Castaway Scores: Who is the Best Survivor Player of All Time?

June 20, 2025 Daniel Oehm 0 Comments

There have been some legendary players over the 25 years and 48 seasons – Boston Rob, Sandra, Tony, Ozzy, Cirie, Colby. Who is the best is largely a subjective question, and there are many ways you could measure a player’s game.

The methodology below is what I’ve chosen and think it makes a good balance between season performance and between season variation i.e. there may be more or fewer tribal or individual challenges given the tribe setup, number of players, time of merge, etc. I think having a good overall measure is still interesting.

I’m focusing on a player’s game for a season. If they have played multiple times, they will have multiple scores. I’ll convert these into career scores at some point, but honestly, I don’t find career type stats very interesting.

The factors included in the castaway score are:

Result score: Final placing. Score of 1 if they make the final tribal council
Jury score: Percentage of jury votes received
Voting score: Tribal council performance. No votes received, and 100% successful boots equals 1.
Tribal challenge score: Probability of achieving the result or lower, assuming equal probability. The higher the score, the better. Max score 1.
Individual challenge score: Probability of achieving the result or lower, assuming equal probability. The higher the score, the better. Max score 1.
Influence: A measure of influence, strategy, or impact on gameplay.

The above metrics are combined to create three new metrics:

Outwit: A combination of vote, jury, and influence score.
Outplay: A combination of the challenge scores.
Outlast: Essentially, the result score.

The above factors are combined to create an overall castaway score ranging from 0 to 100%.

Hot tip: Don’t try too hard to interpret the measures, just know that higher means better.

If statistics is your passion, then read on. Otherwise, feel free to tap out now and follow the links. All you need to know is that higher equals better.

All scores are available in the castaway_scores table in the survivoR package.

Methodology

The calculation of the overall castaway score is a combination of non-parametric and probabilistic measures that assess how well each player has performed in a season.

Result Score

This score is simply the percentage of the way to make the final tribal council based on the boot order. For example, if there are 18 castaways and three finalists, the first boot gets a score of 0, the second gets 1/16, and so on until the three finalists get 1. The penultimate goal is to make the final tribal council, so this makes sense.

You may think that players who don’t do anything and make it deep in the game shouldn’t get a high score. My feeling is that while staying low and keeping the target off your back may not be particularly exciting, it is a valid strategy. And if you manage to do that, kudos to you. Some credit should be given to the players who successfully manage their threat levels and form solid alliances, even if they’re not the ones making the moves.

Jury Score

This is simply the proportion of jury votes received, for example, if the final vote was 7-1-0 the respective players would get a score of 7/8, 1/8, and 0 or 87.5%, 12.5%, 0%. Easy.

Voting Score

This is the more complicated one. The score is designed to measure the success and performance of the tribal council. It consists of three key components:

Tribal council difficulty: A relative measure of successfully surviving the tribal council, for example, if there are 6 people eligible to be voted out at tribal council the rating is 1/6, if there were 10 people the rating would be 1/10. This is because it is, in theory, it’s harder to survive a Tribal Council with fewer people since there are fewer targets. This excludes immune players before the vote, those immune after playing hidden immunity idols are included.
Votes received penalty: If the play copped a vote they receive a penalty. The penalty is a multiplier based on the number of votes they received compared to the total votes cast, e.g., if 5 votes were cast and they received 2, the penalty multiplier is 1-2/5 = 0.6. If they did not receive a vote, their multiplier is 1-0/5 = 1.
Unsuccessful boot penalty: If the player was on the wrong side of the vote but made it through the tribal council, they should receive a lower score than those who were on the right side. The penalty is a multiplier of 0.5 for being on the wrong side of the vote and 1 for being on the right side. If they are voted out, the multiplier is 0 since they didn’t survive the Tribal.

The final calculation is:

$y^{R}_{\text{vote}}=\sum_{k}{\frac{1}{n_p}\left(1-\frac{n_{vr}}{n_{vc}}\right)\alpha_{ubp}}$

where

$n_p$ : Number attending tribal council eligible to be voted out.
$n_{vr}$ : Number of votes received by the player.
$n_{vc}:$ Number of votes cast at Tribal Council, including extra votes
$\alpha_{ubp}:$ Unsuccessful boot penalty.

This score is useful for comparing castaways across seasons, as some may have had an easier time than others, i.e., fewer tribals and fewer people at those tribals, making them more likely to be targeted.

For a given season, I consider the perfect game to be 1) no votes received and 2) 100% successful boot percentage. This can be achieved by only going to a few tribals or all of them, but obviously, it’s harder to maintain a perfect score if you are attending more tribals.

In terms of an overall castaway score for a castaway for the season, I want a measure such that 100% means that player had a flawless tribal performance. Therefore, I drop the difficulty measure to achieve this.

The final score is then:

$y_{\text{vote}} = \sum_{k}{\left(1-\frac{n_{vr}}{n_{vc}}\right)\alpha_{ubp}}$

This means that some players could have a 100% and only attended two tribal councils, like Mike Skupin in season 2 before he was medically evacuated, or Mike Skupin in season 25 who made it to the final tribal council by attending 8 tribals, didn’t receive a vote, and 100% successful boot score. This indicates that there are both easy and challenging paths to achieving a score of 100%.

This is the core reason why we need two vote scores 1) the raw score, which factors in difficulty and 2) the within-season percentage score. It’s worth exploring the vote history table, you’ll see that Amanda Kimmel in S16 has the highest score at 2.5 and a percentage of 97.3% because she copped 4 votes at 13 tribals and was on the right side of the vote 92% of the time. In S48, Eva attended 9 tribals, didn’t receive a vote, was on the right side of the vote 100%, but has a score of 1.7. It was definitely more challenging for Amanda, and that needs to be acknowledged.

If I used the raw score and scaled it so that Amanda had a 100% score, Eva would have a vote score of 68%. That doesn’t seem right to me if she was on the right side of the vote 100% of the time and didn’t receive a vote.

This is how the vote scores are distributed. It’s not bad, it’s a nice shape, but my only gripe is that it is centered around 75%, rather than the average game being around 50%. It’s not that bad, but it’s important to keep in mind. Also, every player that is booted at their first tribal gets 0%.

Tribal and Individual Challenge Scores

This is conceptually a little tricky as well. The score is the probability of achieving the result by considering how many winners there are for the challenge e.g. in a three tribe set up, there are usually two winning tribes, therefore, the probability of winning is 2/3, assuming equal distribution. The residual is $r = \text{wins} - \text{expected} = \sum_{k}{(y_k - 1/n_{p, k}})$ where $y_k$ is 0 or 1 for a win.

I then look at the probability of observing this amount $y_{\text{tribal}} = P(X \leq r)$ by assuming a Normal distribution. In R the calculation is pnorm(r, 0, 1.5*sd(r)). I apply a 1.5 factor to the standard deviation so that there are few with a score of 1 or very close to it, such that it rounds to 100%.

Assuming a distribution and evaluating the probability has the nice property that the scores range between 0 and 1.

It’s important to understand that this isn’t the probability of someone winning the challenge, it’s the probability of achieving the result or lower assuming equal chance. That’s important because different players have different chances.

For example, consider the challenge Wrist Assured, and the players are David (S48), Jonathan (S42), Courtney Yates (S15, S20), and Venus (S46). Not to discredit Courtney and Venus, but they aren’t winning that challenge of raw strength. The probability is more likely 50/50, David/Jonathan.

But if that challenge was Get a Grip, the probability would be more likely 50/50, Courtney/Venus. Not to discredit David and Jonathan, but, gravity.

This is ok if we assume the residual is a score such that the higher the score, the higher the challenge success/performance of the player.

There is an argument to be made for why the score should be on a (0, 1) scale rather than being left as a raw score. I think it’s easier to understand if 100% is a theoretical maximum a player can achieve rather than some arbitrary unknown. There are pros and cons.

The most individual challenges played in a season is 14. The most ever won is 8 – Terry Deitz is S12. There’s only been one person who has won 8 individual challenges, i.e., it’s only happened once in 48 seasons (~2%) or 0.2% of players that made the merge. In his season, he competed in 12 individual challenges. It doesn’t seem right to give him a score of 8/12 (67%). It doesn’t really reward Terry for his amazing challenge performance. But taking a probabilistic approach, it does – he gets a score of 99.9% with room for improvement! I may revisit this if someone ends up winning 12/12, which would be wild.

This all makes sense when you look at the distribution of the raw score. Terry is way out in the tail of the distribution with a score of 5.05. The mean for both the tribal and individual challenge scores is 50%.

Influence Score

This is probably the most controversial factor. There is a lot in the game that can’t be captured by the data, for example, the player’s social game, the bonds and connections they’ve made on the beach, strategy, and influence. However, I think it’s reasonable to use confessionals as a crude measure of those things, particularly the strategic part.

My reasoning is that each season, there is a narrative that tells the story from when they landed on the beach to crowning the Sole Survivor. In each episode, we see the storytelling, who the key players are, the key events, the successful strategies, and the unsuccessful strategies. If the editing team doesn’t tie this together well, the story won’t be coherent and make sense at the end. Most of the strategy is told to the audience through the player confessionals.

In a previous study, I’ve shown a correlation between the proportion of confessionals received and making it to the final tribal council. It makes sense, the editors need to tell the story of how they got there and signpost the key events that ultimately sway the Jury’s votes. Anything that isn’t important to the overall narrative won’t get shown, and there are a number of examples of that.

Therefore, I think it’s reasonable to assume that those who received more confessionals had more influence in the game or were at least strategic players, social threats, antagonistic players, or a combination of those. In season 48, Kyle, Eva, and Joe all received more confessionals than expected. Regardless of what you thought of either of those players and their strategy, they held a lot of influence in the game. By contrast, Star, while loved by the fan base, received fewer confessionals than expected, second only to Chrissy. It’s fair to say that Star had little influence in the game.

Another example is the infamous first season of Russell Hantz. He received over 7 times more confessionals than Natalie, who ultimately won. It is the clearest case of how important the social game is. Russell, love him or hate him, changed the game. There’s no doubt that the narrative for season 19 was about Russell, his influence, and aggressive strategy. If he hadn’t had an enormous impact on the game and players’ decisions, we wouldn’t have seen him so much.

With all that in mind, I think confessionals are a good proxy for the immeasurable part of Survivor (poor choice of words since measuring is what I’m doing). Yes, there are kooky-bing-bong type players, e.g., Phillip Shppard, that are great entertainment, but perhaps not particularly strategic. There’s fewer now than in the early days – although there are exceptions, season 46 gave us Bhanu. I still think it’s a valid point to use it as a proxy for strategy and influence.

So this measure doesn’t dominate the overall castaway score, I assign it a weight of 0.25, or 5% of the total overall castaway score. It’s good to help separate some players who have played similar games. $y_{\text{influence}}$ is calculate from the residual of the confessionals per hour ( $r = y_\text{cph} - \hat{y}_{cph}$ ) and finding the probability $P(X \gt r)$ so the maximum score is 1 (or at least a rounding error away).

Lastly, I’ve noticed that early boots can get a high influence score and I don’t think that’s capturing what I want so I apply a filter to the score to down-weight the early boots and keep the score for the later boots. That filter is a log-normal distribution (in R cumsum(dlnorm(1:16, 1.4, 0.3))).

And here’s the distribution of the influence score. The mean is 39%.

Outwit, Outplay, Outlast

Using the above metrics I wanted to combine these into measures representing the theme of the show – Outwit, outplay, outlast. Each is a linear combination of the above metrics:

Outwit

Outwit is a combination of the vote, jury, and influence scores. This reasoning is the outwit part of the game is usually associated with positioning themselves in the tribe, making alliances, avoiding votes, and blinding others. It is also in jury management – the more jury votes received is a pretty good measure of jury management and outwitting the other players. Finally, the influence score is a crude measure of strategic play and influence on the game.

$y_{\text{outwit}} = 0.5y_{\text{vote}} + 0.3y_{\text{jury}} + 0.2y_{\text{influence}}$

Outplay

I’m interpreting outplay to be the challenge element of the game – tribal and individual challenges. I believe the individual challenges are more indicative of someone’s challenge success and therefore get a higher weight than tribal challenges.

$y_{\text{outplay}} = 0.2y_{\text{tribal}} + 0.8y_{\text{individual}}$

Outlast

Outlast is the simple one – the result metric i.e. how far they made it in the game. It doesn’t matter how they got there, if they made the final tribal council they outlasted everyone else and should receive the highest score. Whether you believe they deserve the win or not is a different story. Outwit and Outlast scores should separate the wheat from the chaff.

$y_{\text{outlast}} = y_{\text{result}}$

Overall Castaway Score

The final calculation is a linear combination of Outwit, Outplay, Outlast:

$y_\text{overall}=0.4y_{\text{outwit}} + 0.3y_{\text{outplay}} + 0.3y_{\text{outlast}}$

The weightings I have chosen are arbitrary based on my gut feel about what is important in the measure and I reserve the right to change them at any time.

The three metrics Outwit, Outplay, Outlast are difficult to interpret at face value but that’s also kind of the point. It difficult if not impossible to capture the entirety of the game in data so any measure is going to miss something. At least with these there is an understanding of the inputs to the metric and that higher means better and 1 is the max value.

The final overall castaway score is distributed as shown below. There is clearly a bi-modal property which aligns very well with those that were booted pre-merge and those that made the merge. I’d prefer it if it was a flatter bell shape but this isn’t bad. Those booted pre-merge should get a lower score.

The mean is 42%

20% for pre-merge boots and
55% for post-merge boots.

I’ll probably apply different weightings to find something that works well, but I’m sticking with this for now.

So, who is the best?

With the methodology established, who is the best?

🏆 Tom Westman 🏆

Here is the top 20 over the past 48 seasons. I won’t share more here as the details are much better viewed on survivorstatsdb.com.

Castaway	season	Rank	Score
Top 20 Survivor Players
Players with the highest overall score
Tom Westman	10	1	93%
‘Boston’ Rob Mariano	22	2	91%
Kim Spradlin-Wolfe	24	3	90%
James ‘JT’ Thomas Jr.	18	4	90%
John Cochran	26	5	88%
‘Boston’ Rob Mariano	8	6	88%
Brian Heidik	5	7	87%
Mike Holloway	30	8	86%
Jenna Morasca	6	9	86%
Oscar ‘Ozzy’ Lusth	13	10	85%
Colby Donaldson	2	11	83%
Tyson Apostol	27	12	83%
Michele Fitzgerald	32	13	83%
Tony Vlachos	40	14	82%
Domenick Abbate	36	15	81%
Jud ‘Fabio’ Birza	21	16	81%
Rachel LaMont	47	17	81%
Chris Daugherty	9	18	80%
Natalie Anderson	29	19	80%
Kelly Wiglesworth	1	20	79%

This is something I’ll continue to revisit and refine, but for now, here is where all 875 castaways ranked.

You may not agree with the results or how I’ve approached it. Maybe you think I’ve given challenges too much weight, perhaps tribal challenges shouldn’t be included at all. That’s fine, the data is there for you to use as you see fit and construct your own score.

As I mentioned above, it’s incredibly difficult to derive a score which works for everything. There are going to be trade-offs, but personally, I think this works well.

Season 50

I’ve pulled together a short analysis of the scores for the season 50 cast into a neat ‘scrollytelling’ format using quarto. It’s designed for mobile devices. You can find it here.

Follow me on social media:

data Statistics survivor