Interactive Data Analysis with MAIHDA

Online Version

You can access a live, cloud-hosted version of the MAIHDA interactive dashboard directly in your browser without installing R: https://hdbt.shinyapps.io/shiny/

Local Version

You can start the interactive dashboard locally by running a single command:

library(MAIHDA)
run_maihda_app()

This will automatically launch the dashboard in your default web browser or the RStudio viewer.

Shiny Dashboard

1. Data Import

• Upload Own Data: Easily upload datasets in .csv, Stata (.dta), or SPSS (.sav) formats.
• Use Included Data: Try out the app instantly by selecting the pre-loaded maihda_health_data or maihda_sim_data.
• View Data: The app includes an interactive data table letting you sort, filter, and inspect variables before analyzing.

2. Variable Selection & Strata Creation

• Choose a categorical/continuous outcome metric from your dataset.
• Select two or more categorical demographic variables (e.g., gender, race, education) to automatically generate intersectional strata.

3. Model Fitting & Settings

• Fit models with the lme4 engine used by the interactive dashboard. Bayesian brms models remain available from R code via fit_maihda(engine = "brms").
• Select covariates to control for within your models.
• Choose whether to calculate bootstrap confidence intervals to get robust uncertainty metrics for your Variance Partition Coefficient (VPC / ICC).

4. Interactive Visualizations

Once a model is fit, you can navigate across multiple tabs:

• Predicted Values: Visually evaluate stratum-level predictions relative to the overall mean with dynamic prediction intervals.
• VPC Decomposition: Examine how much of your outcome’s variance is attributed to between-stratum differences versus within-stratum individual heterogeneity.
• Observed vs. Shrunken Estimates: Compare raw unadjusted group means to your model’s shrinkage estimates to see the protective mechanism of multilevel modeling.

5. Stepwise Variance Analysis (PCV)

The dashboard calculates stepwise Proportional Change in Variance (PCV) tables:

• See how much inequality is “explained away” by adding covariates sequentially.
• Uncover masking/suppression effects directly inside the app by comparing partial PCV values across models.