Skip to content
PaperFren

Research method

Structural Equation Modelling (SEM)

3 studies

Ask PaperFrenWatch this

Structural equation modelling estimates a network of directed paths among latent or observed variables, often after a factor analysis has defined those latents. Fit statistics say whether the covariance matrix is compatible with the diagram; path coefficients say how strongly one node predicts another in that sample. A significant path is not a randomised manipulation, and a mediation path that 'explains' attendance is still a decomposition of correlations in one course.

Education researchers reach for SEM when they have several questionnaire or trace variables and a theory about what mediates what — attendance via engagement, technology-integration quality via engagement, information-evaluation literacy via search literacy. It answers 'is this path diagram consistent with the covariances?' Its main limitation is design: all three papers here are observational (one longitudinal-in-a-course, two cross-sectional surveys), so they cannot prove that requiring attendance or changing a teaching practice would move the outcome.

Studies

3

Findings

3

3 supporting · 0 challenging · 0 qualifying citations

Open tensions

1

Currently

What we know

  1. In a technology-enhanced assessment course, SEM showed full mediation: lecture attendance correlated with final scores until online engagement and twelve formative quizzes were in the model, after which the direct attendance path was no longer a positive predictor. That is a decomposition in one convenience-sampled education course, not a trial of banning absences.
  2. A new Technology Integration Quality Scale (four factors, EFA/CFA in 2,281 upper-secondary students) was then placed in cluster-robust SEM: support for learning and classroom management positively predicted behavioural engagement; cognitive activation predicted digital competencies; use-frequency links were smaller; individualised teaching related negatively to engagement in the reported model. Cross-sectional perceptions cannot prove that changing those practices would raise engagement.
  3. Among Lithuanian pre-service teachers, perceived information-evaluation literacy predicted information-search literacy and ICT self-efficacy, with an indirect path through search literacy. Self-report SEM on a cross-sectional sample cannot prove that literacy training would raise teaching ICT self-efficacy.

Largest unresolved question

What gets to be a mediator is a modelling choice, not a discovery of a unique true path. The TEL course paper puts LMS engagement and quizzes between attendance and grades; the TIQS paper puts quality dimensions and frequency on engagement and competencies, including an unexpected negative path; the teacher-education paper chains evaluation literacy → search literacy → ICT self-efficacy. Different diagrams on different variables are not replications of one mediation law.

Common misconceptions

  • If SEM shows full mediation, attendance no longer matters and can be ignored.

    In that course, attendance’s association with grades ran through engagement and formative scores. That does not prove attendance is useless; it says those process measures carried the covariance. Requiring or banning attendance was not tested.

  • A well-fitting CFA means the SEM paths are causal.

    CFA says the four TIQS factors fit the item covariances. Cluster-robust SEM paths from those factors to engagement are still associations among student perceptions in a cross-section.

  • An indirect path from evaluation literacy through search literacy proves a training sequence.

    It is a decomposition of self-reports in two Lithuanian universities. A trial of teaching evaluation then search skills would be a different design.

Related