I want this to go smooth and by the numbers …

The chart above shows hits on my old blog post A Game of Thrones and the Misogyny of Imagined Worlds between 23rd February and 19th May 2012. The second season started on 1st April, and there’s a clear series of weekly peaks in views on my post, that occur between Tuesday and Thursday each week – corresponding to blog searches in the USA on Monday – Wednesday. The peak varies by week, but they’re a week apart and anyone who has any experience with time series can see in that data a shift in level, probably change in variance, and strong seasonal signal. Additionally, the height of the peak varies from week to week and I soon noticed it corresponds with just how nasty the treatment of women was in the episode of that week. So, in this post I’m going to show the effect, and give a numerical estimate of the extent of misogyny in each episode of A Game of Thrones, using crowd-sourcing based on google hits on my blog. Note that almost all the hits shown in this data series – 245 in April – are from google searches (though I think one week there might have been a link put up on facebook). This post is nearly a year old, and usually my year old posts (bar one or two techy ones that attract continual regular hits) get very few hits, and certainly never attract a pattern.

Methods

(Skip this if statistical methodology makes your eyes bleed).

I built a simple log-linear regression model using time, cyclical pulse functions for Monday, Tuesday and Wednesday (the date for the first, second and third day after the show is released in the USA) and episode. Episode was set to zero for all days before the 1st April. This enabled the peaks to be (partially) fitted, and also allowed a specific magnitude effect for each episode. The model was not adjusted for serial correlation because serial correlation is notoriously hard to fit in series with low numbers of observations. I also suspect that the tight nature of the seasonality (7 days) compared to the unit of observation (1 day) in a short series and the sharp peaks will prevent models from converging, and this is what happened – I tried fitting a generalized estimating equation model with exchangeable correlation and got divergent estimates of correlation. Standard backwards stepwise model-building methods were used to eliminate unnecessary variables, with the usual strict inclusion criteria (a variable was out if it didn’t get a p-value less than 0.05 on a Wald test). Given the small number of observations, small counts in early parts of the series, and the fact I was driving this model like a Westerosian Ox-cart as it was, I figured it was best to avoid interaction terms of any kind.

The exponentiated coefficient for episode 1 incorporates a general effect of increased interest in the show with the onset of new seasons, but by calculating ratios of other exponentiated coefficients, one can estimate the degree of misogyny for all episodes relative to episode 1. Predicted values were also produced and plotted against observed values.

Results

Results of the model are shown in Table 1. Hits increased by a factor of 2.4 for episode 1 compared to pre-release hits, and each week the peak was on Wednesday, when hits were twice the values for Monday or Thursday. Hits declined by 1% per day over the period of data collection.

Table 1: Model Results
Variable Odds Ratio 95 % CI P value
Intercept 4.69 3.44 – 6.38 0.001
Time 0.99 0.98 – 1.00 0.2
Tuesday Peak 1.44 1.09 – 1.91 0.01
Wednesday Peak 2.06 1.61 – 2.63 <0.001
Episode 1 2.38 1.50 – 3.78 <0.001
Episode 2 2.58 1.51 – 4.43 0.001
Episode 3 1.95 1.03 – 3.72 0.04
Episode 4 3.63 1.79 – 7.36 <0.001
Episode 5 1.31 0.56 – 3.08 0.5
Episode 6 2.07 0.82 – 5.20 0.1
Episode 7 3.04 1.12 – 8.27 0.03

Note that episodes 5 and 6 are indistinguishable from background (pre-release) noise in their intensity of misogyny. Figure 1 shows the predicted values derived from this model, plotted with the observed values, and shows relatively good fit, although the model fails to reach the dizzying heights of misogyny displayed in some of the observed hits; but this could be the fault of that facebook link creating outliers, so we won’t get too angsty about that.

Figure 1: Observed and predicted numbers of views

From these results we can estimate the relative degree of misogyny of each episode, shown in Table 2. All estimated misogyny ratings are relative to episode 1.

Table 2: Misogyny Ratings for Season 2, Episodes 1 – 7
Episode Rating relative to episode 1
Episode 1 1
Episode 2 1.08
Episode 3 0.82
Episode 4 1.53
Episode 5 0.55*
Episode 6 0.87*
Episode 7 1.28

*Not significantly different from background noise.

Thus, the most misogynist episode was episode 4, while episodes 5 and 6 were indistinguishable from background noise. Although episode 3 was of lower misogyny rating than episode 1, it did attract significantly more views than during an equivalent period before the release of the show.

Conclusion

Viewers around the world increased their rate of searches about misogyny in A Game of Thrones, and rates of searching increased most in the days immediately following the release of each episode. The most misogynist episodes were:

• Episode 4: Sansa is stripped and beaten in public in the throne room; Joffrey forces one prostitute to beat (or sodomize?) another after they are sent to him to “sap the poison” though he will remain “a cunt”; Malesandre gives birth to a wicked shadowy abomination after an improbable pregnancy
• Episode 7: The strange and overly sexualized encounter between Jon Snow and his prisoner in the North; Sansa relives her near-rape in the previous episode; Jaime successfully taunts Catelyn with the memory of her husband’s infidelity; Daenerys loses her dragons and is reduced from her previous pride to dependency on Ser whatshisface
• Episode 2: Theon Greyjoy uses a woman like an object in his ship, abandons her and then fails to recognize and subsequently tries to fuck his own sister

In contrast, the least misogynist two episodes were:

• Episode 5: Theon is shown no the respect by his men, who obviously fear and obey his sister; Brienne proves herself to be tough and honourable; Daenerys rejects a marriage proposal; Arya is generally excellent
• Episode 6: Talisa the field nurse shows herself a spirited pacifist; Arya is once again generally excellent; Osha the wildling girl saves the Winterfell boys; this episode has some rape scenes and the delicious hypocrisy of Cersei’s attitude towards Sansa compared to her own betrothed daughter, which would be why its misogyny rating is higher than episode 5 – presumably Arya and Talisa save it from being higher

I think the relative ratings are generally quite representative, though perhaps an adjustment for the downward time trend needs to be incorporated to make them more accurate. This model fairly accurately fits to the data on search hits for this topic, and in my opinion sorts nicely between the most and least misogynist offerings. There is strong evidence that web search numbers correspond with the density of common signifiers of misogyny in any one episode (rape, mistreatment of prostitutes, degradation of childbirth through black magic, vicious anti-woman language, use of women as sex toys without regard for their feelings or identity, and women’s sexuality being either used as a tool for personal gain or expressing itself in incongruous neckbeard-fantasy ways). Hits are much lower on weeks where strong female characters take control of their own lives and act sensibly, even where their situation is difficult and/or oppressed.

Obviously, no one believes that google searches are reflective of some underlying truth about what is or isn’t misogynist or sexist. But I think they do at least show that a lot of people are disturbed by the images and themes in the show – disturbed enough to get on the internet and look to see what others think of the issue. This show has a deep streak of misogyny, and it isn’t going unnoticed.