4 Interactive Visualizations

4.0.1 Who handles what: interactive borough–type–agency “pair” plot

To connect complaints to the agencies that act on them, we built an interactive Sankey diagram with four stages:

Borough → Detailed Complaint Type → Agency → Status

This is the plot we think of as a pair-exploration tool: users can focus on one borough–type, type–agency, or agency–status “pair” at a time.

Code

library(dplyr)
library(ggplot2)
library(lubridate)
library(stringr)
library(tidyr)
library(forcats)
library(scales)
library(tidytext)
library(ggalluvial)
library(plotly)
#| label: parcats-311-borough-color
#| echo: true
df_clean_final <- readRDS("complaint_bucket_borough.rds")

d <- df_clean_final |>
  transmute(
    borough        = fct_lump_n(factor(borough), 5, other_level = "Other"),
    complaint_bucket = fct_lump_n(factor(complaint_bucket), 10, other_level = "Other"),
    agency_name    = fct_lump_n(factor(agency_name), 10, other_level = "Other"),
    status         = fct_lump_n(factor(status), 6, other_level = "Other")
  ) |>
  tidyr::drop_na(borough, complaint_bucket, agency_name, status) |>
  filter(
    borough != "Other",
    complaint_bucket != "Other",
    agency_name != "Other",
    status != "Other"
  )

d_counts <- d |>
  count(borough, complaint_bucket, agency_name, status, name = "n") |>
  arrange(desc(n)) |>
  mutate(borough_id = as.integer(borough))

borough_levels <- levels(d_counts$borough)
K <- length(borough_levels)

pal5 <- c("#0072B2", "#D55E00", "#009E73", "#CC79A7", "#F0E442")

discrete_scale <- list(
  list(0/4, pal5[1]), list(0/4, pal5[1]),
  list(1/4, pal5[2]), list(1/4, pal5[2]),
  list(2/4, pal5[3]), list(2/4, pal5[3]),
  list(3/4, pal5[4]), list(3/4, pal5[4]),
  list(4/4, pal5[5]), list(4/4, pal5[5])
)
  
fig_parcats <- plot_ly(
  type = "parcats",
  arrangement = "freeform",
  dimensions = list(
    list(label = "Borough",        values = d_counts$borough),
    list(label = "Complaint Bucket", values = d_counts$complaint_bucket),
    list(label = "Agency",         values = d_counts$agency_name),
    list(label = "Status",         values = d_counts$status)
  ),
  counts = d_counts$n,
  line = list(
    color = d_counts$borough_id,
    colorscale = discrete_scale,
    cmin = 1, cmax = K,
    showscale = FALSE
  )
)%>%
  plotly::layout(
    title = list(
      text = "311 Complaint Flow: Borough → Type → Agency → Status (Colored by Borough)",
      x = 0.02
    ),
    margin = list(l = 40, r = 40, t = 60, b = 30)
  )|>
  layout(autosize = TRUE)

fig_parcats

Once we keep borough, detailed type (e.g. Illegal Parking, Noise – Residential), agency (NYPD, DOT, etc.), and status, the cross-tab becomes extremely high-dimensional.
A static plot would either hide most detail or be unreadable.
Interactivity (hovering to show labels and counts, highlighting a path) allows the reader to drill down without losing the full context.

How the interactivity helps

Hovering on Brooklyn → Illegal Parking reveals a very thick band flowing to NYPD, making it obvious that parking enforcement is overwhelmingly a police workload.
Highlighting Noise – Residential quickly shows which boroughs and agencies share that burden; for example, large flows from Queens and Brooklyn again route to NYPD.
Smaller but meaningful flows (for example, HEAT/HOT WATER to the housing agency) are easy to surface by hovering, even though they would be visually drowned out if everything were static.

Instead of giving one fixed conclusion, this plot serves as a navigation map for responsibility: it lets a reader answer questions like “Who usually deals with this type of complaint in my borough, and how does it tend to end?”.