Predicting Survivor 6: Predictions from an Empirical Model

In previous articles outlining theory and methodology, I created a model to try to predict contestant’s performance in Survivor: The Amazon. This article explains how to apply the model and (most importantly to those of you readers with eyes glazed from trying to understand the other two articles) the predictions.

When performing OLS or discriminant analysis, for each variable a slope coefficient gets produced in this form, for scores on each case all cases aggregated together:

dependent variable = constant + coefficient1 x variable1 + coefficient2 x variable2 ... + error

(Where error is whatever the difference is between the actual placement or category of a particular player was with the predicted placement of category according to the equation – the more often this is closer to 0, the better the model.) Thus, knowing these coefficients, we can take the scores of future players on the variables associated with these slopes, plug them into the equation, and produce a predicted placement (OLS) or category (discriminant analysis).

I compared the various models outlined in the previous article, looking for the best two-group, three-group, and OLS models. I wanted different but related models because of the richness they would bring in seeding the players’ performances. Within each kind, I took for predictive purposes the model that had the best fit (most significant F-score for OLS, most significant Wilks’ lambda for the others). The previous article contained some discussion about the models’ accuracy and in it I essentially gave away my choices of models. These turned out to be the OLS model with significant variables only, the final four-jury-non-jury three group model with only significant variables, and the finals-non-finals group two-model with all variables included.

What we know about from detailed model discussions in the last article leads to a decision rule to place S6 contestants. We saw the high degree of accuracy the discriminant models had in distinguishing between final four and not (the two-group did well, sometimes overestimating chances of those who didn’t finish in the final four, and almost all three-group errors came between the non-final four categories). Thus, the grouped models can tell us whether a person is realistically final four material, and then OLS can distinguish between the rest of the jury or not, and/or in borderline cases.

Bearing this decision calculus in mind, the following lists first names, alphabetically, starting with the females, of the S6 contestants, along with their predicted placements first by category for each group analysis, then by exact placement according to OLS, with the final category my judgment synthesizing the results of the three approaches:

The only real dispute between methods is the fates of Rob and Ryan. OLS ranks them highly while the grouped analyses put them as borderline jury/non-jury members. The fact that Janet and Jeanne were tagged by both grouped methods to be in the final four and their relatively high OLS scores was enough for me to make them more likely final four members than the two males (the females’ OLS scores were barely worse than the males’ in any event). I give Daniel the nod over Deena for another spot given their comparative OLS scores. Roger comes in as the undisputed best scorer in all three models. Using these models, I feel confident in predicting that the final four likely will come from among Daniel, Deena, Janet, Jeanne, and Roger, and that Roger is an odds-on favorite to win.

Now for the disclaimers: Note that the results given above represent predictions, not explanations, because the event has yet to happen. Note also that a model is only as good as the theory behind it, but I believe the theory to be quite plausible as the predictive power of this model shows. Note still that the model was built upon publicly-released information prior to the series’ start; other information may be dramatically improve predictive abilities when incorporated into the model, but if it is not readily available then chance plays a role in correct prediction, insofar as the fortune of getting certain information others cannot. Finally, chance cannot be completely modeled away and predictive models are, by their nature, unable to incorporate chance since it has yet to happen.

But the biggest disclaimer here is this ordering, of course, looks at the 16 as a whole, not in the segmented way that the initial placements actually occur, i.e. influenced by tribal rather than individual immunity challenges. Particularly the format of Survivor 6 also may not lend an analysis of it completely congruent to the past versions, given that this is the first to have sex-segregated tribes. This particularly is apparent in that this prediction would have the women winning four or five of the six tribal immunity challenges (assuming the merge happens after the sixth plank-walking). The dynamics this arrangement will produce may make this series uncomparable to the others and I may have wasted several hours of my time doing this analysis.

Nonetheless, in comparison to more subjective predictions made by my RNO colleagues, I anticipate this model should do quite well. While our editor did tag the winner of S5, the four staffers who picked all places for S4 did not fare so well. Of the 16 picks for the final four contestants in S4, only two correct ones were made, and only one of them was the winner. If past performance is any indications, at the very least, this model ought at worst to get at least as many final four participants right as it does wrong. Time and especially reality, the thing we observe and report on at this site, will tell.

Jeffrey D. Sadow is an associate professor of political science at Louisiana State University in Shreveport where he teaches, among other things, classes in international politics, international organizations, and diplomatic history. He has published in the area of gaming simulations in international politics.

Be sure to sign up for our e-mail update so you can stay informed about new articles on the site! And take a look at the rest of the site. You can find all of our recent Survivor articles at the Survivor: The Amazon page and take a look at our sections on Joe Millionaire and The Osbournes. You can even buy reality show stuff at our Reality TV Store!

For more news about reality TV, be sure to check out RealityTVFans.com and SirLinksALot!