Tag Archives: probability

“Should I Dump My Girlfriend? Will I Find Another?”

Well let’s not get too hasty.

Earlier this week, I posted the Nanaya white paper to ArXiv: “Should I Dump My Girlfriend? Will I Find Another?” Or: An Algorithm for the Forecasting of Romantic Options. The title is a bit melodramatic but it’s a reference to Peter Backus’s “Why I Don’t Have a Girlfriend.” Apologies to everyone, especially Dr. Backus.

So this paper summarizes how the Nanaya algorithm works. I genuinely wonder who will have actually read it. Theory papers are hard enough to get through on their own and then harder yet when it’s an interdisciplinary mush. So we’ll go over some of the key points here.

The first major issue is uniqueness. Did we do anything academically worthy? Well, a lot of people have worked on the Secretary Problem (to be covered in a future post), but no one has really tailored it for a situation like romance. There are simplifications that do little justice to the complexity and information available to people – but those aren’t useful. Let’s just call it overzealous journalism…

We also do something new: comparing a specific relationship to being single and the chance of any other relationship. This is options analysis and rather new to the problem.

The Nanaya process.
The Nanaya process.

The rest is explaining what’s going on in the algorithm. Let’s check out the first figure:

So let’s break it down by each step:

Assessment of Inputs

When the Nanaya Beta is up (pending funding, be sure to call your local, friendly Venture Capital fund and request them to fund Nanaya) we’ll be asking a bunch of questions to understand your life circumstances:

  • What is your personality and social and romantic history. This gives us an idea of how likely you’ll be successful in finding a match, attracting a match, and how happy you are when you don’t have a partner in your life. We also need to know how long you’ve been in the city and job you’re in as that has a huge impact on how many people you’ll be meeting. You can check out this blog post that goes over this detail.
  • What is the ideal This is a mix of personality and specific values that are shared.
  • If you’re with someone, how do you feel about them.
  • The groups you interact with, like at work, geographically, and in socializing.

Determining Match Probabilities

With all of that data we can use our database and others available to us to figure out what the chance is of finding a match in any given encounter, depending on the group.

These numbers are typically low. Something like 1-in-10,000 is not unreasonable. This is where statistics comes in to play and you have something like the birthday paradox to offset those low odds.

Sociological Modeling

This is where we mix the match probabilities with your social behavior to figure out what are the chances of actually meeting someone. This is a bit of sociology and a bit of statistics to solve a problem no one’s really touched before, at least in this context (if I’m wrong, let us know). The sociology is based on personality, group types, and the results of some of our prototype experiments. The probability is based on binomial distributions and a variation of the Urn Problem. This is how we describe our social interactions…with urns!

The Nanaya implementation of the Urn problem
The Nanaya implementation of the Urn problem

We admit we make some simplifications in the paper, namely use of binomial probability distributions and assumptions that populations are very large. For almost everyone, these are pretty good assumptions but not for people in small populations. This gets into a lot of theoretical combinatorics which is well understood but not yet implemented.

Utility Function Valuation

This is the ugly philosophically ugly part where we literally put values to things that ordinarily don’t have numbers associated with them: like compatibility and happiness.

Using personality test results for who you are, who you want, and who your partner is we can estimate how happy you’ll be in any given relationship. We can also use your personality and romantic history to gauge how happy you’ll be in time as you’re single.

To simulate a random match you meet through the course of your life, we run a Monte Carlo simulation based on the groups you interact with. Some groups will have their own types of traits that lead to specific archetypes or segments (in demographic-speak) – we want to respect what’s observed in reality so we use these archetypes in seeding the randomness of the Monte Carlo.

Reporting the Results

With all of the above done, all that’s left is showing the numbers in a way that’s easy to digest!

If you sign up and take our personality tests, we’ll be inviting you to join our Beta program when it’s ready! So be sure to take your personality test and start today!


Does Online Dating Actually Boost Your Odds?

This is a bit of a no-brainer: online dating increases your exposure to more people than you ordinarily would run into each day. Sometimes that’s not so good:

But let’s face it, it’s not always easy to meet people you want to meet and the one might be out there. There’s pretty minimal investment in online dating, for better or worse.  If you use a site like eHarmony or OkCupid, they do a bit of work in figuring out your personality and finding matches that you’d likely be compatible with. That makes things even easier.

But is it worth that minimal investment? The costs of premium services, frustration of having your messages ignored, objectifying remarks, and simply the wherewithal needed to maintain yet another online account?

If you have thick enough skin, yes. It’s all in the math.

We can use a few portions of the Nanaya’s algorithm to show the impact. At a high level, there are two major factors that govern whether or not you will run into a match: how you interact within groups you belong and the chance that, upon any given encounter within a group, that a person will be your type.

I’ve used values from a volunteer who lives a rather normal, post-college life in the same city she was raised. She isn’t too picky but because there isn’t too much variation in social habits, the odds of someone new popping up aren’t distinctly high. In fact, the ideal environment for meeting compatible matches, work, is also one of the worst because of low turnover rate. She isn’t very picky but just enough that people she’ll run into out on the town will probably not be her type but a group of close friends has a different distribution of personalities than the city itself. Distant friends, family, and strangers met traveling are the smallest contribution.

Let’s look at her odds without online dating:

Volunteer's odds of finding a match without online dating broken down by group.
Volunteer’s odds of finding a match without online dating broken down by group.

The relative probability is shown for each group normalized by the “all groups” probability for that given point in time. You can compare this to the “stacked line” plot on the main page of Nanaya.

So what happens if she starts to go on a few online dates from people online starting in a year at a frequency of about one date every two months?

Volunteer's odds of finding a match with and without online dating.
Volunteer’s odds of finding a match with and without online dating.

We can see the group breakdown for the real effect of online dating. It’s pretty dramatic, especially considering how many more people she meets elsewhere!

Volunteer's odds of finding a match with online dating broken down by group.
Volunteer’s odds of finding a match with online dating broken down by group.

Some people might notice that it’s seems incorrect to separate potential matches from the local area from other, similarly local places (i.e. work and online). There’s undoubtedly validity to that argument but biased sampling, i.e. real human interactions, indicate otherwise. There are different personalities demographic, especially considering the size of the local population and the way data is aggregated. We also assume that there exists a sufficient sample of compatible men online for the volunteer to date, but most metropolitan areas are well populated when it comes to online dating and most sites do a good job of filtering. It’s that filtering that really helps the “signal to noise” problem in finding compatible partners.

Fortunately for my volunteer dating, let alone online dating, isn’t an issue. She’s actually happy in a relationship! After running the prototype, Nanaya predicted she should stick with it. A decision she had already confidently made. Good news for the volunteer and good news for continuing validation!

But the impact of online dating isn’t exclusive to that volunteer. All validation studies have shown it to be a rather significant part of improving romantic outcomes, ceteris paribus, anyway.

If you are interested in increasing your odds in finding a match, I recommend having an online dating account. Even better if the site is well trafficked and has personality testing. Log in now and again to search for new people and check on your inbox, but don’t take it too seriously. Why? That gets into ceteris paribus and that’s the subject of a later post.