The developmental effects of media-ideal internalization and self-objectification processes on adolescents’ negative body-feelings, dietary restraint, and binge eating

As shown in Fig. 1 each of the constructs was assessed at a different point in time. Specifically, media-ideal internalization, self-objectification, negative body-feelings (i.e., body shame and appearance anxiety), and disordered eating (i.e., dietary restraint, binge eating) were measured at wave 1, 2, 3 and 4, respectively. Each wave was separated by a 1-year interval during which the variables under investigation can develop or change [13, 33, 50‐53]. This spacing of the assessments across four waves and the statistical control of prior (Time _x -1) levels of each endogenous (dependent) variable would ensure temporal precedence of media-ideal internalization to self-objectification, of self-objectification to negative body-feelings, and of negative body-feelings to disordered eating [51]. However, in contrast to the other four endogenous (continuous) model variables (i.e., self-objectification, body shame, appearance anxiety, and dietary restraint), for binge eating we could not statistically control prior (Time _x -1) relevant levels, as this variable was operationalized categorically (see measures for details) at wave 4 (Fig. 1) [51]. In line with prior longitudinal research [33, 50] we assessed therefore binge eating episodes in each wave, and subsequently participants who reported binge eating episodes at the first three waves were excluded from main analyses (see binge eating in measures section). This strategy would both ensure a more rigorous and a truly prospective test of our hypothesis and prevent over-estimation of model parameters [33, 50, 51, 53], as there is increasing evidence that binge eating (if present) tends to be relatively stable or increase during the developmental period that the current study covers, and adolescents who report binge eating relative to those who did not, showed significantly higher levels of body mass, media-ideal internalization, negative affect, depressed mood, restraint, and body image concerns [4, 14, 50, 53, 54].

Age- and sex-adjusted BMI centiles from the Centers for Disease Control and Prevention (CDC) [56] were used to determine whether participants at baseline were underweight (less than 5th percentile), normal weight (5th percentile to less than 85th percentile), overweight (85th percentile to less than 95th percentile) and obese (equal to or greater than the 95th percentile). However, as BMI percentiles are poorly suited for structural equation modeling analyses [51] and not recommended as a (proxy) measure of change in adiposity in longitudinal studies of adolescents see [57], in our planned statistical analyses BMI was used as a continuous variable that was z-standardized with respect to gender and age according to the CDC standards [56]. This permitted us to include the full scale of weight (z-BMI) and reduce potential measurement error [33, 51].

As the present study was conducted as part of the Mind & Body Project [58]; see also acknowledgments available annual data regarding BMI and depression were used to provide an additionally conservative test of our hypotheses, as prior research suggests that both variables co-vary with ED and objectification theory constructs (Fig. 1) and their values differ over time (i.e., time-varying variables) [12, 32‐34, 50, 52].

Although participants who reported binge eating at the month prior to the first three annual assessments (n = 36), relative to those who did not, showed significantly higher mean scores in all study variables, there were no significant differences in terms of demographics between these two groups. Given that some readers might wonder if the results from the model under investigation would have changed if the 36 participants had been included, the structural model (Fig. 1) was re-estimated including these participants. Because this had the effect of amplifying the range of all model variables, we noted the expected increases in structural parameter estimates (relative to the sample without these participants) (β|Δ| = 0.06–0.16, M|Δ| = 0.10, SD |Δ| = 0.03) and in the proportion of total variation of each endogenous variable (3.1–7.8 %) explained by the model. In line with prior research [33, 50] we reported the more conservative analysis without the inclusion of 36 cases to avoid over-estimation (even minor) of structural parameters and ensure that we conducted a truly prospective test of our hypothesis [51, 53]. Due to space considerations, the detailed results of the analyses briefly reported here are available from the corresponding author upon request.

To ensure that measures assessed at multiple time points (i.e., self-objectification via body surveillance, body shame, appearance anxiety, dietary restraint) were not allowed to change over time, the same items in the three parcels for these measures were included at initial (i.e., self-objectification via body surveillance at T1) and later (i.e., self-objectification via body surveillance at T2) time points [51, 73]. Also, to control for possible systematic error due to the repeated assessment, the measurement error amongst the identical observed indicators of the latent variables was allowed to be correlated over time [51, 73]. For instance, the measurement error for the 1st observed indicator of self-objectification via body surveillance from T1 was allowed to correlate with the measurement error for the same 1st observed indicator of self-objectification via body surveillance at T2. This was also done for the 2nd and 3rd observed indicators of self-objectification via body surveillance from T1 and T2. In the same manner, correlated error for the observed indicators of the other longitudinal latent variables (i.e., body shame, appearance anxiety, dietary restraint) were included.

As the current manuscript includes the maximum permitted number of tables and figures, the correlations among the 10 latent variables and the 28 observed indicators and time-varying covariates stratified by gender are available on request from the corresponding author.

Modification indices provided by Mplus were detected in both the measurement and structural model but their magnitude (<5.0) suggested that any not originally specified parameters did not impact the fit of model to the data [73].

In each model path coefficients from the time-varying covariates and constructs included to control for previous levels of the endogenous variables to the latent variables were allowed to vary across groups, as recommended [51, 73].

As all existing effect sizes measures proposed in the mediation context are limited to the simple cross-sectional mediational models [76] see for details we did not quantify the size of the indirect effects either per se or as compared with some specified value [51].

For contradictory findings and potential explanations especially in terms of methodological and statistical shortcomings, see [12, 81].