Factors Affecting Children's Eating Behavior: Regression Analysis

Introduction

Children are highly dependent on their parents because parents are their sole providers. A parent's primary responsibility is to provide the basic needs — food, shelter, and clothing — of their children. Consequently, parents shape the eating habits of children, especially those under the age of 12. Generally, children are usually ready to learn how to eat new foods, and they observe the eating behavior of adults around them (Reicks et al.). However, their eating behaviors evolve as they grow older.

Numerous studies have identified factors that influence children's eating behavior. These include living conditions, access to food, number of caretakers or family members nearby, employment status, age, gender, and health condition (Savage et al.). This paper estimates the effects that these factors have on the eating habits of children. For a broader overview of children's eating behavior and the variables that shape it, established research provides useful context.

Data and Descriptive Statistics

The data for this project were compiled from various internet sources. All statistical analyses were carried out using Microsoft Excel. Descriptive statistics indicate the mean, median, standard deviation, maximum, and minimum values of each variable. The correlation coefficient, r, measures the strength of the linear relationship between any two variables. Regression analysis predicts the influence of one or more explanatory (independent) variables on the dependent variable.

Descriptive statistics were used to characterize the variables used in this project. The results are displayed in Table A1 (Appendix A). Eating behavior scores range from 41 to 100 (M = 71.07, SD = 17.47). A higher eating behavior score reflects healthier eating behavior. The average age of the subjects is 6.39 years. Most subjects live in a developed area (60.20%) and have employed parents (65.31%). Most households are composed of both parents. Almost half of the subjects were male (53.06%). Furthermore, 54.08% of subjects do not use electronics at mealtimes, and approximately half (56.12%) confirmed that the availability of healthy food was limited.

Correlation Analysis

Correlation results are displayed in Table B1 (Appendix B). The analysis confirms that the independent variables are not substantially correlated with one another, indicating that multicollinearity is not a concern in the regression models that follow.

Regression Models and Interpretations

Regression analysis was performed to predict eating habits among children. Four different regression equations were estimated. Each equation is described below.

Regression Equation 1

Eating Behavior = β₀ + β₁(Living Location) + β₂(Access to Food) + β₃(Age) + β₄(Gender) + β₅(Electronic Use)

Equation 1 is the base regression model for estimating eating behavior in children. It represents the linear relationship between eating behavior and the key explanatory variables: living location, access to food, age, gender, and electronic use. The Excel results for this equation are displayed in Table C1 (Appendix C). The estimated equation is as follows:

EatingBehavior = 75.598 + 2.253 Livloc − 7.643 Foodacc − 0.572 Age − 4.268 Gender + 7.375 Elecuse
(t): (15.11)   (0.62)   (−2.20)   (−1.17)   (−1.23)   (2.10)    R² = 0.1133

All variables are statistically insignificant at the 5% level except access to food (p = 0.03) and electronic use (p = 0.04). A t-test was performed to determine whether access to food has a negative effect on eating behavior and whether electronic use during mealtimes affects eating behavior.

For access to food, the null and alternative hypotheses are:

H₀: β₂ = 0    H_A: β₂ ≠ 0

The test statistic for access to food is −2.20. At the 5% significance level, the critical value of the t-distribution with N − 5 = 93 degrees of freedom is t(0.95, 93) = 0.063. Since the calculated value falls in the rejection region, the null hypothesis that β₂ = 0 is rejected, and it is concluded that the coefficient of access to food is nonzero (Hill et al. 109).

For electronic use, the null and alternative hypotheses are:

H₀: β₅ = 0    H_A: β₅ ≠ 0

Since t = 2.10 is greater than 0.063, the null hypothesis that β₅ = 0 is rejected, and it is concluded that the coefficient of electronic use is statistically significant. This confirms that when children do not use electronics during mealtimes, their eating habits improve.

R² = 0.1133, meaning the regression model explains 11.33% of the variation in eating behavior.

Regression Equation 2

Eating Behavior = β₀ + β₁(Living Location) + β₂(Access to Food) + β₃(Age) + β₄(Gender) + β₅(Electronic Use) + β₆(Household)

In Equation 2, the first proxy variable — household structure — is added to the base model. The Excel results are displayed in Table D1 (Appendix D). The estimated regression equation is:

EatingBehavior = 75.366 + 2.225 Livloc − 7.675 Foodacc − 0.572 Age − 4.283 Gender + 7.414 Elecuse + 0.470 Household
(t): (14.1)   (0.61)   (−2.20)   (−1.16)   (−1.22)   (2.09)   (0.14)    R² = 0.1134

In this model, household structure is statistically insignificant (p = 0.892791). Therefore, household type — whether single-parent or two-parent — does not influence children's eating behavior. The value of R² remained essentially unchanged at 0.1134, indicating that adding family structure did not improve model fit. The coefficients changed only slightly compared to the base model; for example, the coefficient for electronic use increased marginally from 7.375 to 7.414, while the coefficient for age remained unchanged at −0.572.

Regression Equation 3

Eating Behavior = β₀ + β₁(Living Location) + β₂(Access to Food) + β₃(Age) + β₄(Gender) + β₅(Electronic Use) + β₆(Employment Status)

In Equation 3, the second proxy variable — parental employment status — is added to the base model. The Excel results are displayed in Table E1 (Appendix E). The estimated equation is:

EatingBehavior = 76.937 + 2.386 Livloc − 7.630 Foodacc − 0.600 Age − 3.833 Gender + 7.449 Elecuse − 2.307 Empstatus
(t): (14.09)   (0.65)   (−2.19)   (−1.22)   (−1.08)   (2.11)   (−0.62)    R² = 0.1170

The effects of the second proxy are almost identical to those of the first. The coefficient for employment status is insignificant at the 5% level (p = 0.53511). The slight change in R² indicates that including employment status does not meaningfully improve the adequacy of the base model. Coefficient estimates remain close to those in the base model; for instance, the coefficient for access to food is −7.630 in Equation 3 versus −7.643 in Equation 1.

Regression Equation 4

Eating Behavior = β₀ + β₁(Living Location) + β₂(Access to Food) + β₃(Age) + β₄(Gender) + β₅(Electronic Use) + β₆(Employment Status) + β₇(Household)

In Equation 4, both proxy variables are included in the base model simultaneously. The Excel results are displayed in Table F1 (Appendix F). The estimated regression equation is:

EatingBehavior = 76.775 + 2.367 Livloc − 7.650 Foodacc − 0.600 Age − 3.847 Gender + 7.474 Elecuse − 2.282 Empstatus + 0.298 Household
(t): (13.21)   (0.64)   (−2.18)   (−1.21)   (−1.07)   (2.10)   (−0.61)   (0.09)    R² = 0.1170

Both employment status (p = 0.54319) and household structure (p = 0.932424) are statistically insignificant at the 5% level. Including both proxies does not improve model fit; R² increases only marginally from 0.1133 to 0.1170. Overall, neither proxy variable influences children's eating habits in a statistically meaningful way.

Conclusion

Many factors influence a child's eating behavior. This analysis shows that access to food and electronic use during mealtimes are the two variables that significantly influence children's eating behavior. There is a negative relationship between food accessibility and eating behavior: when food availability is limited, a child's eating habits tend to be less healthy. Conversely, a positive relationship exists between electronic use during mealtimes and eating behavior scores, suggesting that refraining from electronics at the table is associated with healthier eating. Other factors — including living location, gender, age, employment status, and household type — are statistically insignificant.

The base regression model (Equation 1) is adequate for explaining variability in children's eating habits. Parental employment status and household type are irrelevant variables that unnecessarily complicate the base model. However, further research should be conducted to clarify how these two proxies may interact with other determinants of eating behavior. Research on parental influence on children's eating behavior from conception through adolescence provides useful grounding for such follow-up investigations.

Works Cited

Cooke, Lucy J., and Jane Wardle. "Age and Gender Differences in Children's Food Preferences." British Journal of Nutrition, vol. 93, no. 5, 2005, pp. 741–746. DOI: 10.1079/BJN20051389.

Hill, R. C., et al. Principles of Econometrics. 4th ed., Wiley, 2011.

Reicks, Marla, et al. "Influence of Parenting Practices on Eating Behaviors of Early Adolescents during Independent Eating Occasions: Implications for Obesity Prevention." Nutrients, vol. 7, no. 10, 2015, pp. 8783–8801. www.ncbi.nlm.nih.gov/pmc/articles/PMC4632451/.

Savage, Jennifer S., et al. "Parental Influence on Eating Behavior: Conception to Adolescence." The Journal of Law, Medicine & Ethics, vol. 35, no. 1, 2007, pp. 22–34. www.ncbi.nlm.nih.gov/pmc/articles/PMC2531152/.

Appendix A — Descriptive Statistics

Table A1: Descriptive Statistics

Eatingbehavior — Eating behavior scores: Mean = 71.07, Median = 72, SD = 17.47, Min = 41, Max = 100

Livloc — Dummy variable = 1 if the subject lives in a developed area, otherwise 0: Mean = 0.60, Median = 1, SD = 0.49, Min = 0, Max = 1

Foodacc — Dummy variable = 1 if healthy food is easily accessible, otherwise 0: Mean = 0.44, Median = 0, SD = 0.50, Min = 0, Max = 1

Age — Actual age in years: Mean = 6.39, Median = 6, SD = 3.58, Min = 1, Max = 12

Gender — Dummy variable = 1 if male, 0 if female: Mean = 0.53, Median = 1, SD = 0.50, Min = 0, Max = 1

Elecuse — Dummy variable = 1 if electronics are made available during mealtimes, otherwise 0: Mean = 0.46, Median = 0, SD = 0.50, Min = 0, Max = 1

Empstatus — Dummy variable = 1 if the parent is employed full-time, otherwise 0: Mean = 0.65, Median = 1, SD = 0.48, Min = 0, Max = 1

Household — Dummy variable = 1 if the household has both parents, otherwise 0: Mean = 0.54, Median = 1, SD = 0.50, Min = 0, Max = 1

Note: Livloc = living location; Foodacc = access to food; Elecuse = electronic use; Empstatus = employment status.

Appendix B — Correlations

Table B1: Correlations

Livloc / Livloc: 1.0000

Foodacc / Livloc: 0.0887   Foodacc / Foodacc: 1.0000

Age / Livloc: 0.1120   Age / Foodacc: 0.0886   Age / Age: 1.0000

Gender / Livloc: 0.1125   Gender / Foodacc: 0.0336   Gender / Age: −0.1331   Gender / Gender: 1.0000

Elecuse / Livloc: −0.2130   Elecuse / Foodacc: 0.0105   Elecuse / Age: −0.0429   Elecuse / Gender: 0.0050   Elecuse / Elecuse: 1.0000

Empstatus / Livloc: 0.0644   Empstatus / Foodacc: −0.0035   Empstatus / Age: −0.1133   Empstatus / Gender: 0.2165   Empstatus / Elecuse: 0.0263   Empstatus / Empstatus: 1.0000

Household / Livloc: 0.0875   Household / Foodacc: 0.0720   Household / Age: 0.0141   Household / Gender: 0.0360   Household / Elecuse: −0.0960   Household / Empstatus: −0.0694   Household / Household: 1.0000

Note: Livloc = living location; Foodacc = access to food; Elecuse = electronic use; Empstatus = employment status.

Appendix C — Regression Results for Equation 1

Table C1: Regression Results for Equation 1

Multiple R: 0.33653   R²: 0.11325   Adjusted R²: 0.06506   Standard Error: 16.8939   Observations: 98

ANOVA: Regression df = 5, SS = 3353.454, MS = 670.691, F = 2.350, Significance F = 0.04691. Residual df = 92, SS = 26257.05, MS = 285.403.

Coefficients:

Intercept: 75.598, SE = 5.004, t = 15.108, p = 1.18E-26, 95% CI [65.660, 85.536]

Livloc: 2.253, SE = 3.634, t = 0.620, p = 0.5368, 95% CI [−4.965, 9.471]

Foodacc: −7.643, SE = 3.467, t = −2.204, p = 0.0300, 95% CI [−14.529, −0.759]

Age: −0.572, SE = 0.489, t = −1.168, p = 0.2457, 95% CI [−1.543, 0.400]

Gender: −4.268, SE = 3.483, t = −1.225, p = 0.2235, 95% CI [−11.185, 2.649]

Elecuse: 7.375, SE = 3.509, t = 2.102, p = 0.0383, 95% CI [0.406, 14.343]

Appendix D — Regression Results for Equation 2

Table D1: Regression Results for Equation 2

Multiple R: 0.33679   R²: 0.11343   Adjusted R²: 0.05497   Standard Error: 16.9847   Observations: 98

ANOVA: Regression df = 6, Residual df = 91. Significance F = 0.08260.

Coefficients:

Intercept: 75.366, SE = 5.316, t = 14.178, p = 9.13E-25, 95% CI [64.807, 85.925]

Livloc: 2.225, SE = 3.660, t = 0.608, p = 0.5448, 95% CI [−5.045, 9.495]

Foodacc: −7.675, SE = 3.494, t = −2.197, p = 0.0306, 95% CI [−14.615, −0.734]

Age: −0.572, SE = 0.492, t = −1.162, p = 0.2481, 95% CI [−1.549, 0.405]

Gender: −4.283, SE = 3.503, t = −1.223, p = 0.2247, 95% CI [−11.242, 2.676]

Elecuse: 7.414, SE = 3.540, t = 2.095, p = 0.0390, 95% CI [0.383, 14.445]

Household: 0.470, SE = 3.477, t = 0.135, p = 0.8928, 95% CI [−6.436, 7.376]

Appendix E — Regression Results for Equation 3

Table E1: Regression Results for Equation 3

Multiple R: 0.34207   R²: 0.11701   Adjusted R²: 0.05879   Standard Error: 16.9504   Observations: 98

ANOVA: Regression df = 6, Residual df = 91. Significance F = 0.07230.

Coefficients:

Intercept: 76.937, SE = 5.462, t = 14.086, p = 1.38E-24, 95% CI [66.088, 87.787]

Livloc: 2.386, SE = 3.653, t = 0.653, p = 0.5152, 95% CI [−4.869, 9.642]

Foodacc: −7.630, SE = 3.479, t = −2.193, p = 0.0308, 95% CI [−14.540, −0.719]

Age: −0.600, SE = 0.493, t = −1.217, p = 0.2266, 95% CI [−1.580, 0.379]

Gender: −3.833, SE = 3.563, t = −1.076, p = 0.2849, 95% CI [−10.911, 3.245]

Elecuse: 7.449, SE = 3.523, t = 2.115, p = 0.0372, 95% CI [0.452, 14.447]

Empstatus: −2.307, SE = 3.706, t = −0.623, p = 0.5351, 95% CI [−9.669, 5.054]

Appendix F — Regression Results for Equation 4

Table F1: Regression Results for Equation 4

Multiple R: 0.34218   R²: 0.11708   Adjusted R²: 0.04841   Standard Error: 17.0436   Observations: 98

ANOVA: Regression df = 7, SS = 3466.924, MS = 495.275, F = 1.705, Significance F = 0.11776. Residual df = 90, SS = 26143.576, MS = 290.484.

Coefficients:

Intercept: 76.775, SE = 5.813, t = 13.209, p = 8.82E-23, 95% CI [65.228, 88.330]

Livloc: 2.367, SE = 3.680, t = 0.643, p = 0.5218, 95% CI [−4.944, 9.678]

Foodacc: −7.650, SE = 3.506, t = −2.182, p = 0.0317, 95% CI [−14.617, −0.684]

Age: −0.600, SE = 0.496, t = −1.210, p = 0.2293, 95% CI [−1.585, 0.385]

Gender: −3.847, SE = 3.587, t = −1.073, p = 0.2863, 95% CI [−10.974, 3.279]

Elecuse: 7.474, SE = 3.553, t = 2.103, p = 0.0382, 95% CI [0.414, 14.533]

Empstatus: −2.282, SE = 3.739, t = −0.610, p = 0.5432, 95% CI [−9.710, 5.146]

Household: 0.298, SE = 3.500, t = 0.085, p = 0.9324, 95% CI [−6.656, 7.252]