tool designed to measure practices in which siblings are positioned as mediators in parents ’ attempts to prompt or persuade a child to eat. Item development was guided by a conceptual model derived from mixed-methods research and refined through expert reviews and cognitive interviews. These interviews were conducted in two phases, where parents responded to the questionnaire primarily to test i) the readability and relevance of each item


Background
Food parenting practices can have profound impact on the development of food preferences and appetitive traits in children (Daniels, 2019).In the literature, food parenting practices have been conceptualised under the three higher order domains of coercive control, structure, and autonomy support or promotion (Vaughn et al., 2015).Coercive control encompasses practices such as pressure to eat and overt restriction that may inadvertently teach a child to eat in response to external factors, rather than their own hunger and fullness cues (Vaughn et al., 2015).They differ from practices within the other two domains that largely encompass responsive feeding, whereby parents encourage eating through role modelling and structured mealtime routines, in a manner that fosters the child's capacity for appetite self-regulation (Vaughn et al., 2015).These practices therefore serve as modifiable behaviours that can shape what and how children eat from a young age, which can have implications for health outcomes including diet quality and weight status (Hernandez et al., 2024;Paul et al., 2018).
The capacity to understand the full implications of food parenting practices relies on the valid and reliable measurement of these constructs in research.To date, the conceptualisation and measurement of food parenting practices in general has been based on research with parent-child dyads.While this research has served the field well, current measurement tools are yet to comprehensively capture the broad scope of food parenting practices used with children from birth to 5 years of age (Heller & Mobley, 2019).One dimension of food parenting yet to be assessed relates to practices specific to feeding siblings.This gap in the literature is despite population-based data in many countries, including Australia, indicating that most parents have two or more children (Australian Bureau of Statistics, 2022).Therefore, new methods are needed to expand the scope of existing measurement tools beyond the parent-child dyad.
Despite recognition of their role in child development and adjustment across a range of disciplinary perspectives (Whiteman, McHale, & Soli, 2011), siblings are often overlooked in child feeding research.However, in early childhood, siblings are typically present at mealtimes in the home (Moding & Fries, 2020).To better understand how children are socialised around food from a young age, future research must therefore position siblings as fundamental constituents of the family unit.This idea aligns with principles of family systems theory, which posits that the family is a complex, interrelated system that must be understood as a whole, rather than as individual components (e.g., parent-child dyads) alone (Broderick, 1993).
Emerging studies in Australia, Europe, and North America have explored similarities and differences in food parenting practices used with siblings by comparing their scores on questionnaire subscales, originally developed for use in parent-child dyads (Ayre, Harris, White, & Bryne, 2023;Ayre, Harris, White, & Byrne, 2022;Kininmonth et al., 2023;Ruggiero et al., 2022 a;Ruggiero et al., 2023a;Ruggiero et al., 2022;Vollmer, 2022).In general, these studies have found that parents may adapt their use of pressure to eat and overt restriction in response to differences in sibling characteristics, such as their weight status and eating behaviours (Ayre et al., 2022).However, questionnaires used in these studies are unable to capture nuanced interactions between a parent and child, in which a sibling is also involved.For example, recent qualitative research revealed that parents may motivate a child to eat by leveraging the competitive nature of their sibling relationship or overtly reward a child for eating with the intention of vicariously conditioning their sibling's behaviour (Ayre, Harris, White, & Bryne, 2023).However, without operationalising these practices, little can be understood about their implications for child dietary, weight, and health outcomes.
The current study aimed to develop and test a theoretically and empirically informed measure to assess food parenting practices involving siblings.Such methods can contribute toward valid and reliable evidence to inform the prioritisation of intervention targets and assessment of intervention effects for promoting responsive feeding in families.Practical research outcomes, such as contributing to healthier diets in children, are in turn dependent on the rigorous conceptualisation and measurement of these practices.While the current study developed and tested this measure in the Australian context, psychometric testing is ideally conducted iteratively over multiple timepoints and samples (Boateng, Neilands, Frongillo, Melgar-Quiñonez, & Young, 2018), providing the opportunity for researchers to modify and expand its application to diverse socioeconomic and cultural contexts in the future.

Participants and procedure
Recruitment and data collection was conducted between August and December 2022, with methods detailed elsewhere (Ayre, Harris, White, & Bryne, 2023).Briefly, digital advertisements were shared via social media (including paid advertising), childcare centres, and emailing lists.Participants included parents with two or more children aged 2-5 years living in Australia.Parents self-enrolled into the online study via REDCap (Research Electronic Data Capture) (P.A. Harris et al., 2019;P. A. Harris et al., 2009), hosted by Queensland University of Technology, and were screened for eligibility criteria at recruitment.Eligible parents were aged 18 years or older and able to read English.Their two children were also born as healthy, term infants (>35 weeks), and living with them full-time.If parents had more than two children within this age range, they were asked to respond to items with reference to their two eldest children.The study comprised two phases: cognitive interviews (n = 5) and a survey (n = 330).While parents who completed an interview were not directly invited to participate in the survey, it is possible that some parents completed both phases.Upon completion of the survey, parents were also invited to partake in a repeated survey two weeks later, which was accessible via a link emailed to them through REDCap.
All procedures were approved by the Queensland University of Technology Human Research Ethics Committee (reference number: 5900).Informed consent was obtained from all parents.To acknowledge their contributions, parents who participated in the cognitive interviews were offered an AU$20 gift voucher, and parents who completed the survey were offered entry into a prize draw to receive one of three AU $200 gift vouchers.An additional prize draw entry was offered to parents who completed the repeated survey.The Checklist for Measure Development and Validation Manuscripts (Holmbeck & Devine, 2009) was used to guide the conduct of the study (see Supplementary Table S1), while the STROBE-nut checklist (Lachat et al., 2016) was used to guide the reporting (see Supplementary Table S2).

Item sources
The Feeding Siblings Questionnaire (FSQ) was designed using systematic methods that correspond with five of the six components of instrument development outlined by Vaughn, Tabak, Bryant, and Ward (2013).These components include: i) conceptualisation of the instrument aims, ii) systematic development of the item pool, iii) refinement of the item pool, iv) validity testing (i.e., factorial and construct validity), and v) reliability testing (i.e., internal consistency and test re-test reliability) (see Fig. 1).The questionnaire is targeted toward parents with children aged 2-5 years.It aims to measure practices in which siblings are positioned as mediators in parents' attempts to prompt or persuade their child to eat.An initial 38-item pool was developed by the first author (SA) using guidelines outlined by DeVellis and Thorpe (2021) to ensure that items were short, simple, and unambiguous.The items align with five hypothesised constructs that were extrapolated from formative mealtime observation and semi-structured interview data (Ayre, White, et al., 2023).The constructs describe mealtime interactions, such as leveraging sibling competitiveness, engaging siblings as active intermediaries, threatening to share food with siblings, modelling sibling behaviour, and vicarious operant conditioning.Definitions for these constructs are outlined in Table 1.

Expert review
The items and associated constructs were independently reviewed by four academics with expertise in food parenting practices and/or scale development, in addition to three authors (RB, MW, and HH).Reviewers provided feedback on the relevance of each item to its associated construct, the readability of each item, and the overall content and structure of the questionnaire.The review resulted in the removal of 11 items, revision of 20 items, and addition of 1 item.Items were removed primarily due to ambiguity (n = 4), repetition (n = 3), complexity (n = 2), use of emotive language (n = 1), and measurement of a different construct (n = 1).Furthermore, items were revised to increase specificity (n = 9), include examples (n = 8), and simplify wording (n = 3).The additional item was an example of parents engaging a sibling as an active intermediary when praising their other child to eat: 'I get [Child A] to agree with me when I praise [Child B] for eating (e.g., "[Child B] is eating well tonight, isn't he/she, [Child A]?").'

Cognitive interviews
Online cognitive interviews were conducted by the first author (SA) using Zoom videoconferencing software (Zoom Video Communication, 2020).These interviews were undertaken with five parents, including four mothers, one father, three born in Australia, and two born overseas.Although the sample size was small due to feasibility constraints, a sample comprising as few as five participants is considered appropriate for cognitive interviews, since these methods form a preliminary phase in questionnaire development and testing (Peterson, Peterson, & Powell, 2017).The interview aims and protocols were modified across two stages of interviewing (see Supplementary Tables S3 and S4).Cognitive interviewing is underpinned by a model introduced by Tourangeau (1984).According to this model, participants engage in four cognitive stages while completing a questionnaire: comprehension, retrieval, judgment, and response selection, with each stage constituting a potential source of error (Tourangeau, 1984).Therefore, verbal probes were designed to target each stage of cognition based on recommendations in the literature (Peterson et al., 2017;Willis & Artino, 2013).
In the first stage of interviews, parents (n = 3) provided concurrent feedback on the readability and relevance of items as they completed the questionnaire.During these interviews, the 'think aloud' method was used, whereby parents were asked to verbalise their thoughts as they read and responded to each item (Peterson et al., 2017).In the second stage of interviews, parents (n = 2) were asked to complete the questionnaire uninterrupted at their own pace, whilst noting items that they perceived as difficult to comprehend or respond to with the designated options.Upon completion, parents provided retrospective feedback, with a primary focus on the overall feasibility of the questionnaire.Verbal responses from all interviews were consolidated into a document and reviewed by the authors (SA, RB, MW, and HH), resulting in the revision of 11 items.Items were revised to increase the relevance and applicability of the examples provided (n = 6), reduce repetition (n = 3), remove emotive language (n = 1), and capture additional dimensions of a construct (n = 1).

Survey
The revised 28-item questionnaire was administered in the current

Table 1
Examples of parent-sibling interactions identified from mealtime observation and semi-structured interview data (Ayre, White, et al., 2023).(Jansen, Williams, Mallan, Nicholson, & Daniels, 2016) were used to measure food parenting practices for each sibling.This questionnaire has been validated in samples of Australian mothers (Jansen et al., 2016;Jansen, Harris, Mallan, Daniels, & Thorpe, 2018) and fathers (Jansen et al., 2018).The subscales included four non-responsive (i.e., coercive control) practices, including persuasive feeding (6 items, e.g., 'Do you say something to show your disapproval of [Child] for not eating?';α = 0.84 and α = 0.86 for earlier and later-born children, respectively), reward for eating (4 items, e.g., 'When [Child] refuses food he/she usually eats, do you encourage him/her to eat by offering a food reward (e.g., dessert)?';α = 0.86 and α = 0.91 for earlier and later-born children, respectively), reward for behaviour (4 items, e.g., 'I reward [Child] with something to eat when he/she is well behaved'; α = 0.80 and α = 0.85 for earlier and later-born children, respectively), and overt restriction (4 items, e.g., 'I have to be sure that [Child] does not eat too many sweet foods (e.g., lollies,1 ice-cream, cake, pastries)'; α = 0.72 and α = 0.78 for earlier and later-born children, respectively).Two structure-related (i.e., responsive) practices were also measured, including structured meal timing (3 items, e.g., 'I decide when it is time for [Child] to have a snack'; α = 0.65 and α = 0.69 for earlier and later-born children, respectively) and structured meal setting (3 items, e. g., 'How often are you firm about where [Child] should eat?'; α = 0.78 and α = 0.81 for earlier and later-born children, respectively).In addition, the single-item indicator measured family meal settings ('[Child] eats the same meals as the rest of the family').All items were scored on a 5-point Likert scale (e.g., ranging from 'never' to 'always' or 'disagree' to 'agree'), with higher scores indicating greater endorsement of the practice (Jansen et al., 2016).The items were averaged to determine mean subscale scores for each child.Difference scores were then calculated for each sibling pair based on the absolute differences between these scores.

Children's Eating Behaviour Questionnaire (CEBQ)
Four subscales of the Children's Eating Behaviour Questionnaire (CEBQ) (Wardle, Guthrie, Sanderson, & Rapoport, 2001) were used to measure eating behaviours in siblings.This questionnaire has been validated in a comparable sample of Australian mothers (Mallan et al., 2013).The subscales included food fussiness (6 items; e.g., '[Child] decides that he/she doesn't like a food, even without tasting it'; α = 0.94 and α = 0.93 for earlier and later-born children, respectively), slowness in eating (4 items, e.g., '[Child] eats slowly'; α = 0.74 and α = 0.82 for earlier and later-born children, respectively), satiety responsiveness ( 5items, e.g., '[Child] cannot eat a meal if he/she has had a snack just before'; α = 0.75 and α = 0.81 for earlier and later-born children, respectively), and food responsiveness (5 items, e.g., 'If allowed to, [Child] would eat too much'; α = 0.76 and α = 0.84 for earlier and later-born children, respectively).Items were scored on a 5-point Likert scale ranging from 'never' to 'always', with higher scores indicating more frequent observation of the behaviour.Consistent with the methods outlined above, mean subscale scores were then calculated for each child, along with difference scores for each sibling pair.

Sociodemographic and anthropometric measures
Sociodemographic data included parental age, gender, ethnicity, education, employment status, marital status, and number of children in the household.Postcodes were also collected to determine Index of Relative Socioeconomic Advantage and Disadvantage (IRSAD) scores (Australian Bureau of Statistics, 2018).For children, data were collected on age and gender, in addition to weight and height.Child body mass index z-scores (BMIzs) were calculated from parent-reported data using SAS Version 9.4 software (SAS Institute Inc., 2020).An accompanying program, developed according to the Centres for Disease Control and Prevention (CDC) age and sex-adjusted growth charts (Centers for Disese Control and Prevention, 2022;Kuczmarski et al., 2002), was employed for this purpose.Due to the reliance on parental reports and the absence of additional anthropometric measurements to validate the data, it was necessary to identify and exclude biologically implausible values (BIVs).Consistent with recommendations in the literature (Freedman et al., 2015(Freedman et al., , 2016)), BIVs were identified based on cut-off points of < -4 and >8 for the modified BMIzs integrated into the program.

Data analysis
Data analysis was conducted in SPSS (Statistical Package for the Social Sciences) Version 27.0.1.0software (IBM Corp, 2020).Available baseline data from 359 participants were screened, and overall, 2 (0.6%) participants were excluded due to incomplete responses (each with 50% of FSQ items missing) and 27 (7.5%)participants were excluded due to invalid responses, resulting in a total sample of 330 participants at baseline.Descriptive statistics were performed on this sample and compared with the subsample of participants who completed the repeated survey at two weeks (n = 133, 40.3%).The validity of the item used to determine the framing of children's names within the FSQ was assessed by comparing the mean subscale scores on the CEBQ for siblings who were rated as the "better" eater versus those who were not.In line with the profile of a 'non-fussy' eater identified in a latent profile analysis (Tharner et al., 2014), it was expected that children perceived as the "better" eater would score lower on the food avoidant subscales (food fussiness, slowness in eating, and satiety responsiveness) and higher on the food approach subscale (food responsiveness).

Factorial validity
Preliminary assessment of the FSQ items revealed no deviations from the assumption of linearity.However, univariate and multivariate outliers were detected via assessment of box plots and Mahalanobis distances, respectively.While the univariate skewness and kurtosis coefficients were within an acceptable range (Curran, West, & Finch, 1996), Mardia's multivariate skewness and kurtosis coefficients were also statistically significant (p < 0.001), indicating non-compliance with the assumption of normality (Mardia, 1970).Therefore, exploratory factor analysis (EFA) was performed on the items using principal axis factoring with oblique rotation (direct oblimin).Using the POLYMAT-C program (Lorenzo-Seva & Ferrando, 2015), a polychoric correlation matrix formed the basis of the EFA due to its increased robustness with ordinal and non-normally distributed variables (Watkins, 2018).Communalities were estimated using squared multiple correlations.The factorability of the data was confirmed via assessment of the correlation matrix, Bartlett's test of sphericity, and Kaiser-Meyer-Olkin (KMO) statistic (Tabachnick & Fidell, 2018).
In line with methods outlined by Watkins (2018), the number of factors retained in subsequent analyses was determined via consultation of the eigenvalues, scree plot, parallel analysis estimates, and minimum average partials (MAPs).Pattern coefficients ≥0.32 were considered salient (Tabachnick & Fidell, 2018).Criteria for assessing factor adequacy included the loading of ≥3 items with salient pattern coefficients, an acceptable internal consistency estimate, and convergence with the hypothesised factor structure (Watkins, 2018).In favour of a simple solution, items that complicated the factor structure (i.e., loaded inadequately onto all factors or cross-loaded onto multiple factors with a loading difference <0.20) were deleted one at a time, until each item retained in the model loaded saliently onto one factor only (Howard, 2016;Watkins, 2018).

Internal consistency and test re-test reliability
For each factor, internal consistency was estimated using Cronbach's alpha coefficients, with a value of ≥0.70 considered acceptable (Johnson, 2018).Subscale scores were calculated by averaging the scores for items that loaded onto similar factors.As estimates of two-week test re-test reliability, intraclass correlation coefficients (ICCs) were calculated for the subscale scores based on a single measure, absolute agreement, two-way mixed-effects model.ICCs ≥0.50, ≥0.75, and ≥0.90 were indicative of moderate, good, and excellent test re-test reliability, respectively (Koo & Li, 2016).

Construct validity
As the FSQ describes instances in which parents feed siblings differently (i.e., one child is used as a mediator in parents' attempts to prompt or persuade the other child to eat), comparing this scale to other relevant measures of sibling discordance would enable assessment of its convergent construct validity.With the absence of other measures that target parent-sibling triads, sibling difference scores on the FPSQ and CEBQ subscales were used as a proxy, with larger scores indicating that one child scored comparatively higher on that subscale compared to their sibling.
Spearman's correlation coefficients were first used to examine how FSQ subscale scores were correlated with FSPQ subscale difference scores.It was predicted that higher scores on each FSQ subscale would be correlated with larger difference scores for coercive control (i.e., persuasive feeding, reward for eating, reward for behaviour, overt restriction) and structure-related food provision (i.e., family meal settings, structured meal timing, and structured meal setting).Second, independent samples t-tests were used to compare FSQ subscale scores between sibling pairs who were discordant and non-discordant on each CEBQ subscale.In line with methods reported elsewhere (Ayre, Harris, White, & Bryne, 2023;Kininmonth et al., 2023), sibling pairs were defined as discordant if they had a difference score ≥1 standard deviation of the mean difference score for that subscale.It was predicted that parents of sibling pairs discordant on food fussiness, slowness in eating, satiety responsiveness, and food responsiveness subscales would have higher scores on each FSQ subscale compared to parents of non-discordant sibling pairs.

Results
The characteristics of the parent-sibling triads who completed the survey are reported in Tables 2 and 3.The survey was repeated by 40.3% of parents after a median of 2 weeks (range: 2-9 weeks).The proportion of parents who repeated the survey was different between education groups (X 2 1 = 5.78, p = 0.016).The highest proportion was evident among parents who had completed a university degree (44.1%), and the lowest proportion was evident among parents who had completed Year 12 or below (17.6%).No other differences were observed in sociodemographic characteristics between the samples at baseline and two weeks.
In response to the single item, most parents (89.4%) were capable of differentiating their children based on their eating behaviours, with more than half (57.3%) of these parents rating their later-born child as the "better" eater.Relative to their sibling, the child who was perceived as the "better" eater scored, on average, lower for food fussiness (t 294 = − 16.91, p < 0.001), slowness in eating (t 294 = − 6.61, p < 0.001), and satiety responsiveness (t 294 = − 10.67, p < 0.001), and higher for food responsiveness (t 294 = 3.37, p < 0.001) (see Table 4).However, based on their effect sizes, only differences in food fussiness and satiety responsiveness were considered practically significant (Cohen's d = 0.98 and 0.62, respectively; see Table 4) (Ferguson, 2009).Differences between mean subscale scores for the eating behaviours were not significant for sibling pairs for whom parents could not differentiate on this basis (ps≥0.114;results not reported).

Factorial validity
Preliminary examination of the correlation matrix for the FSQ revealed that Items 17 and 24 were highly correlated (r = 0.830), and as such, the R determinant was indicative of potential multicollinearity (<0.00001) (Tabachnick & Fidell, 2018).To minimise these effects, Item 17 was excluded from the analysis as this item had stronger correlations with other variables.It was hypothesised that a 5-factor solution would fit the data; however, the eigenvalues and MAPs both indicated that four factors should be retained, while parallel analysis indicated that only three factors were required.In addition, the scree plot demonstrated two points of inflexion that justified the retention of either four or six factors.Therefore, 6, 5, 4, and 3-factor solutions were sequentially examined.
The 6-factor solution was inadequate with only two items loading onto the sixth factor.All other solutions were adequate; however, the 5 and 3-factor solutions had fewer items that cross-loaded onto multiple factors.Due to its higher convergence with the theoretically driven and hypothesised factor structure, the 5-factor solution was accepted as the final model.To simplify the model, Item 15 was first deleted as it failed to load onto any factor with a salient pattern coefficient.Item 2 was then deleted as it cross-loaded onto Factors 2 (− 0.333) and 4 (0.362) with the highest loading ratio.After the deletion of this variable, Item 1 failed to load onto any factor with a salient pattern coefficient.Therefore, this item was also deleted.Finally, Items 12 and 19, which cross-loaded onto Factors 2 (− 0.379) and 4 (0.327), and 1 (0.365) and 5 (− 0.439), respectively, were deleted in this order (highest to lowest loading ratio).
After five iterations, the final 22-item, 5-factor model explained 72% of the total variance (after rotation) and was deemed to reflect the following constructs: sibling competitiveness, active sibling influence, threatening unequal division of food, sibling role modelling, and vicarious operant conditioning.Table 5 includes the descriptive statistics, factor loadings, and communalities for all items.Participant scores for each item ranged from 1 to 5. All items had salient factor loadings (pattern coefficients ≥0.388) and a reasonable proportion of variance within each item was explained by the factor on which it loaded (h 2 ≥ 0.608).The inter-factor correlations ranged from r s = 0.351 to 0.698 (see Table 6).The Bartlett's test was significant, χ 2 (231) = 6450, p < 0.001, confirming sufficient intercorrelation between the items for EFA (Bartlett, 1954).Sampling adequacy for factor analysis was also evidenced by a KMO statistic of 0.938 (with values for each item ≥0.871) (Kaiser, 1974).

Internal consistency and test re-test reliability
The internal consistency and test re-test reliability estimates for the five subscales are reported in Table 7. Cronbach's alpha coefficients were acceptable, ranging from 0.84 to 0.92.ICCs ranged from 0.76 to 0.88 (ps < 0.001), indicating good to excellent test-retest reliability.

Construct validity
Table 8 presents correlations between the FSQ subscale scores and FPSQ subscale difference scores for siblings.Construct validity testing revealed that scores for four of the five FSQ subscales were significantly correlated with mean difference scores on at least two FPSQ subscales.However, the strength of these correlations was often small (see Table 8).For example, positive correlations were observed for sibling competitiveness with differences in persuasive feeding (r=0.114,p = 0.039), reward for eating (r = 0.160, p = 0.004), and reward for behaviour (r = 0.201, p < 0.001).Similarly, positive correlations were evident for sibling role modelling and differences in persuasive feeding (r = 0.134, p = 0.015), reward for eating (r = 0.229, p < 0.001), and reward for behaviour (r = 0.217, p < 0.001), in addition to differences in overt restriction (r = 0.113, p = 0.040) and family meal settings (r = 0.120, p = 0.030).Parents who differed more in the extent to which they used reward for eating and reward for behaviour also scored higher for threatening unequal division of food (r = 0.133, p = 0.015 and r = 0.138, p = 0.012, respectively) and vicarious operant conditioning (r = 0.197, p ≤ 0.0001 and r = 0.264, p < 0.001, respectively).No significant correlations were found for differences in structured meal timing and structured meal settings.Although not reported, differences in FSQ subscale  scores between discordance groups are provided in Supplementary Table S5 for completeness.Table 9 presents differences in FSQ subscale scores between sibling pairs discordant and non-discordant on each CEBQ subscale.Construct validity testing revealed at least one significant difference between the groups across all five FSQ subscales.On average, parents scored higher for threatening unequal division of food if their children were discordant on any of the four eating behaviours, including food fussiness (t 328 = 2.17, p = 0.030), slowness in eating (t 184 = 4.06, p < 0.001), satiety responsiveness (t 137 = 2.15, p = 0.033), and food responsiveness (t 328 = 2.98, p = 0.003).In addition, parents scored higher for sibling role modelling if their children were discordant on food fussiness (t 328 = 3.65, p < 0.001) and slowness in eating (t 328 = 2.66, p = 0.008).Differences in sibling competitiveness (t 328 = 3.02, p = 0.003), active sibling influence (t 328 = 1.99, p = 0.048), and vicarious operant conditioning (t 194 = 2.88, p = 0.004) were only evident between sibling pairs discordant and nondiscordant on slowness in eating, with higher scores observed in the discordant group.However, the effect sizes were generally small, meaning that these results may not represent practically significant differences between the groups (see Table 9) (Ferguson, 2009).Correlations between the FSQ subscale scores and CEBQ subscale difference scores for siblings were not reported; however, these results are provided in Supplementary Table S6 for completeness.The final questionnaire is available as supplementary material (see Supplementary Table S7).

Discussion
This study aimed to develop an instrument to measure food parenting practices with siblings in early childhood, and provide an initial assessment of its validity and reliability in an Australian sample of siblings aged 2-5 years.A 22-item, 5-factor structure was determined through a systematic process of questionnaire development and refinement, using a compilation of methods.Results from the psychometric analyses indicate that the Feeding Siblings Questionnaire (FSQ) may be considered a robust and parsimonious measure for examining mealtime interactions beyond those confined to the parent-child dyad.To the authors' knowledge, this instrument was also the first to measure food parenting practices with two children comparatively within single FSQ, Feeding Siblings Questionnaire; SD, standard deviation.Salient pattern coefficients (≥0.32) are bolded to indicate the primary factor on which the item loads.h 2 refers to the communality value.
a Child A refers to the name of the child nominated as the "better" eater and Child B refers to their sibling (alternatively, if parents could not differentiate their children based on their eating behaviours, Child A refers to the name of the earlier-born child and Child B refers to their sibling).
b Item excluded to minimise multicollinearity.c Item excluded due to low loading (pattern coefficients <0.32) on all factors.d Item excluded due to cross-loading (pattern coefficients ≥0.32) on multiple factors.items, thus expanding the scope of existing parent-report tools (Vaughn et al., 2013).
The framing of the children's names in the FSQ was determined based on responses to a single item, which should be used in conjunction with the 22 items.In previous research, the use of single-item indicators to assess parents' perceptions of their child's eating behaviours (e.g., 'Do you think your child is a picky or fussy eater?') have demonstrated, to some extent, validity in predicting differences in child behavioural, dietary, and anthropometric measures (Byrne, Jansen, & Daniels, 2017;Carruth, Ziegler, Gordon, & Barr, 2004;Jacobi, Agras, Bryson, & Hammer, 2003).However, the current study was novel in that parents were asked to differentiate one sibling as the "better" eater, thereby forming a broad assessment of how parents evaluate these behaviours.Although parents were prompted to consider behaviours such as food refusal when responding to this item, the child who was regarded as the "better" eater typically scored lower than their sibling not only for food fussiness, but also for slowness in eating and satiety responsiveness, and higher for food responsiveness.This finding is consistent with a latent profile analysis of eating behaviours in 4-year-old children, where the profile of a 'non-fussy' eater was characterised by lower scores for food fussiness, slowness in eating, and satiety responsiveness, and higher scores for food responsiveness and enjoyment of food (Tharner et al., 2014).Therefore, the current study provides validation for the use of this item.
Responses to the single item also highlighted discrepancies between community and public health concerns, whereby children perceived as the "better" eater tended to have lower responsiveness to internal appetite cues and heightened sensitivity to external food cues.Although these are adaptive traits from an evolutionary point of view, they may increase risks of overweight and obesity in the modern environment which is no longer dominated by food scarcity but abundance (Kininmonth et al., 2021).In addition, the "better" eater tended to exhibit lower food fussiness.Despite this perception among parents, food fussiness is regarded as a developmentally normal and transient trait in toddlers (Cardona Cano et al., 2015).As concerns about fussiness can motivate parental use of persuasive feeding strategies (Burnett, Russell, Lacy, Worsley, & Spence, 2023;H.A. Harris, Jansen, Mallan, Daniels, & Thorpe, 2018), there is a need to normalise food avoidant behaviours, not as acts of deviance, but as developmentally appropriate expressions of appetite and food preferences, in order to minimise undue stress and the use of counterproductive food parenting practices in families (Walton, Kuczynski, Haycraft, Breen, & Haines, 2017).Moreover, the perception of a "better" eater was not related to other child characteristics such as age.Behaviours like food fussiness tend to be overt in nature; therefore, it is evident that parents may be more attuned to differences in these types of characteristics (Ayre, Harris, White, & Bryne, 2023).

Preliminary validity and reliability of the FSQ
The FSQ comprised five factors that measured practices reflecting sibling competitiveness, active sibling influence, threatening unequal division of food, sibling role modelling, and vicarious operant conditioning.While other factor solutions also demonstrated adequate fit to the data, the final model was consistent with constructs extrapolated from formative mealtime observations and interviews in a comparable sample (Ayre, White, et al., 2023).Hence, the factors directly correspond with the constructs defined in Table 1.However, one exception is Factor 2 (active sibling influence), which expands on the original definition of the construct.Items 7 and 11, originally conceptualised as sibling role modelling practices (Factor 4), loaded more saliently onto Factor 2. These items describe practices in which parents actively ask or direct their child to role model desired eating behaviours.Thus, the revised factor encompasses practices in which siblings exert influence, whether it be verbal (e.g., praise, encouragement) or non-verbal (e.g., role modelling).A second exception is Factor 5 (vicarious operant conditioning), which originally included the use of both tangible (e.g., dessert) and non-tangible (e.g., social praise) rewards.However, with Item 21 loading more saliently onto Factor 4 (sibling role modelling), the focus of Factor 5 was narrowed to include tangible rewards only.Although the factors were conceptually distinct, two items were observed to cross-load onto Factors 2 (active sibling influence) and 4 (sibling role modelling).There was also a relatively strong correlation between these subscales, suggesting that future research should test whether sibling role modelling may be conceptualised more suitably as a subtype of active sibling influence.
The resulting scale describes food parenting practices in which siblings are positioned as mediators in parents' attempts to prompt or persuade their child to eat.Similarities in parents' motivations for applying these practices are indicated in the positive correlations between the subscales.However, as shown in Fig. 2, the five subscales are presumed to fall under different domains of food parenting practices as described by Vaughn et al. (2015), including coercive control, structure, and autonomy support or promotion, with one subscale spanning across two domains.Therefore, this study not only contributes a novel tool for measuring food parenting practices in the field, but also expands on the broader conceptualisation of this construct in the literature.
Comparing responses on the FSQ with existing measures of food parenting practices and child eating behaviours provided some evidence of its construct validity.The findings partially support the hypothesis that parents would score higher on the FSQ subscales if there were greater differences in their use of coercive control (i.e., persuasive feeding, reward for eating, reward for behaviour, and overt restriction) and structure-related food practices (i.e., family meal settings, structured meal timing, structured meal settings) between siblings.For example, parents used sibling competitiveness and sibling role modelling practices more often when they differed more in their use of persuasive and instrumental (i.e., reward for eating and behaviour) feeding for each child.This finding may demonstrate an overall attempt by parents to prompt or persuade one child to eat.
In addition, sibling role modelling was used more often by parents when they differed more in their use of overt restriction and family meal settings (i.e., providing family foods).It is plausible, for example, that when a child refuses to eat, parents may provide them with alternative foods, but continue to reinforce expected or preferred behaviours using their sibling as a positive role model.This example aligns with other research indicating that many parents are reluctant to cater to individual preferences, yet often resort to this practice when faced with food refusal (Fraser, Markides, Barrett, & Laws, 2021).In the current study, greater use of vicarious operant conditioning and threatening unequal division of food were also reported when parents differed more in their use of instrumental feeding.It is possible, for example, that differences in instrumental feeding are the direct result of vicarious operant conditioning.For instance, parents may reward the behaviour of one sibling in an attempt to teach the other child that by eating, they too can obtain the same reward (Ayre, White, et al., 2023).
In contrast to other food parenting practices, no significant correlations were observed between the FSQ subscales and differences in structured meal timing and structured meal settings.This finding may be partly due to parents in the current sample reporting, on average, a relatively high degree of mealtime structure for both children (mean subscale scores ≥3.43).However, this finding also serves as evidence of discriminant validity, as greater differences in these subscale scores indicate that siblings tend to eat separately more often, with fewer opportunities for parents to directly or indirectly use sibling dynamics to influence child eating behaviours.
Differences in the FSQ subscale scores were also observed in relation to child eating behaviours, including food fussiness, slowness in eating, satiety responsiveness, and food responsiveness.In the current sample, parents of siblings discordant on any of the four eating behaviours threatened to give their child's food away more often, compared to parents of non-discordant siblings (determined by a higher score on threatening unequal division of food).By using this practice, parents were favouring their "better" eater by offering the food to them.This observation aligns with the notion that threatening to serve a child's unwanted food to their sibling can serve as motivation for that child to eat (Ayre, White, et al., 2023).Research demonstrates that young children are often willing to take costs to avoid being at a perceived disadvantage to others (Sheskin, Bloom, & Wynn, 2014).However, the effectiveness of this practice may also rely on the willingness of their sibling to eat the food.Therefore, there may be increased opportunity and motive for parents to implement this practice if siblings are discordant on their eating behaviours (e.g., if one child is comparatively slower at eating).
In the current study, parents of siblings discordant on food fussiness and slowness in eating used sibling role modelling more frequently (i.e., by modelling the behaviour of the "better" eater), compared to parents of non-discordant siblings.Social learning, which involves observing and imitating others, is a widely recognised process through which children are socialised around food in early childhood (Bandura, 1977).Siblings, particularly if older in age, can serve as prominent role models for children during mealtimes (Ayre, White, et al., 2023;Ruggiero, Moore, & Savage, 2023).Therefore, parents may leverage on this dynamic by directing attention toward the sibling who is eating, when the other is not, as a source of modelling and reinforcement (Ayre, White, et al., 2023).Parents of siblings discordant on slowness in eating also used other practices, including sibling competitiveness, active sibling influence, and vicarious operant conditioning more often, compared to parents of siblings non-discordant on this behaviour.With multiple children, discordance in eating pace may serve as a constraint on coordinating mealtimes in the context of competing schedules and routines (Brannen, O'Connell, & Mooney, 2013).Hence, in waiting for a child to finish eating, parents may implement various practices in an attempt to execute the mealtime more efficiently, whilst keeping their sibling engaged (Ayre, White, et al., 2023).Contrary to this finding, however, research shows that children may consume more fruits and vegetables when mealtimes are longer in duration (Dallacker, Knobl, Hertwig, & Mata, 2023).

Implications for research and practice
Within the last decade, there have been repeated calls to expand the current operationalisation of food parenting practices in the literature (de Lauzon-Guillain et al., 2012;Heller & Mobley, 2019;Vaughn et al., 2013).In their seminal paper, Vaughn et al. (2015) invited researchers in this field to continue refining their conceptualisation of food parenting practices, toward establishing a comprehensive and cohesive model.The FSQ, which has been mapped onto this model, has found preliminary support in the current sample as a valid and reliable measure of food parenting practices with siblings.In this sample, there was a skewed distribution of the subscale scores, with most parents implementing these practices relatively infrequently.However, due to their nature, these practices may also indicate broader dimensions of family functioning, for example, the presence of favouritism or bias towards a particular child.Therefore, it is necessary to explore these practices within the context of family system processes, such as sibling comparison and differentiation (McHale, Updegraff, & Whiteman, 2012).This area of research may be particularly relevant in families where one child lives with obesity, considering that parents are often encouraged to adopt a whole-of-family approach to treatment.While additional psychometric testing of the FSQ is needed, further research is needed to examine how these practices may be implicated in the trajectories of eating behaviours and growth among children over time, in addition to their overall development and adjustment.If relevant, this knowledge can then be integrated into public health guidelines and interventions on responsive feeding.For example, if the practices are associated with increased overweight and obesity risk in children, they could serve as potentially relevant intervention targets and outcome measures in responsive feeding interventions.

Limitations
There are several limitations to note.Firstly, the proposed factor structure could not be verified using confirmatory factor analysis (CFA) due to limitations in sample size.To the authors' knowledge, there are also no existing measures that capture parents' responses for two children within a single item to offer direct comparison with the FSQ.Therefore, construct validity of the proposed factor structure was assessed using sibling difference scores on the FPSQ and CEBQ.It is recognised, however, that instrument evaluation is a systematic process that requires multiple points of data collection (Boateng et al., 2018;Vaughn et al., 2013).Further psychometric testing is therefore needed before the instrument can be effectively used in practice.Testing should include, but is not limited to, verification of the factor structure using CFA and assessment of criterion validity using comparisons with observational data (Vaughn et al., 2013).Exploring how these subscales relate to food parenting practices and child eating behaviour scores for children separately also provides a different angle through which construct validity may be assessed.As proposed by Vaughn et al. (2013), the sixth component of instrument developmentresponsiveness testing should also be undertaken to ascertain the extent to which the FSQ can detect changes in food parenting practices, to inform subsequent power and sample size calculations.
Another limitation was that the sample was relatively homogenous, with a large proportion of participants identifying as female, Australian, university educated, and married.University educated participants were also more likely to complete the repeated survey at two weeks, compared to participants with lower educational attainment.Therefore, estimates of test re-test reliability were subject to attrition bias.Hence, it is necessary to test the applicability of the instrument in more diverse samples.There was also a risk of response bias due to parents selfreporting their data.For example, responses may have been affected by the ordering of the children's names within the FPSQ and CEBQ, with items referring to the earlier-born child always listed first.Additionally, parents may have been sensitive to social desirability bias, particularly when asked to disclose partiality towards one child.Finally, while the FSQ demonstrated validity and reliability in the current sample, its scope is limited in that it focuses only on two children, and captures behaviours potentially only relevant to parents of siblings who differ in their eating behaviours.

Conclusion
This study describes the systematic development and testing of the FSQ, a potentially robust and parsimonious measure of food parenting practices with siblings.In the current sample, a 22-item, 5-factor structure demonstrated adequate fit, and provided an interpretable solution that mapped onto constructs identified in mealtime observation and interview data.The instrument was reliable and provided some evidence of construct validity.While its factor structure should be verified using CFA in a different sample, the FSQ offers a novel tool for assessing, monitoring, and evaluating feeding interactions with siblings beyond those confined to the parent-child dyad.

Fig. 1 .
Fig. 1.Flowchart for the development and psychometric testing of the Feeding Siblings Questionnaire (FSQ).

Fig. 2 .
Fig. 2. Sibling-specific food parenting practices mapped onto the conceptual model by Vaughn et al., 2015.a a The figure represents a modified and simplified version of the original conceptual model.b Factor identified within the current study describing a sibling-specific food parenting practice.

Table 2
Sociodemographic characteristics of the parents participating in the online survey at baseline (n = 330) and two weeks (n = 133).
IRSAD, Index of Relative Socioeconomic Advantage and Disadvantage; IQR, interquartile range.a In Sample 1, 12 (3.6%)parents had missing or invalid data on residential state and IRSAD score; 7 (2.1%) parents had missing data on marital status; (1.5%) parents had missing data on gender and work and study status; and (1.2%) parents had missing data on cultural identity, indigenous status, education, marital status, and number of children.b In Sample 2, 3 (2.3%)parents had missing data on marital status; and (0.8%) parent had missing or invalid data on residential state and IRSAD score.c Calculated based on 2016 Statistical Area Level 1 (SAL1).

Table 3
Demographic and anthropometric characteristics of siblings reported by parents participating in the online survey at baseline (n = 330).
a Determined based on the age gap between siblings.bForearlier-bornsiblings, data were missing for 19 CEBQ, Children's Eating Behaviour Questionnaire; CI, confidence interval; FF, food fussiness; FR, food responsiveness; SD, standard deviation; SE, slowness in eating; SR, satiety responsiveness.aSingleitem:'At this point in time, which of your children would you generally consider to be the "better" eater?' b Excludes parents who were unable to differentiate siblings based on their eating behaviours (n = 35).S.K.Ayre et al.
3I ask [Child A] to convince [Child B] that he/she will like the food (e.g., "Tell [Child B] how yummy the sauce is").[Child A] about ways to convince [Child B] to eat (e.g., "How can we get [Child B] to eat his/her dinner tonight?").c I use [Child A] as a positive example when encouraging [Child B] to eat (e.g., "Look, [Child A] is eating up all his/her vegetables!").d I look to [Child A] for backup when trying to encourage [Child B] to eat (e.g., "It's delicious, isn't it [Child A]?").d When [Child A] eats most of his/her meal, I reward him/her with something other than food (e.g., sticker, toy, screen time) to try and convince [Child B] to also eat more.

Table 7
Cronbach's alpha coefficients for subscale scores at baseline (n = 330) and intraclass correlation coefficients (ICCs) between the subscale scores at baseline and two weeks (n = 133) for the Feeding Siblings Questionnaire (FSQ).

Table 8
Spearman's correlations between the Feeding Siblings Questionnaire (FSQ) subscale scores and Feeding Practices and Structure Questionnaire (FPSQ) subscale difference scores for siblings (n = 330).

Table 9
Independent samples t-tests comparing subscale scores on the Feeding Siblings Questionnaire (FSQ) between sibling pairs who were discordant and non-discordant on the Children's Eating Behaviour Questionnaire (CEBQ) subscales (n = 330).Discordant sibling pairs were defined as those with a difference score >1 standard deviation of the mean difference score for that particular subscale.
a S.K.Ayre et al.