Skip to contents

Interactive NBA Net Rating

Saiem Gilani
@saiemgilani @saiemgilani

Source: vignettes/recipe-nba-net-rating-interactive.Rmd
recipe-nba-net-rating-interactive.Rmd
On this page

The brief: the season-review blog post wants its net-rating chart to be explorable: point at a team and read its offensive, defensive and net rating instead of estimating them from the axes. It is the interactive version of the NBA Net Rating Quadrant recipe, with the same ratings from hoopR’s ESPN team box scores. The chart is built natively with plotly’s plot_ly(): ggplotly() drops sdvplotR’s logo layers (see the interactive cookbook), so add_sdv_logos() places the logos as plotly layout images instead.

library(sdvplotR)
library(ggplot2)
library(dplyr, warn.conflicts = FALSE)
library(plotly, warn.conflicts = FALSE)

season <- 2026 # the 2025-26 season: NBA seasons are named by the year they end
label <- paste0(season - 1, "-", season %% 100)

1. Get the data

The same ratings as the static recipe: points scored and allowed per 100 possessions, with possessions estimated from the box score (FGA - OREB + TOV + 0.44 FTA) and averaged with the opponent’s, so both teams in a game share one count.

box <- hoopR::load_nba_team_box(season) |>
  filter(season_type == 2) # regular season

games <- box |>
  mutate(
    poss = field_goals_attempted - offensive_rebounds + total_turnovers +
      0.44 * free_throws_attempted
  )
filter(games, is.na(poss)) |>
  select(game_id, game_date, team_abbreviation, opponent_team_abbreviation)
#> # A tibble: 1 × 4
#>     game_id game_date  team_abbreviation opponent_team_abbreviation
#>       <int> <date>     <chr>             <chr>                     
#> 1 401810169 2025-12-01 CHI               ORL

One box score is missing a stat, so that game has no possession count. It is dropped for both teams together, as in the static recipe, so no NA reaches the season sums.

opponents <- select(games, game_id, opponent_team_id = team_id, opp_poss = poss)
stopifnot(identical(class(games$opponent_team_id), class(opponents$opponent_team_id)))

ratings <- games |>
  inner_join(opponents, by = c("game_id", "opponent_team_id")) |>
  mutate(game_poss = (poss + opp_poss) / 2) |>
  filter(!is.na(game_poss)) |>
  group_by(team = team_abbreviation) |>
  summarise(
    ortg = 100 * sum(team_score) / sum(game_poss),
    drtg = 100 * sum(opponent_team_score) / sum(game_poss)
  ) |>
  mutate(net = ortg - drtg)
stopifnot(!anyNA(ratings$net))

ESPN files the All-Star Game’s teams as regular-season games, so sdv_resolve() returns NA for them (with a warning naming the values) and they are dropped. sdv_teams() adds each team’s full name and conference for the hover text, and sdv_team_colors() its primary color.

ratings <- ratings |>
  mutate(team_id = sdv_resolve(team, league = "nba"))
filter(ratings, is.na(team_id))$team
#> [1] "STARS"   "STRIPES" "WORLD"

conferences <- sdv_teams("nba") |>
  transmute(
    team_id, team_abbr = abbr, name,
    conference = sub("ern Conference$", "", conference)
  )
ratings <- ratings |>
  filter(!is.na(team_id)) |>
  inner_join(conferences, by = "team_id") |>
  mutate(color = unname(sdv_team_colors("nba", team_abbr))) |>
  arrange(desc(net), team_abbr) |>
  mutate(tip = sprintf(
    "<b>%s</b> (%s)<br>Offensive rating: %.1f<br>Defensive rating: %.1f<br>Net rating: %+.1f",
    name, conference, ortg, drtg, net
  ))
stopifnot(nrow(ratings) == 30)
select(ratings, team_abbr, conference, ortg, drtg, net) |> head()
#> # A tibble: 6 × 5
#>   team_abbr conference  ortg  drtg   net
#>   <chr>     <chr>      <dbl> <dbl> <dbl>
#> 1 OKC       West        116.  105. 10.9 
#> 2 DET       East        114.  106.  7.93
#> 3 SA        West        117.  109.  7.86
#> 4 BOS       East        117.  109.  7.85
#> 5 NY        East        117.  110.  6.40
#> 6 HOU       West        115.  110.  5.21

The page embeds this 30-row table, not the box scores.

2. Hover first

Start with the part a reader touches. A scatter trace carries one marker per team, and text with hoverinfo = "text" replaces plotly’s default hover (raw x and y values) with the tooltip built above: name, conference and the three ratings. hoverlabel colors each tooltip in the team’s color. A range given high-to-low reverses the axis, so the best defenses are on top, as in the static recipe.

x_lim <- range(ratings$ortg) + c(-1, 1)
y_lim <- range(ratings$drtg) + c(-1, 1)

plot_ly(
  ratings, x = ~ortg, y = ~drtg, type = "scatter", mode = "markers",
  marker = list(size = 12, color = ~color),
  text = ~tip, hoverinfo = "text",
  hoverlabel = list(bgcolor = ~color, font = list(color = "white"))
) |>
  layout(
    title = list(text = paste("NBA offense vs defense,", label, "regular season"), x = 0.02),
    xaxis = list(title = "Offensive rating (points per 100 possessions)", range = x_lim),
    yaxis = list(title = "Defensive rating (allowed per 100)", range = rev(y_lim))
  )

The dots already answer “who is that?”, but a reader still has to hover each one to find a team.

3. Logos, average lines and net-rating guides

add_sdv_logos() puts each team’s logo at its point, in data units. Its size is the logo height as a share of the y-axis span. The markers stay as the hover targets, now transparent (opacity = 0), and plotly has no geom for reference lines, so they go in as layout shapes. Dashed lines mark the league-average offense and defense, and the faint diagonals are the static recipe’s net-rating guides: teams on one line share a net rating, in steps of 5. Each diagonal is cut to the axis box first.

# the line drtg = ortg - net, cut to the box x_lim by y_lim
guides <- data.frame(net = c(-10, -5, 0, 5, 10)) |>
  mutate(
    x0 = pmax(x_lim[1], y_lim[1] + net),
    x1 = pmin(x_lim[2], y_lim[2] + net)
  ) |>
  filter(x0 < x1) |>
  mutate(
    y0 = x0 - net, y1 = x1 - net,
    lx = x0 + 0.88 * (x1 - x0), ly = lx - net, # a label near the lower-right end
    label = if_else(net > 0, paste0("+", net), as.character(net))
  )

line_shape <- function(x0, y0, x1, y1, color, dash = "solid") {
  list(
    type = "line", x0 = x0, y0 = y0, x1 = x1, y1 = y1, layer = "below",
    line = list(color = color, width = 1, dash = dash)
  )
}
shapes <- c(
  Map(line_shape, guides$x0, guides$y0, guides$x1, guides$y1, "#e0e0e0"),
  list(
    line_shape(mean(ratings$ortg), y_lim[1], mean(ratings$ortg), y_lim[2], "#999999", "dash"),
    line_shape(x_lim[1], mean(ratings$drtg), x_lim[2], mean(ratings$drtg), "#999999", "dash")
  )
)
guide_labels <- lapply(seq_len(nrow(guides)), function(i) {
  list(
    x = guides$lx[i], y = guides$ly[i], text = guides$label[i],
    showarrow = FALSE, font = list(color = "#999999", size = 10)
  )
})
quadrant <- plot_ly(
  ratings, x = ~ortg, y = ~drtg, type = "scatter", mode = "markers",
  marker = list(size = 30, opacity = 0),
  text = ~tip, hoverinfo = "text",
  hoverlabel = list(bgcolor = ~color, font = list(color = "white"))
) |>
  layout(
    title = list(
      text = paste0(
        "NBA offense vs defense, ", label, " regular season",
        "<br><sup>Up and right is better. Dashed: league average. Diagonals: net rating.</sup>"
      ),
      x = 0.02
    ),
    xaxis = list(title = "Offensive rating (points per 100 possessions)", range = x_lim, zeroline = FALSE),
    yaxis = list(title = "Defensive rating (allowed per 100)", range = rev(y_lim), zeroline = FALSE),
    shapes = shapes, annotations = guide_labels,
    margin = list(t = 70)
  ) |>
  add_sdv_logos(ratings$ortg, ratings$drtg, ratings$team_abbr, "nba", size = 0.07)
quadrant

Hovering works as before, since the transparent markers sit where the logos are. plotly’s toolbar adds zoom: drag a box around the crowded middle of the chart and the logos spread apart; double-click to zoom back out.

4. Net rating as bars, logos on the axis

The quadrant shows two numbers per team; a bar chart ranks the one that sums them up. Order the teams by net rating with a factor, color each bar in its team’s color, and sdv_axis_logos() replaces the category labels with logos (it hides the tick labels and puts a logo under each bar). textposition = "none" keeps the hover text out of the bars.

bars <- ratings |>
  mutate(team_abbr = factor(team_abbr, levels = team_abbr)) # already sorted by net rating

plot_ly(
  bars, x = ~team_abbr, y = ~net, type = "bar",
  marker = list(color = ~color),
  text = ~tip, hoverinfo = "text", textposition = "none",
  hoverlabel = list(bgcolor = ~color, font = list(color = "white"))
) |>
  layout(
    title = list(text = paste("NBA net rating,", label, "regular season"), x = 0.02),
    xaxis = list(title = ""),
    yaxis = list(title = "Net rating (points per 100 possessions)"),
    margin = list(t = 50, b = 50)
  ) |>
  sdv_axis_logos(levels(bars$team_abbr), "nba")

5. Share it

htmlwidgets::saveWidget() writes a chart to one HTML file, and selfcontained = TRUE inlines plotly’s JavaScript so no folder sits beside it (it needs pandoc, which ships with RStudio). plotly’s logos are layout images that load from their URLs, so the file is small but needs the network when it is opened.

out <- tempfile("sdvplotR-recipe-")
dir.create(out)
html <- file.path(out, "nba_net_rating_interactive.html")
htmlwidgets::saveWidget(quadrant, html, selfcontained = TRUE, title = "NBA offense vs defense")
round(file.size(html) / 1024) # kilobytes, most of it plotly.js
#> [1] 3705

6. The static version

plotly has no ggsave(), and a screenshot of the widget needs a headless browser (the cookbook’s last recipe shows it with webshot2). For a post or a slide, draw the static recipe’s ggplot instead, from the same ratings and guides tables: geom_sdv_logos() for the logos, geom_mean_lines() for each conference’s averages, and a panel per conference.

ggplot(ratings, aes(ortg, drtg)) +
  geom_segment(
    aes(x = x0, y = y0, xend = x1, yend = y1),
    data = guides, inherit.aes = FALSE, color = "grey88", linewidth = 0.4
  ) +
  geom_text(
    aes(x = lx, y = ly, label = label),
    data = guides, inherit.aes = FALSE, color = "grey60", size = 2.6
  ) +
  geom_mean_lines(aes(x0 = ortg, y0 = drtg), color = "grey60", linewidth = 0.5) +
  geom_sdv_logos(aes(team = team_abbr), league = "nba", width = 0.1) +
  facet_wrap(vars(conference)) +
  scale_x_continuous(expand = expansion(mult = 0.06)) +
  scale_y_reverse(expand = expansion(mult = 0.08)) +
  labs(
    title = paste("NBA offense vs defense,", label, "regular season"),
    subtitle = "Up and right is better. Dashed: conference average. Diagonals: net rating.",
    x = "Offensive rating (points per 100 possessions)",
    y = "Defensive rating (allowed per 100)",
    caption = "Data: ESPN via hoopR  |  Chart: sdvplotR"
  ) +
  theme_minimal(base_size = 10) +
  theme(
    plot.title.position = "plot",
    plot.title = element_text(face = "bold", size = 13),
    plot.subtitle = element_text(color = "grey40", size = 9),
    plot.caption = element_text(color = "grey45", size = 7.5),
    strip.text = element_text(face = "bold", size = 11, hjust = 0),
    panel.grid.minor = element_blank()
  )

Static two-panel chart of NBA offensive against defensive rating by conference, 2025-26, with team logos, dashed conference-average lines, faint diagonals marking net ratings from -10 to +10, and the defensive axis reversed so up and right is better.

Our Authors

Citation

To cite the sdvplotR R package in publications, use:

BibTeX Citation

@misc{gilani_carl_lee_kay_weatherman_sdvplotR,
  author = {Gilani, Saiem and Carl, Sebastian and Lee, Jared and Kay, Camden and Weatherman, Andrew},
  title = {sdvplotR: The SportsDataverse's R Package for Multi-League Sports Plotting.},
  url = {https://sdvplotR.sportsdataverse.org},
  year = {2026}
}