Skip to contents

Interactive Shiny Application

Source: vignettes/shiny-app.Rmd
shiny-app.Rmd

Overview

The uncounted package includes an interactive Shiny application that provides a point-and-click interface for estimating unauthorised population sizes. The app wraps the package’s core functions in a multi-tab interface suitable for exploring data, fitting models, and assessing model quality.

Installation

The app requires three additional packages:

install.packages(c("shiny", "DT", "bslib"))

Launching the App

This opens the application in your default browser.

Tabs

1. Data

Upload a CSV file or use the built-in irregular_migration dataset (1,382 observations of border apprehensions across 145 countries, 2019–2024).

Map your columns to the model variables:

  • Observed (m): count of detected unauthorised migrants
  • Auxiliary (n): auxiliary detection count (e.g., police identifications)
  • Reference pop (N): size of the lawful reference population
  • Countries: country identifier for clustering

2. Model Fitting

Specify and fit multiple models. For each model, choose:

  • Method: Poisson (recommended), Negative Binomial, OLS, or NLS
  • cov_alpha: formula for community anchor covariates (e.g., ~ year * ukr + sex)
  • cov_beta: formula for detection elasticity covariates (e.g., ~ year)
  • Gamma: estimate jointly, fix at a value, or exclude
  • Constrained: enforce alpha in (0,1) and beta > 0

Click “Fit Model” to estimate. The model is stored and can be compared with others. The summary panel shows coefficients with robust standard errors.

3. Population Size

Select a fitted model and specify a grouping formula (e.g., ~ year or ~ year + sex). The app computes population size estimates with delta-method confidence intervals and optional bias correction.

Results can be downloaded as CSV.

4. Model Comparison

When two or more models are fitted, this tab shows a comparison table with:

  • AIC, BIC, log-likelihood
  • Deviance, Pearson chi-squared
  • Pseudo R-squared measures (correlation, explained deviance, Cameron-Windmeijer)

A likelihood ratio test can be run for any pair of nested models.

5. Diagnostics

Select a fitted model and choose a diagnostic:

  • Residual plots: 4-panel Zhang (2008) diagnostics
  • Rootogram: count distribution fit (hanging, suspended, or standing)
  • Gamma profile: sensitivity of population size to gamma
  • LOO (by country): leave-one-out influence analysis
  • Exploratory: log-log diagnostic scatter plots

Example Workflow

  1. Select “Built-in (irregular_migration)” and verify variable mapping
  2. In Model Fitting, set method = “poisson”, cov_alpha = “~ year + sex”, cov_beta = “~ year”, name = “Poisson S3”. Click Fit Model.
  3. Change method to “nb”, name = “NB S3”. Click Fit Model.
  4. Change cov_alpha to “~ year * ukr + sex”, method = “poisson”, name = “Poisson S8”. Click Fit Model.
  5. Go to Model Comparison to see AIC differences.
  6. Go to Population Size, select “Poisson S8”, type “~ year” in the stratify field, click Compute.
  7. Go to Diagnostics, select “Poisson S8”, and explore residuals, rootograms, and gamma profiles.