Can a dating site help you make better decisions?
The site OKCupid has become known for its innovative approach to dating.
To match daters to each other, OKCupid asks members to answer questions, then indicate how they’d like their ideal match to answer the same question.
But they take it one step further.
Daters are then asked how important the correct answer is to them. Daters rate the answer using the following scale:
0  Irrelevant 
1  A Little Important 
10  Somewhat Important 
50  Very Important 
250  Mandatory 
Notice the differences between each level aren’t equal. That’s because the weights assigned to each level indicate relative worth.
Having matching answers to a question marked as Very Important is equivalent to having 5 matching answers on a Somewhat Important question. Likewise, it takes 25 Somewhat Important matches to equal a single Mandatory match.
Unbalanced scales like these, where the interval between levels is not equal, can be used to improve decision techniques like the weighted procon list.
Consider a typical scale of 15:
1  Irrelevant 
2  A Little Important 
3  Somewhat Important 
4  Very Important 
5  Mandatory 
Using this scale, a factor rated Very Important is equal to only two factors rated A Little Important.
But you rarely want two pros rated A Little Important canceling out a con rated Very Important. You want 10 or 20 A Little Important pros before you’ll consider ignoring a Very Important con.
Unbalanced scales allow us to better map the value differences between different levels of importance when weighing factors.
But don’t just rely on OKCupid’s scale. Feel free to make your own based on your own differences in value. Maybe 10 Somewhat Important factors should be equal to 1 Very Important factor. In which case make your Somewhat Important level equal to 10 and your Very Important level equal to 100.
Adjust the labels and number of levels to your needs. For instance, you might use:
1  Trivial 
5  Minor 
50  Major 
250  Critical 
5,000  Showstopper 
Bottom line: when you need to weight factors that you later average or sum together, consider using an unbalanced scale to account for differing levels of importance rather than a balanced scale with equal intervals.
What alternative scales have you used besides the typical linear 15 or 110 scales?
Credits: The photo used in this article was taken by Courtney Carmody.

Hi Trevor, thanks for the post. This sort of decision is proving important to me at the moment. I’m working through some similar scoring using Chris Guillebeau’s $100 Start Up methodology. He suggests ranking possible businesses using the 1 – 5 scale, but this doesn’t take into account your typical scale point above.
How about using one of two methods.
The first one is simply doubling up for each increment. Sticking with your 5 levels;
Level 1 = 1
Level 2 = 2
Level 3 = 4
Level 4 = 8
Level 5 = 16
If you use this, then two level 1’s = one level 2, two level 2’s = one level 3 & it would take four level 1’s to equal one level 3 etc.
If you want to get a little more scientific, use a log scale
Level 1 = 1
Level 2 = 10
Level 3 = 100
Level 4 = 1,000
Level 5 = 10,000
Andy,
The scale can vary depending on your decision and what you’re aiming to accomplish.
Log scales are good when you’re trying to figure out the order of magnitude. Assuming nearly unlimited upside, figuring out how good a business idea is would fall into this category. You generally don’t have the information to figure out if an idea is a $1 million idea or a $2 million idea, but you can figure out it’s not a $10 million or $100 million idea.
Read Derek Sivers’s post Ideas are just a multiplier of execution for a good example of using a logarithmic scale to estimate the value of ideas.
Doubling each increment can work; though it’s better for programmers who are used to thinking in multiples of 2. And it assumes that your next level up is only twice as valuable as your previous level, which might not be a large enough margin of error if your estimation is wrong (one of the benefits of a logarithmic scale is that it allows for larger margin of errors).
Though to rank a business, you may want to rank it across multiple dimensions: market size, access to distribution channel, competition, etc. With a linear scale you can just add all these together and get a score. With a nonlinear scale, you need to multiply the factors together and take the nth root, where n is the number of factors.
For instance, if you have a business scored as 10,000 for market size, 10,000 for customer value, but 1 for team member experience, your total rating would be the cube root of 10,000 x 10,000 x 1, or 464. Whereas a business scored at 1,000 for all three dimensions would have a cube root of 1,000, significantly better.
OKCupid talks about this in their article. Though they translate the ratings to percentages and treat them like probabilities.
Which brings up another idea: think in probabilities. Treat your top level as the highest confidence level you have that you’ll hit some goal. So say Level 5 means you have a 90% confidence this is an idea that can be generating $10 million a year in revenue in 5 years. What is your Level 4 confidence?
So your scale may be:
Level 1 = 1%
Level 2 = 5%
Level 3 = 20%
Level 4 = 50%
Level 5 = 90%
Then just drop the percentage and use these as your weights.
Be careful with probabilities though. It’s easy to deceive ourselves. I recommend thinking in terms of odds, then translating. 1% sounds small, but it’s a 1 in 100 chance, which is pretty high if you’re talking about something like hitting $10 million in 5 years.
Depending upon how ambitious your goal is, your highest level may only have a 20% or 50% probability of success.
Hi Trevor,
Thanks for including the other articles for reference.
I like your idea of probabilities & the idea of thinking of them as odds. As you correctly suggest given any uncertainty starting a new business, if someone starting up was offered 20% odds of getting $10 million revenue in 5 years, they’d surely take them…
Hello, Trevor!
As a big fan of the OKCupid algorithm, I really liked learning more about the weighting in this article. I’ve long wondered how the distribution shook itself out.
However, as the tightlipped grammarian that I am, I lost confidence in the source when I read this: “Notice the amount of points assigned to each level aren’t equally spaced. ” An oversight, I am sure.
To review, points can be counted and so occur in numbers of, while pointage cannot and would occur in amounts of.
Thanks. I feel better, and can now return to my review of the OKCupid Press.
Lana,
You are absolutely correct about the grammar.
Though while fixing it, I realized there was another issue with that sentence. The number of points assigned to each level aren’t equally spaced, it’s the differences between the levels that aren’t equally spaced.
I’ve fixed the text so it reads correctly. Thanks for noticing.