jamovi vs. SPSS: Which Is Better for Teaching Statistics?

How to Run Advanced Regression Models in jamovi (Step‑by‑Step)Regression analysis is a cornerstone of quantitative research. jamovi provides an approachable, GUI-driven environment for running standard and advanced regression models without extensive coding. This article walks through setting up your data, choosing appropriate models, running analyses, checking assumptions, interpreting outputs, and extending jamovi with add‑modules for advanced techniques.

---

Overview: What counts as “advanced” regression in jamovi

Advanced regression analyses include:

• Multiple and hierarchical (multilevel) linear regression
• Generalized linear models (GLMs) (logistic, Poisson, negative binomial)
• Mixed‑effects models (linear mixed models, generalized mixed models)
• Regularized regression (LASSO, ridge)
• Robust regression and bootstrapped inference
• Model selection, comparison, and diagnostics

jamovi supports many of these natively and via add‑modules (e.g., GAMLj, MixedModels, glmnet, psych/robust packages). Below we cover step‑by‑step workflows for several common advanced scenarios.

---

1. Prepare your data

1. Open jamovi and load your dataset: File → Open → choose CSV, Excel, or jamovi (.omv).
2. Inspect variable types in the Data tab. Ensure:
   • Continuous variables are numeric.
   • Categorical variables are nominal/factor (set as “nominal” or “ordinal” as appropriate).
3. Create derived variables if needed: Transform → Compute.
   • Center predictors (mean‑center) for interaction terms to reduce multicollinearity: Compute → (X – mean(X)).
4. Handle missing data:
   • Use Filter to run analyses on complete cases, or consider multiple imputation (use the jamovi module mice or conduct imputation externally).
5. Check distributions with Descriptives and Plots to spot outliers and skewness.

---

2. Multiple linear regression (with interactions and centering)

Use this when your outcome is continuous and residuals are approximately normal.

Steps:

1. Analyses → Regression → Linear Regression.
2. Move the dependent variable into “Dependent Variable.”
3. Add predictors to “Covariates” and categorical predictors to “Factors.”
4. For interaction terms: click the little cog (model) → Model → add interaction(s) between predictors.
5. To mean‑center a predictor, create a new variable: Transform → Compute → X_center = X – mean(X).
6. Under Options, request:
   • Estimates, Confidence intervals, Standardized coefficients (if desired).
   • Model fit statistics: R², adjusted R², AIC.
   • Plots: residuals vs. fitted, Q–Q plot.
7. Interpret coefficients: unstandardized b (unit change), p-values, and CIs. For interactions, probe simple slopes or plot predicted values.

Diagnostics:

• Check Variance Inflation Factor (VIF) for multicollinearity: Options → Collinearity diagnostics.
• Examine residual plots for heteroscedasticity and nonlinearity.
• If heteroscedasticity appears, use robust standard errors (Options → Robust standard errors).

---

3. Generalized linear models (logistic, Poisson)

Use GLMs when outcomes are non‑normal (binary counts, rates).

Logistic regression (binary outcome):

1. Analyses → Regression → Generalized Linear Model.
2. Put outcome in “Dependent Variable” and set “Family” = Binomial, “Link” = Logit.
3. Add predictors as covariates/factors.
4. Options: display Odds Ratios (exp(B)), CIs, model fit (AIC), classification table, ROC curve.
5. Check for separation and sparse categories; consider Firth correction via add‑modules if needed.

Poisson / negative binomial (count outcomes):

1. In the GLM dialog, choose Family = Poisson for counts.
2. If overdispersion exists (variance > mean), try negative binomial (if available via module) or use quasi‑Poisson alternatives.
3. Include an offset term for rates (log of exposure) if modeling rates: in GLM specify Offset = log(exposure variable).

Model assessment:

• Use Deviance, AIC, and residual diagnostics (e.g., deviance residuals).
• For logistic models, report classification metrics and the ROC/AUC.

---

4. Mixed‑effects models (multilevel / hierarchical)

Use mixed models when data are clustered (e.g., students nested in schools, repeated measures).

Install the MixedModels or GAMLj module if not present: Modules → jamovi library → search and install (MixedModels, GAMLj).

Linear mixed model (continuous outcome):

1. Analyses → Mixed Models → Linear Mixed Model (or GAMLj → Linear Mixed Models).
2. Dependent Variable → outcome; Fixed Effects → predictors.
3. Specify random effects: add grouping variable (e.g., Subject, School) and choose random intercepts and/or slopes.
4. Choose estimation method (REML for comparing random structures, ML for comparing fixed effects).
5. Request covariance structures and report intraclass correlation (ICC).
6. Interpret fixed effects like regression coefficients; random effect variance components tell how much variability is at each level.

Generalized mixed model (non‑normal outcomes):

1. Analyses → Mixed Models → Generalized Mixed Model; choose appropriate family (Binomial, Poisson).
2. Configure random effects similarly.
3. Assess model convergence warnings and consider simplifying random structure if convergence fails.

Model checking:

• Plot residuals by group; examine random effects BLUPs.
• Compare nested models with likelihood ratio tests (ML) or use AIC/BIC.

---

5. Regularized regression (LASSO, ridge)

Useful when predictors are many and multicollinearity or variable selection is needed.

Install the glmnet or jfa package via Modules → jamovi library.

Steps:

1. Analyses → glmnet (or Regularization module).
2. Specify outcome and predictor matrix.
3. Choose penalty: LASSO (L1), ridge (L2), or elastic net.
4. Use cross‑validation (CV) to pick lambda (regularization strength). Request k‑fold CV (commonly k=10).
5. Extract selected predictors (nonzero coefficients at chosen lambda) and CV error plots.

Interpretation:

• Coefficients are biased toward zero; focus on predictive performance (CV error) rather than unbiased effect estimation.

---

6. Robust regression and bootstrapping

Robust regression reduces influence of outliers; bootstrapping provides empirical CIs.

Robust regression:

• Use Modules → Robust regression / GAMLj’s robust options.
• Choose robust estimator (Huber, MM) and compare with OLS.

Bootstrapping:

1. Many jamovi analyses include a Bootstrapping option in the options panel—enable it.
2. Specify number of bootstrap samples (e.g., 1000).
3. Use bootstrap CIs for coefficients and other statistics (percentile/bootstrap‑t).

---

7. Model comparison, selection, and reporting

Model comparison:

• Compare nested models with Likelihood Ratio Tests (for ML/REML appropriately) or ANOVA-like model comparison where supported.
• Use information criteria (AIC, BIC) for non‑nested comparisons.

Model selection:

• Stepwise selection is discouraged; prefer theory‑driven models or regularization with CV to select predictors.
• Report effect sizes, CIs, and model performance (R² or pseudo‑R² for GLMs, AUC for binary outcomes).

Reporting checklist (concise):

• Data and pre‑processing (missing data handling, transformations).
• Model family and link function.
• Fixed and random effects structure.
• Estimation method (ML/REML) and software/module used (include jamovi version and modules).
• Key coefficients with CIs, p‑values, and model fit indices.
• Diagnostics performed (collinearity, residuals, influence).

---

8. Example: Step‑by‑step — logistic mixed model for repeated binary outcomes

1. Load dataset with repeated responses per Subject.
2. Modules → install MixedModels (if needed).
3. Analyses → Mixed Models → Generalized Mixed Model.
4. Set Family = Binomial, Link = Logit. Dependent = binary outcome.
5. Fixed effects: time, treatment, covariates.
6. Random effects: add Subject as random intercept; add random slope for time if theory suggests.
7. Choose estimation (Laplace or adaptive Gauss‑Hermite where available).
8. Run model; inspect fixed effect coefficients (log odds), convert to odds ratios (exp(coef)), and check random effect variances.
9. Assess convergence, plot predicted probabilities by group and time, and report AIC/BIC.

---

9. Extending jamovi: useful modules and resources

Useful jamovi modules:

• GAMLj — flexible linear, generalized linear, and mixed models with friendly output.
• MixedModels — for more advanced multilevel models.
• glmnet — LASSO/ridge.
• mirt / psych — psychometric and factor analysis tools.
• mice — multiple imputation.

Resources:

• jamovi user guide and module documentation (within jamovi library).
• Tutorials and example datasets in the jamovi community.

---

10. Troubleshooting tips

• Convergence failures: simplify random structure, increase iterations, or rescale predictors.
• Overdispersion in counts: switch to negative binomial or use quasi‑Poisson.
• Separation in logistic models: check sparse cells; consider Firth regression or penalized likelihood.
• High VIFs: center predictors, combine correlated predictors, or use regularization.

---

Advanced regression in jamovi combines a point‑and‑click interface with powerful modules to run multilevel, generalized, robust, and regularized models. Follow theoretical guidance for model building, use diagnostic checks, and when needed, extend jamovi via modules or export data to R for highly specialized models.