h2>Dating : Most Data is Normal: Dating is Anything But
You could plot the surface area:volume ratio of whales from low to high (more spherical to more oblong), and I’d bet it’s fairly normal.
However, scientists hypothesize that Beluga whales’ lower relative surface area & lack of dorsal fins are adaptations to minimize heat loss in their colder arctic environment.
Many powerful statistical models such as Naive Bayes, Linear and Logistic Regression & LDA assume that the data is normally distributed. Many calculations, such as the sigmoid activation functions used in most neural networks, are much easier to perform on normally distributed data.
As such, data scientists often perform transformations (linear, log, etc) to convert datasets into normal-approximate distributions.
Interestingly, even if your data isn’t normally distributed, if you take many sub-samples from the original set and distribute their averages (sample means), you’ll begin to approach a normal curve, due to the strange power of the Central Limit Theorem.
Statistics itself is more about estimating values rather than exactly predicting them with full certainty, and knowing whether a variable is normal makes calculating these probabilities much easier (especially when standard deviation & variance are thrown in).
However, even though these normal bell curves appear often in nature, there’s plenty of situations in social and psychological sciences where the data is more uneven.
Before we check for normal distributions in online dating, let’s first ask:
Is normally distributed romantic data a “good” thing?
Depends on the context. Let’s take a look, courtesy of the aforementioned archived articles. Chris Rudder specifically mentions that he didn’t normalize any data:
The dotted line denotes men on OkCupid ranking women’s attractiveness from 0 to 5 based on their profile pictures.
This is somewhat normal, although it has very low kurtosis — thick “tail” distributions, or high probabilities of extreme values relative to the mean.
The blue line denotes “messages received by women” relative to the same male-perceived attractiveness rating.
This is much less normal; as the author puts it, “two thirds of male messages go to the best-rated third of women”.
The same chart with women ranking men illustrates a very different distribution.
Women rate 80% of guys’ looks as below average. If male attractiveness (as a physical trait) was normally distributed, this would be closer to 50%.
They also send messages to men in a more “realistic” manner compared to men’s discrepancy in rating and messaging women; women send far more messages to guys they consider a 2 than a 5.
Rudder then looks beyond ‘messages received’ to actual reply rates for users’ own messages, which should be more accurate measurements of desirability.
The relative evenness of these two lines initially appears to show a balance between the genders — they both gain about 30% reply rate going from to 5 rankings.
But the previous graph seems to show that ladies send more messages to ‘average’ dudes than Abercrombie models.
So why do we see so many complaints about online dating?
The New York Times quotes a Tinder study, finding that women swipe right on 14% of men on average, and men swipe right on 46% of women.
Aviv’s article can help us interpret this further. He mentions that the difference between men and women’s mutual attractiveness distributions plays out in ‘likes’ received by both sexes:
On Hinge, men send a first like more than three times as often as women. But more than that, when women do initiate, they tend to do so on a smaller segment of the male population.
For example, while about half of all likes sent to women go to about 25 percent of women, half of all likes sent to men go to a much smaller segment — about 15 percent.
As he puts it, “the Brad Pitts of the world take the lion’s share of the likes from an already like-deficient sex”.
Aviv also mentions a Medium article by ‘worstonlinedater’ from 2015 that analyzes Tinder data from the perspective of the author, the sample size being one straight male interviewing 27 women who liked his (artificial) profile.
The author calculated that “the bottom 80% of men are fighting over the bottom 22% of women and the top 78% of women are fighting over the top 20% of men”, and graphed the following percentiles against each other.
The extremely small sample size (n=27) is usually enough to discredit such findings, but ‘worstonlinedater’ calculated Tinder’s straight male Gini coefficient at 0.58.
Using magnitudes more data from Hinge’s servers, Aviv calculated the same Gini at 0.54. Both would reflect the same level of economic inequality as South Africa, Brazil & Mozambique.
Earlier I wrote “these sites don’t often redact blog articles”, but while writing this article I discovered two other deleted posts from OKC, “The Mathematics of Beauty” and “Race and Attraction, 2009–2014”, both of which use large swaths of data and reach some fascinating — but potentially upsetting — conclusions.
There may be a pattern here. Rudder has written a book, “Dataclysm: Who We Are (When We Think No One’s Looking)”, that seems to cover these topics more in general, so perhaps I’ll look for a copy.
But these trends are here to stay. Online dating is on the rise as shutdowns prevent traditional opportunities to meet new people. The genie is firmly out of the bottle, so how should we consider its effects?
Normal probability densities are found in nature, but so are poisson, log and every other distribution under the sun. Something naturally appearing doesn’t necessarily mean it’s “good”, either. Mosquitoes are natural.
If people’s romantic experiences are normally distributed, does that make things more “fair”?
Is fair something we should hope to artificially engineer, given that love is inherently a personal experience?
At any rate: If this data reflects reality, then love is indeed a battlefield.
