In Short
This post looks at the differences in popularity between gender and age groups while accounting for screen time (total confessional time).
Key Takeaways
- Age: Players aged 40 and over are less popular than younger contestants (18-39).
- Gender: There is some evidence that female contestants tend to be more “Loved” and male contestants are more likely to be in the “Not a Fan” category.
Acknowledgments
- 🙏 Kosta, the Survivor Fact Checker for the popularity polling data
During my previous post using the Survivor Fact Checkers popularity poll data, a few things stood out. There was at least one group of players that was less loved.
If you haven’t read my previous post, worthwhile doing that first for some background info. I’m going to take a quick look at two factors: Gender and age.
Gender
There is some evidence that women tend to be more ‘Loved’ than men, and men are more likely in the ‘Not a Fan’ category.
glm(formula = p ~ time * gender, data = mutate(filter(filter(df,
score == "Love"), gender != "Non-binary"), wt = 1/sqrt(time)),
weights = wt)
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.094359 0.025585 3.688 0.000363 ***
time 0.020829 0.002882 7.228 8.84e-11 ***
genderMale -0.004112 0.035997 -0.114 0.909273
time:genderMale -0.005641 0.003914 -1.441 0.152492
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for gaussian family taken to be 0.007650569)
Null deviance: 1.45200 on 106 degrees of freedom
Residual deviance: 0.78801 on 103 degrees of freedom
AIC: -107.61
As the confessional time increases, the love percentage for both men and women also increases, but the women’s slope is slightly greater than the men’s.
Similarly, the ‘Not a Fan’ line is greater for men than it is for women. More interestingly, the intercept of the men’s line is higher than the women’s. This suggests that before viewers really get to see and hear from the player, they will form an opinion of whether they like them or not, and men are more likely to fall in ‘Not a Fan’ category than women.
It’s worth pointing out that 3 out of the 4 outliers in the ‘Not a Fan’ chart are men – Sean (quit), Bhanu (did not win a million hearts), and Rome (checks out!). These guys could be dragging up the intercept.
For completeness, here is the model output for the ‘Not a Fan’ model.
glm(formula = p ~ time * gender, data = filter(filter(df, score ==
"Not a Fan"), gender != "Non-binary"))
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.209491 0.033328 6.286 7.96e-09 ***
time -0.005044 0.002797 -1.803 0.0743 .
genderMale 0.100059 0.048309 2.071 0.0408 *
time:genderMale -0.003009 0.003884 -0.775 0.4402
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for gaussian family taken to be 0.02180243)
Null deviance: 2.6148 on 106 degrees of freedom
Residual deviance: 2.2457 on 103 degrees of freedom
AIC: -99.777
This shows that the intercepts are different and supports the above.
Age: Over 40s need more love
One thing I instantly noticed when looking at this data (maybe because I’m an elder millennial) was that very few players over 40 were high on the ‘Love’ rating. This is where we start to see some big differences.
glm(formula = p ~ time * age_grp_bd, data = filter(df, score ==
"Love"))
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.074573 0.027656 2.696 0.00818 **
time 0.019854 0.002116 9.383 1.61e-15 ***
age_grp_bd40+ 0.067081 0.074702 0.898 0.37127
time:age_grp_bd40+ -0.016674 0.007517 -2.218 0.02871 *
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for gaussian family taken to be 0.02427131)
Null deviance: 4.8609 on 107 degrees of freedom
Residual deviance: 2.5242 on 104 degrees of freedom
AIC: -89.179
There is a big difference between 18-39 and 40+ years based on the model output, and it is easy to see this in the chart. The green line is essentially flat, meaning there is no relationship between confessionals and Love percentage for those over 40. But there clearly is for those 18-39.
Similarly, for the ‘Not a Fan’ model. The slope is positive for those over 40, meaning the more we see players over 40 on the screen, the more likely they are to fall in the ‘Not a Fan’ category. This could be a Bhanu effect, regardless, it cuts deep.
glm(formula = p ~ time * age_grp_bd, data = filter(df, score ==
"Not a Fan"))
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.253900 0.026532 9.570 6.16e-16 ***
time -0.006565 0.002030 -3.234 0.00164 **
age_grp_bd40+ -0.058881 0.071665 -0.822 0.41317
time:age_grp_bd40+ 0.012891 0.007211 1.788 0.07675 .
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for gaussian family taken to be 0.02233769)
Null deviance: 2.6295 on 107 degrees of freedom
Residual deviance: 2.3231 on 104 degrees of freedom
AIC: -98.145
Let’s zoom in on the love.
The only player over 40 with a ‘Love’ percentage higher than 26% is Sol on 62%! Wild. And we all know why that’s the case, it’s mostly because of that fine leather vest he was wearing on jury night.
What a hunk.
Also, you’ll notice the line is much shorter for over 40s and that’s because Joe has the most confessional time for that group at about 17 mins where as the most someone has in the 18-39 group is Rachel at 29 mins.
To simplify the analysis, I split the players into those two broad age groups, which is commonly done in the social sciences. If I chart age against the rating percentage, you can see the trend on a continuous basis. Just look at that void in the ‘Love’ category for players aged over 40. At around age 35, it starts to drop off a cliff!
What is it about age in Survivor?
I have written about this before, that when it comes to making it deep in the game, the most influential factor out of race/ethnicity, gender, and age is age, by a long way. Those who are over 40 are much less likely to win the game.
Assuming equal chance, we would expect 12 players over 40 to have won out of 48 seasons, but have only observed 6 – 4 men and 2 women:
- Tina Wesson S2 (40)
- Tom Westman S10 (40)
- Bob Crowley S17 (57)
- Denise Stapley S25 (41)
- Tony Vlachos S40 (45)
- Mike Gabler S43 (51)
Three of those players only just make the bracket as well at 40 and 41. And season 40 was unique since 9/20 players were over 40.
It’s worth doing a deep dive on this because I’m sure there are many contributing factors, such as the perception of challenge performance, the perception of their financial position, and whether they need the prize money more than someone else, and therefore are voted out early.
I reckon a big factor is connection and relating to other players. On average, only 3-5 players in a season are over 40, and only 1-2 players are over 50. There’s an age imbalance. With that in mind, I guess I’m not surprised that there have been fewer winners over 40 than expected.
Just for comparison, in Survivor Australia, there have been 4 winners out of 11 seasons. We would expect 3. Out of those 4 winners:
- Pia Miranda S3 (45)
- Shane Gould S4 (61)
- Mark Wales S7 (42)
- Parvati Shallow S11 (42)
Is there something about the US version and how age is viewed?
I’m also writing this just after episode 4 of season 49, where Matt was voted out. There was a choice between Matt and Jason. Jason was arguably the bigger threat, but the aspect of trust came into it. Players didn’t think they could trust Matt. Age is probably a factor in that trust equation.
On top of that, Annie was the 2nd boot for the season. Nate was the other player over 40 and considered to be on the bottom of the Uli tribe. So, in 4 episodes, 2/3 players over 40 were voted out, and the remaining player over 40 is on the bottom.
This was bit of an aside but I found this fascinating since the same trends that are seen in the game are also seen in the polls.
Follow me on social media:
