Data Analysis

Structural equation modeling (SEM) using a diagonally weighted least squares (DWLS) estimator was applied to test our hypothesized theoretical structures. All types of weight stigma (including experienced weight stigma, perceived weight stigma, and weight-related self-stigma) are predictors of QoL; experienced and perceived weight stigma are predictors of weight-related self-stigma; and experienced weight stigma is a predictor of perceived weight stigma (see Figure 1 for the conceptual model). Moreover, the experienced weight stigma was constructed using all EWS items; the perceived weight stigma using items 7–12 in the WSSQ; the weight-related self-stigma using all WBIS items and items 1–6 in the WSSQ; the QoL using different child-rated and parent-rated QoL instruments.

Proposed models evaluating different types of weight bias on quality of life (QoL) with standardized path coefficients. (A) Model 1: QoL assessed using child-reported generic instrument (Kid-KINDL). (B) Model 2: QoL assessed using parent-reported generic instrument (Kid-KINDL). (C) QoL assessed using child-reported weight-related instrument (Sizing Me Up). (D) QoL assessed using parent-reported weight-related instrument (Sizing Them Up). All models controlled age, gender, and body mass index. CFI, comparative fit index; TLI, Tucker-Lewis index; RMSEA, root mean square error of approximation; SRMR, standardized root mean square residual. *p < 0.05; **p < 0.01; ***p < 0.001.

The hypothesized structure was tested in four models. Specifically, all models used the same weight stigma measures but different QoL measures: Model 1 used the child-rated Kid-KINDL; Model 2 used the parent-rated Kid-KINDL; Model 3 used the SMU; and Model 4 used the STU. Additionally, all models have adjusted for age, gender, and BMI. The following fit indices with suggested cutoff were used to determine whether our hypothesized models are supported: comparative fit index (CFI) and Tucker–Lewis index (TLI) >0.9; root mean square error of approximation (RMSEA) and standardized root mean squared residual (SRMR) <0.08. Moreover, a non-significant χ² indicates a good fit between the data and model; however, given that the χ² index is notorious in its oversensitivity to sample size (i.e., χ² easily will be significant in a large sample size such as the size in the present study) (Wu et al., 2015), the fit between the data and model depends on CFI, TLI, RMSEA, and SRMR more.

Referencing the findings of Guardabassi et al. (2018), several mediation models were conducted to explore the mediated effects in different weight stigma types. More specifically, Hayes' Process macro (Model 4) was then carried out to understand the mediated effects (Hayes, 2018) of different types of weight stigma on the association between body weight and QoL. In the Hayes' Process macro, 5,000 bootstrapping samples were generated to examine whether each type of weight stigma is a significant mediator in the association between body weight and QoL. The lower limit of confidence interval (LLCI) and upper limit of confidence interval (ULCI) at 95% were used to examine whether the mediated effect is significant. Specifically, when LLCI and ULCI do not cover 0, the mediated effect is significant. Moreover, age and gender were controlled in all the mediation models.

Moreover, given our sample's relatively wide age range, Pearson correlation coefficients were used to examine the bivariate associations between studied variables, including age, gender, experienced weight stigma, perceived weight stigma, weight-related self-stigma, and all the QoL.

The SEM was conducted by the lavaan package (http://lavaan.ugent.be/) in the R program. Descriptive correlational analyses and Hayes' mediation models were conducted by the IBM SPSS 20.0 (IBM Corp., Armonk, NY).