Live Demos

Small Shiny applications running entirely in your browser — no server, powered by WebAssembly.

These are real Shiny apps, compiled to WebAssembly with Shinylive and served as static files — the same technology stack we use for client dashboards, minus the server. The first load downloads the R runtime (~15 MB), so give it a few seconds.

Interactive statistics

1. Two-group comparison simulator

Move the sliders and watch the t-test react. Notice how small samples make even a real effect easy to miss.

#| '!! shinylive warning !!': |
#|   shinylive does not work in self-contained HTML documents.
#|   Please set `embed-resources: false` in your metadata.
#| standalone: true
#| viewerHeight: 560

library(shiny)

ui <- fluidPage(
  tags$style(HTML("
    h4 { color:#1b2a4a; font-weight:700; }
    .irs-bar, .irs-handle { border-color:#2f6fed; background:#2f6fed; }
  ")),
  h4("Simulate a two-group study"),
  fluidRow(
    column(4, sliderInput("n", "Participants per group", 5, 100, 20, step = 5)),
    column(4, sliderInput("d", "True effect size (Cohen's d)", 0, 1.5, 0.5, step = 0.1)),
    column(4, br(), actionButton("resample", "Draw a new sample", class = "btn-primary"))
  ),
  plotOutput("plot", height = "320px"),
  htmlOutput("result")
)

server <- function(input, output, session) {
  dat <- reactive({
    input$resample
    n <- input$n
    data.frame(
      group = rep(c("Control", "Treatment"), each = n),
      score = c(rnorm(n, 0, 1), rnorm(n, input$d, 1))
    )
  })

  output$plot <- renderPlot({
    d <- dat()
    par(mar = c(4, 4, 1, 1), family = "sans")
    boxplot(score ~ group, data = d, col = c("#2f6fed22", "#17a2b822"),
            border = "#1b2a4a", outline = FALSE,
            xlab = "", ylab = "Score", frame = FALSE)
    pts <- jitter(as.numeric(factor(d$group)), amount = 0.12)
    points(pts, d$score, pch = 19, cex = 0.8,
           col = ifelse(d$group == "Control", "#2f6fed88", "#17a2b888"))
  })

  output$result <- renderUI({
    d <- dat()
    tt <- t.test(score ~ group, data = d)
    sds <- tapply(d$score, d$group, sd)
    ms  <- tapply(d$score, d$group, mean)
    d_obs <- (ms[2] - ms[1]) / sqrt(mean(sds^2))
    sig <- tt$p.value < 0.05
    HTML(sprintf(
      "<p style='font-size:1.05rem'>t(%.1f) = %.2f, p = %s, observed d = %.2f — <b style='color:%s'>%s</b></p>",
      tt$parameter, -tt$statistic,
      ifelse(tt$p.value < .001, "&lt; .001", sprintf("%.3f", tt$p.value)),
      d_obs,
      ifelse(sig, "#17a2b8", "#b02a37"),
      ifelse(sig, "statistically significant at α = .05",
                  "not statistically significant at α = .05")
    ))
  })
}

shinyApp(ui, server)

2. Power curve explorer

The planning question every study should start with: how many participants do I need?

#| '!! shinylive warning !!': |
#|   shinylive does not work in self-contained HTML documents.
#|   Please set `embed-resources: false` in your metadata.
#| standalone: true
#| viewerHeight: 540

library(shiny)

ui <- fluidPage(
  tags$style(HTML("h4 { color:#1b2a4a; font-weight:700; }")),
  h4("Power for a two-group t-test"),
  fluidRow(
    column(4, sliderInput("d", "Assumed effect size (d)", 0.1, 1.2, 0.5, step = 0.05)),
    column(4, selectInput("alpha", "Significance level α",
                          c("0.05" = 0.05, "0.01" = 0.01), selected = 0.05)),
    column(4, sliderInput("target", "Target power", 0.5, 0.95, 0.8, step = 0.05))
  ),
  plotOutput("curve", height = "330px"),
  htmlOutput("answer")
)

server <- function(input, output, session) {
  n_needed <- reactive({
    ceiling(power.t.test(delta = input$d, sig.level = as.numeric(input$alpha),
                         power = input$target)$n)
  })

  output$curve <- renderPlot({
    ns <- seq(5, 200, by = 1)
    pw <- sapply(ns, function(n)
      power.t.test(n = n, delta = input$d,
                   sig.level = as.numeric(input$alpha))$power)
    par(mar = c(4.5, 4.5, 1, 1), family = "sans")
    plot(ns, pw, type = "l", lwd = 3, col = "#2f6fed", ylim = c(0, 1),
         xlab = "Participants per group", ylab = "Power", frame = FALSE)
    abline(h = input$target, lty = 2, col = "#1b2a4a")
    abline(v = n_needed(), lty = 2, col = "#17a2b8")
    points(n_needed(), input$target, pch = 19, cex = 1.4, col = "#17a2b8")
  })

  output$answer <- renderUI({
    HTML(sprintf(
      "<p style='font-size:1.05rem'>To detect <b>d = %.2f</b> with %.0f%% power at α = %s you need ≈ <b style='color:#17a2b8'>%d participants per group</b> (%d in total).</p>",
      input$d, input$target * 100, input$alpha, n_needed(), 2 * n_needed()
    ))
  })
}

shinyApp(ui, server)

3. What “95% confidence” actually means

Each vertical line is a confidence interval from a fresh sample of the same population. Intervals that miss the true mean (dashed line) are shown in red — in the long run, about 5% of them at the 95% level.

#| '!! shinylive warning !!': |
#|   shinylive does not work in self-contained HTML documents.
#|   Please set `embed-resources: false` in your metadata.
#| standalone: true
#| viewerHeight: 540

library(shiny)

ui <- fluidPage(
  tags$style(HTML("h4 { color:#1b2a4a; font-weight:700; }")),
  h4("100 confidence intervals, 100 fresh samples"),
  fluidRow(
    column(4, sliderInput("n", "Sample size per study", 5, 100, 20, step = 5)),
    column(4, sliderInput("conf", "Confidence level", 0.80, 0.99, 0.95, step = 0.01)),
    column(4, br(), actionButton("go", "Resample all 100", class = "btn-primary"))
  ),
  plotOutput("cis", height = "330px"),
  htmlOutput("coverage")
)

server <- function(input, output, session) {
  sims <- reactive({
    input$go
    t(replicate(100, {
      x <- rnorm(input$n, mean = 50, sd = 10)
      tt <- t.test(x, conf.level = input$conf)
      c(tt$conf.int[1], tt$conf.int[2])
    }))
  })

  output$cis <- renderPlot({
    ci <- sims()
    miss <- ci[, 1] > 50 | ci[, 2] < 50
    par(mar = c(4, 4.5, 1, 1), family = "sans")
    plot(NA, xlim = c(1, 100), ylim = range(ci),
         xlab = "Study number", ylab = "Estimated mean", frame = FALSE)
    segments(1:100, ci[, 1], 1:100, ci[, 2],
             col = ifelse(miss, "#b02a37", "#2f6fed88"), lwd = 2)
    abline(h = 50, lty = 2, lwd = 2, col = "#1b2a4a")
  })

  output$coverage <- renderUI({
    ci <- sims()
    hits <- mean(ci[, 1] <= 50 & ci[, 2] >= 50)
    HTML(sprintf(
      "<p style='font-size:1.05rem'>This round, <b style='color:#17a2b8'>%.0f%%</b> of the %.0f%% intervals captured the true mean. Resample to see the long-run behaviour.</p>",
      hits * 100, input$conf * 100
    ))
  })
}

shinyApp(ui, server)

4. Diagnostic test calculator

A test’s sensitivity and specificity are fixed, but its predictive values are not — they swing with how common the disease is. Set the test’s properties and the prevalence, and watch the positive and negative predictive values respond.

#| '!! shinylive warning !!': |
#|   shinylive does not work in self-contained HTML documents.
#|   Please set `embed-resources: false` in your metadata.
#| standalone: true
#| viewerHeight: 560

library(shiny)

ui <- fluidPage(
  tags$style(HTML("h4 { color:#1b2a4a; font-weight:700; }
    .irs-bar, .irs-handle { border-color:#2f6fed; background:#2f6fed; }")),
  h4("Predictive values vs. disease prevalence"),
  fluidRow(
    column(4, sliderInput("sens", "Sensitivity", 0.50, 0.999, 0.90, step = 0.01)),
    column(4, sliderInput("spec", "Specificity", 0.50, 0.999, 0.95, step = 0.01)),
    column(4, sliderInput("prev", "Disease prevalence", 0.001, 0.50, 0.10, step = 0.005))
  ),
  plotOutput("curve", height = "320px"),
  htmlOutput("result")
)

server <- function(input, output, session) {
  ppv <- function(sens, spec, p) (sens * p) / (sens * p + (1 - spec) * (1 - p))
  npv <- function(sens, spec, p) (spec * (1 - p)) / (spec * (1 - p) + (1 - sens) * p)

  output$curve <- renderPlot({
    ps <- seq(0.001, 0.6, length.out = 200)
    par(mar = c(4.5, 4.5, 1, 1), family = "sans")
    plot(ps, ppv(input$sens, input$spec, ps), type = "l", lwd = 3,
         col = "#2f6fed", ylim = c(0, 1), xlab = "Disease prevalence",
         ylab = "Predictive value", frame = FALSE)
    lines(ps, npv(input$sens, input$spec, ps), lwd = 3, col = "#17a2b8")
    abline(v = input$prev, lty = 2, col = "#1b2a4a")
    legend("right", c("PPV", "NPV"), col = c("#2f6fed", "#17a2b8"),
           lwd = 3, bty = "n")
  })

  output$result <- renderUI({
    v_ppv <- ppv(input$sens, input$spec, input$prev)
    v_npv <- npv(input$sens, input$spec, input$prev)
    lr_pos <- input$sens / (1 - input$spec)
    lr_neg <- (1 - input$sens) / input$spec
    HTML(sprintf(
      "<p style='font-size:1.05rem'>At <b>%.1f%%</b> prevalence: PPV = <b style='color:#2f6fed'>%.1f%%</b>, NPV = <b style='color:#17a2b8'>%.1f%%</b>. Likelihood ratios: LR+ = %.1f, LR&minus; = %.2f.</p>",
      input$prev * 100, v_ppv * 100, v_npv * 100, lr_pos, lr_neg
    ))
  })
}

shinyApp(ui, server)

---

Want a dashboard like this wired to your own data — with a real server, authentication and automated updates? That’s our Shiny practice.