Formula 1 Visualizations (2000-2021)
N.Dzielawa
10/15/2021
Introduction
Formula 1 has been a beloved sport to its loyal fan base for over 70 years now, but the sport is considered one of the most data-driven professional sports in the world. Each race team has a data team that comes with them to each race weekend, and one that stays back at their respective factories. They are constantly combing through the thousands of data points that a car returns every lap and then meeting with their team’s principal, engineers, and drivers to determine the greatest strategy for every race weekend. Along with this, Formula 1 and their broadcast partners in recent years teamed up with AWS to deliver insights and predictions during the race broadcast. However, one of the greatest periods in the history of the sport is from 2000-2021 did not have data for the public to see until a few years ago.
There is plenty of data but little insight into the powerhouses of the sport so fans can understand how one one-hundredth of a second can make a difference in your outcome. The success of the Netflix docuseries Formula 1: Drive to Survive and the documentary Schumacher based on the career of the legendary Michael Schumacher have attracted a large new fan base to the sport. As a lifelong F1 fan myself, I wanted to take a look back at how the sport has evolved over the past twenty years and look at the teams, drivers, and car performance to see how they have evolved. This is because, for the upcoming 2022 season, there is a brand-new car design and new rules that will make the sport a competitive field for all teams that are racing. Since the end of what has been one of if not the greatest era of Formula 1 yet, I wanted to take a look back and see what made it truly remarkable but also see how the sport can improve. I hope you all enjoy it!
Dataset
For my data set, I have chosen to use the Ergast Developer API’s(http://ergast.com/mrd/db/) data files that have data that look at Formula 1 from 1950-the present. This data set came as multiple .csv files located in a zip file that would have to be joined together when needed for each visualization. When looking at the structure of the data files, each has one or multiple possible connections that can be made to connect different sheets. For this reason, I plan on using sqldf in building out the data frames for each visualization.
The first table that I used in the data set is the driver’s file. In this file, you will have the drivers which will be used to link to other data sets, the driver’s reference which is the driver’s last name. These two columns will be used for my wins per driver visualization.
summary(df_drivers)## driverId driverRef number code
## Min. : 1.0 Length:853 Length:853 Length:853
## 1st Qu.:214.0 Class :character Class :character Class :character
## Median :427.0 Mode :character Mode :character Mode :character
## Mean :427.1
## 3rd Qu.:640.0
## Max. :854.0
## forename surname nationality
## Length:853 Length:853 Length:853
## Class :character Class :character Class :character
## Mode :character Mode :character Mode :character
##
##
## The next data set that I used was the races data file from Ergast. In this data set, I used the raceId to link this table to others as this is the key for this table. Next, I used the year column to help determine which year the races were held to help filter out the thousands of rows of data and look at a specific period. Finally, I used the circuitId as a foreign key to connect to the circuits table which I will talk about next. This table will be used in all of my visualizations to help understand the years that I am looking at and look at results for those races. One Modification that was made in this data frame was to set the time column to not appear as it was creating an issue with joining other tables with the same column name in it.
summary (df_races)## raceId year round circuitId
## Min. : 1.0 Min. :1950 Min. : 1.000 Min. : 1.00
## 1st Qu.: 265.0 1st Qu.:1976 1st Qu.: 4.000 1st Qu.: 9.00
## Median : 529.0 Median :1992 Median : 8.000 Median :18.00
## Mean : 530.7 Mean :1991 Mean : 8.369 Mean :22.19
## 3rd Qu.: 793.0 3rd Qu.:2008 3rd Qu.:12.000 3rd Qu.:32.00
## Max. :1073.0 Max. :2021 Max. :22.000 Max. :77.00
## Grand_Prix_name time
## Length:1057 Length:1057
## Class :character Class :character
## Mode :character Mode :character
##
##
## The circuits data set is showing the information regarding each race track that has been used in Formula 1. I will be using the circuitId number to connect my race data set, and the circuit name to help show what racetracks team are the most successful at.
summary(df_circuits)## circuitId circuitRef track_name location
## Min. : 1 Length:77 Length:77 Length:77
## 1st Qu.:20 Class :character Class :character Class :character
## Median :39 Mode :character Mode :character Mode :character
## Mean :39
## 3rd Qu.:58
## Max. :77
## country lat lng alt
## Length:77 Min. :-37.85 Min. :-118.189 Min. : -7.0
## Class :character 1st Qu.: 33.45 1st Qu.: -9.203 1st Qu.: 18.0
## Mode :character Median : 41.17 Median : 4.327 Median : 129.0
## Mean : 33.72 Mean : 3.551 Mean : 247.5
## 3rd Qu.: 47.20 3rd Qu.: 19.249 3rd Qu.: 332.0
## Max. : 57.27 Max. : 144.968 Max. :2227.0
## url
## Length:77
## Class :character
## Mode :character
##
##
## The constructor’s data gives us the constructorId and the constructor_name which will be used to connect with other tables to determine how successful the teams have been since 2000. One note about this data frame has been renamed to df_teams as a way to simplify the naming convention.
summary(df_teams)## constructorId constructorRef constructors_name nationality
## Min. : 1.0 Length:211 Length:211 Length:211
## 1st Qu.: 54.5 Class :character Class :character Class :character
## Median :107.0 Mode :character Mode :character Mode :character
## Mean :107.0
## 3rd Qu.:159.5
## Max. :214.0The pitstop data in the dataset uses the raceId to tell the data what race the pitstop came from. I will be using that and the duration of the pitstop to determine the average pitstop time per year.
summary(df_pit_stops)## raceId driverId stop lap
## Min. : 841.0 Min. : 1.0 Min. :1.000 Min. : 1.00
## 1st Qu.: 881.0 1st Qu.: 17.0 1st Qu.:1.000 1st Qu.:13.00
## Median : 937.0 Median :813.0 Median :2.000 Median :25.00
## Mean : 939.9 Mean :487.2 Mean :1.768 Mean :25.13
## 3rd Qu.: 993.0 3rd Qu.:826.0 3rd Qu.:2.000 3rd Qu.:35.00
## Max. :1065.0 Max. :854.0 Max. :6.000 Max. :78.00
## time duration milliseconds
## Length:8518 Length:8518 Min. : 12897
## Class :character Class :character 1st Qu.: 21896
## Mode :character Mode :character Median : 23492
## Mean : 65200
## 3rd Qu.: 26055
## Max. :2077164Next, the qualifying data set looks at the qualifying time set by a driver at each race. For this dataset, I used the raceId column to connect to other data sets to connect my qualifying data to a race, the constructor id to connect to a certain team name, and the q1 column which gives the best lap time given by each driver during the first of three qualifying stages.
summary(df_qualifying)## qualifyId raceId driverId constructorId
## Min. : 1 Min. : 1.0 Min. : 1.0 Min. : 1.00
## 1st Qu.:2245 1st Qu.: 109.0 1st Qu.: 15.0 1st Qu.: 4.00
## Median :4488 Median : 353.0 Median : 41.0 Median : 9.00
## Mean :4495 Mean : 543.3 Mean :276.6 Mean : 40.04
## 3rd Qu.:6752 3rd Qu.: 951.0 3rd Qu.:815.0 3rd Qu.: 22.00
## Max. :9013 Max. :1065.0 Max. :854.0 Max. :214.00
## number grid_position q1 q2
## Min. : 0.00 Min. : 1.00 Length:8973 Length:8973
## 1st Qu.: 7.00 1st Qu.: 6.00 Class :character Class :character
## Median :14.00 Median :11.00 Mode :character Mode :character
## Mean :17.33 Mean :11.31
## 3rd Qu.:21.00 3rd Qu.:17.00
## Max. :99.00 Max. :28.00
## q3
## Length:8973
## Class :character
## Mode :character
##
##
## The results table shows us where a driver has started in a race and subsequently where they have finished. This table acts as almost a master table since it houses constructorsId,raceId, and driverId which are all crucial in understanding the performance of a driver and team in a race. I have used these to help connect other tables using sqldf for my visualizations, as well as position and grid position to help show the starting grid placement plays a major role in a driver’s chance to win. One note about this table is that I made the time column set to NULL as it was not needed and was interfering with other tables with the same name.
summary(df_results)## resultId raceId driverId constructorId
## Min. : 1 Min. : 1.0 Min. : 1.0 Min. : 1.00
## 1st Qu.: 6311 1st Qu.: 287.8 1st Qu.: 56.0 1st Qu.: 6.00
## Median :12620 Median : 503.0 Median :158.0 Median : 25.00
## Mean :12621 Mean : 518.4 Mean :251.2 Mean : 47.51
## 3rd Qu.:18930 3rd Qu.: 763.0 3rd Qu.:347.0 3rd Qu.: 58.00
## Max. :25245 Max. :1065.0 Max. :854.0 Max. :214.00
## number grid position positionText
## Length:25240 Min. : 0.0 Length:25240 Length:25240
## Class :character 1st Qu.: 5.0 Class :character Class :character
## Mode :character Median :11.0 Mode :character Mode :character
## Mean :11.2
## 3rd Qu.:17.0
## Max. :34.0
## positionOrder points laps milliseconds
## Min. : 1.00 Min. : 0.000 Min. : 0.00 Length:25240
## 1st Qu.: 6.00 1st Qu.: 0.000 1st Qu.: 21.00 Class :character
## Median :12.00 Median : 0.000 Median : 52.00 Mode :character
## Mean :12.93 Mean : 1.801 Mean : 45.79
## 3rd Qu.:19.00 3rd Qu.: 2.000 3rd Qu.: 66.00
## Max. :39.00 Max. :50.000 Max. :200.00
## fastestLap rank fastestLapTime fastestLapSpeed
## Length:25240 Length:25240 Length:25240 Length:25240
## Class :character Class :character Class :character Class :character
## Mode :character Mode :character Mode :character Mode :character
##
##
##
## statusId
## Min. : 1.00
## 1st Qu.: 1.00
## Median : 11.00
## Mean : 17.72
## 3rd Qu.: 14.00
## Max. :139.00Finally, I used the lap_times data to help look at a team’s lap times in a race. I used this table to join the column raceId in hopes to look at average lap times in a race versus average qualifying lap time for several teams but I was not able to due to a lack of ability to join the constructor’s table.
summary(df_lap_times)## raceId driverId lap position
## Min. : 1.0 Min. : 1.0 Min. : 1.00 Min. : 1.000
## 1st Qu.: 120.0 1st Qu.: 15.0 1st Qu.:14.00 1st Qu.: 5.000
## Median : 343.0 Median : 33.0 Median :29.00 Median : 9.000
## Mean : 517.4 Mean :258.8 Mean :29.97 Mean : 9.646
## 3rd Qu.: 943.0 3rd Qu.:813.0 3rd Qu.:44.00 3rd Qu.:14.000
## Max. :1065.0 Max. :854.0 Max. :87.00 Max. :24.000
## time milliseconds
## Length:505806 Min. : 55404
## Class :character 1st Qu.: 82018
## Mode :character Median : 90606
## Mean : 95557
## 3rd Qu.: 102389
## Max. :7507547Findings
Wins Per Team on Each Track since 2000
df_team_race <- sqldf(
"select *
from df_teams inner join df_results using (constructorId)"
)
df_team_race <- sqldf("select *
from df_team_race inner join df_races using (raceId)")
df_team_race <- sqldf("select *
from df_team_race inner join df_circuits using (circuitId)")
df_team_race_win <- sqldf("select constructors_name, count(position) as 'Wins',track_name
from df_team_race
where year >= 2000
and position == 1
group by track_name, constructors_name")
getPalette = colorRampPalette(brewer.pal(8, "Set2"))
team_race_win <- ggplot(df_team_race_win, aes( x = Wins, y= track_name, fill = constructors_name))+
geom_bar(stat= "identity", position = position_stack(reverse = TRUE)) +
geom_text(aes(x = Wins,
y = track_name,label = Wins),
position = position_stack(vjust=0.5,reverse=TRUE),
size = 4) +
labs(title = "Wins per Team by Track Since 2000", x= "Wins", y = "Track Name", fill = "Team Name")+
theme_light() +
theme(plot.title = element_text(hjust=0.5)) +
scale_fill_manual(values = getPalette(length(unique(df_team_race_win$constructors_name))))
team_race_win
This stacked bar chart is looking at Wins per Team by Track Since 2000. This graph shows us not only what teams are successful at which track but also how many times a race has been held there. The dominance of the Ferrari F1 team is easily shown in the blue color. This is due to their exceptional cars and driver pairings that are brought out each year. One reason in particular for their success is thanks to Michael Schumacher and his amazing time at Ferrari in the early to late 2000s. Another team that has shown great success is McLaren due to their driver pairing of now seven-time world champion Lewis Hamilton and Jenson Button. Looking at the other end of the spectrum the teams with fewer wins in total and by track are most likely because they are the “minor league’ teams that have a smaller budget which means less experienced drivers, and a less successful R&D department.
Wins per Driver since 2000
df_race_year <- sqldf("select*
from df_races inner join df_results using (raceId)")
df_wins_per_driver <- sqldf("Select driverRef as 'Driver Name', count(position) as 'Wins'
from df_drivers inner join df_race_year using (driverId)
where position == 1
and year >= 2000
group by driverRef
ORDER BY count(position) DESC, driverRef")
ggplot(df_wins_per_driver, aes(x= Wins, y= reorder(`Driver Name`,Wins)))+
geom_bar(stat= "identity", position = position_stack(reverse = FALSE),color= "darkgreen", fill = "lightgreen")+
labs(title = "Wins per Driver Since 2000", x= "Wins", y = "Drivers Last Name")+
theme_light()+
theme(plot.title = element_text(hjust = 0.5))+
geom_text(data = df_wins_per_driver, aes(x= round(Wins,digits = 0), y = `Driver Name`, label = Wins, fill =NULL), hjust = -0.1,size = 4)
This bar chart is looking at Career wins per driver since 2000, and you can see that is skewed in one direction and highlights the best drivers of the last 20 years. The skewed this way because both Lewis Hamilton and Michael Schumacher are considered the greatest drivers in F1 history. The drivers that have more wins than Max Verstappen have another thing in common; they were all at one time driving for Ferrari, Mercedes, or McLaren.
Average Pit Stop Time
avg_pit_stop<- sqldf("select year, avg(stop) as 'Average Pit stops per Race', avg(duration) as 'Average Pit Stop Time'
from df_pit_stops inner join df_races using (raceId)
where year >= 2011
group by Year")
x_axis_years <- min(avg_pit_stop$year): max(avg_pit_stop$year)
pit_stop <- ggplot(data = avg_pit_stop,aes(x = year, y = `Average Pit Stop Time`))+
geom_line(color = 'black', size = 1.5)+
geom_point(shape = 21, size = 3, color= 'white', fill = 'red')+
labs(title = "Average Pit Stop time (2011-2021)", x= "Years",
y = "Average Pit Stop Time
(Seconds)")+
theme_light()+
theme(plot.title = element_text(hjust = 0.5))+
scale_x_continuous(labels = x_axis_years, breaks = x_axis_years)+
geom_label_repel(aes(label= round(`Average Pit Stop Time`,digits =2)),
box.padding = 1,
point.padding = 1,
size =3,
color ='black',
segment.color = 'red')
pit_stop
The timing of a pit stop is crucial in F1. This is because teams will use it to execute what is known as an “undercut” to get ahead of them before they are expected to pit. If the undercut is performed perfectly it can change the tide in a race, but one small mistake from your pit crew can derail that. The average pitstop time that is measured here is the time from when the car enters the pit lane until it exits the pit lane again. The graph is showing us that since 2011 which is the first year of the pitstop data that there is an overall increase in the average pit stop time. This could be due to a few different reasons. First, the aspect of human error will play, for example, one wheel gunner cannot get his wheel off fast enough, etc. Second is the car has sustained damage to a part that can be replaced on the car such as the front wing. The final possibility is the issue of Drivers having to serve timed penalties by serving a drive-through penalty or a penalty to be served during a pit stop. One thing to note this year’s data shows roughly six-tenths of a second difference because the 2021 season is not finished yet.
Average Qualifying time since 2010
df_all_teams <- sqldf("select year, constructors_name, avg(q1) as 'Average Q1 Lap Time'
from df_lap_times inner join df_races using (raceId)
inner join df_qualifying using (raceId)
inner join df_teams using (constructorId)
where year >= 2010
group by constructors_name, year")
trellis_df <- df_all_teams %>%
filter(constructors_name %in% c("Mercedes", "McLaren", "Red Bull", "Ferrari")) %>%
data.frame()
team_trellis<-ggplot(trellis_df, aes(x= year, group=constructors_name)) +
geom_line(stat ="identity", aes(y=(Average.Q1.Lap.Time)), color="steelblue")+
geom_point(aes(y= Average.Q1.Lap.Time),shape = 21, size = 2, color= 'white', fill = 'red')+
labs(title= "Average Qualifying Times for Four Teams Since 2010", x="Year",y= "Average Q1 Lap Time
(Minutes)")+
theme_light()+
facet_wrap(~constructors_name, nrow=2, ncol=2)
team_trellis
When looking at this trellis chart, we are observing the average qualifying pace for the first Qualifying Session. In this chart, I wanted to look at the top four performing teams over the last 11 years; Mercedes, Ferrari, McLaren, and Red Bull. Between the four teams, each follows a fairly similar plot to them, except for McLaren which had similar trends as Red Bull but drastically increased followed by a radical decrease the following year. Some key takeaways here are that after the 2014 season, each team had a gradual decrease. This is most likely due to changes in the engine regulations that improved horsepower. The increase of the average qualifying time could be attributed to poor weather and track conditions, worn-down engine parts that will hurt performance. However, this helps us to explain why teams want to go as fast as possible in qualifying, to gain the best position on the grid for the race.
Race Results Based on Qualifying Position
library(plotly)
df_wins_based_grid <- sqldf("select grid, count(position) as 'Wins'
from df_results inner join df_races using (raceId)
where position == 1
and year >= 2000
group by grid")
Race_Result_Qualifying<-plot_ly(df_wins_based_grid, labels = ~grid, values = ~Wins)%>%
add_pie(hole=0.6)%>%
layout(title =" Total Wins Based on Starting Grid Position",legend=list(title=list(text='Starting Grid Position',font = 4)) )%>%
layout(annotations=list(text=paste0("Total number of races since 2000:
", (sum(df_wins_based_grid$Wins))),
"showarrow"=FALSE))
Race_Result_QualifyingContinuing off of the trellis chart based on qualifying pace, this donut chart looks at the probability of winning the race depending on their starting grid position. Out of the 403 total races over the last 21 years. We see that the data is extremely skewed to the front row of the grid. If you are in pole position when starting the race, there is a little over 50% probability that you would win the race, and if you are in second place to start; you have a 25% chance of winning. This shows that there is a 75% chance that the drivers starting in first or second will win the race.
Conclusions
In conclusion, when looking at Formula 1 over the last 21 years we are seeing a similar trend. The best drivers who drive for the best teams are usually more successful, which is no surprise. Along with this, it shows that the more successful team has a better-performing car and team that will help them win. All of these visualizations show why the FIA (the F1 governing body) is working to drastically change the sport next year to level the fight for the worse teams to compete. These improvements will make the sport more interesting for fans and make the sport even more competitive.