1. Exploratory and Confirmatory Factor Analysis
of Measures of Controlling Feeding Practices
Jansen E1*, Mallan K1, Nicholson J 2,3, Daniels L1
1 Institute of Health and Biomedical Innovation, School of Exercise and Nutrition Sciences, Queensland University of Technology, Brisbane, Australia
2 Queensland University of Technology, Brisbane, Australia
3 Murdoch Children’s Research Institute, Melbourne, Australia
* t: 61 (0) | e:
Rationale
A large number of studies have examined the role of parental feeding, notably controlling feeding practices (CFPs), as a determinant of children’s weight gain[1-3]. However, key constructs of controlling feeding practices such as pressure to eat, restriction, monitoring, overt and covert control have only been validated/evaluated in older children (mostly > 4 years)[4-6].
In a sample of Australian first-time mothers with 2 year old children, the factor structure of 24 items commonly used to assess controlling feeding practices was examined.
Results continued
EFA: The factors and their items identified by the Exploratory Factor Analysis are shown in Table 1. EFA on the 24 items revealed a 7-factor solution, fulfilling most of the criteria listed in Figure 1. Two of the 5 above factors were replicated exactly (i.e. ‘covert control’ and ‘monitoring’). New factors consisted of a) 2 items assessing ‘reward’, b) 2 items assessing restriction of sweets and high fat foods, and c) the remaining 4 ‘restriction’ items. ‘Overt control’ and ‘pressure’ were partially replicated with 1 item originally loading on the former, now loading on the latter factor.
Table 1: Factor structure of controlling feeding practices (N=466)
Methods
Participants
Participants from the intervention and control groups of the NOURISH RCT (in
2008)[7] – 698 mother-child dyads allocated at baseline, child age 4-7 months
NOURISH promoted protective early feeding practices, healthy child eating
behaviours & growth
Data used here were collected during the 3rd assessment time point (follow-up)
when children were approximately 2 years of age
Valid data were available from 466 first-time mothers
Measures
Self-administered questionnaire
24 items, measured on 5 point Likert scales – 1 (lowest) to 5 (highest) – sourced
from 2 questionnaires (including Cronbach’s α from original validation studies):
Child Feeding Questionnaire[4]
Restriction – 8 items (α = 0.73)
Pressure – 4 items (α = 0.70)
Monitoring – 3 items (α = 0.92)
Overt vs. Covert Control Questionnaire[6]
Overt control – 4 items (α = 0.71)
Covert control – 5 items (α = 0.79)
Data Analyses
Multi-stage validation analysis of the factor structure of controlling feeding practices items (see Figure 1)
Exploratory factor analysis (EFA): using principal axis factoring and oblique rotation; conducted on 224 randomly selected cases (i.e. sample 1)
Confirmatory factor analysis (CFA): carried out on the remaining 242 cases; to confirm factor structure identified in EFA (i.e. sample 2)
Internal consistency was examined via Cronbach’s alpha & coefficient H combining sample 1 and 2
Factors and items composing each factor
Cronbach’s α
Coefficient H
Factor 1 – Covert Control: How often do you avoid...
CC1: ... going with your child to cafes or restaurants which sell unhealthy foods
CC2: ... buying lollies and snacks and bringing them into the house
CC3: Not buy foods that you would like because you do not want your children to have them
CC4: Try not to eat unhealthy foods when your child is around
CC5: ... buying biscuits and cakes and bringing them into the house
.820
873
Factor 2 – Rewards: I offer...
Rest5: Sweet foods to my child as a reward for good behaviour
Rest6: My child his favourite foods in exchange for good behaviour
.833
.869
Factor 3 – Monitoring: How much do you keep track of the...
Mo1: ... sweet foods that your child eats
Mo2: ... snack food that your child eats
Mo3: ... high fat foods that your child eats
.931
.976
Factor 4 – Restriction: high fat & sweets: I have to be sure that my child does not eat too...
Rest1: ... many sweet foods
Rest2: ... much of his favourite foods
.865
.939
Factor 5 – Restriction: general: If I did not guide or regulate my child’s eating, he would eat too...
Rest3: I have to be sure that my child does not eat too much of his favourite foods
Rest4: I intentionally keep some foods out of my child’s reach
Rest7: ... many junk foods
Rest8: ... much of his favourite foods
.703
.782
Factor 6 – Overt Control: How often are you firm about...
OC1: ... what your child should eat
OC2: ... when your child should eat
OC3: ... where your child should eat
.691
.768
Factor 7 – Pressure
Pr1: My child should always eat all of the food on his plate
Pr2: I have to be especially careful to make sure my child eats enough
Pr3: If my child says “I’m not hungry” I try to get him to eat anyway
Pr4: If I did not guide or regulate my child’s eating, he would eat much less than he should
OC4: How often are you firm about how much your child should eat
.783
.842
Abbreviations: CC = Covert Control, Rest = Restriction, Mo = Monitoring, OC = Overt Control, Pr = Pressure
CFA: Confirmatory Factor Analysis of this 7-factor model indicated ‘reasonable’ model fit to the data (see Figure 2 and Table 2) once paths between error terms of items restriction2 + monitoring3 and overt control4 + pressure1 respectively were included.
Item OC4 (How often are you firm about how much your child should eat) loaded more strongly on the construct ‘pressure’ than its original ‘overt control’ factor (standardized regression weight, no modification indices: .49 and .33 respectively).
Figure 2: Final CFA of the CFP items showing standardized estimates
Figure 1: Procedure of multi-stage validation analysis followed in this study
Table 2: CFA ‘reasonable’ fit indices
Data preparation
Cases with valid data
Systematic missing data (Little's MCAR test)
Outliers
Case allocation
(EFA & CFA)
EFA & checking of assumptions
Sample size
(case/item; missing data; homogeneous constructs across sample)
Items
(items/construct; metric or dichotomous)
Factorability
(R-matrix r>0.3; Kaiser-Meyer-Olkin overall measure of sampling adequacy (MSA) >0.6; individual MSA on item level >0.5; Bartlett’s test of sphericity p<0.05)
Outliers among items
(poor MSA <0.6; low communality <0.1; low factor loading; split/cross-loading)
Non-linearity
Multicollinearity
(r> between items;
determinant of R-matrix: >.00001)
Extraction (Principal Components Analysis, Principal Axis Factoring etc.)
Rotation (orthogonal or oblique)
Factor correlation matrix r>0.3
Theory
Number of factors to retain
Kaiser criterion: factors with eigenvalues >1.0
% of variance explained
Scree plot
Item/factor loading >0.4 on a single factor
Final factor structure based on theoretical conceptualisation + simple structure
High loading on only 1 factor
Several highly loading items for each factor
No split loading (cross-loading)
CFA
Missing data imputation
Regression imputation with EM (expectation-maximization) estimation adjustment
Bollen-Stine bootstrap p-value if non-normal (i.e. kurtosis critical ratio >±1.96)
Check full model for:
Item/face validity (standardized regression weight >0.7; estimates are positive & significant)
Item reliability (squared multiple correlations=R2>0.5  more explained than unexplained variance = low measurement error)
Fit indices (χ2 + df; (Bootstrap) p-value; Normed χ2; SRMR; GFI & AGFI; TLI; CFI; RMSEA )
Decision for change
Standardized residual covariances >1.96  problem with these items
Modification indices (except for negative covariances of error terms)
Reliability
Cronbach’s alpha + Coefficient H
Fit indices
Cut-offs
7 factor model
Item validity
Standardized regression weight >0.7
15 items 
9 items x
Item reliability
Squared multiple correlations (R2) >0.5
χ2 (df) +
Normed χ2
1) p- value
2) bootstrap
>0.05
(229) +  x
GFI
>.90
x (.87)
TLI
CFI
RMSEA
SRMR
<.050
x (.063)
Abbreviations: GFI = Goodness-of-Fit Index, TLI = Tucker Lewis Index, CFI = Comparative Fit Index, RMSEA = Root Mean Square Error of Approximation, and SRMR = Standardized Root Mean Square Residual
Conclusions
The new 7-factor-structure for measuring controlling feeding practices in
children ≤ 2 years of age was stable across 2 samples.
Internal consistency of factors was better than in the original studies[4,6] and
particularly coefficient H indicated good reliability of all factors.
Construct validity can be further assessed by associating individual factors
and child outcomes (i.e. weight status).
Further development and validation of new items may be required, given
some low fit-indices.
Results
Children (48% boys) in the sample were 24 (SD±1) months old with weight-for-age z-score = 0.6 (SD±0.9). Their mothers were 33 (SD±5) years old, 65% had a university degree, and 97% were married or in a defacto relationship.
Strengths and Limitations
+ Simultaneous usage of various commonly applied controlling feeding practice
measures
+ The relatively large sample size allowed several additional steps in the validation analysis and supported the robustness of the result
Maternal controlling feeding practices were self-reported
References
[1] Clark et al. J Public Health 2007;29(2): ; [2] Faith et al. Obesity Research 2004;12 (11): ; [3] Ventura & Birch. Int J Behav Nutr Phys Act 2008;5:15; [4] Birch et al. Appetite 2001;36(3):201-10; [5] Musher-Eizenman & Holub. J Pediatr Psychol 2007;32(8):960-72; [6] Ogden et al. Appetite 2006;47(1):100-6; [7] Daniels et al. Bmc Public Health 2009;9:387
Acknowledgements
NOURISH is funded by NHMRC ID
K Mallan’s post doctoral fellowship is funded by HJ Heinz Co.