WHAT IT IS
The analysis a thesis has to pass,
run by people who are not statisticians.
01
What is SEM?
A statistical method that tests the relationships between concepts in a theoretical model against data
e.g. it tests the hypothesis "service quality raises revisit intention through customer satisfaction" against data.
Concepts you cannot see are measured with several survey items
02
Who uses it
Graduate students writing an empirical thesis
If a business, education, psychology or nursing thesis has a research-model figure and hypotheses H1, H2, it is usually this analysis.
Thesis · Ch. 3 Research model and hypotheses
- H1Quality → Satisfaction
- H2Satisfaction → Revisit
- H3Quality → Revisit
03
Why it pays
Required for the degree,
done by people who are not statisticians
The model, the survey and the analysis sit in different tools, so you either shuttle between them alone or pay someone to do it.
The hard part is not SEM itself — it is that the researcher has to do all of this alone, moving between several tools.
THE PROBLEM
To run one study,
every stage needs a different analysis tool.
Paper · documents
The model is drawn while talking it through
Survey platform
Service quality item 1
Staff responded kindly
The concept is written again as items
Spreadsheet
Responses downloaded, columns renamed
SEM software
Quality =~ sq1 + sq2 + sq3 # and the rest, again…
The paper model is rebuilt from scratch
The same "service quality" is drawn on paper, written as items, renamed as columns and typed into the software again. The four places are not connected.
HOW WE SOLVE IT
Type the research topic,
and the model, the survey and the report follow.
Research topic
How café service quality affects revisit intention through customer satisfaction
You can start from a research question or an existing theory
- Service quality → Satisfaction
- Satisfaction → Revisit intention
- Service quality → Revisit intention
- Service qualityStaff responded kindly and promptly
- SatisfactionOverall I am satisfied with this café
- Revisit intentionI will use this café again
| Latent variable | CR | AVE | Verdict |
|---|---|---|---|
| Service quality | .805 | .580 | met |
| Satisfaction | .848 | .651 | met |
| Revisit intention | .842 | .640 | met |
Or, if the model is already drawn by hand, you can upload it as an image and analyse it.
A survey built in Survey Lab brings its constructs and items with it.
HOW IT DIFFERS
Other tools cover one stretch;
SKARI covers it from start to finish.
Survey platforms
online survey tools
Statistics · SEM software
SPSS, AMOS, R…
General-purpose AI
chat assistants
Outside statistics services
analysis done for you
SKARI
one flow
WHAT CHANGES
One model carries you from the survey to the results.
Managing the model
Built separately in each tool and copied between them.
One model connects the survey, the analysis and the report.
Checking paths
A missing path is found only after you see the results.
Checked before running, and fixed right away.
Reading the fit
The numbers alone leave the next step unclear.
Points out the indices outside the cut-offs and what to look at first.
Writing up the results
Results are screenshotted and re-typed into the paper.
Results in paper order, together with the R code.
THE REPORT
The report follows
the order of a thesis results chapter.
Same order as the results chapterthe report chapters become the thesis skeleton
Thesis · IV. Results
SKARI analysis report
4.1Sample characteristics
01Sample and data
4.2Descriptive statistics
02Descriptive statistics
4.3Reliability and validity
03 · 04Reliability · measurement
4.4Model fit
05Model fit
4.5Hypothesis testing
06Structural model
4.6Mediation
07Effect decomposition
4.7Discussion and limitations
08Diagnostics and limits
Down to the R code used for the analysisreport chapter numbers sit in the code comments
# SKARI · SEM Lab · Service quality → Satisfaction → Revisit intentionlibrary(lavaan); library(semTools)data <- read.csv("responses.csv") # 01 sample and data model <- ' Quality =~ sq1 + sq2 + sq3 # 04 measurement model Satisfy =~ cs1 + cs2 + cs3 Revisit =~ ri1 + ri2 + ri3 Satisfy ~ a*Quality # 06 structural model Revisit ~ b*Satisfy + c*Quality indirect := a*b # 07 effect decomposition'fit <- sem(model, data = data, se = "bootstrap")summary(fit, fit.measures = TRUE, standardized = TRUE) # 05htmt(model, data = data) # 04 discriminant validity
Thesis appendix
Attach the analysis as code
Answering the committee
Show at once how a number was produced
Sharing data and code
R is free, so anyone can run it again
CFI .953 · RMSEA .101
evidence is mixedSRMR .051 · TLI .938 in range, RMSEA to review
1 pair to review
discriminant validityDemocracy1960–1965 HTMT .98 · AVE .63–.85
3 / 3 significant
no non-significant pathStrongest Democracy1960 → 1965 β .91
| Construct | CR | AVE | min β | status |
|---|---|---|---|---|
| VisualAbility | .626 | .371 | .424 x2 | review |
| TextualAbility | .885 | .720 | .838 x6 | in range |
| SpeedAbility | .691 | .430 | .570 x7 | review |
CR .70 and AVE .50 are the usual reference points. An AVE below .50 means the construct explains less than half the variance of its own indicators, on average.
| Path | β | B [95% CI] | p |
|---|---|---|---|
| Industrialization → Democracy1960 | .448 | 1.474 [0.708, 2.240] | < .001 |
| Industrialization → Democracy1965 | .146 | 0.453 [0.009, 0.898] | .046 |
| Democracy1960 → Democracy1965 | .913 | 0.864 [0.647, 1.082] | < .001 |
β is the standardised coefficient; the interval is on the unstandardised B. "Supported / rejected" appears as its own column only when hypotheses are attached.
WHAT IT FITS
From model specification to estimation,
explore the analyses that fit your research design.
Measurement models and structural paths, multi-group comparison and longitudinal designs, through to estimation methods — all in one place.
Checks that each item measures the construct it was meant to.
Explains several lower-order constructs with a higher-order factor. The example goes up to a third-order factor.
Estimates a general factor and domain factors together, set to be uncorrelated with each other.
Explores the factor structure with cross-loadings allowed, then rotates it for interpretation.
When several raters assess several traits, shows how much of each score reflects the trait and how much the rater.
Splits each score into trait effect, method effect and measurement error.
A model in which several indicators together form the construct.
WHAT IT CONNECTS TO
The model is one step in a longer study.
Start from one sentence about your study.
The model, the survey and the report follow from it. Already drew the model? Upload the image instead.