Dissertation Topics Using SEM and Mediation Analysis

SEM · CFA · Mediation · PLS-SEM · Dissertation research

50 dissertation topics organised by statistical model, from scale validation and mediation to measurement invariance and latent growth. Each topic includes a suggested design, dataset and related university thesis.

50Research topics
10Model groups
10Data sources
9Universities cited
2020–26Thesis years

By Siddharth Gupta · Statssy

SEM projects often stall because a model does not fit, the measurement scales are unsuitable, or the sample is too small for the complexity proposed.

In my 12 years of helping students with structural equation models, two decisions have consistently mattered: choosing a model that genuinely answers the research question and obtaining data that contain the necessary measured items.

The following 50 research ideas include a method, accessible source and related university thesis. Check each dataset’s codebook before confirming the construct measurements.

Which model do you need?

Start with the research question, measurement model and temporal design. Some techniques overlap, so treat this as a selection guide rather than a substitute for model diagnostics.

Your research questionStarting point
Does my scale measure its intended constructs?Confirmatory factor analysis and scale validation
Which individual test items work?Rasch and item response theory
Does X relate to Y through one mediator?Simple mediation
Are there two or more parallel or sequential mediators?Parallel and serial mediation
Does an association depend on a moderator?Moderation
Does an indirect effect vary by another variable?Moderated mediation and conditional process models
Can multiple latent constructs be tested together?Full structural equation models (CB-SEM)
Is prediction with composite-based paths the primary aim?PLS-SEM and technology acceptance models
Can measurement be compared across groups?Multi-group SEM and measurement invariance
Are there latent subgroups or developmental trajectories?Latent class, latent profile and growth models

How to use this list

  1. Choose your research question type and identify the corresponding model.
  2. Pick two or three topics in a domain you know.
  3. Check the data for item-level responses, grouping variables and longitudinal waves where required.
  4. Read the linked university thesis and examine its model fit, measurement and indirect-effect reporting.
  5. Review the proposed model with your supervisor or book a free consultation.
Model group 01 / 10

Confirmatory factor analysis and scale validation

CFA tests whether a scale measures what it claims. It is the first step in almost every SEM dissertation, and often a full dissertation on its own.

Model group 02 / 10

Rasch and item response theory

IRT and Rasch models look at each item, not only the total score. They suit test design and scale shortening topics.

Model group 03 / 10

Simple mediation

Simple mediation tests whether X affects Y through one middle variable. It is the most requested model in psychology and business dissertations.

TOPIC 12

Does job satisfaction explain the link between overtime and leaving?

Study design
Cross-section
Model
Mediation with a binary outcome
Dataset
IBM HR Analytics Attrition (Kaggle)
Model group 04 / 10

Parallel and serial mediation

Multiple mediation tests two or more middle variables, side by side or in a chain. It turns a simple model into a stronger dissertation.

TOPIC 20

Does sleep partly explain the link between social media hours and teen wellbeing?

Study design
Longitudinal panel
Model
Longitudinal mediation (cross-lagged)
Dataset
Understanding Society
Model group 05 / 10

Moderation

Moderation tests when or for whom a link is stronger. Interaction terms and simple slopes are all you need to start.

See the printable 50-topic guide

Model group 06 / 10

Moderated mediation and conditional process models

Moderated mediation asks whether an indirect effect changes across groups or levels. Examiners see it as advanced, so it lifts your marks.

TOPIC 26

Does work-life balance change the indirect path from overtime to leaving through job satisfaction?

Study design
Cross-section
Model
Moderated mediation (PROCESS model 7)
Dataset
IBM HR Analytics Attrition (Kaggle)
Model group 07 / 10

Full structural equation models (CB-SEM)

Full SEM combines a measurement model and a structural model in one test. Use it when your variables are latent constructs with several items each.

Model group 08 / 10

PLS-SEM and technology acceptance models

PLS-SEM suits prediction, smaller samples and models with many constructs. It is standard in business, marketing and information systems.

Model group 09 / 10

Multi-group SEM and measurement invariance

Invariance tests check that a scale means the same thing in different groups. You need it before you compare men and women, or countries.

Model group 10 / 10

Latent class, latent profile and growth models

Latent class and growth models find hidden groups and change over time. They suit large longitudinal datasets well.

See the printable 50-topic guide

Datasets for SEM, mediation and psychometric research

These ten sources provide potential study data. Verify that your chosen constructs are measured by enough appropriate items, rather than assuming that every source directly supports a full latent-variable model.

Open Psychometrics

Raw responses from personality and psychological instruments; inspect each dataset for instrument version, response quality and sampling limitations.

MIDUS

Adult health, wellbeing, stress and personality data, with repeated measures and diary studies.

European Social Survey

Cross-national measures of trust, attitudes and wellbeing; verify each round’s item availability.

Add Health

Longitudinal adolescent-to-adult data; public-use access and restricted-use applications differ.

Understanding Society

UK household panel with repeated measurements of work, wellbeing and household characteristics.

PISA Data (OECD)

Student questionnaire item responses and assessment outcomes; survey weights and plausible values require specialist handling.

OSMI Mental Health in Tech (Kaggle)

Technology-worker survey responses on workplace support and mental health; check whether selected questions form validated multi-item constructs.

Printable SEM dissertation research guide

Get all 50 topics in one PDF

The printable guide will include all 50 questions, study designs, models, datasets and related thesis links. The download link will be enabled once the PDF is generated and uploaded to WordPress.

PDF download coming soon

How to turn one of these into your own topic

  1. Draw your model first. Specify constructs, indicators and paths using theory.
  2. Check measurement quality. Evaluate factor structure and reliability before interpreting structural paths.
  3. Plan sample size. Base it on model complexity, indicators, effect sizes, missing data and desired power rather than a universal threshold.
  4. Evaluate indirect effects carefully. Report appropriate confidence intervals and distinguish statistical mediation from causal evidence, particularly in cross-sectional studies.
  5. Report fit and model changes transparently. Include relevant indices, confidence intervals and a justification for modifications.

See EFA versus CFA, handling cross-loading items and regression dissertation topics for related guidance.

Related topic lists

Need a second pair of eyes?

Choose a defensible SEM model.

Bring your model sketch and dataset to a free 15-minute call. We can discuss your constructs, available measures and sample size.

Book a free 15-minute call

15 minutes · No cost · No pressure

Popular questions

What sample size do I need for SEM?

Aim for at least 200 cases for a simple model. Complex models with many constructs need more. Free datasets like MIDUS and the ESS give you thousands.

Should I use CB-SEM or PLS-SEM?

Use CB-SEM to test a theory with good model fit. Use PLS-SEM for prediction, smaller samples or many constructs.

Which software is best for SEM?

AMOS and Mplus are common in psychology. SmartPLS is standard for PLS-SEM. The R package lavaan is free and handles almost everything.

What model fit values should I report?

Report chi-square, CFI, TLI, RMSEA and SRMR. CFI above 0.90 and RMSEA below 0.08 are common cut-offs.

How do I test mediation properly?

Use bootstrapped confidence intervals for the indirect effect. The older Baron and Kenny steps are no longer enough on their own.

Is there an SEM dissertation topics PDF?

Yes. Download the full Topic and Method list from this page. It includes every topic, model and dataset link.

Perfect for students, researchers, and professionals looking to build real statistical skills.