Six-Stage Regression Workflow
The shared paper follows the regression workflow from Hair et al. This course turns that workflow into a practical SmartPLS checklist.
Stage 1: Select the Objective
Decide whether the analysis is for:
- Explanation
- Prediction
- Both explanation and prediction
Output:
Research objective statement
Example:
The objective is to explain customer satisfaction using customer perceptions of product quality, e-commerce, complaint resolution, product line, and salesforce image.
Stage 2: Design the Regression Analysis
Define:
- Dependent variable
- Independent variables
- Controls
- Sample size
- Measurement level
- Linear or logistic regression
Output:
Model specification table
Stage 3: Test Assumptions
For multiple linear regression, check:
- Linearity in parameters
- Random sampling
- No perfect multicollinearity
- Exogeneity
- Homoscedasticity
- Independence of residuals
- Approximate normality of residuals for inference
For logistic regression, check:
- Binary dependent variable
- Independent observations
- No perfect multicollinearity
- Linearity between predictors and log-odds
- Adequate sample size
Output:
Assumption assessment table
Stage 4: Estimate the Model and Assess Overall Fit
For multiple linear regression, assess:
- ANOVA or F-test
- R-square
- Adjusted R-square
- Durbin-Watson test if relevant
- Residual plots
For logistic regression, assess:
- Log-likelihood
- Deviance
- AIC
- BIC
- McFadden's R-square
- Cox and Snell's R-square
- Nagelkerke's R-square
- Confusion matrix
Output:
Model fit assessment table
Stage 5: Interpret the Regression Variate
For multiple linear regression, interpret:
- Unstandardized coefficients
- Standardized coefficients
- t-values
- p-values
- Confidence intervals
- Relative importance
For logistic regression, interpret:
- Logit coefficients
- Wald test
- p-values
- Odds ratios
- Direction of group membership probability
Output:
Hypothesis decision table
Stage 6: Validate the Results
Validation can include:
- Holdout sample
- Cross-validation
- Prediction accuracy
- Sensitivity checks
- Robust standard errors
- Alternative model specification
Output:
Validation and limitation paragraph
Final Course Deliverable
At the end of the workflow, students should produce:
- Model diagram
- Exported SmartPLS tables
- Assumption checklist
- Model fit table
- Coefficient table
- Written result section
- Limitations paragraph