Moderation effects of perceived resilience on the relationship between screen time, unstructured socializing, and self-perceived overweight

Abstract

Objective Research has shown that social factors like peer networks and screen time exposure have a critical role in personal perceptions of weight. This study examined the relationships between television, computer/video games exposure, unstructured socializing (UnS), perceived resilience (PR), self-perceived overweight (SPO), and misperceptions of overweight (MO) in early adulthood. Method Data were obtained from Add Health, a public use sample of 2033 men and 2336 women 18–26 years old (M = 21.81 years; SD = 1.80). Binary logistic regression was used for all data analyses. Hesmer-Lemeshow and Wald test statistics were reported to compare binary logit models. Nagelkerke pseudo-R2 was computed for effect sizes. Results The results showed that TV had a positive effect on SPO and MO. Gender-specific patterns were found in SPO and MO that women were more likely to overestimate their weight than men, and men were more likely to have MO compared to women. The results yielded that the effect of PR on SPO is significant at .001 level. The moderator effect of PR was statistically significant only for the relationship between computer/video games exposure and SPO. Conclusion Reducing TV exposure might be a protective factor to prevent weight misperceptions. Further studies are needed to examine the effects of low, moderate, and high levels of leisure-based ST and gender-specific resilience strength programs on weight misperceptions.

Keywords:

• Screen time
• unstructured socializing
• perceived resilience
• perceived weight

1. Introduction

The prevalence rate of overweight status and obesity has become a critical concern for adolescents and young adults (World Health Organization (WHO), 2021). According to the 2017–2018 National Health and Nutrition Examination Survey (NHANES), approximately 19.3% of adolescents aged 2–19 years had obesity, while 6.1% of them had severe obesity, and 16.1% were categorized as overweight in the U.S. (Fryar et al., 2020). On a global scale, obesity has tripled since 1975, and 1.9 billion people who were 18 years or older were overweight in 2016 (World Health Organization (WHO), 2021). Consistent with the increased rate in overweight status (Foti & Lowry, 2010; Hales et al., 2018; Han et al., 2019) and perceptions of being overweight (Wardle et al., 2006), the prevalence rate of misperceptions of body weight has been significantly increased (Dorsey et al., 2009; Edwards et al., 2010; Fan et al., 2014; Johnson et al., 2008; Ver Ploeg et al., 2008). Edwards et al. (2010) reported that the overall proportion for misperceptions of body weight ranged from 29% to 33% in 1999–2007, and 23% of overweight girls and 40% of overweight boys were categorized as misperceivers in the US, according to the Youth Risk Behavior Surveillance System.

Self-perceived weight refers to how a person identifies themselves in terms of weight and size (Robinson, 2017). Weight perceptions might have a more critical influence than actual body mass regarding mental health outcomes (Skrove et al., 2016). However, most overweight individuals have difficulty correctly identifying their weight status (Robinson, 2017). Such misperceptions are linked to physical and mental health problems (McEvoy, 2009), such as unnecessary weight-loss behaviors (Choi & Kim, 2017), depression, and stress about obesity (Cho et al., 2012).

Research has shown that weight perceptions are associated with ethnicity, age, and gender (Ali et al., 2012; Kenny et al., 2017; Robinson, 2017). Boys tend to perceive themselves as underweight; girls tend to perceive themselves as overweight (McCreary & Sadava, 2001). It is also found that weight misperceptions show gender-specific patterns (Binkley et al., 2009; Wardle et al., 2006) and is highly frequent among ethnic minorities (Standley et al., 2009) and individuals with lower education levels (Dorsey et al., 2009).

Social dynamics, including social capital, social norms, and social stress, significantly influence personal perceptions of body weight in adolescence and adulthood (Burke et al., 2010; Christensen & Jæger, 2018; Hammond, 2010). Burke et al. (2010) examined the changes in weight perceptions in a sample of individuals aged 17–74. Data were obtained from individuals in the US using the NHANES in the periods 1988–1994 and 1999–2004. They reported that there was a generational shift in weight perceptions aligned with the social norms that were affected by social factors like popular media and public health campaigns.

1.1. Unstructured socializing

Friends, family, and mass media are influential social agents in developing a sociocultural “ideal” (Dohnt & Tiggemann, 2006; Hutchinson & Rapee, 2007; Vincent & McCabe, 2000). In adolescence, people shift their focus from their parents to their friends. Their friends assume a critical role in how a person develops their body image and perception (Helfert & Warschburger, 2011; McCabe & Ricciardelli, 2003).

Festinger’s social comparison theory (Festinger, 1954) highlighted the similarities in peer groups and how peers influence adolescents’ perceptions. According to this theory, friends and social groups might influence weight perceptions (Christensen & Jæger, 2018), particularly for overweight individuals (Ramirez & Milan, 2016). By observing and comparing their physical attributes with peers, adolescents identify their “ideal” body type against which they compare themselves (Carey et al., 2014; Kenny et al., 2017).

Since Festinger’s work, other researchers have established that young adults and adolescents look to their peer group for image validation. Valente et al. (2009) examined the weight status in social networks in a sample of 617 adolescents in the US and reported that weight status was a shared characteristic among adolescent peer groups. Kenny et al. (2017) found that “good-looking” and popular peers were copied by the adolescents because aligning one’s looks with popular peers helped support self-confidence and self-esteem.

The structure of social network systems can influence overweight status and obesity (Koehly & Loscalzo, 2009). The frequency of informal socializing with peers is referred to as unstructured or unsupervised socializing (UnS) in the literature (Augustyn & McGloin, 2013; Haynie & Osgood, 2005). Research has shown that overweight status is a risk factor for peer surveillance, exclusion, teasing (Kenny et al., 2017), and social isolation (Ali et al., 2012; Jacobs et al., 2020). As peer groups share weight-specific expectations (Valente et al., 2009), adolescents who perceive themselves as overweight might isolate themselves from their peer networks and spend less time in person with their friends compared to their non-overweight counterparts. UnS is an important area of inquiry; however, relatively little is known about UnS role in misperceptions of weight status; therefore, there is no directionality in H_1a and H_3a.

1.2. Screen time

Media is a social agent that has significantly shaped the development of sociocultural ideals (Clark & Tiggemann, 2007) and set generic expectations for what bodies should look like, that adolescents compare themselves against (Bailey et al., 2017). According to Gerbner’s cultivation theory, individuals who spend more time in front of the television are more likely to have critical discrepancies between the real world and the world displayed on the screen(Gerbner et al., 2002). The work of McCreary and Sadava (1999) demonstrated that those who spent a lot of time in front of television (TV) were more likely to see themselves as overweight.

Research has shown that screen time (ST) was associated with body mass index (BMI; Dumith et al., 2012) and internalization of thin-ideal (Dohnt & Tiggemann, 2006; Tiggemann & Slater, 2013). The study carried out by Keel et al. (2020) revealed that perceived weight and changes in ST exposure showed an upward trend in a sample of 90 undergraduate students. The vast majority of the work in weight perceptions has focused on gender differences. The results yielded that increased ST is negatively associated with overestimation of weight for adolescent boys, while positively associated with overestimation of weight for girls (Archbell et al., 2020; Dumith et al., 2012; Lim & Wang, 2013).

1.3. Perceived resilience

Resilience is a multifactorial concept (Olsson et al., 2003) that includes personal, interpersonal, and community dimensions (Resnick, 2000) associated with positive outcomes despite adversities (Luthar et al., 2000; Masten, 2001). Current resilience theories emphasize the promotive role of prosocial involvement (Zimmerman et al., 2013), prosocial attitudes (Dyer & McGuinness, 1996), and activities (Eccles et al., 2003). Despite its critical role during adolescence and emerging adulthood, there is limited research investigating the role of perceived resilience (PR) on weight perceptions. Borinsky et al. (2019) carried out an extensive study on the association between resilience and perceived overweight/obesity in a sample of 85 adolescents. Despite a nonsignificant relationship between perceived weight and low resilience, they pointed out that resilience training might be an effective way for weight misperceptions.

1.4. Purpose of the present study

Understanding the effects of social dynamics of factors associated with being overweight requires continuous effort (Hammond, 2010). Despite the critical influence of social relationships on the development of overweight status and obesity, social context of overweight status has not been frequently studied (Koehly & Loscalzo, 2009). The current study seeks to answer the following research questions:

RQ1: Do television (TV), computer /video games (CV) exposure, and unstructured socializing (UnS) predict self-perceived overweight (SPO)?

RQ 2: Does PR effectively moderate the relationship(s) between SPO, UnS, TV, and CV?

RQ 3: Do TV, CV and UnS predict misperceptions of overweight (MO)?

1.5. Hypotheses

H_1a: UnS will be significantly related to SPO.

H_1b: TV exposure will be positively related to SPO.

H_1c: CV exposure will be positively related to SPO.

H_1d: PR will be negatively related to SPO.

H_2a: The relationship between UnS and SPO will be effectively moderated by PR.

H_2b: The relationship between TV, CV exposure, and SPO will be effectively moderated by PR.

H_3a: UnS will be significantly related to MO.

H_3b: TV exposure will be positively related to MO.

H_3c: CV exposure will be positively related to MO.

2. Method

2.1. Sample

Data for the present study were drawn from the Wave III public-domain National Longitudinal Study of Adolescent Health (Add Health). Add Health Wave III was collected between 2001–2002 in a sample of young adults ages between 18-26 years (“Add Health Research Design,” n.d.). Missing values and other response options, including “not applicable,” “refused,” and “don’t know,” were less than 5% for all variables in the dataset, so they were excluded from the data analysis. The final sample consists of 4369 young adults, with females representing 53.47% of those and males representing 46.53%. The descriptive statistics are shown in Table 1. This research protocol has been approved by the Institutional Review Board (IRB) at the University of North Dakota (IRB Protocol No: IRB0004408).

Table 1. Frequencies, means, and standard deviations of dependent, independent, and control variables

Variable	M (SD)	n (%)
Self-perceived overweight (SPO)
Not overweight		2641 (60.45%)
Overweight		1728 (39.55%)
Misperceptions of overweight (MO)
Not overweight		2541 (58.16%)
Overweight		1828 (41.84%)
Screen time (ST)
Television (TV)	12.73 (13.11)
Computer/ video games (CV)	4.74 (9.16)
Unstructured Socializing (UnS)	4.45 (2.35)
Perceived Resilience (PR)	0.83 (0.13)
Age	21.81 (1.80)
Gender
Male		2033 (46.53%)
Female		2336 (53.47%)
Education (In formal Years)	13.23 (1.98)
Ethnic Group
White		3101 (70.98%)
Black or AfricanAmerican		1045 (23.92%)
American Indian orNative American		198 (4.53%)
Asian or PacificIslander		211 (4.83%)
Multiracial		163 (3.73%)
Body Mass Index (BMI)
Underweight		125 (2.86%)
Normal weight		1993 (45.62%)
Overweight		1199 (27.44%)
Obesity		1052 (24.08%)

Note. N = 4369

2.2. Measures

2.2.1. Self-Perceived overweight

Self-perceived overweight (SPO) was the dependent variable of interest of this study. It has been frequently measured with single-item instruments and has acceptable concurrent and predictive validity (Atlantis & Ball, 2008; Christensen & Jæger, 2018; Gillison et al., 2006; Keel et al., 2020; McCreary & Sadava, 2001). Respondents were asked: “How do you think of yourself in terms of weight?” The response options included five categories very underweight (1), slightly underweight (2), about the right weight (3), slightly overweight (4), and very overweight (5) (Add Health Codebook Explorer (ACE), n.d.). The scope of the study is overweight perceptions; therefore, SPO was coded as a binary dependent variable, which equals 1 if an individual reported overweight perception, and otherwise equals 0.

2.2.2. Unstructured socializing

Unstructured socializing (UnS) is defined as the weekly frequency of informal meetings with friends. It was frequently measured with single items in the literature (Augustyn & McGloin, 2013; Li & Kanazawa, 2016). Respondents were asked, “During the past week, how many times did you just hang out with friends?” The response values ranged from not at all (0) to seven or more times (7) (M_uns = 4.44, SD_uns = 2.35).

2.2.3. Screen time

Screen time (ST) consists of two separate variables, including TV and computer/ video games (CV) exposure. ST variables were calculated from the following survey questions, “How many hours a week do you watch television?” (TV: M_TV = 12.73, SD_TV = 13.11), and “On the average, how many hours a week do you spend playing video or computer games or using a computer for something other than school work?” (M_CV = 4.74, SD_CV = 9.16).

2.2.4. Perceived resilience

Perceived resilience (PR) was used as a moderator variable of the relationship between explanatory variables (UnS, TV, and CV) and SPO. Ali et al. (2010) identified 31 Add Health items related to resilience and examined the factor structure in three dimensions: self-resilience, self/family resilience, and overall resilience. Nguyen (2012) developed the individual resilience scale with 35 Add Health items and reported four dimensions: optimistic perspectives, emotional coping, problem-focused coping, and perceived social support (α = .88). Ozturk and Mohler (2021) developed the PR scale using 11 Add Health items (α = .79). In this study, we used the PR scale developed by Ozturk and Mohler (2021), as the current study is limited to Add Health Wave III. The sample PR scale questions are: “You felt that you could not shake off the blues, even with help from your family and your friends.” and “You feel you are doing everything just about right.” As the PR scale items include 4-point and 5-point Likert scale items, transforming was used to have a common scale. We used the same procedures with Ozturk and Mohler (2021): Four-point items were coded from 1 to 0, 2 to 0.33, 3 to 0.67, and 4 to 1; and five-point items were coded from 1 to 0, 2 to 0.25, 3 to 0.5, 4 to 0.75, and 5 to 1 (min = 0, max = 1) (M_PR = 0.83, SD_PR = 0.13).

2.2.5. Misperceptions of overweight

Misperceptions of overweight (MO) were measured by calculating the discrepancy between perceived weight (PW) and standard body mass index (BMI) categories. BMI was computed by “dividing weight in pounds (lbs) by height in inches (in) squared and multiplying by a conversion factor of .703” (Central for Disease Control and Prevention [CDC], 2014). BMI includes four categories underweight (BMI ≤ 18.5), normal or healthy weight (18.5 ≤ BMI ≤ 24.99), overweight (25 ≤ BMI ≤ 29.99), and obesity (BMI ≥ 30 or greater; Central for Disease Control and Prevention [CDC], 2020). PW categories were coded into four by combining “very underweight” and “slightly underweight” categories. The binary dependent variable, MO, equals “1” if there is a discrepancy between PW and BMI; otherwise equals “0”.

2.2.6. Control variables

Age (in years), ethnic group, gender (0 = male, 1 = female), and education (in formal years) were used as control variables. The dummy variables were created for ethnic group variables. White ethnic group was used as a reference category.

2.3. Statistical analyses

Binary logit regression models were used to explore the effect of explanatory variables (UnS, TV, and CV) on SPO and MO. Figure 1 shows the conceptual model of moderating role of PR in the relationships between UnS, CV, TV and SPO. We first fit the main effects model and then fit the full model for SPO. The main effects model (Model 1) tested the relationship between UnS, TV, CV, and SPO by controlling age, gender, ethnic group, and education. The full model (Model 2) tested the moderator effect of PR on the relationship between UnS, TV, CV, and SPO. We fit the null and main effects model for MO. Wald test statistics were used to test the hypotheses for each regression coefficient and to measure the overall model fit. Likelihood-ratio (LR) tests were reported to compare binary logistic regression models. The Hosmer-Lemeshow (HL) test statistics and Nagelkerke pseudo-R² were reported to compare the quality of the fit of the models and effect sizes. The Brant test was implemented to determine whether parallel regression assumptions hold and to calculate variance inflation factors (VIF) to examine multicollinearity. All predictors (UnS, TV, and CV) and control variables were mean-centered before computing the interactions with PR. Mean centering is required when examining the interaction effects (Robinson & Schumacker, 2009), which takes away the non-essential multicollinearity (Aiken & West, 1991). All regression analyses were carried out using R statistical software.

Figure 1. Conceptual Model of Moderated Relationship between UnS, TV, CV, and SPO. In this figure, UnS represents unstructured socializing, TV represents television exposure, CV represents computer /video games exposure, PR represents perceived resilience, and SPO represents self-perceived overweight.

3. Results

Binary logit regression analysis was performed to investigate whether PR might moderate the effects of UnS, TV, and CV on SPO. The correlations between independent and control variables are presented in Table 2. A two-stage model was run in which main effects and control variables were used in Model 1. The moderator variable, PR, and the interactions were entered as predictors into the full model (Model 2).

Table 2. Correlations with confidence intervals of independent, and control variables

Variable	1	2	3	4	5	6
1. Gender
2. Age	−.06**
	[−.09, −.03]
3. Education	.09**	.20**
	[.06, .12]	[.17, .23]
4. TV	−.04*	.01	−.14**
	[−.07, −.01]	[−.02, .04]	[−.17, −.11]
5. CV	−.18**	−.02	.01	.16**
	[−.21, −.15]	[−.05, .01]	[−.02, .04]	[.13, .19]
6. UnS	.04**	−.15**	.09**	.03*	.07**
	[.01, .07]	[−.18, −.12]	[.06, .12]	[.00, .06]	[.04, .10]
7. PR	−.10**	.02	.14**	−.06**	−.02	.03*
	[−.13, −.07]	[−.01, .05]	[.11, .17]	[−.09, −.03]	[−.05, .01]	[.00, .06]

Note. TV = Television exposure; CV = Computer/ video games exposure, UnS = Unstructured socializing, PR = Perceived resilience. Values in square brackets indicate the 95% confidence interval for each correlation. * indicates p < .05. ** indicates p < .01.

Both binary logit models were statistically significant at .001 level (LR test: Model 1: χ²(10) = 213.5, p < .001; Model 2: χ²(4) = 37.71, p < .001). The Hosmer-Lemeshow (HL) test statistics yielded a χ² (8) of 13.956 (p = .083) for Model 1, and a χ² (8) of 9.703 (p = .286) and were nonsignificant for Model 2, indicating that our models fit the data well. Wald test statistics were reported in Table 3 (Model 1: χ²(10) = 202.2, p < .001; Model 2: χ²(10) = 37.3, p < .001). It shows that including PR, and moderation effects results in an improvement in the fit of the model. The Nagelkerke pseudo-R² showed approximately %7 and 8% of the variance in SPO were accounted by the predictors overall, respectively, in Model 1 and Model 2. The larger pseudo-R² values provide evidence for the best model (Long & Freese, 2006), suggesting Model 2 fits better than Model 1.

Table 3. Results of binary logistic regression models explaining self-perceived overweight

Predictor	Model 1						Model 2
	β	SE β	W	p	OR($e^{\beta })$	95 % CI for OR	β	SE β	W	p	OR($e^{\beta })$	95% CI for OR
Constant	−0.886	0.054	271.3	<.001	0.41	.37, .46	−0.868	0.054	256.8	<.001	0.42	.38, .47
TV	0.011	0.003	17.8	<.001	1.01	1.01, 1.02	0.010	0.003	15.1	<.001	1.01	1.00, 1.02
CV	0.007	0.004	4.1	<.05	1.01	1.00, 1.01	0.009	0.004	5.3	<.05	1.01	1.00, 1.02
UnS	−0.018	0.014	1.6	.203	0.98	.96, 1.01	−0.018	0.014	1.7	.184	0.98	.96, 1.01
Age	0.075	0.018	16.9	<.001	1.08	1.04, 1.12	0.075	0.018	16.4	<.001	1.08	1.04, 1.12
Gender (1 = f)	0.877	0.067	174.3	<.001	2.40	2.11, 2.74	0.845	0.067	158	<.001	2.33	2.04, 2.66
Education	−0.021	0.017	1.5	.216	0.98	.95, 1.01	−0.007	0.017	0.15	.673	0.99	.96, 1.03
Ethnicity (Ref = White)
Black	−0.136	0.077	3.2	.076	0.87	.75, 1.01	−0.128	0.077	2.7	.096	0.88	.76, 1.02
American Indian	0.295	0.183	2.6	.106	1.34	.94, 1.92	0.278	0.184	2.3	.139	1.32	.91, 1.89
Asian/Pacific Islander	0.009	0.154	.004	.953	1.01	.74, 1.36	−0.04	0.155	.066	.772	0.96	.71, 1.30
Multiracial	−0.273	0.208	1.7	.189	0.76	.50, 1.14	−0.274	0.209	1.7	.200	0.76	.50, 1.14
PR							−1.417	0.258	30.1	<.001	0.24	.15, .40
UnS*PR							−0.205	0.106	3.7	.055	0.81	.66, 1.00
TV*PR							−0.012	0.018	0.49	.937	099	.95, 1.02
CV*PR							0.05	0.022	5.1	<.05	1.05	1.01, 1.10
Overall Model Evaluation	χ^2	df	p				χ^2	df	p
Likelihood ratio test	213.5	10	<.001				37.71	4	<.001
Wald test	202.2	10	<.001				37.3	4	<.001
GOF test
Hosmer-Lemeshow (H-L)	13.956	8	.083				9.703	8	.286
Nagelkerke R^2	0.065						0.076

Note. N = 4369. TV = Television exposure, CV = Computer /video games exposure, UnS = Unstructured socializing; PR = Perceived resilience, BMI = Body Mass Index, GOF = Goodness of fit test.

As shown in Model 1, the log of the odds of having SPO was positively related to TV (b = .011, SE = .003, p < .001, OR = 1.01, 95%CI [1.01, 1.02]) and CV exposure (b = .007, SE = .004, p < .05, OR = 1.01, 95%CI [1.00, 1.01]). According to Model 2, the predictor variables, TV and CV exposure, were positively related to the log of the odds of having SPO (TV: b = .010, SE = .003, p < .001, 95%CI [1.00, 1.02]; CV: b = .009, SE = .005, p < .05, 95%CI [1.00, 1.02]). For a one-unit increase TV viewing, and CV exposure, the odds of having SPO increase by a factor of 1.01, holding all other variables constant. As hypothesized in H_1b and H_1c, the higher hours of TV viewing, computer use and playing video games, the more likely it is that an individual would have overweight perceptions.

Both binary logit models showed that the effect of UnS on having SPO was not statistically significant (Model 1: b = −.018, SE = .014, p = .203; Model 2: b = −.018, SE = .014, p = .184). It indicates that there is not enough evidence to reject the null hypothesis (H_1a).

As hypothesized H_1d, the log of the odds of having SPO was negatively related to PR scores (b = −1.417, SE = .258, p < .001, 95%CI [.15, .40]). For a one-unit increase in PR scores, the odds of having SPO decrease by a factor of 0.28, holding all the variables constant. The results showed a significant interaction between CV exposure and PR in predicting SPO, b = .05, SE = .022, p < .05, 95%CI [1.01, 1.10]). Simple effects coefficients were examined for 1 SD below the mean, at the mean, and 1 SD above the mean of PR to explore the form of the interaction. The results showed that at a low level of PR, a one-unit increase in CV exposure is associated with a 1.002 unit increase in SPO (95% CI [−0.006, 0.010]); at the mean level of PR, a one-unit increase in CV is associated with 1.008 unit increase in self-perceived overweight (95% CI [.0013, 0.016]); and at high PR, a one-unit increase in CV is associated with 1.015 unit increase in self-perceived overweight (95% CI [0.0047, 0.025]). These results indicate that CV has a different relationship to SPO depending on the level of PR.

It can be inferred from Table 3 that there was an important gender differences in SPO, with females approximately two and a half times more likely to perceive themselves as overweight in comparison with males, holding all other predictors constant (Model 1: b = .877, SE = .067, p < .001, OR = 2.40, 95%CI [2.11, 2.74]; Model 2: b = .845, SE = .067, p < .001, OR = 2.33, 95%CI [2.04, 2.66]). The results yielded that the log of the odds of having SPO was positively related to age; for each additional year, the odds of having SPO increased by a factor of 1.08, holding all other predictors constant (Model 2: p < .001, 95%CI [1.04, 1,12]). The results indicated that ethnicity does not have a significant effect on SPO (see, Table 3).

4. Binary logistic regression results for misperceptions of overweight

To identify factors that predict MO, the null and main effects binary logit models were fitted. The main effects model was statistically significant, likelihood ratio χ²(6) = 391.94, p < .001, Wald test ratio χ²(6) = 366.4, p < .001. The HL goodness of fit test (χ² (8) of 12.289, p = .139), indicating the model fits the data well. The Nagelkerke pseudo-R² was 0.115, suggesting 11.5% of the variance in overweight misperceptions was accounted by the predictors.

As hypothesized by H_3b, the odds of having MO was positively related to TV exposure, b = .006, SE = .003, p < .05, suggesting for each unit increase in TV hours, the odds of having MO increase by a factor of 1.01, holding all other variables constant. The effect of CV exposure on MO was not statistically significant (CV: b = .002, SE = .004, p = .53). This result suggests that there is not enough evidence to reject the null hypothesis H_3c. A possible reason for this finding might be related to the low exposure to self-reported computer and video games. The effect of UnS on MO was not statistically significant (H_3a: p = 0.93, see, Table 4).

Table 4. Results of binary logistic regression model explaining misperceptions of overweight

Predictor	β	SE β	Wald’s χ^2	p	OR($e^{\beta })$	95 % forOR
						Lower	Upper
Constant	0.235	0.46	26.7	< .001	1.27	1.16	1.38
TV	0.006	0.003	6.4	<.05	1.01	1.00	1.01
CV	0.002	0.004	0.4	.53	1.00	1.00	1.01
UnS	−0.024	0.014	2.8	.093	0.98	.95	1.00
Age (in years)	0.059	0.019	10.1	< .01	1.06	1.02	1.10
Gender (1 = female)	−1.11	0.065	288.1	< .001	0.33	.29	.37
Education (in formal years)	−0.087	0.017	26.5	< .001	0.92	.89	.95
Overall model Evaluation	χ^2	df	p
Likelihood ratio test	391.94	6	< .001
Wald test	366.4	6	< .001
Goodness-of-fit test
Hosmer-Lemeshow	12.289	8	.139
Nagelkerke R^2	0.115

Note. N = 4369. TV = Television exposure; CV = Computer /video games exposure, UnS = Unstructured socializing; PR = Perceived resilience.

The results indicated that there was a significant gender differences in MO, b = −1.11, SE = .065, p < .001, OR = .33, 95%CI [.29, .37]), suggesting that females more than three times less likely to report MO in comparison with males after controlling other predictors. The log of the odds of reporting MO was positively related to age, b = 0.059, SE = .019, p < .001, OR = 1.06, 95%CI [1.02, 1.10]), and negatively related to education, b = −0.087, SE = .017, p < .001, OR = .92, 95%CI [.89, .95].

5. Discussion

This study used binary logit models to test whether UnS, TV, and CV exposure are associated with SPO (RQ1), PR might moderate the effects of UnS, TV, and CV exposure on SPO (RQ2), and examined relations between UnS, TV, CV exposure and MO (RQ3). Our data provide evidence for the cultivation theory (Gerbner et al., 2002) that the higher the TV viewing hours, the more likely it is that an individual would have MO (p < .05), and SPO (p < .001). Contrary to our expectations, the effect of CV exposure on MO is not statistically significant (p = .53). However, these findings are less surprising if we consider the positive effect of moderate levels of internet use and video games on mental health outcomes (Suchert et al., 2015, 2016) and low exposure to computer and video games in our dataset. While it is generally agreed that the higher ST is associated with negative outcomes, there is less consensus over whether or not the moderate and low levels of ST exposure have a negative impact (Dubicka et al., 2019; Falbe et al., 2013). Future research will have to investigate to what extent ST including educational (e.g., homework on electronic devices), interactive ST (e.g., video games), passive ST (e.g., TV) (Sanders et al., 2019), leisure ST (Graff et al., 2013) affect weight concerns.

Furthermore, the correlations between education and TV viewing time were negative and statistically significant. This finding is congruent with Snoek et al. (2006) findings that adolescents with higher education levels showed lower TV viewing than those with lower education levels. However, Tsuji et al. (2018) discussion of the association between TV viewing and educational attainment openly acknowledges that environmental and socioeconomic factors (e.g., income) need to be considered.

Our findings are congruent with previous results (Skrove et al., 2016), showing a significant relationship between perceived body mass and resilience. The effect of PR on overweight perceptions was statistically significant (p < . 005), indicating that the higher the PR score, the less likely it is that an individual would have overweight perceptions. The moderation effect of PR on the relationship between CV exposure and SPO was found to be significant at .05 level. This result suggests that we have weak evidence that CV exposure has a different relationship to overweight perceptions depending on the level of PR. Overall, the significant interaction suggests that exposure to computer use and playing video games have a slightly different relationship to perceived overweight depending on the level of resilience, but the magnitude of increased risk associated with computer and video games exposure is minimal.

Existing research provided evidence that obese adolescents showed lower socializing and had fewer friends than their non-obese counterparts (Ali et al., 2012) to avoid being teased about their appearance and negative social comparison (Barker & Galambos, 2003). Our results revealed that the effect of UnS on SPO and MO was not statistically significant (p = .98). These findings might point out that the structure and quality of the relationship with friends might play a more important role than the frequency of meeting with friends.

Previous research highlighted that weight perception varies across gender (McCreary & Sadava, 2001; Ramirez & Milan, 2016; Wardle et al., 2006), ethnic groups, education level (Dorsey et al., 2009; Standley et al., 2009), and socioeconomic status (Lim & Wang, 2013). As in previous studies (Edwards et al., 2010; McCreary & Sadava, 2001), the results of this analysis confirm that females were more likely to have overweight perceptions than males (OR = 2.33, p < .001) and males more likely to have misperceptions compared to females (OR = .33, p < .001). A possible interpretation of this finding is that males and females have gender-specific body ideals (Gattario & Frisén, 2019). The results yielded a significant positive relationship between age, SPO (p < .001), MO (p < .01), and a negative correlation between education and MO (p < .001). The results demonstrated that ethnicity has not a significant effect on SPO (see, Table 3).

There are several limitations in our study that needs to be acknowledged. First, TV, computer, and video games exposure and UnS were measured with self-reported instruments. Second, the Add Health dataset for Wave III spans the period from 2001 to 2002. Although it is limited to higher TV exposure compared to computer and video games exposure, the general picture emerging from the recent data is that watching TV is the most common leisure-time activity in the United States (United States Bureau of Labor Statistics (2020, 2021). According to the USBLS 2021 data, spending time in front of the TV was increased 3.1 hours per day in 2020, and it increased by 19 minutes per day for people aged 15 years and older. Third, research has shown that the specific type of media content like computer homework (Añez et al., 2018) and image-focused media (Bair et al., 2012) have a more substantial impact than exposure (Calado et al., 2011). Fourth, relatively, little is understood about the structure of virtual socializing on weigh-related concerns. Therefore, future studies will have to investigate to what extent ST and virtual socializing affect weight concerns. In this context, we would encourage researchers to examine generational differences in socializing behaviors to provide more generalizable findings beyond the study sample. Finally, ST exposure and UnS are time-variant predictors. Future studies could examine how changes between ST and socializing behaviors influence weight perceptions.

5.1. Conclusion

Our findings provide strong evidence for the cultivation theory (Gerbner et al., 2002), which explains the critical effect of TV exposure on body image distortion. Overall, our study provides support for the validity of reducing TV exposure might be a protective factor for poor weight perceptions. The results yielded that the effect of PR on SPO was found to be highly significant (p < .001). Our findings are consistent with previous studies showing there were gender-specific patterns in misperceptions of weight status (Blashill & Wilhelm, 2014; McCreary & Sadava, 1999, 2001). Further studies will have to further our understanding of the levels of ST exposure, and gender-specific resilience strength intervention programs on weight concerns in longitudinal designs to explore causal relations.

Acknowledgements

We thank Dr. Jeffrey Haring, Yi Feng, and Nancy Hathaway for their assistance and thoughtful comments.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data Availability Statement

The data that support the findings of this study are available from the three resources: The Odum Institute at UNC, the Inter-university Consortium for Political and Social Research (ICPSR), and the Association of Religion Data Archives (ARDA) https://addhealth.cpc.unc.edu/data/ The data was derived from the following resource available in the public domain: https://www.icpsr.umich.edu/web/pages/

Additional information

Funding

The authors received no direct funding for this research.

Notes on contributors

Emine Ozturk

Emine Ozturk earned her doctorate at Purdue University and is a postdoctoral research associate in the Department of Teaching & Leadership and Professional Practice and College of Engineering & Mines at the University of North Dakota. Her research interests include gender dynamics, resilience, underrepresented groups in STEM, quantitative, and mixed research methods.

James L. Mohler is a professor of Computer Graphics Technology (CGT). Dr. Mohler is a Faculty Scholar, a member of the Purdue University Teaching Academy, and a past faculty fellow for the Discovery Learning Center. Dr. Mohler is a member of Purdue’s ADVANCE team and has served as a Diversity Catalyst. Dr. Mohler has authored, co-authored, or contributed to over 21 texts related to computer graphics and media development and over 71 articles for refereed, reviewed, or trade publications.