How to Choose a Goddess (Using a Spreadsheet)

After years spent perfecting my OkCupid game, my efforts hit home in April 2015. Two wonderful girls replied to me on the same day, and two first dates were set up for the following week. Sarah, a redhead Jewish community organizer. Terese, a blonde biologist. My Biblical namesake also knew a thing or two about choosing among two women, but unlike in olden times I didn’t think that marrying both would be a viable option. By May, both relationships had progressed to the point where I was going to have to choose one and break up with the other.

Hey ladies, did you see how long my “staff” is?

I knew two things to be true:

  1. This could potentially be the most important decision of my life. I absolutely had to get it right.
  2. Rationality taught me that this is exactly the kind of decision that humans are terrible at.

The study of rationality interlaces two threads: the math of good decision making and the cognitive psychology of bad decision making. Our brains aren’t well suited to the former, instead we use simple heuristics that work for simple, commonplace decisions.

Unfortunately, the most important decisions we make in our lives are neither simple nor quotidian: what to study, where to work, which house or car to buy. These decisions come up infrequently. They involve comparing several options that differ on many variables and carry long-term and uncertain effects. Our brains are short-term oriented, clouded by transient emotions, and often fixate on a single factor instead of evaluating the broad picture.

That’s not a promising way to pick a girlfriend.

The study of heuristic decision making became a science in the last 50 years, but a story from roughly Biblical times illustrates the issue perfectly: the judgment of Paris. Paris is a (secret) Trojan prince living as a cowboy on a  lovely Mediterranean island, passing his days by staging bull fights and cavorting with nymphs. Paris is known for his good looks, big cojones, and for his great judgment. When the three great Greek goddesses quarrel over which one of them is the most beautiful, even Zeus declines to pass a verdict lest the two spurned goddesses turn against him. The decision is delegated to Paris.

Paris doesn’t shrink from the perilous task, he relishes it. He appraises Athena, Hera and Aphrodite, then makes the goddesses strip and parade in front of him naked, then finally he asks them for bribes. I told you the dude was ballsy.

Hold on, let me call some from “Miss Universe” for advice

Hera offers Paris a great empire stretching across the continent. Athena offers him skill and wisdom. Aphrodite offers him Helen. Let’s step into Paris’ head for a second and jot down a quick list of pros and cons Re: Aphrodite’s offer.


  • By not picking Hera or Athena, Paris misses out on emprire, skill and wisdom.
  • Paris is already married.
  • He’s married to Oenone, a sexy nymph with magical healing powers. And they have a son.
  • Helen is already married.
  • Helen is married to Menelaus, the king of Sparta, a city known for holding grudges and resolving them with violence.
  • Menelaus has 44 Greek kings and warlords sworn to support him if anyone messes with him or his wife, up to and including declaring war on Paris’ hometown of Troy.
  • Among the Greeks is Achilles, an invincible(ish) warrior, about whom it is prophesied that if he joins the war Troy will defeated. This means a likely violent death for Paris, his entire family, and his city.


  • Helen is smokin’ hot.

Paris could have really used Athena’s gift here, or even modest common sense. Instead, he picks Aphrodite and 10 years later his family is dead, his city is in ruins, and Paris dies of multiple stab wounds after begging Oenone in vain to save him.

In my opinion, there are two reasons why Paris makes the wrong call here. First, he just watched three goddesses dance naked in front of him. He’s so aroused that his brain can only consider a single decision variable – lust. Second, Paris is unfamiliar with the wonderful method of evaluation matrices. A simple spreadsheet could have saved him a world of hurt, and maybe help me make my mind up about Terese and Sarah.

The Why and How of Matrixcraft

An evaluation matrix, also called “decision matrix” when used for boring decisions like buying a car, allows you to compare multiple alternatives on multiple attributes at once.

Here’s an evaluation matrix of pets:

pet evaluation matrix.png

The main goal of the evaluation matrix is to measure each option on a single, quantifiable scale. This is the same logic behind shopping for happiness: by measuring each purchase on a single $/😊 scale you can make the smartest $-😊 trade offs.

An evaluation matrix allows you to integrate a lot of information in as objective a manner as humans are capable of. In Kahnemannian terms, the matrix allows your System 2 to take part in a decision making process that would normally be governed by System 1 alone. The steps and skills used in crafting a useful matrix are all about making the most of available data and staying unbiased.

Step 0 – If you know your destination, you are already there

Needless to say, don’t bother evaluating choices with a spreadsheet if you know which option you’re going to choose ahead of time. If your gut (System 1) has settled on a course of action, a matrix will not change your fate – the numbers will magically come out to support the decision you have already made. The math will only hurt you: a pretty spreadsheet makes it all the easier to rationalize the choice you have made, and makes it harder to revisit the choice later.

Decide ahead of time what you will use the matrix for. Do you commit to picking the option that will score the highest? Will you choose it only if the gap is large enough? Is your purpose to discard all the low scoring options to narrow down the list of choices to two or three? The spreadsheet won’t hold a gun to your head, but precommitment will keep you honest.

Step 1 – Generate a list of attributes

You want an exhaustive list of features to compare your options on, so that two options that are equal based on the features you made explicit are in fact equal in their attractiveness. If you end up with too many attributes you can always combine similar ones into a single factor later. The advanced skill that can be employed for feature listing is goal factoring.

Goal factoring is a CFAR technique designed to evaluate whether a goal is worth pursuing, but it can be co-opted for our purposes. It’s a complex skill that requires training, but at its core is the question: If I had all this, would I be content?

For example, let’s factorize a car:

Car factor 1.png

Now ask yourself: if I had a cardboard box that cost $2,000 a year and that teleported me to work after I sat in it for 30 minutes, would I be content? Actually imagine yourself sitting in the gray box and popping up at your destination some time later. Yes, it will get you from point A to point B, but it’s not a very fun experience. Also, you’re a weirdo that sits in a box. If the cardboard box leaves you wanting, you need to add more factors:

Car factor 2.png

Spoiler alert: social desirability (what others will think of your choice) is almost always going to be an important factor, whether you’re buying a watch or choosing a date. Don’t fight it, this exercise is all about being honest with yourself.

Step 2 – Determine the range and scale of each attribute

To measure a feature, you must decide on a scale for it along with some reference points to anchor the scale. The reference points should cover the entire range of the attribute, and should be based on options other than the ones you’re ultimately choosing from.

Some attributes come with an obvious scale, like a price that is measured in dollars. It’s better to stick with those than to convert everything to a 1-5 scale, the point is to make the differences on each attribute obvious and emotionally salient.

For example, the cuddly-fuzziness scale for pets starts with a snail at 1/5 fuzzies, a short haired dog sets the midpoint at 3/5 and the fluffiest bunny is at 5/5. Once the scale is set, it’s easy to rate the options consistently. Rating a hedgehog on an anchorless 1-5 scale is hard, but it’s easy to see that a hedgehog has 1 cuddly side which is more than a snail (0 cuddly sides) but less than a dog (2 sides). So, we rate the hedgehog at 2/5 fuzzies. A dragon can fly by flapping its wings laboriously, which falls short of the hedgehog’s effortless levitation and earns the dragon a 4/5 rating on flight.

Step “Hold on a second, hedgehogs can’t fly!”

flying hedgehog

Sure they can.

Step 3 – Assign weights to a unit difference in each attribute

The weights should represent the importance of each factor to the final decision. The critical thing to remember is that the weight is relative to a specific increment of the attribute. In the car example, instead of abstractly comparing price to acceleration to coolness you would weigh $1,000 vs. a second gap in going 0-60 vs. the coolness increment from a BMW to a Tesla.

If you’re ready to get serious about this, you can use a method that companies employ to determine pricing: pairwise ranking and regression. The process of pricing a product is similar to what we’re doing. It consists of converting a bundle of product features into a single number: the customer’s willingness to pay for the product in dollars.

Here’s how it works, in a nutshell:

  1. Come up with a list of options that differ from each other on each attribute, such that every one rates high on some attributes and low on others.
  2. Rank the options by comparing them two at a time. For example, would you rather pay $25,000 for a car that goes 0-60 in 5 seconds and looks like an old Toyota or $22,000 for a car that goes 0-60 in 8 seconds and looks like a new Mercedes? Move the option you choose higher in the rankings and the other one lower.
  3. Run a linear regression on the resulting list, with the final ranking as the dependent variable and the attributes (price, acceleration, looks) as the explanatory variables. The regression coefficients for the attributes are your weights. If the coefficient for for acceleration (measured in 0-60 seconds) is twice as high as the one for price (measured in thousands), a second of 0-60 acceleration weighs twice as much as $1,000. This means that shaving a second off 0-60 is worth $2,000 to the person doing the ranking.

If you’re confused but curious and live in North Carolina, you can ask the professor who taught me this yourself. He’s brilliant and fun to hang out with, buy him a beer. If you’re confused and don’t care, just ignore this part and make up weights for the factors yourself.

In all the steps, this one in particular, remember the credo of Putanumonit: a guesstimated number is better than not using numbers at all.

Step 4 – Score all on the options on one attribute at a time

As opposed to scoring one option at a time on all the attributes. The purpose of this is to avoid the halo effect, which can infect even the great scientist who discovered it. If you rate a single option on all the attributes, the scores you give on one aspect will affect your scores on the others. You’ll find yourself “adjusting” and “rounding up” a lot based on your overall (semi-conscious) desire for one alternative or another to come out on top.

Scoring by attribute keeps the measuring scale foremost in your mind, which makes your rating more accurate. If Paris wants to make Aphrodite “win” the matrix, he will be inclined to say that having a hot wife rates high on categories like “life fulfillment” and “personal success”. However, if he had to compare having a hot wife directly to ruling an Empire and not getting stabbed in the “personal success” category, it would be harder to avoid the correct conclusion.

Step 5 – Multiply each score by the attribute weights, and sum up the scores for each option

pet evaluation matrix

This is the part with the actual matrix. A fox scores 5 on “breadth of knowledge”, and knowledge breadth is weighted at 2 points, so we write in 5 * 2 = 10 in the Fox/breadth cell of the table. The product, 10, combines both the importance of an attribute and the animal’s score on it. We add up these products to arrive at the total rating of each animal, in the fox’s case – 50.

Step 6 – Stare at the bottom row. What do you feel?

Does anything surprise you? Did some option score a lot lower or higher than you expected? Are you ready to implement the matrix-based decision procedure you decided on in Step 0 or are you having regrets?

The matrix has done what it can for you, now you’re on your own.

hedgehog in car
The clear choice in any decision matrix

The Life Partner Matrix

Sarah and Terese, Terese and Sarah. I kept changing my mind back and forth, several times each day. I asked my friends for advice. I asked my mom. I asked my ex. I had trouble sleeping.

Both are gorgeous. I had more immediate chemistry with Sarah, Terese was slower to open up but more intriguing. Sarah perfectly fit my daily life in New York, Terese resonated with my philosophy and worldview. It was impossible to compare them.

So, I decided to do the impossible. I opened a spreadsheet.

Step 0 – Mindset

I definitely wasn’t precommitted to either one, but I didn’t really settle on a decision rule ahead of time either. I wasn’t sure what this exercise was going to accomplish, and I was ready to bet that after hours of work both would end up with the exact same score down to the decimal. I was hoping that I would at least clarify to myself what I liked best about each one.

Step 1 – Attributes

This was the critical part, figuring out what I’m actually looking for in a girlfriend. I goal factored my previous long-term relationships, thinking about every interaction, the positive and negative. I planned ahead through the stages of a relationship, from the initial crush to the routine of living together. I devoured several books. I read a couple of blog posts by a single dude on the internet, which made more sense to me than all the books written by married PhDs.

I ended up with 5 main categories, which I then broke down into 15 distinct characteristics.

  1. 20,000 Wednesdays – Based on Tim Urban’s idea that a relationship is made out of 20,000 mundane Wednesdays together, not 2 or 3 highlight vacations. Great Wednesdays are made out of great friendship. This category included “Can talk on my level about big things I care about”, “Honest and emotionally aware”, “Shared interests and pursuits” and “Cracks me up + gets my jokes”.
  2. Hotness and sex. 
  3. Social life – How easy we would integrate into each other’s social life and how much the people whose opinion I care about (family, close friends) will like her.
  4. Building a future – How compatible are our work/life plans, will she be a great mom, is she likely to become rich enough that I could quit my job and write Putanumonit full time.
  5. Reciprocity – Are the attraction and admiration one-sided or mutual.

Step 2 – Scaling

I did this mostly by ranking ex-girlfriends and female friends I knew well. This also increased my confidence in the exercise: the women I liked the most and had the best relationships with scored higher than short-lived flings.

Step 3 – Weighting

I weighted the attributes by spreading 100 points among them. Reciprocity should technically be worth 50/100 points since it’s half the relationship, but I knew a lot less about Terese and Sarah’s impressions of me than vice versa. I ended up making this category worth 27 points, joint-highest along with Wednesdays. Same went for Future: I lowered the weight of factors like “will be a great mom” because they’re very hard to judge and differentiate after a month of dating. Hotness and sex are very important, but the variance wasn’t as high in the other categories since I thought that every girl I ever dated was really hot and sexy. Finally, Social got the lowest weight with 8/100. My friends’ and parents’ reactions are hard to judge, and ultimately not as important to me as things that affect our relationship daily.

Step 4,5 – Scoring and multiplying

The exact ratings I assigned are known only to me and the 17 intelligence agencies that spy on our every keystroke.

Step 6 – Reckoning

I expected a close race with a slight edge for Sarah, but when I looked at the totals Terese came out way on top. I was quote surprised, and dug into the numbers.

The biggest difference was in Futures and Wednesdays, the categories related to partnering up for a life together and the last thing on your brain when it is flooded by the testosterone and dopamine of limerence. My ratings for Terese on these were based on limited thin slices of information, like a single conversation. These bits of information would not have even come to mind had I not made myself focus on one attribute at a time.

Also, when forced to putanumonit, I had to admit that Terese was hotter.

Two days later I broke up with Sarah. It’s hard to say how large of a role the matrix played, whether it tipped the balance by itself or just cleared my mind. It certainly helped me understand what I like and admire about Terese: as we kept dating and her shyness dissipated she justified and then exceeded the high marks I gave her intellect, humor and emotional maturity.

After a couple months, in summer 2015, I told her about Sarah and the matrix. She reacted with excitement and curiosity. That was the first time I knew for certain that I had made the right choice. By 2016, we had moved in together and bought a hedgehog. Last month I proposed.

She said “yes”.

terese milan.jpg.png

53 thoughts on “How to Choose a Goddess (Using a Spreadsheet)

  1. Great story. Yay, everyone!

    Twenty years from now, you’ll look back on this post. I’m hoping you (and us onlookers) do so with fondness and a sense of winning.


  2. My new headcanon is that Paris really liked Wednesdays with Helen. Suddenly the decision matrix looks interesting, maybe even if you apply a ‘surviving to enjoy it’ multiplier.


  3. Nice post! How do you deal with diminishing returns (or non-linearities in general)? To be concrete, let’s say I only care that my car accelerates faster than any other car, but any more than that is unimportant.


    1. You can always force your measurement to be linear by converting it to a 1-5 or 1-10 desirability scale. So for example you’d give any car faster than a certain limit 5 points, and then assign 4-1 for various accelerations that are equidistant in terms of your enjoyment of them. The downside of this is having to convert scales, but the upside is linearity.

      In the end, you shouldn’t worry too much about this unless the “natural” scale is strongly non-linear and your options vary a lot on this scale and this attribute has a very high weight. Otherwise, it’s not going to impact your bottom line noticeably.


  4. I wonder if there’s a simpler way to assign weights to attributes (and the scores themselves) that still mitigates the effect of sub-/semiconsciously tweaking the numbers to get the result you expect


    1. You can visualize two options that are equal on all other attributes and fall in the general range of what you’re looking for, and differ only on the two features. Then you can mentally play with the “sliders” on those features to figure their relative weights. Keep playing with the features two at a time until you cover all of them, and remember that you only care about the ratios of the weights to each other.


    2. Rate attributes on a scale where 0 is neutral, positive is good, negative is bad. Then for each attribute score x, calculate f(x) = exp(x) – exp(-x). Sum the results.


  5. This is a wonderful piece that I shall strongly recommend to people who need to think more system2ishly about their relationships.

    Still grumpy about the hedgehog, though. Dragon should have won.


    1. Yes, Helen’s engagement and marriage were the event of the decade with all of Greece attending. He also probably noticed that she was married when he had to abduct her from Menelaus’ palace :)


      1. Another reason for the war is that Helen took her personal staff, or ladies in waiting, with her, and that was a real economic unit in terms of what they produced.


  6. Hi Jacob, your 5 categories are so well thought out! Would you mind sharing the 15 distinct characteristics you came up with, e.g. “Cracks me up + gets my jokes”? Thanks for writing this!


    1. I would mind. First, because that’s actually private and would cause adverse selection in the women looking to date me. Second, because it would only satisfy your curiosity and not help you build you own matrix. You actually have to sit down honestly with yourself and do the goal factoring thing.


  7. Hi Jacob :)

    I’m taking your advice to stop lurking. Thanks for the encouragement.

    You write: “Some attributes come with an obvious scale, like a price that is measured in dollars. It’s better to stick with those than to convert everything to a 1-5 scale”.

    I’m currently trying this method out for picking a place to live and there are attributes with obvious scales like distance or price. If you don’t convert this into a 1-5 scale doesn’t it look weird in the total score? For example 500$*5 comes out as 2500. Doesn’t this mean the score of this factor will dominate all the other factors that are on a 1-5 scale if they have similar weights?

    Hope you can help clear up my misunderstanding.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s