Validation of a brief food parenting measure for fathers: A test of factorial validity, measurement invariance, internal reliability, and concurrent validity

Abstract

Fathers are underrepresented in food parenting research partly due to the lack of succinct, theory-informed, and father-mother equivalent food parenting measurement tools. To address this, we 1) tested the factorial validity of a brief food parenting measure utilizing a subset of items from the Comprehensive Feeding Practices Questionnaire (CFPQ) to represent coercive control, structure, and autonomy support, 2) assessed the extent to which the brief tool works similarly in fathers and mothers (i.e., measurement invariance), and 3) evaluated its internal reliability and concurrent validity. Participants included 1,071 fathers of children aged 2–6 years and 487 of their co-parents (mothers) from the Fathers & Families study. Fathers and mothers responded to 16 CFPQ items, and fathers reported on children’s diets. Confirmatory factor analysis (CFA) was used to test factorial validity. Multi-group CFA was used to examine measurement invariance across fathers and mothers. Internal reliability was examined using Cronbach’s alpha and the Spearman-Brown coefficient. Concurrent validity was assessed utilizing multiple logistic regressions to examine associations between the three food parenting factors and children’s intakes of fruit, vegetables, sugar-sweetened beverages (SSBs), and fast food. CFA confirmed a three-factor model with 11 items, including coercive control (four items), structure (five items), and autonomy support (two items). Multi-group CFA indicated measurement invariance across fathers and mothers. Internal reliability was established. Concurrent validity was strong for structure, but weaker for coercive control and autonomy support. Further refinements are encouraged to enhance items’ utility in characterizing fathers’ food parenting practices, including in different social and cultural contexts.

1. Introduction

Food parenting practices consist of a range of parent-child interactive behaviors around food and are categorized around three parenting dimensions: coercive control, structure, and autonomy support (Grolnick & Lerner, 2023; Mâsse et al., 2020; Musher-Eizenman et al., 2019; O’Connor et al., 2017; Vaughn et al., 2016). Coercive control-related food parenting practices reflect parents’ feeding-related goals and desires without considering the child’s emotional or psychological needs (Vaughn et al., 2016). Such practices include pressure to eat, restriction, bribes and threats, and using food to control a child’s emotions (Musher-Eizenman et al., 2019; O’Connor et al., 2017; Vaughn et al., 2016), and are often linked with negative diet-related behaviors and outcomes among children, such as disordered eating behaviors and unhealthy weight status (Loth, 2016; Ventura & Birch, 2008). Structure-related food parenting practices provide a structured and consistent environment to facilitate children’s learning and maintenance of healthy eating habits (Vaughn et al., 2016). Examples include setting mealtime rules and limits, routine monitoring, role modeling, and creating healthy home food environments (Musher-Eizenman et al., 2019; O’Connor et al., 2017; Vaughn et al., 2016). Autonomy support-related food parenting practices cultivate children’s independence in food choices and exploration (Vaughn et al., 2016), with examples including encouragement, nutrition education, involving children in food work, praise, reasoning, and negotiation (Musher-Eizenman et al., 2019; O’Connor et al., 2017; Vaughn et al., 2016). Both structure- and autonomy support-related food parenting practices are linked with positive diet-related outcomes and behaviors among children, such as lower consumption of energy-dense foods and healthier Body Mass Index (BMI) (Balantekin et al., 2020; Shloim et al., 2015; Ventura & Birch, 2008).

Currently, the majority of food parenting research has been conducted with mothers. In a 2016 systematic review of 667 observational studies of weight-related parenting practices, including food parenting, fathers represented only 17% of the participants across all studies (Davison et al., 2016). Moreover, while 36% of studies focused exclusively on mothers, only 1% focused on fathers (Davison et al., 2016). Ways to intentionally include fathers in research are much needed to capture fathers’ food parenting perspectives and experiences because emerging evidence suggests that fathers may use different food parenting practices than mothers (Daniels et al., 2020; Davison et al., 2020; De-Jongh González et al., 2021; Douglas et al., 2023; Hendy et al., 2009; Loth et al., 2013; Philippe et al., 2021, 2022; Pratt et al., 2019; Tschann et al., 2013), and fathers’ food parenting practices are linked to children’s diet-and-weight-related outcomes, (Davison et al., 2020; Fraser et al., 2011; Khandpur et al., 2014), even when accounting for mothers’ feeding practices (Penilla et al., 2017). As fathers are increasingly more involved in child feeding (Davison et al., 2020; Lo et al., 2022; Statistics Canada, 2017; Swenson, 2020) including grocery shopping, cooking, and setting mealtime rules (Lundquist et al., 2022; Rahill et al., 2020), a comprehensive understanding of fathers’ food parenting practices will support the development of programs and policies to equitably promote child and family health.

Current literature suggests that, in order to meaningfully capture fathers’ food parenting, food parenting instruments need to be succinct, theory-informed, demonstrate measurement invariance across fathers and mothers, and be relevant to children’s diets. It has been found that while fathers are interested in participating in research, they prefer participations with minimal burden such as the use of succinct surveys (Davison et al., 2017). However, not only are many of the widely utilized food parenting instruments mostly validated among mothers, but they also contain an exhaustive list of items. For example, the Food Parenting Practice (FPP) item bank contains 129 items across 17 subscales (Mâsse et al., 2020), the original Comprehensive Feeding Practices Questionnaire (CFPQ) contains 49 items across 12 subscales (Musher-Eizenman & Holub, 2007), and the Child Feeding Questionnaire (CFQ) contains 31 items across 7 subscales (Birch et al., 2001). Among the few food parenting instruments that have been validated among fathers, such as the Spanish version of CFPQ (Del Valle et al., 2023), the Swedish version of CFPQ (Morris et al., 2023), and the Feeding Practices and Structure Questionnaire (FPSQ-28) (Jansen et al., 2018), they are still relatively long containing 19, 28, and 28 items, respectively. This indicates that there is a real need to develop succinct food parenting measurement instruments to facilitate fathers’ participation in food parenting research. A more succinct instrument can also benefit mothers’ participation in research because minimal participants’ burden has been consistently identified as a key facilitator for parents’ participation in research (Nathe et al, 2022).

In addition, to capture essential and distinct food parenting factors, food parenting conceptual frameworks need to be used to guide survey item selection and latent factor identification. However, this theory-informed approach to selecting survey items is not commonly practiced. For example, while the Spanish and Swedish versions of the CFPQ have been validated for use among fathers, items selection and latent factor identification of these translated-versions of CFPQ primarily relied on exploratory factor analysis without adequate consideration of food parenting conceptual frameworks (Del Valle et al., 2023; Morris et al., 2023). These data-derived latent factors, therefore, may contribute to similar, overlapping, and/or identical constructs being labeled differently across studies.

Furthermore, it is equally important that the food parenting instrument utilized in research involving fathers can represent the same underlying food parenting structure and the results can be interpreted similarly for fathers and mothers (i.e., measurement invariance). This will ensure food parenting differences reported in studies, in fact, originate in substantive differences between fathers and mothers instead of due to the instrument’s inadequate and inaccurate food parenting assessment for one or both groups of parents (Adamsons & Buehler, 2007). Among the many food parenting instruments developed, however, only a handful of them tested measurement invariance across fathers and mothers (Del Valle et al., 2023; Jansen et al., 2018; Morris et al., 2023).

Lastly, it is essential to understand if the newly developed instrument relates to children’s diets as expected, supporting its practical utility. Prior tool development studies, however, did not examine how fathers’ reported food parenting practices relate to children’s diets (Del Valle et al., 2023; Jansen et al., 2018; Morris et al., 2023), providing limited evidence on the concurrent or predictive validity of their tools among fathers.

1.1. Study aims

To address these gaps, this study aimed to 1) test the factorial validity of a brief food parenting measure (using a subset of CFPQ items) representing three overarching food parenting constructs (coercive control, structure, and autonomy support), 2) examine measurement invariance of the brief scale across fathers and mothers, and 3) assess the scale’s internal reliability and concurrent validity with children’s diets among fathers.

2. Methods

2.1. Study sample

The present study used baseline data from Fathers & Families (F&F), a cohort study examining the role of fathers in childhood obesity prevention (https://gutsweb.org/key-information-for-participants/fathers-families/). Fathers were initially recruited from an existing cohort, the Growing Up Today Study (GUTS) (https://gutsweb.org/), which aims to examine factors influencing health throughout our lives by following more than 27,000 participants since childhood. Male participants in GUTS who had an email address on file were invited to participate in GUTS F&F via email from July 2021 to June 2022. Those that were interested completed an eligibility screener. Eligible fathers included those who were a biological, adoptive, or social father of a child aged 1–6 years and who were living with the child at least 50% of the time. Due to the demographic makeup of the original GUTS cohort, over 90% of GUTS F&F participants self-identified as non-Hispanic White.

To increase the racial and ethnic diversity of participants, a supplemental cohort of a fathers was recruited from Michigan Medicine (MI F&F), the health system of the University of Michigan, from June 2022 to June 2023. Eligibility criteria were the same as those for GUTS F&F. During the recruitment process for MI F&F, we focused on fathers who met inclusion criteria but self-identified as a race or ethnicity other than non-Hispanic White. A total of 1,272 fathers were enrolled in F&F, 750 from GUTS F&F (93.4% non-Hispanic White) and 522 from MI F&F (30.3% non-Hispanic White). Except for race/ethnicity, household income, and child age, there were no significant differences between GUTS F&F and MI F&F in terms of fathers’ relationship status or employment status. Participating fathers were invited to identify their co-parents and co-parents living with the study fathers were invited to also participate in the study, with 582 co-parents enrolling (574 female co-parents, 6 male co-parents, and 2 co-parents with undisclosed gender identity). Participants were provided with study information and indicated their consent to participate by proceeding with the study surveys. Study activities were approved by the Mass General Brigham Institutional Review Board (protocol #: 2020P002688) and Boston College Institutional Review Board (protocol #: 22.201.01).

For analysis, we excluded 201 households with a child below the age of 2 (n = 201 fathers and 87 co-parents) because the CFPQ was originally designed for use among children 2 years and above (Musher-Eizenman & Holub, 2007) and some CFPQ items and subscales may not be applicable to parenting infants and toddlers (e.g., involvement in meal planning and meal preparation). We also excluded male co-parents (n = 6) and those with undisclosed gender identity (n = 2) because our aim was to assess measurement invariance between fathers and mothers. In the present study, we describe male parents as fathers and female parents as mothers. Thus, our final analytic sample included 1,071 fathers and 487 mothers. Participants’ demographics are outlined in Table 1.

Table 1.

Descriptive characteristics of parents

Characteristicsa	Fathers (n = 1,071)	Mothers (n = 487)
Age (years; mean, SD)	36.6 (4.1)	35.4 (4.0)
Race/ethnicity (n, %)
Hispanic/Latino	60 (5.6)	39 (6.3)
Non-Hispanic Asian	162 (15.1)	77 (16.1)
Non-Hispanic Black	97 (9.1)	24 (5.0)
Non-Hispanic American Indian or Alaska Native	11 (1.0)	3 (0.6)
Non-Hispanic White	721 (67.3)	325 (68.0)
Non-Hispanic Native Hawaiian	4 (0.4)	1 (0.2)
Education (n, %)b
High school diploma or less	31 (2.9)	7 (1.5)
Trade/vocational training/Associate’s degree/some college	137 (12.8)	56 (11.6)
Bachelor’s degree	387 (36.1)	162 (33.7)
Postgraduate degree	509 (47.5)	255 (53.0)
Annual household income in USD reported by fathers (dollars; n, %)
$0 – $49,999	121 (11.5)	-
$50,000 – $99,999	172 (16.4)	-
$100,000 – $149,999	262 (25.0)	-
$150,000 or above	494 (47.1)	-
Child biological sex reported by fathers (n, %)
Female	494 (46.1)	-
Male	577 (53.9)	-
Child age reported by fathers (months; mean, SD)	46.2 (15.9)	-

Notes: SD: Standard deviation; USD: United States Dollars;

^aIf Ns in each category do not add to the total N, the difference is due to missing data;

^bMothers’ education was reported by fathers in MI F&F and self-reported by mothers in GUTS.

2.2. Food parenting

Fathers and mothers reported their food parenting practices using 16 items selected from the original 49-item (12 subscale) CFPQ (Musher-Eizenman & Holub, 2007) (see Supplemental Table 1). In recent work, the CFPQ authors categorized the 12 subscales under the three higher-order domains of coercive control, structure, and autonomy support based on the self-determination theory in the parenting context (Musher-Eizenman et al., 2019) (see Supplemental Table 1). Using this typology, we selected six items representing coercive control (two items from the emotion regulation CFPQ subscale, two items from the food as reward subscale, and two items from the restriction for health subscale), seven items reflecting structure (two items from the monitoring subscale, two items from the environment subscale, one item from the modeling subscale, and two items from the child control subscale), and three items reflecting autonomy support (one item from the involvement subscale and two items from the encourage balance and variety subscale) for fathers in the present study to complete. We selected these nine food parenting practices subscales because they are commonly labelled across different food parenting instruments and are commonly used in research with fathers (Penilla et al., 2017; Tschann et al., 2015; Watterworth et al., 2017; Parada et al., 2016). We did not include items in the “teaching about nutrition” subscale (e.g., I discuss with my child the nutritional value of foods) because they are less commonly examined and its link with preschool-age children’s diets and weight is unclear (Larsen et al., 2015; Yee et al., 2017). We excluded items from the “pressure” subscale as its meaning (i.e., parents pressure the child to consume more food at meals) overlaps in the opposite direction of the “child control” subscale meaning (i.e., parents allow the child control of his/her eating behaviors). We also excluded items related to the “restriction for weight control” subscale because the subscale’s meaning (i.e., parents control the child’s food intake with the purpose of decreasing or maintaining the child’s weight) somewhat overlaps with the meaning of the “restriction for health” subscale (i.e., parents control the child’s food intake with the purpose of limiting less healthy foods and sweets). As we wanted to produce a brief tool, we selected one or two items that are most representative of each of the nine selected subscales based on authors’ food parenting expertise.

All items used a four-point Likert response scale with 1 = “Strongly Disagree”, 2 = “Disagree”, 3 = “Agree” and 4 = “Strongly Agree”.

2.3. Children’s diets

Fathers reported their children’s intake of fruit, vegetables, sugar-sweetened beverages (SSBs), and fast food using validated questions from the National Health and Nutrition Examination Survey (NHANES) Dietary Screener Questionnaire (DSQ) (National Cancer Institute Division of Cancer Control & Population Science, n.d.; Thompson et al., 2017). We included these dietary factors for analysis because of their links with childhood overweight and obesity (Jakobsen et al., 2023; Liberali et al., 2020). Fathers indicated how often, on average, in the past 30 days their child: (a) ate fruit, including fresh, frozen, or canned fruit; (b) ate vegetables, including raw, cooked, canned, or frozen vegetables; (c) drank regular soda or pop, excluding diet soda; (d) drank sweetened fruit, sports, or energy drinks, including fruit juices made at home with added sugar to, excluding diet ones; and (e) ate something from a fast food restaurant, such as McDonald’s, Burger King, and Taco Bell. Response options were “never,” “less than once per week,” “once per week,” “2–4 times per week,” “nearly daily or daily,” “2–4 times per day,” and “5 or more times per day.” When answering, fathers were asked to include meals and snacks that their child ate at home, at school/daycare, at restaurants, and anywhere else. We combined regular soda/pop and sweetened drinks to generate a SSBs variable for analysis. The Dietary Guidelines for Americans recommend that children between 5 and 8 years old should consume at least 1 cup of fruit and 1.5 cups of vegetables per day and minimize SSBs consumption (U.S. Department of Agriculture & U.S. Department of Health and Human Services, 2020). Based on this, we dichotomized children’s diets into whether they were consuming fruit/vegetables more than once per day and whether they were consuming SSBs less than once per week. In addition, the Dietary Guidelines for Americans also recommend limiting the consumption of added sugar, saturated fat, and sodium (U.S. Department of Agriculture & U.S. Department of Health and Human Services, 2020). We dichotomized children’s fast food intake into whether they were consuming fast food less than once per week.

2.4. Demographics

Fathers reported their age, race/ethnicity, employment status, education, and household income, along with their children’s age, biological sex, and childcare attendance in facilities and school settings. Fathers also reported mothers’ age and race/ethnicity. Mother education was reported by fathers among those recruited from Michigan Medicine and self-reported by mothers recruited from GUTS.

2.5. Statistical analyses

2.5.1. Creating independent subsamples

To support the integrity and independence of model building and testing in Aims 1 and 2, we created a number of independent subsamples. This was possible as we had a total sample of over 1,500 fathers and mothers. Figure 1 explains how three independent subsamples were created using random splitting. The overall objectives in creating the random-split subsamples were to ensure equivalent but independent samples were used for the initial (subsample 1) and updated (subsample 2) confirmatory factor analysis (CFA) in Aim 1, and that the fathers and mothers included in the assessment of multi-group CFA (subsample 3) for Aim 2 were not from the same household. Subsample 1 and subsample 2 included two independent sets of 535 participants (subsample 1 = 417 fathers, 118 mothers; subsample 2 = 413 fathers, 122 mothers). Supplemental Table 2 summarizes the demographic characteristics of the two independent subsamples. Subsample 3 included 481 participants (240 fathers and 241 mothers). For Aim 3, we used the entire sample of fathers and mothers (N = 1,558; 1,071 fathers and 487 mothers) to assess the scale’s internal reliability and the father sample (n = 1,071) to assess concurrent validity with children’s diets among fathers.

Figure 1:

Participants flow for subsamples

2.5.2. Aim 1: Factorial validity

We conducted CFA in two steps to validate the three-factor food parenting structure using a subset of CFPQ items. First, using subsample 1, we specified a three-factor CFA using the 16 food parenting items. We allowed correlations among the three factors and fixed each factor’s variance to 1.0 to identify the model. Missing data were treated using full information maximum likelihood estimation (Enders & Bandalos, 2001). Maximum likelihood estimation with robust (Huber-White) standard errors was used to treat non-normally distributed data (Knekta et al., 2019).

Because chi-square statistic (χ²/df) is sensitive to sample sizes, we considered additional fit indices to assess model fit. We used comparative fit index (CFI), which compares the fit of the specified model with that of a null model assuming the factors in the model are uncorrelated; CFI values ≥ 0.95 indicate good, and values ≥ 0.90 indicate acceptable fit (Byrne, 2010; Hu & Bentler, 1999). Root mean square error of approximation (RMSEA) assesses how well the model fits the population covariance matrix while considering sample size and model complexity. RMSEA values > 0.10 are considered to indicate poor fit, values between 0.06 and 0.10 represent acceptable fit, and values < 0.06 indicate good fit (Hu & Bentler, 1999; Kline, 2016; MacCallum et al, 1996). The standardized root means square residual (SRMR) measures discrepancies between covariance matrices of the model. SRMR values < 0.08 are considered acceptable (Hu & Bentler, 1999).

In this initial CFA step, items were removed if they had an extremely low standardized factor loading (i.e., <0.20) (Vahip et al., 2005). We also used modification indices to identify components in the model that could be modified to improve overall goodness-of-fit. Next, using subsample 2, we specified an updated three-factor CFA informed by the results of the initial CFA in the previous step to validate the three-factor food parenting structure using a subset of retained CFPQ items.

2.5.3. Aim 2: Measurement invariance

Next, using subsample 3, we used multi-group CFA to compare the factor structure (i.e., test for measurement invariance) across fathers and mothers. As previously noted, fathers and mothers were not from the same household (Figure 1). Using standard rigorous measurement invariance procedures (Putnick & Bornstein, 2016), we compared a series of measurement models sequentially to analyze four types of measurement invariance: configural, metric, scalar, and residual invariance (Putnick & Bornstein, 2016). Configural invariance refers to the same pattern of factor structures between fathers and mothers. If configural invariance was supported, metric invariance would be tested next to examine if the factor loadings of the items on the latent factors were similar between fathers and mothers. If metric invariance was supported, we would test scalar invariance to examine if the intercepts of the metric invariant items were similar between fathers and mothers. When the scalar invariance is met, means of the latent constructs can be compared across groups (Jiang et al., 2017). Lastly, if scalar invariance was supported, residual invariance would be tested to examine if the measurement error of individual items was similar across fathers and mothers.

Guidelines for violating metric invariance are a change in CFI ≥0.010 coupled with either a change in RMSEA ≥0.015 or a change in SRMR ≥0.030 (Chen, 2007). Guidelines for violating scalar and residual invariances are a change in CFI ≥0.010 coupled with either a change in RMSEA ≥0.015 or a change in SRMR ≥0.010 (Chen, 2007).

2.5.4. Aim 3: Internal reliability and concurrent validity

Utilizing data of the entire sample with both fathers and mothers, we assessed the scale’s internal reliability using Cronbach’s alpha (for factors with more than two items) and the Spearman-Brown coefficient (for factors with two items) (Eisinga et al., 2013) with ≥0.60 indicating acceptable reliability (Nunnally & Bernstein, 1994).

We assessed each factor’s concurrent validity by examining the association between fathers’ food parenting factor scores and children’s diets. Specifically, multiple logistic regression was used to examine the links between fathers’ food parenting factors (independent variables) and the likelihood of children consuming fruit/vegetables more than once per day and consuming SSBs/fast food less than once per week (dependent variables). Each dependent variable (i.e., fruit, vegetables, SSBs, and fast food) was analyzed separately. The three food parenting factors (i.e., coercive control, structure, and autonomy support) were analyzed simultaneously in the same model. Factor scores were calculated by taking the mean of the representing items so that higher scores indicate greater levels of coercive control, structure, and autonomy support. When selecting covariates, we examined review papers to identify demographic characteristics that are linked to both fathers’ food parenting and children’s diets. The relevant covariates included in concurrent validity testing were children’s age (Hamner et al., 2023; Khandpur et al., 2014; Rahill et al., 2020), children’s biological sex (Hamner et al., 2023; Khandpur et al., 2014; Rahill et al., 2020), and fathers’ education (Hamner et al., 2023; Khandpur et al., 2014; Rahill et al., 2020). We also controlled for recruitment sites to account for potential differences between fathers recruited from GUTS and those recruited from Michigan Medicine. We did not control for fathers’ race/ethnicity because evidence illustrating critical racial/ethnic differences in fathers’ food parenting on preschool-age children is limited within our sample and others (Lora et al., 2016; Wang et al., 2024).

3. Results

Participants were mostly non-Hispanic (NH) White (67.3%) or NH Asian (15.1%) fathers with a mean age of 36.6 (SD=4.1). The majority of fathers had completed a bachelor’s degree or beyond (83.6%). The demographics of the fathers were similar to those of the mothers (Table 1). Close to three-quarters (72.1%) of households had an annual income ≥100,000 USD. Half of the children in the study were male (53.9%) with a mean age of 46.2 months (SD=15.9; Range=24 to 85 months) at enrollment. There were no demographic differences between parents in the initial CFA (i.e., subsample 1) and updated CFA (i.e., subsample 2) (Supplementary Table 2).

3.1. Aim 1: Factorial validity

The initial CFA model using 16 food parenting items to represent coercive control, structure, and autonomy support did not fit the data well as evidenced by poor model fit indices (robust χ²[101] = 1105.94; robust CFI = 0.54; robust RMSEA = 0.137, 90% CI = 0.129–0.145; robust SRMR = 0.115).

Based on the standardized loadings output, we removed five items that had extremely low loadings (i.e., < 0.20). This decision was approved by food parenting experts on the team, who determined that removing these five items would not compromise the theoretical basis of the remaining items representing the three higher-order factors. Although the widely recommended threshold for factor loading is 0.7 (Shrestha, 2021), evidence in the literature suggests that items with low factor loadings can still be retained if they represent an essential part of the factor (e.g., Knekta et al., 2019). Based on the authors’ expertise and consensus, and supported by previous studies using a similar approach (Afulani et al., 2017), we decided to include all remaining 11 items in the analysis. The standardized factor loadings for the finalized items ranged 0.26 to 0.68 (Table 2).

Table 2.

Standardized item loadings from the final updated confirmatory factor analysis for the eleven retained Comprehensive Feeding Practices Questionnaire (CFPQ) items representing coercive control, structure and autonomy support^a

Item #	CFPQ item	Original CFPQ construct	3-factor, higher order construct	Final updated CFA standardized factor loadingc
1	I give my child something to eat or drink (excluding water) if they are bored even if I don’t think they are hungry.	Emotion regulation	Coercive control	0.45
2	I give my child something to eat or drink (excluding water) if they are upset even if I don’t think they are hungry.			0.47
3	I offer sweets (candy, ice cream, cake, pastries) to my child as a reward for good behavior.	Food as reward		0.34
4	I withhold sweets/dessert from my child in response to bad behavior.			0.26
5	I keep track of sweets (candy ice cream, cake, pies, pastries) that my child eats.	Monitoring	Structure	0.48
6	I keep track of snack food (potato chips, Doritos, cheese puffs) that my child eats.			0.51
7	I keep a lot of snack food (such as potato chips, Doritos, and cheese puffs) in my house.b	Environment		0.30
8	I keep a lot of sweets (candy, ice cream, cake, pies, pastries) in my house.b			0.27
9	I try to eat healthy foods in front of my child.	Modeling		0.62
10	I tell my child that healthy food tastes good.	Encourage balance & variety	Autonomy support	0.66
11	I encourage my child to eat a variety of foods.			0.68

Notes: CFPQ: Comprehensive Feeding Practices Questionnaire; CFA: Confirmatory Factor Analysis;

^aThe 16 items selected from the original 49-item (12 subscale) CFPQ and item mean scores are listed in Supplemental Table 1;

^bIndicates reverse-coded items;

^cFactor loadings of the updated CFA model utilizing data from subsample 2 with 413 fathers and 122 mothers.

To identify potential model misspecification, we examined modification indices. Large modification indices were found in four pairs of error covariances and each pair belonged to the same subscale in the original CFPQ (i.e., emotion regulation, food as reward, monitoring, and environment). We accordingly modified the model to allow these error covariances to be estimated (Byrne et al., 1989). The updated three-factor CFA using 11 items demonstrated adequate model fit, as evidenced by robust fit indices (robust χ²[37] = 95.87; robust CFI = 0.965; robust RMSEA = 0.053, 90% CI = 0.039–0.068; robust SRMR = 0.049; (Table 2). The CFI exceeded 0.95, indicating a good fit, while the RMSEA and SRMR values fell within the range considered acceptable fit. Collectively, these model fit indices suggest that the model appropriately captures the underlying structure of the data.

3.2. Aim 2: Measurement invariance

The results of the measurement invariance testing are summarized in Table 3. The configural invariance model (M1) suggested that factor structures were invariant between fathers and mothers (robust χ² [74] =116.502, robust CFI = 0.977, robust RMSEA = 0.045, robust SRMR = 0.056). When data were fitted separately, adequate model fit was also observed for fathers (robust χ²[37] = 47.379; robust CFI = 0.990; robust RMSEA = 0.031, 90% CI = 0.000–0.059; robust SRMR = 0.044) and mothers (robust χ²[37] = 65.281; robust CFI = 0.961; robust RMSEA = 0.055, 90% CI = 0.030–0.078; robust SRMR = 0.068).

Table 3.

Goodness-of-fit indices of the three-factor fathers’ food parenting model and measurement invariance across fathers and mothers

Model	Robust χ2 (df)	Robust CFI	Robust RMSEA (90% CI)	Robust SRMR	Model compared	Δ Robust χ2 (df)	Δ Robust CFI	Δ Robust RMSEA	Δ Robust SRMR	Decision
M1: Configural	116.502 (74)	0.977	0.045 (0.025–0.062)	0.056	-	-	-	-	-	Accept
M2: Metric invariance	127.218 (82)	0.973	0.046 (0.028 – 0.063)	0.058	M1	11.379 (8)	0.004	0.001	0.002	Accept
M3: Scalar invariance	150.685 (90)	0.966	0.049 (0.033 – 0.065)	0.063	M2	28.232 (8)	0.007	0.003	0.005	Accept
M4: Residual invariance	173.993 (101)	0.957	0.053 (0.037 – 0.067)	0.069	M3	22.799 (11)	0.009	0.004	0.006	Accept

Notes: M1: Configural model; M2: Metric invariance model; M3: Scalar invariance model; M4: Residual invariance model; χ²: Satorra-Bentler scaled chi-square test; df: degree of freedoms; CFI: Comparative Fit Index; RMSEA 90% CI: Root Mean Square Error of Approximation and its 90% confidence interval; SRMR: Standardized Root Mean Square Residual; Δ: values difference.

The metric invariance model (M2) also showed that the model fit was acceptable, and the changes in fit indices from M1 to M2 met the criteria, illustrating that factor structures are invariant between fathers and mothers (ΔCFI = 0.004; ΔRMSEA = 0.001; ΔSRMR = 0.002).

Similarly, scalar invariance model (M3) indicated an acceptable model fit. The changes of fit indices between M2 and M3 (ΔCFI = 0.007; ΔRMSEA = 0.003; ΔSRMR = 0.005) also illustrated that all factor loadings and intercepts are invariant between fathers and mothers.

Residual invariance was also established given the changes in fit indices in residual invariance model (M4) when compared to M3 (ΔCFI = 0.009; ΔRMSEA = 0.004; ΔSRMR = 0.006).

3.3. Aim 3: Internal reliability

There was adequate internal reliability across the three factors. Cronbach’s alpha for coercive control and structure were 0.60 and 0.69 respectively. The Spearman-Brown coefficient for autonomy support was 0.61.

3.4. Aim 4: Concurrent validity

Results of the concurrent validity testing are summarized in Table 4. The 11-item, three-factor, overarching food parenting structure demonstrates strong concurrent validity. We found that for fathers practicing more coercive control, their children had decreased odds of consuming fruit more than once per day (adjusted odds ratio [AOR] = 0.68, 95% CI = [0.52, 0.90]), decreased odds of consuming vegetables more than once per day (AOR = 0.62, 95% CI = [0.45, 0.86]), and decreased odds of consuming fast food less than once per week (AOR = 0.63, 95% CI = [0.46, 0.85]). In contrast, for fathers providing more structure, their children had increased odds of consuming fruit more than once per day (AOR = 1.54, 95% CI = [1.12, 2.13]), increased odds of consuming vegetables more than once per day (AOR = 2.51, 95% CI = [1.70, 3.71]), increased odds of consuming SSBs less than once per week (AOR = 3.15, 95% CI = [2.17, 4.57]), and increased odds of consuming fast food less than once per week (AOR = 3.34, 95% CI = [2.34, 4.78]). For fathers providing more autonomy support, their children had increased odds of consuming fruit more than once per day (AOR = 1.47, 95% CI = [1.10, 1.95]), increased odds of consuming vegetables more than once per day (AOR = 2.13, 95% CI = [1.48, 3.05]), and decreased odds of consuming SSBs less than once per week (AOR = 0.72, 95% CI = [0.52, 0.99]).

Table 4.

Multiple logistic regression for concurrent validity testing linking fathers’ food parenting with child diets (N = 1,071)

Variables	Fruita (n=1,033)		Vegetablesb (n=1,034)		Sugar-sweetened beveragesc (n=1,030)		Fast foodd (n=1,034)
	AOR^	CI (95%)	AOR^	CI (95%)	AOR^	CI (95%)	AOR^	CI (95%)
Father-reported food parenting practices
Coercive control	0.68**	(0.52 – 0.90)	0.62**	(0.45 – 0.86)	0.83	(0.61 – 1.14)	0.63**	(0.46 – 0.85)
Structure	1.54**	(1.12 – 2.13)	2.51***	(1.70 – 3.71)	3.15***	(2.17 – 4.57)	3.34***	(2.34 – 4.78)
Autonomy support	1.47**	(1.10 – 1.95)	2.13***	(1.48 – 3.05)	0.72*	(0.52 – 0.99)	1.03	(0.76 – 1.39)

Notes: AOR: Adjusted odds ratio; CI (95%): 95% confidence interval;

^aOutcome: consuming fruit more than once per day;

^bOutcome: consuming vegetables more than once per day;

^cOutcome: consuming sugar-sweetened beverages less than once per week;

^dOutcome: consuming fast food less than once per week;

^*p < .05;

^**p < .01;

^***p < .001;

^{^}Controlled for child age in month (continuous), child biological sex (male, female), fathers’ education (some college or less, bachelor’s degree or less, graduate degree or more), recruitment site (GUTS, Michigan Medicine).

4. Discussion

To develop an efficient and valid measure of food parenting for fathers, we tested the factorial validity of using a subset of items from the CFPQ to represent three overarching food parenting factors: coercive control, structure, and autonomy support (Mâsse et al., 2020; Musher-Eizenman et al., 2019; O’Connor et al., 2017; Vaughn et al., 2016). Factor analyses revealed acceptable fit for utilizing 11 items to represent the three factors, with measurement invariance established across fathers and mothers. All three factors exhibited adequate internal reliability. We also observed that scores on the brief food parenting measure were associated with certain children’s dietary intakes, thus providing some evidence of the concurrent validity of the brief tool.

Our study showcased the ability to use only a small subset of CFPQ items to capture coercive control, structure, and autonomy support (Mâsse et al., 2020; Musher-Eizenman et al., 2019; O’Connor et al., 2017; Vaughn et al., 2016) among fathers. Such a discovery has several important and timely contributions. First, considering fathers have been historically underrepresented in food parenting research (Davison et al., 2016), and time constraints have been noted as a barrier for fathers to participate in research (Davison et al., 2017), using fewer items to measure food parenting might be useful to encourage fathers to participate in research. Involving fathers in research will substantially expand our knowledge of fathers’ food parenting practices and how that might influence their children’s diets, allowing the development of targeted interventions that equitably support children’s healthy eating. Second, the items selected and the factors identified in our study were based on the three overarching parenting domains (i.e., coercive control, structure, and autonomy support) derived from the self-determination theory in the context of food parenting (Grolnick & Lerner, 2023; Mâsse et al., 2020; Musher-Eizenman et al., 2019; O’Connor et al., 2017; Vaughn et al., 2016). Using a theory to inform our tool development minimizes the creation of similar or overlapping factors, which will allow for potential comparison of findings with other studies. Furthermore, measurement invariance was established across fathers and mothers, suggesting the possibility of using this tool to comprehensively understand the impacts of food parenting synergies within father-mother households, where mothers can also benefit from having a brief survey instrument. Studies have found that regardless parents’ gender, parents prefer participating in studies with minimal burden such as time commitment (Nathe et al., 2022). Lastly, we observed some concurrent validity relating the three food parenting factors to children’s fruit, vegetables, SSBs, and fast food intakes, demonstrating their potential utility in predicting children’s diets.

Across all three factors, only structure had significant associations with all of the children’s diets, providing strong evidence of the concurrent validity of the brief measure of structure. Similarly, we also observed strong evidence of concurrent validity between our measure of coercive control and children’s diets. There were significant associations between coercive control and children’s fruit, vegetables, and fast food intakes, except for SSBs. These collectively suggest that fathers’ food parenting structure and coercive control captured by our tool are relevant to children’s diets tested in our study.

As for autonomy support, we only found significant associations with children’s fruit, vegetables and SSBs intakes, demonstrating weak concurrent validity. The direction of the association between autonomy support and SSBs was in the opposite direction of what we expected. The weaker concurrent validity for autonomy support suggests that while autonomy support is a theory-driven factor, items in our tool may not have fully captured this factor adequately, or autonomy support captured by our tool is less relevant to children’s diets among fathers. In our study, we only retained two items to represent autonomy support. Among the two items used to capture autonomy support, they were highly rated by fathers in our study. 92.9% indicated agreement for “I tell my child that healthy food tastes good,” and 99.1% indicated agreement for “I encourage my child to eat a variety of foods.” The lack of variability in responses supports the need to further refine the items that may capture autonomy support more fully among preschool children and test the relevance of autonomy support in relation to children’s diets among fathers. We also recommend to test items within autonomy support and the relevance of this factor in relation to children’s diets in different social and cultural contexts.

While our study is among the first to use a large sample of fathers to successfully demonstrate using a subset of CFPQ items to represent coercive control, structure, and autonomy support food parenting practices, some limitations exist. First, many of the participants were non-Hispanic White and had high socioeconomic status with young children, limiting the generalizability of findings to other populations. As non-White fathers and those with a low socioeconomic status may utilize food parenting practices differently (Rahill et al., 2020), future research needs to test if items are still applicable to fathers with these different backgrounds. Second, we used a brief dietary screener to only capture children’s frequency of fruit, vegetables, SSBs, and fast food intakes without distinguishing food preparation sources (e.g., fast casual vs full-service restaurants). By only capturing intake frequencies of four food categories, our measures could only serve as dietary intake proxies, and it limited our understanding of our measurements’ concurrent validity beyond these dietary factors. Future studies can use more comprehensive dietary measurements, such as 24-hour dietary recalls, to examine different food types and food preparation sources, thereby providing greater clarity on the influences of food parenting practices on child dietary outcomes in different contexts. Lastly, further refinement of items included for each factor is encouraged. We only found weak concurrent validity for autonomy support with children’s diets partly due to the limited number of relevant items included. Further considerations of which CFPQ items to be included in autonomy support is required as others have found links between autonomy support-related constructs and children’s diets (Yee et al., 2017). Attention is also warranted for coercive control as we did not observe a link with SSBs while others did (Lora et al., 2016).

5. Conclusion

Our study was the first to provide an example of using 11 CFPQ items to represent three overarching food parenting factors equivalently among fathers and mothers: coercive control, structure, and autonomy support. These findings provide a blueprint for a succinct food parenting measurement tool that can facilitate fathers’ participation in research. Further refinements of items are encouraged, particularly coercive control and autonomy support, in other social and cultural contexts.

Abbreviations

BMI

Body Mass Index

CFQ

Child Feeding Questionnaire

CFI

Comparative fit index

CFPQ

Comprehensive Feeding Practices Questionnaire

CFA

Confirmatory factor analysis

DSQ

Dietary Screener Questionnaire

FPSQ

Feeding Practices and Structure Questionnaire

FPP

Food Parenting Practice

NHANES

National Health and Nutrition Examination Survey

RMSEA

Root-mean-square error of approximation

SRMR

Standardized root-mean-square residual

SSBs

Sugar-sweetened beverages

Data availability:

Upon reasonable request to the senior author: Kirsten K. Davison (kirsten.davison@bc.edu).