Shiny Dashboards for Development Data

Two halves: ui and server

Every Shiny app has two parts.

• The ui describes the page: headings, the controls a reader can change (inputs) and empty slots where results will appear (outputs). Each input and output has an id, a short name in quotes such as "state".
• The server is a function that fills the output slots. It reads the inputs by id and recomputes an output whenever an input it uses changes.
• The last line joins the two: shinyApp(ui, server) in R, App(app_ui, server) in Python.

In R the server reads input$state and writes to output$summary. In Python it reads input.state() (note the brackets: it is a function call) and the output is a function decorated with @render.text whose name is the output id.

This app embeds a ten-row data frame of district results. The data is invented for teaching Illustrative data; module 5 shows where the numbers come from. Pick a state and read the sentence.

Try it: change pct_toilet to report the highest value instead of the mean (max() in R, .max() in Python), and edit the sentence to match.

Inputs and outputs

Inputs and outputs come in pairs of functions: one in the ui to place them, one in the server to fill an output. The ones a monitoring dashboard needs most:

• Drop-down menu: selectInput() in R, ui.input_select() in Python.
• Slider: sliderInput() in R, ui.input_slider() in Python.
• Tick box: checkboxInput() in R, ui.input_checkbox() in Python.
• Text box: textInput() in R, ui.input_text() in Python.
• Text output: textOutput() with renderText() in R, ui.output_text() with @render.text in Python.
• Table output: tableOutput() with renderTable() in R, ui.output_data_frame() with @render.data_frame in Python.
• Chart output: plotOutput() with renderPlot() in R, ui.output_plot() with @render.plot in Python.

This app uses a slider, a tick box and a table. Move the slider: the table lists the districts whose toilet coverage is below the threshold.

Try it: change the slider's default value to 70 and its step to 1, then re-run the app in the Shinylive editor.

Reactivity: write the filter once

Shiny keeps track of which inputs each output reads. When an input changes, Shiny reruns only the outputs that depend on it. This is reactivity, and it is why you never write "when the menu changes, redraw the chart" yourself.

When two outputs need the same filtered data, do not filter twice. Put the filter in a reactive expression: reactive({ ... }) in R, a function decorated with @reactive.calc in Python. Call it like a function, selected(), from each output. Shiny computes it once per change and both outputs share the result.

• The menu feeds selected().
• selected() feeds both the count and the table.
• Change the menu, and Shiny marks selected() out of date, then recomputes it and both outputs.

Try it: add a third output that shows the highest mean_pc_exp among the selected districts. Place it in the ui and fill it from selected().

Prototype the summary before it goes into the app

A dashboard is an analysis with controls attached. Get the analysis right first, in ordinary code you can run and check, and only then wrap it in Shiny. These cells run on this page, on households.csv and districts.csv. Illustrative data Both tables are invented for teaching; the district names are real places, but no number describes them.

Step 1: one row per district

Join households to districts, turn the Yes/No columns into 0 or 100, and average by district. Run it in R, then switch the tab to Python and run again.

Ten rows, one per district, and the same numbers in both languages. R lists the rows by district and Python by state; the values agree. Kozhikode has the highest mean expenditure, 6,758 rupees a month, and Rewa the lowest toilet coverage, 50 per cent. These ten rows are the data frame the apps in this course carry.

Step 2: the filter and the chart, with the inputs as plain variables

Before there is a menu, the reader's choices are just two variables. Write the filter and the chart with them. When this works, each variable becomes an input, and the code moves into the server almost unchanged.

The three Madhya Pradesh districts, sorted: Indore 83.3, Betul 75.0 and Rewa 50.0 per cent, and a bar chart with the highest at the top. The rows are reversed just before plotting because a horizontal bar chart draws the first row at the bottom.

Try it: set state to "Kerala" and run again. Then follow the same steps for has_bank_account.

The district dashboard in R

This app puts modules 2 to 5 together. The sidebar has three inputs: a state (or All), an indicator and a sort switch. The main panel has a horizontal bar chart and a table, both fed by one reactive expression. The data frame at the top holds the ten rows that the Step 1 cell printed, typed in, because a Shinylive app cannot read households.csv from this page.

How the code maps to what you see

• sidebarLayout() puts the inputs on the left and the outputs on the right; on a narrow screen they stack.
• setNames(names(labels), labels) shows readable labels in the menu while the server receives the column name, such as pct_bank.
• selected() is the same filter-and-sort as the Step 2 cell, with input$state and input$indicator in place of the plain variables.
• The chart uses base R graphics, so the app needs no package beyond shiny and starts faster in Shinylive.

Try it: add a fifth indicator. Put a column into the data frame (for example pct_shg, the share in a self-help group, computed the way Step 1 computes the others), then add a line for it to labels. The menu, chart and table pick it up with no other change.

The district dashboard in Shiny for Python

The same dashboard in Shiny for Python, using the same ten rows. The structure is identical; the spelling differs.

Differences from the R version

• ui.page_sidebar() with ui.sidebar() plays the part of sidebarLayout().
• ui.input_select() accepts a dictionary for choices: the keys are what the server receives, the values are what the reader sees.
• Inputs are read with brackets, input.indicator(), and outputs are functions named after their ids, decorated with @render.plot or @render.data_frame.
• @render.plot displays the matplotlib figure the function returns.

Try it: add ax.axvline(districts[input.indicator()].mean(), color="#B45309") before return fig to draw the ten-district average as a line, and re-run.

Sharing a dashboard, and what not to put in one

Three ways to get a dashboard to the people who need it, from lightest to heaviest:

• A Shinylive link. The link in your browser's address bar, after you open an app from this page, contains the whole app. Anyone who opens it runs the app in their own browser. Good for small apps with small, non-sensitive data.
• A static Shinylive site. The shinylive package turns an app folder into plain web files you can put on any static host (GitHub Pages, Netlify, your organisation's web server). It needs a web server: opening the files straight from disk does not work.
• A Shiny server. For large data, private data, or heavy computation, run the app on a server with R or Python installed, such as Posit Connect or a server your organisation runs. The data then stays on the server.

Export an app to a static site

On your own computer, with the app saved as myapp/app.R or myapp/app.py, the commands from Posit's documentation are:

install.packages("shinylive")
shinylive::export("myapp", "site")
httpuv::runStaticServer("site/")   # preview it locally

uvx shinylive export myapp site

Sources: the shinylive R package site and the Shiny for Python Shinylive guide, both read in October 2026.

Try it: add a households column (24 for every district here) to the dashboard's data frame and show it in the table, so every bar comes with its sample size.

Where next