In murrayjw/nbaWinLossExplorer: NBA Win Loss Explorer

knitr::opts_chunk$set(echo = TRUE)

Data for analysis

This project uses the results of all NBA seasons from 1948-2021. We use the nbastatR package, and in particular the game_logs() function to extract the results of every game in that period. We also extract playoff results to determine which teams actually made the playoffs.

Below I extract and save the raw data using nbastatR.

# load libraries ----------------------------------------------------------
library(tidyverse)
library(nbastatR)
library(future)
library(janitor)

# the seasons we want to extract data for
seasons <- 1948:2020

# for parallel processing
plan(multiprocess)

nba_season_data <- nbastatR::game_logs(seasons = seasons,
                             result_types = "team",
                             season_types = "Regular Season",
                             return_message = F)
nba_playoff_data <- nbastatR::game_logs(seasons = seasons,
                              result_types = "team",
                              season_types = "Playoffs",
                              return_message = F)

# save the raw data

save(nba_season_data, file = here::here("raw-data", "nba_season_data.Rda"))
save(nba_playoff_data, file = here::here("raw-data", "nba_playoff_data.Rda"))

Next we do some data cleaning and summarizing. First we use the janitor package to clean up the variable names. Then we summarize the playoff game logs to determine which teams made the playoffs.

For the regular season game logs, we:

• create a game number variable
• create a cumulative wins variable
• create a cumulative loss variable
• create a current record variable (format W-L).

We then do a right join with the playoff data. If a team has no playoff data, then they did not make the playoffs (setting made_playoffs = 0).

Finally we create summary variables for team and season: - id: paste together the team name and season - final_record: the team's final win loss record at the end of the season - final_wins: the team's total win count at the end of the season - final_losses: the team's total loss count at the end of the season - final_pm : the plus minus differential at the end of the season. - plus_minus_pre: the cumulative plus minus differential prior to the game result represented in the current row of the data. - plus_minus_post: the cumulative plus minus differential after the game result represented in the current row of the data.

The data is saved as a .csv file.

# clean var names
nba_season_data <- janitor::clean_names(nba_season_data)
nba_playoff_data <- janitor::clean_names(nba_playoff_data)

# which teams made the playoffs each season
nba_made_playoffs <- nba_playoff_data %>%
  distinct(year_season, id_team, name_team) %>%
  mutate(made_playoffs = 1)

# add some additional variables
nba_season_data <- nba_season_data%>%
  group_by(year_season, name_team) %>%
  arrange(date_game) %>%
  mutate(game_number = 1:n(),
         wins = cumsum(is_win),
         losses = game_number - wins,
         record = paste0(wins, "-", losses)) %>%
  select(name_team, year_season, slug_season,
         number_game_team_season, date_game,
         id_team, location_game,
         slug_matchup,
         url_team_season_logo,
         wins, losses, record, plus_minus = plusminus_team) %>%
  ungroup()


# join season data with playoffs data and create playoff indicator
nba_season_data <- nba_season_data %>%
  left_join(nba_made_playoffs, by = c("name_team", "year_season", "id_team")) %>%
  mutate(made_playoffs = ifelse(is.na(made_playoffs), 0, 1))


# add some season long information into the data
nba_season_data <- nba_season_data %>%
  group_by(id_team, year_season) %>%
  arrange(date_game) %>%
  mutate(id= paste(name_team, year_season), 
         final_record = record[n()],
         final_wins = wins[n()],
         final_losses = losses[n()],
         final_pm = sum(plus_minus),
         plus_minus_pre = cumsum(plus_minus) - plus_minus,
         plus_minus_post = cumsum(plus_minus)) %>%
  ungroup()

# write as .csv file
readr::write_csv(nba_season_data, here::here('data', 'nba_season_data.csv'))
usethis::use_data(nba_season_data)

Data Exploration

Here we explore the data a little, to get a better understanding of what is available.

First a look at the first observations:

DT::datatable(head(nba_season_data), options=list(scrollX=T))

We can plot the total number of wins agains the game number of the season for each team and season.

season_plot <- nba_season_data %>%
            ggplot(aes(number_game_team_season, wins, color = id)) +
            geom_line() +
            labs(x = "Game number of season",
                 y = "Number of wins")  +
            theme_minimal()+
            theme(legend.position='none') +
            theme(axis.text.x = element_text(colour = "#006bb8"),
                  axis.text.y = element_text(colour = "#006bb8"),
                  axis.title.x = element_text(colour = "#032f4f"),
                  axis.title.y = element_text(colour = "#032f4f"))

season_plotly <- plotly::ggplotly(season_plot)
season_plotly

The team with the best record of all time are the 2016 Golden State Warriors (73 wins), while the team with the worst record are the 1973 Philadelphia 76ers with 9 wins.

Results to end of year

What we are ultimately interested in, is cutting off the a season at some mark (say the 10 game mark), and determining likely outcomes for teams based on

1. Their record at the time
2. Their plus minus differential at that time.

The reason for including plus minus is because not all bad teams are the same. Some teams catch a lot of bad breaks and end up with a losing record despite scoring as many or more points than their opponents over the course of the season (see 2021 Raptors).

Let's look at one example. Let's look at all teams that started their season with 5 wins and 5 loses, with a plus minus between -10 and + 10.

 acceptable_records <- nba_season_data %>%
            mutate(contains_record = (wins == 5) &
                       (losses == 5) &
                       (plus_minus_post >= -10 ) &
                       (plus_minus_post <= 10)) %>%
            group_by(name_team, year_season) %>%
            arrange(date_game) %>%
            mutate(keep = max(contains_record)) %>%
            filter(keep > 0) %>%
            ungroup()

Now let's recreate the same plot as above

example_season_plot <- acceptable_records %>%
            ggplot(aes(number_game_team_season, wins, color = id)) +
            geom_line() +
            labs(x = "Game number of season",
                 y = "Number of wins")  +
            theme_minimal()+
            theme(legend.position='none') +
            theme(axis.text.x = element_text(colour = "#006bb8"),
                  axis.text.y = element_text(colour = "#006bb8"),
                  axis.title.x = element_text(colour = "#032f4f"),
                  axis.title.y = element_text(colour = "#032f4f"))

example_season_plotly <- plotly::ggplotly(example_season_plot)
example_season_plotly

We can see the lines cross where every team was 5-5. Now we can calculate some sample statistics.

season_summary <- acceptable_records %>% 
  group_by(id) %>% 
  slice(1) %>% 
  ungroup()

# how many teams match the requirements
nrow(season_summary)

# what proportion make the playoffs
sum(season_summary$made_playoffs)/nrow(season_summary)

# distribution of wins and losses at the end of the season
summary(season_summary$final_wins)
summary(season_summary$final_losses)

We can see that r nrow(season_summary) teams match these requirements, with 65.8% making the playoffs. The median number of wins is 40 (IQR 34.5-48.0) and median number of losses of 39 (33.0-45.0). So on average, teams finish with roughly the same number of wins and losses.

Let's make a small change, where we look at teams starting with a 5-5 record, with a plus minus of 20 or greater. So in this case, teams that have scored many more points than their opponents.

 acceptable_records <- nba_season_data %>%
            mutate(contains_record = (wins == 5) &
                       (losses == 5) &
                       (plus_minus_post >= 20)) %>%
            group_by(name_team, year_season) %>%
            arrange(date_game) %>%
            mutate(keep = max(contains_record)) %>%
            filter(keep > 0) %>%
            ungroup()

example_season_plot <- acceptable_records %>%
            ggplot(aes(number_game_team_season, wins, color = id)) +
            geom_line() +
            labs(x = "Game number of season",
                 y = "Number of wins")  +
            theme_minimal()+
            theme(legend.position='none') +
            theme(axis.text.x = element_text(colour = "#006bb8"),
                  axis.text.y = element_text(colour = "#006bb8"),
                  axis.title.x = element_text(colour = "#032f4f"),
                  axis.title.y = element_text(colour = "#032f4f"))

example_season_plotly <- plotly::ggplotly(example_season_plot)
example_season_plotly

We can see the lines cross where every team was 5-5. Now we can calculate some sample statistics.

season_summary <- acceptable_records %>% 
  group_by(id) %>% 
  slice(1) %>% 
  ungroup()

# how many teams match the requirements
nrow(season_summary)

# what proportion make the playoffs
sum(season_summary$made_playoffs)/nrow(season_summary)

# distribution of wins and losses at the end of the season
summary(season_summary$final_wins)
summary(season_summary$final_losses)

In this case we get 69 records, with 75.3% making the playoffs, median number of wins of 44 (IQR 39.0 - 49.0) and median number of losses of 36 (IQR 30.0-41.0). So we found a subset of teams that have the same starting record, but a higher end of season expectation.

Using the simple method above we can build a simple shiny application that allows users to play with those simple inputs and explore how historical teams have performed.

murrayjw/nbaWinLossExplorer documentation built on Dec. 21, 2021, 11:02 p.m.