In doehm/survivoR: Data from all Seasons of Survivor (US) TV Series in Tidy Format

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = '#>',
  fig.path = 'README-',
  error = TRUE,
  eval = TRUE,
  fig.height = 5
)

suppressPackageStartupMessages(library(tidyverse))
suppressPackageStartupMessages(library(dplyr))
suppressPackageStartupMessages(library(forcats))
suppressPackageStartupMessages(library(survivoR))
suppressPackageStartupMessages(library(paletteer))
suppressPackageStartupMessages(library(glue))
suppressPackageStartupMessages(library(rvest))

good_pal <<- c("#ffffff", "#f2fbd2", "#c9ecb4", "#93d3ab", "#35b0ab")
bad_pal <<- rev(c("#ef6351", "#f38375", "#f7a399", "#fbc3bc", "#ffe3e0", "white"))

max_eps <- viewers |> 
  filter(episode_title != "Reunion") |> 
  nrow()

max_seasons <- season_summary |> 
  nrow()

# in progress seasons
in_prog_eps <- survivoR::castaways |> 
  filter(version_season == "US43") |> 
  pull(episode) |> 
  max(na.rm = TRUE)

n_people <- survivoR::castaways |> 
  distinct(version_season, castaway_id) |> 
  nrow()

version <- str_replace(readLines("DESCRIPTION")[4], 'Version: ', 'v')

url <- "https://cran.r-project.org/web/packages/survivoR/index.html"
cran_version <- read_html(url) |> 
  html_text() |> 
  str_extract("Version:\n[:digit:]{1}\\.[:digit:]+\\.[:digit:]+") |> 
  str_remove("Version:\n")
cran_version <- paste0("v", cran_version)

survivoR

r max_seasons seasons. r n_people people. 1 package!

survivoR is a collection of data sets detailing events across r max_seasons seasons of Survivor US, Australia, South Africa, New Zealand and UK. It includes castaway information, vote history, immunity and reward challenge winners, jury votes, advantage details and a lot more.

For analysis and updates you can follow me on Bluesky @danoehm.bsky.social or Threads @_survivordb

Installation

Install from CRAN (r cran_version) or Git (r version).

If Git > CRAN I'd suggest install from Git. We are constantly improving the data sets so the github version is likely to be slightly improved.

install.packages("survivoR")

devtools::install_github("doehm/survivoR")

News: survivoR 2.3.5

• Adding complete US47 data
• Adding new castaway_scores dataset
• Adding new add_* functions:
  • add_alive(): Adds a logical flag if the castaway is alive at the start or end of an episode
  • add_bipoc(): Adds a BIPOC to the data frame. If any African American (or Canadian), Asian American, Latin American, or Native American is TRUE then BIPOC is TRUE.
  • add_castaway(): Adds castaway to a data frame. Input data frame must have castaway_id.
  • add_demogs(): Add demographics that includes age, gender, race/ethnicity, and LGBTQIA+ status to a data frame with castaway_id.
  • add_finalist(): Adds a winner flag to the dataset.
  • add_full_name(): Adds full name to the data frame. Useful for plotting and making tables.
  • add_gender(): Adds gender to a data frame
  • add_jury(): Adds a jury member flag to the data set.
  • add_lgbt(): Adds the LGBTQIA+ flag to the data frame.
  • add_result(): Adds the result and place to the data frame.
  • add_tribe(): Adds tribe to a data frame for a specified stage of the game e.g. original, swapped, swapped_2, etc.
  • add_tribe_colour(): Add tribe colour to the data frame. Useful for preparing the data for plotting with ggplot2.
  • add_winner(): Adds a winner flag to the data set.
• Adding new filter_* functions:
  • filter_alive(): Filters a given dataset to those that are still alive in the game at the start or end of a user specified episode.
  • filter_final_n(): Filters to the final n players e.g. the final 5.
  • filter_finalist(): Filters a dataset to the finalists of a given season.
  • filter_jury(): Filters a dataset to the jury members of a given season.
  • filter_new_era(): Filters a dataset to all New Era seasons.
  • filter_us(): Filter a dataset to a specified set of US season or list of seasons. A shorthand version of filter_vs() for the US seasons.
  • filter_vs(): Filters a data set to a specified version season or list of version seasons.
  • filter_winner(): Filters a data set to the winners of a given season.

The Sanctuary

The Sanctuary is the survivoR package's companion. It holds interactive tables and charts detailing the castaways, challenges, vote history, confessionals, ratings, and more. Confessional counts from myself, Carly Levitz, Sam, Grace.

Confessional timing

Included in the package is a confessional timing app to record the length of confessionals while watching the episode.

To launch the app, first install the package and run,

library(survivoR)
launch_confessional_app()

To try it out online 👉 Confessional timing app

More info here.

Dataset overview

There are 19 data sets included in the package:

1. advantage_movement
2. advantage_details
3. boot_mapping
4. castaway_details
5. castaway_scores
6. castaways
7. challenge_results
8. challenge_description
9. challenge_summary
10. confessionals
11. jury_votes
12. season_summary
13. survivor_auction
14. tribe_colours
15. tribe_mapping
16. episodes
17. vote_history
18. auction_details
19. screen_time
20. season_palettes

See the sections below for more details on the key data sets.

Season summary

Season summary

A table containing summary details of each season of Survivor, including the winner, runner ups and location.

season_summary

Castaways

Castaways

This data set contains season and demographic information about each castaway. It is structured to view their results for each season. Castaways that have played in multiple seasons will feature more than once with the age and location representing that point in time. Castaways that re-entered the game will feature more than once in the same season as they technically have more than one boot order e.g. Natalie Anderson - Winners at War.

Each castaway has a unique castaway_id which links the individual across all data sets and seasons. It also links to the following ID's found on the vote_history, jury_votes and challenges data sets.

• vote_id
• voted_out_id
• finalist_id

castaways |> 
  filter(season == 45)

Castaway details

A few castaways have changed their name from season to season or have been referred to by a different name during the season e.g. Amber Mariano; in season 8 Survivor All-Stars there was Rob C and Rob M. That information has been retained here in the castaways data set.

castaway_details contains unique information for each castaway. It takes the full name from their most current season and their most verbose short name which is handy for labelling.

It also includes gender, date of birth, occupation, race, ethnicity and other data. If no source was found to determine a castaways race and ethnicity, the data is kept as missing rather than making an assumption.

african_american, asian_american, latin_american, native_american, race, ethnicity, and bipoc data is complete only for the US. bipoc is TRUE when any of the *_american fields are TRUE. These fields have been recorded as per the (Survivor wiki)[https://survivor.fandom.com/wiki/Main_Page]. Other versions have been left blank as the data is not complete and the term 'people of colour' is typically only used in the US.

I have deprecated the old field poc in order to be more inclusive and to make using the race/ethnicity fields simpler.

castaway_details

Castaway scores

I have created a measure for challenge success, vote history or tribal council success and advantage success. For more details please see follow the links:

Challenge score methodology

Vote history mothodology

castaway_scores

Vote history

Vote history

This data frame contains a complete history of votes cast across all seasons of Survivor. This allows you to see who who voted for who at which Tribal Council. It also includes details on who had individual immunity as well as who had their votes nullified by a hidden immunity idol. This details the key events for the season.

There is some information on split votes to help calculate if a player engaged in a split vote but ultimately hit their target. There are events which influence the vote e.g. Extra votes, safety without power, etc. These are recorded here as well.

vh <- vote_history |> 
  filter(
    season == 45,
    episode == 9
  ) 
vh

vh |> 
  count(vote)

Challenges

Challenge results

Note: From v1.1 the challenge_results dataset has been improved but could break existing code. The old table is maintained at challenge_results_dep

There are 3 tables challenge_results, challenge_description, and challenge_summary.

Challenge results

A tidy data frame of immunity and reward challenge results. The winners and losers of the challenges are found recorded here.

challenge_results |> 
  filter(season == 45) |> 
  group_by(castaway) |> 
  summarise(
    won = sum(result == "Won"),
    lost = sum(result == "Lost"),
    total_challenges = n(),
    chosen_for_reward = sum(chosen_for_reward)
  )

The challenge_id is the primary key for the challenge_description data set. The challange_id will change as the data or descriptions change.

Challenge description

Note: This data frame is going through a massive revamp. Stay tuned.

This data set contains the name, description, and descriptive features for each challenge where it is known. Challenges can go by different names so have included the unique name and the recurring challenge name. These are taken directly from the Survivor Wiki. Sometimes there can be variations made on the challenge but go but the same name, or the challenge is integrated with a longer obstacle. In these cases the challenge may share the same recurring challenge name but have a different challenge name. Even if they share the same names the description could be different.

The features of each challenge have been determined largely through string searches of key words that describe the challenge. It may not be 100% accurate due to the different and inconsistent descriptions but in most part they will provide a good basis for analysis.

If any descriptive features need altering please let me know in the issues.

challenge_description

challenge_description |> 
  summarise_if(is_logical, ~sum(.x, na.rm = TRUE)) |> 
  glimpse()

See the help manual for more detailed descriptions of the features.

Challenge Summary

The challenge_summary table is solving an annoying problem with challenge_results and the way some challenges are constructed. You may want to count how many individual challenges someone has won, or tribal immunities, etc. To do so you'll have to use the challenge_type, outcome_type, and results fields. There are some challenges which are combined e.g. Team / Individual challenges which makes this not a straight process to summarise the table.

Hence why challenge_summary exisits. The category column consists of the following categories:

• All: All challenge types
• Reward
• Immunity
• Tribal
• Tribal Reward
• Tribal Immunity
• Individual
• Individual Reward
• Individual Immunity
• Team
• Team Reward
• Team Immunity
• Duel

There is obviously overlap with the categories but this structure makes it simple to summarise the table how you desire e.g.

challenge_summary |> 
  group_by(category, version_season, castaway) |> 
  summarise(
    n_challenges = n(), 
    n_won = sum(won)
    )

How to add the challenge scores to challenge summary.

challenge_summary |>
  group_by(category, version_season, castaway_id, castaway) |>
  summarise(
    n_challenges = n_distinct(challenge_id),
    n_won = sum(won),
    .groups = "drop"
  ) |>
  left_join(
    castaway_scores |>
      select(version_season, castaway_id, starts_with("score_chal")) |>
      pivot_longer(c(-version_season, -castaway_id), names_to = "category", values_to = "score") |>
      mutate(
        category = str_remove(category, "score_chal_"),
        category = str_replace_all(category, "_", " "),
        category = str_to_title(category)
      ) |>
      select(category, version_season, castaway_id, score),
    join_by(category, version_season, castaway_id)
  )

See the R docs for more details on the fields. Join to challenge_results with version_season and challenge_id.

Jury votes

Jury votes

History of jury votes. It is more verbose than it needs to be, however having a 0-1 column indicating if a vote was placed or not makes it easier to summarise castaways that received no votes.

jury_votes |> 
  filter(season == 45)

jury_votes |> 
  filter(season == 45) |> 
  group_by(finalist) |> 
  summarise(votes = sum(vote))

Advantages

Advantage Details

This dataset lists the hidden idols and advantages in the game for all seasons. It details where it was found, if there was a clue to the advantage, location and other advantage conditions. This maps to the advantage_movement table.

advantage_details |> 
  filter(season == 45)

Advantage Movement

The advantage_movement table tracks who found the advantage, who they may have handed it to and who the played it for. Each step is called an event. The sequence_id tracks the logical step of the advantage. For example in season 41, JD found an Extra Vote advantage. JD gave it to Shan in good faith who then voted him out keeping the Extra Vote. Shan gave it to Ricard in good faith who eventually gave it back before Shan played it for Naseer. That movement is recorded in this table.

advantage_movement |> 
  filter(advantage_id == "USEV4102")

Confessionals

Confessionals

A dataset containing the number of confessionals for each castaway by season and episode. There are multiple contributors to this data. Where there are multiple sets of counts for a season the average is taken and added to the package. The aim is to establish consistency in confessional counts in the absence of official sources. Given the subjective nature of the counts and the potential for clerical error no single source is more valid than another. So it is reasonable to average across all sources.

Confessional time exists for a few seasons. This is the total cumulative time for each castaway in seconds. This is a much more accurate indicator of the 'edit'.

confessionals |> 
  filter(season == 45) |> 
  group_by(castaway) |> 
  summarise(
    count = sum(confessional_count),
    time = sum(confessional_time)
    )

The confessional index is available on this data set. The index is a standardised measure of the number of confessionals the player has received compared to the others. It is stratified by tribe so it measures how many confessionals each player gets proportional to even share within tribe e.g. an index of 1.5 means that player as received 50% more than others in their tribe.

The tribe grouping is important since the tribe that attends tribal council typical get more screen time, which is fair enough. I don't think we should expect even share across everyone in the pre-merge stage of the game.

The index is cumulative with episode, so the players final index is the index in their final episode.

confessionals |> 
  filter(season == 45) |> 
  group_by(castaway) |> 
  slice_max(episode) |> 
  arrange(desc(index_time)) |> 
  select(castaway, episode, confessional_count, confessional_time, index_count, index_time)

Screen time

Screen time [EXPERIMENTAL]

This dataset contains the estimated screen time for each castaway during an episode. Please note that this is still in the early days of development. There is likely to be misclassification and other sources of error. The model will be refined over time.

An individuals' screen time is calculated, at a high-level, via the following process:

1. Frames are sampled from episodes on a 1 second time interval

2. MTCNN detects the human faces within each frame

3. VGGFace2 converts each detected face into a 512d vector space

4. A training set of labelled images (1 for each contestant + 3 for Jeff Probst) is processed in the same way to determine where they sit in the vector space. TODO: This could be made more accurate by increasing the number of training images per contestant.

5. The Euclidean distance is calculated for the faces detected in the frame to each of the contestants in the season (+Jeff). If the minimum distance is greater than 1.2 the face is labelled as "unknown". TODO: Review how robust this distance cutoff truly is - currently based on manual review of Season 42.

6. A multi-class SVM is trained on the training set to label faces. For any face not identified as "unknown", the vector embedding is run into this model and a label is generated.

7. All labelled faces are aggregated together, with an assumption of 1-5 full second of screen time each time a face is seen and factoring in time between detection capping at a max of 5 seconds.

screen_time |> 
  filter(version_season == "US45") |> 
  group_by(castaway_id) |> 
  summarise(total_mins = sum(screen_time)/60) |> 
  left_join(
    castaway_details |> 
      select(castaway_id, castaway = short_name),
    by = "castaway_id"
  ) |> 
  arrange(desc(total_mins))

Currently it only includes data for season 42. More seasons will be added as they are completed.

Boot mapping

Boot mapping

A mapping table to detail who is still alive at each stage of the game. It is useful for easy filtering to say the final \code{n} players.

# filter to season 45 and when there are 6 people left
# 18 people in the season, therefore 12 boots

still_alive <- function(.version, .season, .n_boots) {
  survivoR::boot_mapping |>
    filter(
      version == .version,
      season == .season,
      final_n == 6,
      game_status %in% c("In the game", "Returned")
    )
}

still_alive("US", 45, 6)

Episodes

Episodes

Episodes is an episode level table. It contains the episode information such as episode title, air date, length, IMDb rating and the viewer information for every episode across all seasons.

episodes |> 
  filter(season == 45)

Survivor Auction

Survivor Auction

There are 2 data sets, survivor_acution and auction_details. survivor_auction simply shows who attended the auction and auction_details holds the details of the auction e.g. who bought what and at what price.

auction_details |> 
  filter(season == 45)

Issues

Given the variable nature of the game of Survivor and changing of the rules, there are bound to be edges cases where the data is not quite right. Before logging an issue please install the git version to see if it has already been corrected. If not, please log an issue and I will correct the datasets.

New features will be added, such as details on exiled castaways across the seasons. If you have a request for specific data let me know in the issues and I'll see what I can do.

Showcase

Survivor Dashboard

Carly Levitz has developed a fantastic dashboard showcasing the data and allowing you to drill down into seasons, castaways, voting history and challenges.

Data viz

This looks at the number of immunity idols won and votes received for each winner.

Contributors

A big thank you to:

Package contributor and maintainers

• Carly Levitz for ongoing data collection and curation

Data contributors

• Sam for contributing to the confessional counts
• Dario Mavec for developing the face detection model for estimating total screen time
• Matt Stiles for collecting and contributing the acknowledgment features on the castaways data frame.
• Camilla Bendetti for collating the personality type data for each castaway.
• Uygar Sozer for adding the filming start and end dates for each season.
• Holt Skinner for creating the castaway ID to map people across seasons and manage name changes.
• Kosta Psaltis for the original race data.

References

Data was sourced from Wikipedia and the Survivor Wiki. Other data, such as the tribe colours, was manually recorded and entered by myself and contributors.

doehm/survivoR documentation built on March 1, 2025, 3:52 p.m.