Description
Introduction. In longitudinal studies it may be needed to model change from wave to wave to properly study dynamic interrelations between the growth of different abilities. It may also be unavoidable to switch to more difficult items when the abilities are growing rapidly, by using items that partly overlap across successive waves. Latent Change Score (LCS) modeling combined with Item Response Theory (IRT) modeling offers a powerful way accommodate these needs. LCS modeling is better suited for studying dynamic relations than either latent growth modeling or cross-lagged panel modeling (McArdle, 2009). Latent growth modeling uses a single parameter to model growth over time, which has the advantage of increased power because information from all waves is combined in this one estimate. However, a drawback is that possibilities to detect time varying interrelations with another grower are rather limited, since the single slope parameter is time-invariant. Cross-lagged panel models are well suited to address wave to wave relations, but less appropriate for detecting overall trends. LCS modeling, integrating both these models, provides a useful technique for assessing time-varying interrelations between growth of two variables, while also providing information about general growth trends (McArdle, 2009). IRT modeling allows for the linkage of response accuracy on a given item to an underlying latent ability of the participant and the difficulty of the item. The larger the positive difference between this ability and the characteristic difficulty level for a particular item, the larger the likelihood of a correct response, and vice versa. An advantage of IRT modeling is therefore that it can deal with differences in item difficulty. As long as there are items administered at any two successive waves -so the same difficulty level can be assumed, and increases in ability level for these items can be estimated- it is possible to reconstruct the underlying growth in ability from first to last wave. Research aims and predictions. Using as example a large data set pertaining to nonword repetition and vocabulary growth during the preschool years we will demonstrate the feasibility of this combined model. We expect better fit as can be obtained with either latent growth modeling or cross-lagged panel modeling. Method. Participants were 3607 children who participated at least once in a longitudinal study (Pre-COOL, cf. Mulder et al., 2014) and were assessed at two, three, four and five years. At each wave, children completed a shortened version of the Dutch Peabody Picture Vocabulary Test (PPVT-III-NL, Dunn & Dunn, 2005) and a nonword repetition test. Mplus 7.4 (Muthén & Muthén, 1998-2016) with the weighted least squares estimator and a PROBIT link was used for analysis. Results. Our bivariate LCS/IRT model showed better fit than a traditional LGM. Surprising reciprocal influences were found that partly cancelled each other out such that they might have gone unnoticed with other methods. Discussion. Latent Change Score modeling opens up new ways to study dynamic interrelations in longitudinal development, can be combined with Item Response Theory, and can reveal hitherto undetected mutual influences in bivariate growers.| Period | 7 Apr 2017 |
|---|---|
| Event title | SRCD Biennal Meeting |
| Event type | Conference |
| Location | Austin, United StatesShow on map |
| Degree of Recognition | International |