Ideal Weight and Weight Discrepancy: A Study of Life Course Trajectories and Intercohort Change in the Netherlands

Objective: This study examined how individuals’ ideal weight and weight discrepancy (between ideal and actual weight) changed over the life course and across cohorts. Methods: The study used population-representative longitudinal data collected in the Netherlands (N = 61,431 observations between 2007 and 2018 among N = 13,409 individuals aged 16 to 80 and born 1927–2000). Results: Ideal weight increased linearly with age. Weight discrepancy showed a bell-shaped age pattern. Approximately half of the age-related increase in ideal weight was associated with concurrent increases in actual weight. Ideal weight and weight discrepancy increased slightly across cohorts. The cohort-related increase in ideal weight vanished after adjusting for change in actual weight. Analyses of population heterogeneity showed similar patterns of change in both outcomes across groups, although levels differed by gender, education, and migration status even after adjusting for differences in actual weight between these groups. Conclusion: These results show that ideal weight and weight discrepancy in the Netherlands change substantially with age and modestly across cohorts. Potential explanations include changes in physical appearance and in the importance of physical appearance.


Supplement S1: Statistical modelling
The HLM estimation allows within-person age trajectories to differ in their starting levels (random intercepts) and rates of change (random slopes).The estimation of HLM provides information about mean ideal weight and satisfaction with weight trajectories (growth curves) as well as individual variation around the average curves.The models allow data to be unbalanced in time and incorporate all respondents, regardless of the number of waves in which they were observed.
The growth curves for each of the outcomes Y (ideal weight and weight discrepancy) of respondent i at time t are calculated as follows (see 13, 26): Level 1: (1) where i = 1, . . ., N are individuals in the sample,  #! is an individual-specific intercept, and  $! is the growth rate for individual i.This model estimates the intercepts ( #! ) and slopes ( $! ) of Level-1 variables.
Level 2: (2) where  &' are the effect of cohort on intercept  %! and slope  $! , and  &! is an error term for unmeasured time-constant characteristics of individual i.
Combining (1) and (2) yields: (3) All analyses were performed separately for men and women.The parametrizations of age and cohort effects on each of the outcomes were based on three criteria, (a) similarity between observed and fitted data examined by diagnostic plots, (b) BIC comparison between models, and additionally (c) model parsimony if models were similar on criterion (a) and did not differ by more than 10 BIC points (28).This resulted in different functional forms of age and cohort and interactions between these two variables for different outcomes as well as slightly different functional forms for men and women.Inclusion of multiple polynomials of age and cohort terms as well as interaction terms between them complicates a straightforward interpretation of regression model coefficients.Therefore, studies on life course trajectories and intercohort change in physical and mental health outcomes (13,29,30) typically visualize their results as age-vector graphs, which allow for a detailed inspection of life course and intercohort patterns as well as of heterogeneity in these patterns across study populations.In line with previous research, I visualize all main results in Figures 1, 2 and 3.
In addition to these main analyses, three sets of additional models have been performed.First, the role of changes in individual BMI for ideal weight was assessed in models controlling for individual BMI (M1c and M1d, Table 2).Second, the role of BMI in reference groups for ideal weight was assessed in M1e and M1f (Table A4 in Appendix), which controlled for both individual and reference group BMI.Finally, demographic indicators for education, immigrant status and civil were included into the models M1g and M1h (Table A4 in Appendix) for ideal weight and into the models for weight discrepancy M2c, M2d, M3c and M3d, as previous research has suggested that weigh perceptions may vary between demographic groups independently of differences between these groups in BMI (20).The results of these additional analyses are presented in Tables 2 and 3 (main text), A4 and A5 (Appendix), described in the results section and are visualized in Figures 1 and 2 (main text) and Figures A1 and A2 (Appendix).

Supplement 2: Control variables
Education was measured as highest level of education attended, distinguishing between (1) primary and intermediate secondary education, (2) higher secondary and intermediate vocational education, (3) higher vocational education, and (4) university education and higher.Migration status indicator distinguished between those with migration background -including (1) 1 st generation migrants (not born in the Netherlands) and (2) 2 nd generation migrants (being born in the Netherlands but having either mother or father who were not born in the Netherlands), and (3) those without migration background (born in the Netherlands and both parents born in the Netherlands).The indicator of civil status distinguished between four categories: (1) married, (2) separated or divorced, (3) widowed, (4) never been married.A4.In the plots visualizing M1c-M1f indiv.BMI and reference group BMI were fixed at the sex specific means.In the plots visualizing M1g and M1h, education was fixed at "University"; immigrant status was fixed at "non-migrant"; civil status was fixed at "Married".
BMI adj.for BMI and demographics, Men

Table A1 .
Age and Cohort Overlaps in the Analytic Sample Data from 12 waves of the Longitudinal Internet Studies for the Social Sciences (LISS) collected between 2007 and 2018.Number of observations is presented in the cells.

Table A4 :
Hierarchical Linear Regression Models for Change in Ideal BMI for men and women

Table A5 Hierarchical
Linear Regression Models for Change in Weight Discrepancy for Men and Women with Controls

Table A6 :
HLPM Models for Change in Categories of Weight Discrepancy for Men and WomenFIGURE A1.LIFE COURSE AND COHORT PROFILES OF IDEAL WEIGHT Note.Data from 12 waves of the Longitudinal Internet Studies for the Social Sciences (LISS) collected between 2007 and 2018.Curves are based on models (M1a-M1d) shown in Table2 and Table