Data Analysis Using SPSS
Reliable analysis, accurate results — from raw responses to a defensible results chapter.
Read more
Advanced PLS-SEM: measurement model, structural model, bootstrapping and mediation.
Partial least squares structural equation modelling has become the default in management, information systems and behavioural research — often for the wrong reason, which is that it tolerates a small sample.
PLS-SEM is a genuinely good choice when your model is complex, your constructs are formative, your data is not normally distributed, or your aim is prediction rather than theory testing. It is the wrong choice when you have a well-established theory, a large sample and reflective constructs, where covariance-based SEM in AMOS or Lisrel gives you stronger evidence and proper fit indices.
We will tell you which situation you are in before running anything. That conversation takes twenty minutes and saves people from a reviewer question they cannot answer, which is: why did you choose PLS here? "Because my sample was small" is not a defensible answer on its own, and reviewers in the better journals now say so explicitly.
The six blocks on the poster are the standard assessment sequence: measurement model assessment, structural model evaluation, reliability and validity testing, bootstrapping and significance, mediation and moderation analysis, and report generation with interpretation. We follow that order because each stage is only meaningful once the previous one has passed.
Five service steps on the poster, and every statistic each one produces.
The stage that stops more models than any other is discriminant validity. Two constructs that your theory says are distinct turn out to be measured by items that respondents treated as the same thing. The HTMT ratio comes back above the threshold, and the model as specified cannot proceed.
There are honest responses to this and dishonest ones. The honest responses are to drop or rewrite the overlapping items, to merge the two constructs into one if the theory allows it, or to report the problem as a limitation. The dishonest response is to delete indicators one at a time until the number falls below the threshold, which is common, detectable, and something we will not do for you.
We report the assessment as it comes out. If your model has a problem you will hear it from us first, with the options laid out, rather than from a reviewer eight months later.
You send the conceptual model, the questionnaire and the data. We check whether PLS is the right method and whether the model is identified.
Loadings, reliability, convergent and discriminant validity, with a clear decision recorded for every item retained or dropped.
Collinearity, path coefficients, R², f², Q² and the significance of every relationship in your hypotheses.
Resampling with reported settings, producing confidence intervals and t-statistics for each path.
Indirect effects with bootstrapped intervals, interaction terms and simple-slope plots where a moderator is involved.
Tables in the format the journals expect, the model diagram, and a session in which we take you through every number.
The reporting bar in PLS-SEM has risen sharply. Reviewers now expect the HTMT ratio rather than Fornell–Larcker alone, they expect Q² from a blindfolding or PLSpredict procedure rather than R² by itself, and they expect the bootstrap settings — number of subsamples, sign-change option, confidence interval method — to be stated explicitly.
They also expect a justification for the method itself, a statement of how the minimum sample size was determined, and, increasingly, some evidence about common method bias when all constructs were measured with one instrument at one time. We include all of this as standard because leaving it out invites exactly the revision request you were hoping to avoid.
You receive the SmartPLS project file itself, not only the results. If a reviewer asks for an additional control variable or a multi-group split, you or we can re-run it in an afternoon rather than rebuilding the model from the questionnaire.
| What you receive |
|---|
| SmartPLS project file you keep and can re-run |
| Measurement model assessment table |
| Discriminant validity tables including HTMT |
| Structural model results with R², f² and Q² |
| Bootstrapping output with confidence intervals |
| Mediation and moderation results |
| Publication-quality model diagram |
| Written interpretation and a walk-through call |
The old rule of ten times the largest number of arrows pointing at a construct is no longer considered adequate on its own. We run a power analysis based on your model's complexity and the smallest effect you want to detect, and give you a defensible number with the calculation shown. Small samples are possible in PLS but they are not free.
AMOS and covariance-based SEM when you are testing an established theory with reflective constructs and a reasonable sample, because you get proper fit indices. SmartPLS when the model is complex, the constructs are formative, the data is badly non-normal, or the aim is prediction. We will tell you which applies to your study before any work begins.
Then two of your constructs are not being distinguished by respondents. The options are to revise or drop the overlapping items, to merge the constructs if theory permits, or to report it honestly as a limitation. We will not delete indicators one at a time purely to push a number under a threshold.
Yes. You get a clean vector model diagram with every path coefficient, significance level and R² labelled, sized to fit your page and readable in black and white as well as colour.
Reliable analysis, accurate results — from raw responses to a defensible results chapter.
Read more
From raw data to real insight — descriptive, inferential, regression and multivariate work.
Read more
End-to-end support for PhD and M.Tech thesis writing, analysis and publication.
Read moreSend your topic, your dataset or one draft chapter. We will tell you honestly what it needs — before you pay anything. The first consultation is free.