Fig 1.
Directed acyclic graph showing the structure of causal relationships between variables in simulated scenarios A, B and C. Growth represents the growth rate of an individual and is not simulated or measured in the scenario. U is an unknown and unmeasured variable. The age in years at which known variables are measured is shown in subscript. Arrows show the direction of causal relationships and numbers attached to these arrows show the correlations induced by them.
Table 1.
Parameters of latent variables and error terms used to simulate data in section 3.
Table 2.
Summary of simulated variables in scenario A.
Table 3.
Summary of simulated variables in scenario B.
Table 4.
Summary of simulated variables in scenario C.
Fig 2.
Z-score plots of weight from birth to age 2 years for scenarios A, B and C. Dotted lines show the group diagnosed with diabetes at age 40 and dashed those without a diagnosis. Error bars show empirical 95% confidence intervals.
Fig 3.
Fitted weight values from multilevel models (outcome as covariate) and average mean weight values for scenarios A, B and C. Dotted lines (fitted values) and circular points (average mean weight values) represent fitted values for the group with a diabetes diagnosis at age 40. Dashed lines (fitted values) and triangular points (average mean weight values) represent those without a diagnosis. The grey ribbon represents an empirical 95% confidence band around the fitted values.
Table 5.
Average parameter estimates from multilevel models of weight (outcome as covariate).
Table 6.
Average parameter estimates from the logistic regression model of diabetes status on weight growth rate.