Coffee Meets Bagel Meets Me – Statistical analysis of online dating
medium.com
medium.com
> The purpose of this analysis is to understand how to predict whether or not I’ll be compatible with someone based on his profile and the messages that he sends.
One of the best online dating studies was published in Psychological Science in the Public Interest[1].
Its basic conclusion: online dating is great because it expands one's access to potential partners but the algorithms don't really work and the profile-based structure creates an "assessment mindset" that leads individuals to be more picky, judgmental and non-committal. When there's a bunch of new potential matches available every day, you don't have to accept the fact that nobody is perfect. The nice guy or gal you had dinner with last night might pale in comparison to the Mr. or Ms. Perfect who could be in today's batch.
Ironically, the OP's experience demonstrates this dynamic more than it demonstrates merit to her own analysis. After 45 first dates, and what sounds like a reasonable number of second, third, fourth and n dates beyond that, the OP seems content to continue what she calls an "active dating life."
[1] http://www.psychologicalscience.org/pdf/PSPI-online_dating-p...
Also the very concept of "dating" is very much tied to American culture, and one's perception of it (and its goals/effectiveness) will vary greatly depending on culture/religion/upbringing/etc. There's a version of you in a parallel universe brought up in a culture where arranged marriages are the norm, and you'd be posting in this thread about how ridiculous it is that one should date at all - your parents should pick your partner for you!
(I've been in a monogamous relationship for the past ~3 years)
Using the author of this post as an example, I wonder how well aligned her expectations and intentions were with those of her 45 first dates. Note that in one part of her post, the author refers to "meet[ing] someone" while in another part she refers to leading an "active dating life." Although you can obviously expect to meet multiple people before you meet someone with whom you want to start a monogamous relationship, these might be two different pursuits.
The questions around expectations/intentions are even more intriguing in light of the fact that the author says almost all of her dates looked like their photos and that she didn't have "any especially bad CMB/Hinge dates."
That is not necessarily a bad thing.
People change, society changes, is it so surprising the dating scene would change with the advent of the internet? This is but a small step in the direction of Huxley's Brave New World, for example.
Perhaps that's what you meant, perhaps I was too eager to find negativity in an objective post. My apologies.
Adorable.
I am curious why she would use word length and exclamation/emoji/question mark ratios but not check for spelling or punctuation? Surely they are more indicative of someone's level of education and reading habits.
For a variety of reasons (e.g., multi-collinearity), following this procedure would potentially have you tossing the most important contributors to your model. I would use a different mechanism to evaluate the contributors to your model.
A more classical way to tell if something contributes to your model is by evaluating a model with that value compared with the model without it. How do the AIC (or BIC, or LR, or other metric that you like) of the N plausible models compare?
As an aside, the article's approach to evaluating the "significance" of predictors doesn't account for multiple testing. You have ~10 variables in your model, and your best P value is 0.01, which is essentially 0.1 after accounting for multiple testing (Bonferroni), which is not significant classically.
1) R includes the F-test output that accounts for multiple variables in the regression: in this case, P = 0.089, which is a problem.
2) For variable/model selection, R's step() function is easy to use and tests using AIC too: https://stat.ethz.ch/R-manual/R-devel/library/stats/html/ste...
if this guy's only creepy, what does it make the person who tracks punctuation marks, runs linear regressions, and trains learning algorithm using data collected on dates?
Even the Jezebel post itself, with its tracking down and interviewing people in the spreadsheet, is more creepy than this blog post...
If spreadsheet guy looked like Benedict Cumberbatch instead of being "creepy" he'd be, you know, innovative. The kind of guy who thinks outside the box in new and exciting ways. Particularly if he had money.
Lately, a lot of guys have repeated this misconception about "creepy," -- that it just means you're unattractive, and not genuinely unsettling. I'm sure that shields a lot of guy's egos. And maybe, in some cases, creepy really does mean unattractive. But creepy still has it's other meaning too.
That spreadsheet thing would be creepy no matter who did it. Because the guy is trying to approach dating like a math problem (is he socially awkward?). Because he's keeping detailed notes of everyone he's met (who does that?). Because he probably sees women like entries in a spreadsheet.
That's all you needed.
profoundly depressing for both the guy (the crap odds that come from being 1-out-of-45) and for the girl (can't be happy with any one from out of 45)
Besides, it's the communication age, isn't it? Date, my children!
nice sorry hiking house outdoors harvard google drink math cornell
Edit: PS: you're 1 out of 45 either way, but if you never date, that's when your odds are real crap. Dating only a small subset per year, just because you happened to meet; or getting the girl to stick with you because she just happened not to have met that guy who's perfect for her; that sounds much more depressing to me. Unless you believe in fate?
(It's greener because they use more manure over there, which means that here there's manure all over...)
If you're ever in Miami, let's discuss this over a drink.
As a methodological point, I would also have taken the log of all the ratios, especially when doing a linear regression. A ratio of .01 looks like .00001 to a linear regression, but they are quite different. Of course, if the dynamic range is relatively small (probably within 2 fold either direction) maybe it wouldn't matter too much.
Take this with a grain of salt. The statistics of course (well on a larger scale than this article could take aren't wrong) but that an individual is ready primarily to evaluate their own choices in terms of them is rather scary.
I was commenting more on the general approach, of evaluating ones actions or choices within the context of a statistical framework.
If you know a bagel maker, and they have not tried adding extra-fine ground coffee to their bagel dough, might you suggest is please?
My goto breakfast of champions is the Carmans crunchy clusters[1] (honey roasted nut), using a Dare double espresso as the milk.[2]
Soooo gooood
1 http://www.carmanskitchen.com.au/our-products/clusters#Honey...