Load your dataset
A plain numeric CSV: one row per respondent, one column per indicator (survey item), with a header row. Non-numeric columns are ignored; rows with missing values on used indicators are dropped listwise. Nothing is uploaded — parsing happens in this tab.
…or paste CSV text
Mediation or moderation on observed scores
PROCESS-style analysis on your columns. Pick one or more items per role — multiple items are averaged into a scale score. Mediation is Hayes Model 4 (X → M → Y with a bootstrapped indirect effect); moderation is Model 1 (X × W interaction with simple slopes at W = mean ± 1 SD).
Multiple linear regression
Plain OLS on your columns — the workhorse every methods course starts with. Pick one dependent variable and any number of predictors.
Define constructs and paths
Each latent construct gets a name and its reflective indicators (Mode A). Click a column chip to assign it — or drag any chip onto another construct's card to move it there. An indicator may sit in more than one construct (that's how repeated-indicators higher-order models work). The structural model must be recursive — no loops.
Run the model
Method notes: PLS-SEM uses reflective (Mode A) measurement with the Lohmöller algorithm; bootstrap uses percentile confidence intervals with construct-level sign alignment; p-values use the normal approximation to the t statistic. Results can differ slightly from SmartPLS 4 (initialization, sign-correction, and stopping details vary between implementations). Mediation and moderation run PROCESS-style Models 4 and 1 (Hayes) on observed scale scores — for other model numbers use dedicated software. Not included, honestly: covariance-based SEM (CB-SEM needs maximum-likelihood covariance fitting — use lavaan or AMOS and cite them), GSCA, and formative (Mode B) measurement within PLS. Independent tool, not affiliated with SmartPLS GmbH; all methods implemented from the published literature.
Methods & attribution
AI PLS is an independent implementation of methods from the published literature. Cite the original methodologists in your write-up — these are the canonical sources for what each part of your results rests on.
| Method in this tool | Canonical source |
|---|---|
| PLS path modeling algorithm | Wold (1982); Lohmöller (1989) |
| Bootstrap resampling & CIs | Efron (1979) |
| Cronbach's α | Cronbach (1951) |
| Composite reliability, AVE, Fornell–Larcker criterion | Fornell & Larcker (1981) |
| ρA (consistent reliability) | Dijkstra & Henseler (2015) |
| HTMT discriminant validity | Henseler, Ringle & Sarstedt (2015) |
| f² effect-size thresholds | Cohen (1988) |
| Mediation & moderation (Models 4 / 1), simple slopes | Hayes (2022); Aiken & West (1991) |
| TAM model template | Davis (1989) |
| PLS-SEM reporting practice (loadings ≥ .708, 5,000 resamples…) | Hair, Hult, Ringle & Sarstedt (2022) |