Abstract

Piecewise linear growth model(PLGM) is useful to model for non-linear trajectories. In PLGM, entire assessment period is split into multiple phases at points called “knots’, and separate linear growth model is applied to each phase. Because linear growth model is used at each phase, the interpretation of growth factors is more straightforward and theoretically meaningful compared to other methods for modeling non-linear growths. In addition, radical changes at the knot can lead to disjointed trajectories (referred to as ”jump“ in the follwing) at knots, and PLGM can model the jump. However, such advantage of PLGM　is often overlooked in applications of PLGM. In this study, we reviewed parameterizations of PLGM that allow the estimation of the jump in disjointed trajectories, and examined consequences, in terms of estimation bias and model fit, of model misspecification by omitting the jump. For this purpose, we generated datasets with trajectories with various degrees of jumps and analyzed the datasets using the PLGM proposed by Harring et al. (2006), which estimates the location of the knot, instead of setting it at an a priori point. Thus, we were also able to examine the estimation of the knot locations in the presence of the model misspecification. In our results, with increasing degrees of the jump, in general, the bias of parameter estimates increased and the model fit declined. The results showed that, in most situations, it is a good idea to include the jump in the applications of PLGM, unless there is a strong theoretical background to omit the jump. We also provided practical strategies in the applications of PLGM based on our results.