How Predictive is Opening Day?

It’s Opening Day in America, the start of another MLB season. We’ve had a month of Spring Training, but now’s the time when the real games start. We love the first game, but how excited should we be if our team wins - is that a sign of a good season? If they lose, should we abandon hope?

To find out, we’ll look at all seasons from 2000 to 2022. We can find whether a team won or lost their first game, then compare to the rest of the season to see how they did.

*Technically, we should look at the rest of season winning percentage excluding opening day. But with 162 games per season (except 2020), I’m not going to worry about it too much here.

library(baseballr)
library(broom)
library(dplyr)
library(hrbrthemes)
library(ggtext) 
library(plotly)
library(RcppRoll)
library(retrosheet)
library(scales)
library(tidyverse)

get_limited_box_score <- function(season){
  getRetrosheet(type = "game", season) %>%
    as_tibble() %>%
    select(VisTm, HmTm, HmTmGNum, VisTmGNum, VisRuns, HmRuns, DayNight) 
}


rs_data <- 2000:2022 %>%
  as_tibble() %>%
  mutate(season = value) %>%
  group_by(value) %>%
  nest() %>%
  mutate(box_scores = map(data, ~get_limited_box_score(.x$season))) %>%
  select(-data) %>%
  unnest(c(box_scores)) %>%
  rename(season = value)

rs_data_vis <- rs_data %>%
  rename(team = VisTm, opp = HmTm, team_game_num = HmTmGNum, opp_game_num = VisTmGNum,
         team_runs = VisRuns, opp_runs = HmRuns) %>%
  mutate(loc = 'Visitor')

rs_data_home <- rs_data %>%
  rename(team = HmTm, opp = VisTm, team_game_num = VisTmGNum, opp_game_num = HmTmGNum,
         team_runs = HmRuns, opp_runs = VisRuns) %>%
  mutate(loc = 'Home')

rs_data_merged <- rs_data_vis %>%
  bind_rows(rs_data_home) %>%
  mutate(w = case_when(team_runs > opp_runs ~ 1, TRUE ~ 0)) %>%
  group_by(season, team) %>%
  mutate(g = n(),
         season_win_perc = mean(w),
         wins = sum(w),
         losses = g - w) 

final_win_perct <- rs_data_merged %>%
  group_by(season, team) %>%
  filter(team_game_num == max(team_game_num)) %>%
  select(season, team, season_win_perc) %>%
  rename(wp_season = season, prior_season_win_perc = season_win_perc)

rs_data_final <- rs_data_merged %>%
  inner_join(final_win_perct, by = join_by(closest(season > wp_season), team == team)) %>%
  rename(team_prior_season_win_perc = prior_season_win_perc) %>%
  inner_join(final_win_perct, by = join_by(closest(season > wp_season), opp == team)) %>%
  rename(opp_prior_season_win_perc = prior_season_win_perc) %>%
  mutate(is_favorite = case_when(team_prior_season_win_perc > opp_prior_season_win_perc ~ 'y', TRUE ~ 'n'))

rs_data_final %>%
  filter(team_game_num == 1) %>%
  mutate(won_first_game = case_when(w == 1 ~ 'Won', TRUE ~ 'Lost')) %>%
  group_by(won_first_game) %>%
  group_by(season, won_first_game) %>%
  summarise(games = sum(g),
            wins = sum(wins)) %>%
  mutate(season_win_perc = wins / games) %>%
  ggplot(aes(x = season, y = season_win_perc, color = won_first_game))+
  geom_line()+
  theme_ipsum()+
  scale_color_manual(values = c('#9e9d99', '#0c8a1d'))+
  labs(title = "Season Win Percentage",
       subtitle = "Teams that <span style = 'color:#0c8a1d'>won</span> or 
       <span style = 'color:#9e9d99'>lost</span> on Opening Day")+
  xlab('Season')+
  ylab('Win Percentage')+
  theme(
    plot.title = element_markdown(),
    plot.subtitle = element_markdown(),
    legend.position = "none"
  )+
  scale_y_continuous(labels = label_number(accuracy = 0.001))

This seems to indicate that as a whole, teams that win opening day are more likely to go on to have winning records than those that don’t, at least recently. Before 2010 it was a bit more of a mixed bag. But aggregates can hide detail - if one team had a .600 record and another had a .400, then together they played .500 ball, even though the first team is clearly better.

opening_day_boxplot <- rs_data_final %>%
  filter(team_game_num == 1) %>%
  mutate(won_first_game = case_when(w == 1 ~ 'Won', TRUE ~ 'Lost')) %>%
  ggplot(aes(x = won_first_game,, 
             y = season_win_perc, 
             color = won_first_game))+
  geom_boxplot()+
  geom_jitter(aes(alpha = 0.6, text = paste(season, team, round(season_win_perc, 3), sep = ' ')))+
  theme_ipsum()+
  scale_color_manual(values = c('#9e9d99', '#0c8a1d'))+
  labs(title = "Season Win Percentage")+
  ylab('')+
  xlab('Win Percentage')+
  theme(
    plot.title = element_markdown(),
    plot.subtitle = element_markdown(),
    legend.position = "none"
  )+
  scale_y_continuous(labels = label_number(accuracy = 0.001))

ggplotly(opening_day_boxplot, tooltip = 'text')

Seeing the 2003 Tigers and their .265 win percentage triggered some trauma, but I’m fine now. Teams that win opening day tend to do better over the season, though it’s hardly conclusive. The 2018 Red Sox lost Opening Day and finished the season with a .667 record, while the same year the Orioles won and yet finished with a .290 record.

Of course, Opening Day is just one game, and we have a full 162 to look at. So we’ll take the aggregate win % of all teams by game - i.e. over our 22 season sample, what was the final win % for teams that won or lost game 1, game 2, game 3., etc. We’ll exclude the pandemic-shortened 2020 season from our data.

rs_data_final %>%
  filter(team_game_num <= 162) %>%
  mutate(won_first_game = case_when(w == 1 ~ 'Won', TRUE ~ 'Lost')) %>%
  group_by(won_first_game) %>%
  group_by(team_game_num, won_first_game) %>%
  summarise(games = sum(g),
            wins = sum(wins)) %>%
  mutate(season_win_perc = wins / games) %>%
  ggplot(aes(x = team_game_num, y = season_win_perc, color = won_first_game))+
  geom_line()+
  theme_ipsum()+
  scale_color_manual(values = c('#9e9d99', '#0c8a1d'))+
  labs(title = "Season Win Percentage",
       subtitle = "Teams that <span style = 'color:#0c8a1d'>won</span> or 
       <span style = 'color:#9e9d99'>lost</span> a given game")+
  xlab('Season')+
  ylab('Win Percentage')+
  theme(
    plot.title = element_markdown(),
    plot.subtitle = element_markdown(),
    legend.position = "none"
  )+
  scale_y_continuous(labels = label_number(accuracy = 0.001))

Surprise - teams that win more games win more games. What’s important here is that the first game follows the same pattern as the rest of the season, so there’s nothing especially special nor predictive about it*. So enjoy the game, but however it goes, there are 161 games left, so don’t put too much on the outcome.

*Statements like this are the reason analysts don’t get invited to a lot of parties.