Changing the Rules of Survivor

by Henry Farrell on February 18, 2005

“Kevin Drum”: takes a break from politics-blogging to opine on the new series of _Survivor_. It’s shaping up to be more fun than last season, because there are new – and unpredictable – rules; Kevin suggests that the show’s designers ought to make the rules more unpredictable still.

bq. The appeal of the show is in the human interaction. How do you keep from being voted off? How do you make and break alliances? Who gets betrayed this week? That’s where they need to throw in a few curveballs. The contestants need to learn that the standard way of forming alliances and screwing competitors is subject to change.

He’s probably right – although one of the fun things about _Survivor_ is that there has usually been a high level of unpredictability, even under stable rules. Last season’s show was the very dull exception that proved the rule – the producers threw together tribes in such a way as to generate stable, gender-based cooperation for most of the game. They later made a rather desperate _post-hoc_ attempt to mix things up and weaken alliances, but it didn’t work very well. This season, they’ve deliberately generated tribes in a way that mixes up the sexes.

Anyway, I talked about some of these issues at greater length in a long post on my old blog about the applicability of sociology and game theory to _Survivor_ a couple of years ago. I reproduce it below the fold, if anyone’s interested.

*The Sociology of Survivor*

I spent yesterday evening with friends watching the three hour Survivor end-of-season finale. Much fun, with many upsets over the course of the evening, and an entirely unexpected winner. But – and here’s the sign that I’m an incurable academic – it set me to thinking about the difference between economic and sociological models of human action.

I expect 90% of my readers to switch off at this point. But you shouldn’t; at least not if you’re a Survivor fan (and Survivor‘s about as much fun as junk tv gets). Survivor is all about strategy, and making and breaking alliances. One person gets voted off by the other players every week; nobody wants to be that person. In order to avoid that fate (and to make sure that other, threatening players are voted off), each player tries to create an alliance with other players, where people agree not to vote each other off, or to target other people to vote off, or both. But there’s no way to enforce these agreements. Furthermore, there’s a substantial random element to the game; one person can win “immunity” every week, so that he or she can’t be voted off that week; this can throw pre-arranged strategies into disarray, especially towards the end of the game. Thus, in order to prosper at Survivor, you not only need to be good at working with other people and forming teams; you need to be good at breaking alliances, stabbing people in the back, and finding new allies when circumstances so demand.

Now clearly, you have to behave strategically to prosper in the game. But what does “strategically” mean? Game theorists and sociologists have very different answers to that question. A game theorist would assume that everyone would have a clear map in their head of the different kinds of players that she might encounter, the different kinds of strategies that she could use (perhaps dependent on which kinds of player she is dealing with), and the different kinds of outcomes that will likely result from the various strategies. There’s room for some uncertainty in the game (random acts of “nature” can intervene here and there), but everyone knows that if they are at point x in the game, their interest is best served by a particular strategy y (or perhaps by a set of strategies with identical payoffs; I’m simplifying a little here for the purpose of popularization). Here, it’s a matter of playing within a fixed structure (the parameters of the game), which you completely understand, and where your interest is dictated at any particular moment by the specific point of that structure you’re at.

Sociologists tend to take a different approach to social structures. By and large, they’re interested in networks rather than games. Network theory suggests that a player’s power and influence depends on her position within the network of social interactions that she finds herself in. Again, simplifying wildly, some actors can be at the centre of a spiderweb of relations; everyone comes to them in order to get things done. Others may be gatekeepers or intermediaries between two groups of people who don’t otherwise have much contact with each other; these actors too can be quite important. Here, strategy is all about positioning yourself well within the network, and then manipulating information and resource flows in order to maintain or improve your position. It’s much more open-ended than game theory – the universe of possibilities isn’t fixed at the outset, but changes over time, and can be affected by the conscious action of the players.

Which of these conceptions of strategy best fits Survivor? I think that the answer is obvious to anyone who watched the series. It’s the second, sociological conception. Pretty well everyone who saw this series of Survivor would agree that Rob was the smartest and canniest player. He didn’t win; but this was in large part because he was quite unlucky at the end (Jenna, who did win, survived by a fluke). Both of the two finalists agreed that he should have been there instead of them.

How did Rob play? Not by having a rote set of strategies at the beginning of the game. Instead, as he explained it, he was always at pains to keep his options open; he maintained friendly relations with as large a group as possible (and was rather good at convincing others that he had their best interests at heart, even when he didn’t). He then chopped and changed his strategy as circumstances demanded. When he needed to zig, he zigged, and when he needed to zag, he zagged. He didn’t seek to lead alliances, but instead made himself into the pivotal player, who could move from one alliance to the next, and thus swing the vote in one direction or another. By so doing, he shaped the strategic context which everyone else had to play in.

Compare his behavior with that of Cosimo de Medici in early Renaissance Florence, as depicted in John Padgett and Christopher Ansell’s classic piece, “Robust Action and the Rise of the Medici.” Padgett and Ansell are sociologists, who want to argue against rational choice notions of strategic behavior, by offering their own notion of “robust action.” Robust action is all about trying to maintain your own flexibility and freedom of choice over as wide a range of options as possible, while narrowing the options of everyone else. It implies that there aren’t any fixed interests – all interests are positional and actors are less interested in pursuing specific goals at any point in time, than in maintaining discretionary options against the day when they do have a definite end to pursue. Thus Cosimo positioned himself at the center of a web of influence, without ever wanting fully to commit to anything or anyone. As Padgett and Ansell describe it,

bq. in nasty strategic games like Florence or chess, positional play is the maneuvering of opponents into the forced clarification of their (but not your) tactical lines of action. Locked-in commitment to lines of action, and thence to goals, is the product not of individual choice, but at least as much of others’ successful ‘ecological control’ over you.

This is a more or less exact description of how best to play ‘Survivor’ (or, at least, to play Survivor as Rob played it). Keep your options open – maintain ties with everyone. Don’t get frozen into a group with a specific agenda; instead, try to be a key player (or potential key player) for every possible group. You’re more powerful (in the sense of maintaining options and contacts) as a swing vote than as tribal chief. Keep your end-goals and specific strategies mysterious – try to be all things to all men and women. Maintain flexibility at all costs. And then go for broke when the opportunity arises.

I reckon that two important lessons flow from this. First, that sociological approaches to the understanding of human behavior capture certain things that economic approaches can’t. They’re much better at dealing with fluid situations, where the future is unknowable, and people want to keep as many options open as possible for dealing with unanticipated problems. And many other things besides, I’m sure. Second, that an academic with time on his hands during the summer break is a dangerous thing indeed.



Dirk 02.18.05 at 11:59 pm

This is very interesting, thank you.

What I am curious about is whether your argument is a general critique of game theory as a methodology, or more of the way game theory is most commonly used (and taught). Assuming I understand you correctly, you are saying that there are many situations in which the subsequent parts of the game are so uncertain that any attempt to write down the “game tree” would be completely futile, and any notion of solution by backward induction would be pointless. I think this observation is certainly correct.

But does this really invalidate game theory? I would argue that positioning yourself in a way that maximizes your flexibility and options is exactly the kind of “optimal response” game theory would prescribe in this situation. Hence I am inclined to interpret your post as saying that game theory should focus more on games of bounded rationality / lack of knowledge about the game structure, than saying that game theory isn’t the best framework of analysis.


Kevin Drum 02.19.05 at 1:53 am

“It’s shaping up to be more fun than last season, because there are new – and unpredictable – rules…”

But that’s not really true. The 20-player thing was pretty trivial and lasted for only a day. After the first 24 hours, we were back to the standard 18-player, 2-tribe game. Seems to me that this counts as only the slightest of twists, not really anything genuinely new.

But you’re right: past seasons have had interesting dynamics even with static rules. Last season was an exception, and maybe things will pick up this year.


Lee Scoresby 02.19.05 at 4:30 am

Henry, did you ever read the Jervis’ piece on Survivor and game theory?

Anyway, an important point, and one I’ve been on about for a while (but haven’t published on, sadly, unless one of my current pieces actually sees the light of day). There’s a good deal of sociological argumentation – not just by Padgett and Ansell but also in various work on exchange theory, symbolic interaction, and framing – that suggests the virtues of ambiguity in strategic interaction. Indeed, Schelling even makes this point at various times in Arms and Influence: making it difficult to parse your interests and intentions can give one an advantage.

In a lot of network theory, this advantage can be understood, P&A suggest, as deriving from network centrality within a segmented network. Indeed, one of the reasons why centrality is often used as a proxy for power is that it provides actors with asymmetric access to information (why, for example, the departmental administrator is often the most powerful person in an academic department).

Yet there’s a systemic ambiguity (duh-duh) in empirical work, such as P&A’s famous piece, about whether robust action derives from structural location, personal attributes, or some combination of the two. I do find it interesting, in this respect, that a large number of important political actors have been described as, like Cosimo, Sphinxes (William the Silent, who was not silent but was rather difficult to pin down, is one example).

Anyway, where was I? Oh, yes. The important point: game theory doesn’t necessarily suggest the opposite of robust action – after all, information asymmetries are a big deal in a lot of formal models. The problem, rather, at least in IR, is that this fad for “credible commitments” and for seeing imperfect information as a source of conflict tends to obscure the older recognition in IR theory that ambiguity is a powerful strategic weapon.


John Quiggin 02.19.05 at 9:01 am

Game theory does yield the non-obvious prediction that unpopular players (whatever that means) should do well, since each player’s optimal penultimate outcome is to reach the final vote along with a highly unpopular fellow-player.

Of course, that assumes that unpopularity is an inherent characteristic. With perfectly rational players, perfectly capable of dissimulation, everything is random.


Henry 02.19.05 at 3:04 pm

Dirk – you’re probably right, and I should have dwelt on this more. But of course, games of incomplete information are relatively intractable – Harsanyi’s trick and others aren’t likely to lead us to be able to make very useful predictions. So really, game theory would direct us to the beginning point of network theory in this kind of situation. There was a special issue of _Rationality and Society_ a few years ago on the complementarities of game theory and network theory – I really should dig it up.

Lee – a very interesting point. A lot of it does come down to the personality – which is something that we’ve nothing useful to say on that Machiavelli didn’t say usefully several centuries ago. I haven’t read Jervis on this – would love the cite. I do know of at least one very prominent game theorist who is a survivor buff.

Kevin – the method of picking the teams differed significantly from previous seasons too, and, I would think, will have some interesting long term implications. First, most obviously, it split up the sexes, by obliging women to choose men and vice versa. Second, it created the basis for dyadic loyalties rather than group loyalties – if _x_ picked _y_, it suggests at least the foundation of a pair-based loyalty that may come into play later. Or not, as when the hairdresser whose name I can’t remember didn’t pick the bartender, whose name I also can’t remember. I do think this season looks promising – and it certainly won’t be hard to beat last season, which was the worst ever imo.


WeSaferThemHealthier 02.19.05 at 4:36 pm


“Game theory does yield the non-obvious prediction that unpopular players […] should do well”

I don’t watch Survivor. Can that actually be observed most of the time?


Alex 02.20.05 at 11:36 pm

John- Sorry, you brought up a pet peeve of mine. I hate it when people say penultimate when they really mean ultimate. Penultimate means second to last.


Henry 02.21.05 at 2:43 am


John is actually right here. The penultimate stage in Survivor is the one where there are only two players left, whom the jury (composed of several previously eliminated players) chooses between in the very final stage. So he’s using the word in its correct sense – each player wants to end up at the penultimate stage with an unpopular player, so that he/she is the one chosen in the final stage. At this point of the decision tree at least, backward induction works just fine (although certain players don’t seem to have followed the game tree back, either through doltishness or sincerity/niceness).


jlw 02.22.05 at 6:12 pm

The Jervis piece–actually co-authored with his daughter, who was a fan of the show–can be found in this rather long pdf. (The pdf is worth downloading, as it gives you the entire Nov/Dec 2000 issue of The Sciences. Even if I am biased–and I am–I gotta say that was a great magazine.)

Tidbit: This review was lined up at the last minute–it fell into our laps, really. But we had to round up tapes of the show to send to him, as he hadn’t watched it when the first season was broadcast. Does it show?

Comments on this entry are closed.