Describe what you want to find out
'Does trust go with greater satisfaction?' is different from 'Can high satisfaction occur without trust?' The first asks about a relationship. The second asks whether trust is a requirement. A third question, 'Which combinations of trust and service conditions accompany high satisfaction?', asks about combinations. They can use similar subject matter and still need different methods.
Write your question in one sentence, then check how your variables were measured and what your study design allows you to conclude. Choose the method from those choices. A small sample, a fashionable method or a low p-value is not a sufficient reason on its own.
Match your question to a starting point
Find the row that matches your question, then open the corresponding guide below the table. Check its measurement assumptions and data requirements before setting up the calculation.
On a small screen, swipe the table sideways to see every column.
| What you want to find out | Where to start | What to decide first |
|---|---|---|
| How survey-based concepts such as trust and satisfaction relate | PLS-SEM, consistent PLS, GSCA or CB-SEM | Decide what each concept means and how its survey questions measure it. The methods make different measurement assumptions. |
| How recorded variables relate to a numerical or yes/no outcome | OLS regression for a numerical outcome; binary logistic regression for a yes/no outcome | Check the outcome's scale and coding, then the assumptions in the regression guide. |
| How accurately a model predicts answers in new cases | PLSpredict and CVPAT | Decide which cases to hold back for testing and what prediction to use as the comparison. |
| Which combinations of conditions accompany an outcome | fsQCA and its related set-analysis methods | Define what each condition means and how cases qualify for it. This is the calibration step. |
| Whether a condition must reach a minimum before an outcome can be high | NCA, or set necessity when the question concerns membership in a set | Decide whether you want a minimum level on a measurement scale or a requirement stated in terms of set membership. |
| Whether relationships differ between groups | Multigroup analysis and MICOM; segmentation for groups explored from the data | Distinguish groups you defined in advance from groups suggested by the model. Check that the measures are comparable. |
- Work through the survey-based PLS model
- Read the Consistent PLS assumptions and task
- Read the GSCA question and requirements
- Work through the CB-SEM example
- Set up numerical-outcome regression
- Set up binary logistic regression
- Set up a held-out prediction comparison
- Start the fsQCA combination exercise
- Set up the NCA minimum-requirement exercise
- Check measurement comparability with MICOM
Be clear about what your measurements represent
Some variables are recorded directly, such as age or a purchase amount. Other concepts, such as trust, are measured using several questions. The app may show both as shapes in a diagram, but that does not make them the same kind of measurement.
A common-factor model treats an underlying attribute as the reason its indicators vary together. A composite combines measurements into a score. The choice comes from the concept and the purpose of the measurements. Use the method guide to work through it; a high reliability value cannot choose the measurement model for you.
For fsQCA, you need to define a set such as 'high capability'. Calibration connects the recorded scores to that definition. Standardising a score changes its scale, but does not by itself tell you which cases count as having high capability.
Add a second method only when it answers another question
A study might use PLS-SEM to examine relationships and fsQCA to examine combinations. Explain what each analysis contributes. Prepare and interpret them separately because they answer different questions.
Two analyses pointing in a similar direction can be informative, but agreement does not automatically validate either one. Check the assumptions and evidence for each. If you change the planned analysis because the data or software cannot support it, explain how that changes the question you can answer.
Try a small worked example before your own study
Open the method guide and follow a teaching example with its stated data and settings. Practise explaining the main table before using your participant data. A saved example project lets you explore the output first; a CSV exercise lets you build the analysis yourself.
For your research study, write down the planned model and important analysis choices before inspecting its findings. You can still investigate unexpected results, but keep those later decisions visible in your record.