Applied Statistics for Educational Researchers (DUH165)

This PhD course will introduce educational researchers to SEM and LGC and enable the successful candidate to apply those analyses in her own research using the software package Mplus and/or R.

Course leader: Ulrich Dettweiler (UiS)

Instructors: Ulrich Dettweiler (UiS), Knud Knudsen (UiS), Thormod Idsøe (UiO og UiS), Lars-Erik Malmberg (Oxford University)

By completion of this course, the PhD candidate will have gained the following:

Knowledge

• • of measurement theory
  • a good understanding of multiple regression and factor analysis
  • a good understanding of hierarchical structures in data, and how to address them in analysis
  • a good understanding of SEM and LGC in complex survey data

Skills:

• • running SEM and LGC analyses in MPlus and R (optional)
  • preparing results of such analyses for publication

General competences:

• • being able to choose and apply the right analyses for the given data
  • developing advanced strategies for further research

The students are expected to

• have the software package Mplus ready installed on their personal computers (demo-version only if the candidate is experienced with R)
• know the data structure of their projects and the research questions. and get themselves acquainted with MPlus and/or R prior to the course so that they master to prepare the data for import in Mplus,
• know some basic Mplus syntax (chapters 1 + 2 in Muthén & Muthén),
• have some datasets prepared in the right format (.dat) for MPlus or .csv in R.

Fagperson(er)

Course teacher:

Course teacher:

Course coordinator:

Course teacher:

Course teacher:

Study Program Director:

Course teacher:

Seminar: In the seminar, we will introduce CFA and SEM and see how Latent Growth Curve modelling can be understood as an extension of SEM with intercepts and/or slopes being modelled as latent variables, first as an unconditional latent curve model. We will then look at conditional Latent Growth Curve models (including mediation models, cross-lagged models, hierarchical/multilevel models), and comparison of (latent) groups with different approaches of testing measurement invariance, also in the Bayesian SEM framework. This will be extended to Dynamic Structural Equation Models with autoregressive slopes (DSEM), i.e. accounting for the influence of the respective previous time points on the outcome variable (time-lagging). Those models are important to analyze intensive longitudinal data, where many observations are nested in individuals. In contrast to latent growth modelling, borrowing logic from time-series analyses, DSEM is interested in the dynamics over time, in terms of autoregressive associations between the same variable at Times T (concurrent time-point) and T-1 (the previous time-point), and cross-lagged associations between variables between T and T-1. Such models are possible to implement using Maximum Likelihood with user specified lagged variables. More complex models (e.g., multiple random slopes) are possible to implement using the Bayesian estimator in Mplus.

The working format is a blending of lectures, group discussions, and hand-on analyses in Mplus/R.

Tutors: Lars-Erik Tutors: Knud Knudsen (UiS), Thormod Idsøe (NUBU), Ulrich Dettweiler (UiS), Malmberg (Oxford University)