Chapeau for the enormous level of autism on display here.
To add my own autistic point, these frequently used (1/N)^n calculations are extremely stupid and wrong as the traits in question are clearly not independent.
I don't think you should treat them as independent. I think I'm quite clear here that you need to make your estimates, assuming that someone meets all the previously listed requirements
I'm saying you should choose your percentages based on the idea that they already meet the criteria above on the list. If you just choose the percentages based on the general population, I agree. But I'm specifically advising against that.
This the relevant thing I've written
Then move onto the next trait. Write that down and what proportion of people have that second trait given that they also have the other traits you’ve already written down. If you’re looking for someone who is an NBA player and also over 6ft you don’t need to multiply the proportion of NBA players by the number of men over 6ft. You can safely say that 99% of people in the NBA are tall enough.
But also this is the presumption for when I say you can change the percentage of men who are interested in things and stuff if you are already filtering for having a good profile
I think you a wrong that Aria is unaware of statistical dependence (she clearly states that she is estimating conditional probabilities). I agree that she is probabably underestimating the number of men fitting that criteria. If you just go to a Oxbridge graduation ceremony, there’s probably dozens of bright-eyed potential husbandos from each class every year.
Edit: moreover, the naive Bayes model is not so stupid, if you can’t estimate the conditional probabilities. You would need to point out why you think it is higher than 0.1, or otherwise you may as well argue that the true number is 0.001.
It says a lot about the dating market that there's a 0.014% chance you'd want to date any particular appropriately-aged man in London and a 50% chance that that man you pick would want to date you.
I agree this is a more rigorous and correct model to dating than the secretary problem. Thanks for writing it up! However, I think one of its major downsides is the risk of filtering out the super interesting candidates in favor of mediocre ones (i.e., taking the "settling" part too literally). This is because (1) many criteria you have are a sliding scale instead of a binary choice, so someone could really excel on one criteria and lacking on several others but still be a better overall candidate than another person who is okay on all fronts, and (2) while many criteria correlate, they don't perfectly correlate and the exceptions to the rule could end up being a candidate with a lot of "X factor."
Two examples I've experienced in real life to illustrate my points: (1) I normally wouldn't date someone who frequently consumes alcohol and recreational drugs because this behavior strongly correlates with many other undesirable traits; however, my second longest relationship is actually with such a personal who's extremely agentic, intelligent, and creative. They were also able to keep substance use separate from our time together. (2) Someone who is ambitious and agentic usually have high earning potential, both of which would be in my normal set of criteria, but in the EA circle, many of these people end up founding non-profits which decreases their earning potential. I personally think it'd be worth giving someone who starts the next big idea in the animal welfare/AI safety space (my priority cause areas) a shot over generic tech/PE/VC/quant bros.
That said, both of my objections can be easily incorporated into your existing model if (1) the user is scoring the candidates for each criterion instead of using the criteria as a simple checklist (probability assigned to each criterion becomes a probability distribution) and (2) designate a subset of the total criteria as true "dealbreakers" (i.e., will not continue dating if score < Y) but allow the rest to trade off against each other.
Chapeau for the enormous level of autism on display here.
To add my own autistic point, these frequently used (1/N)^n calculations are extremely stupid and wrong as the traits in question are clearly not independent.
I don't think you should treat them as independent. I think I'm quite clear here that you need to make your estimates, assuming that someone meets all the previously listed requirements
Hate to be autistic again but the calculations you showed evidently assume independence.
There are probably 100s of men (not 0.1 man) that meet the criteria your husband meets BECAUSE the criteria are not independent.
Worth a read https://towardsdatascience.com/statistical-independence-for-beginners-69534ce04641/
I'm saying you should choose your percentages based on the idea that they already meet the criteria above on the list. If you just choose the percentages based on the general population, I agree. But I'm specifically advising against that.
This the relevant thing I've written
Then move onto the next trait. Write that down and what proportion of people have that second trait given that they also have the other traits you’ve already written down. If you’re looking for someone who is an NBA player and also over 6ft you don’t need to multiply the proportion of NBA players by the number of men over 6ft. You can safely say that 99% of people in the NBA are tall enough.
But also this is the presumption for when I say you can change the percentage of men who are interested in things and stuff if you are already filtering for having a good profile
I think you a wrong that Aria is unaware of statistical dependence (she clearly states that she is estimating conditional probabilities). I agree that she is probabably underestimating the number of men fitting that criteria. If you just go to a Oxbridge graduation ceremony, there’s probably dozens of bright-eyed potential husbandos from each class every year.
Edit: moreover, the naive Bayes model is not so stupid, if you can’t estimate the conditional probabilities. You would need to point out why you think it is higher than 0.1, or otherwise you may as well argue that the true number is 0.001.
It says a lot about the dating market that there's a 0.014% chance you'd want to date any particular appropriately-aged man in London and a 50% chance that that man you pick would want to date you.
Women must be the discerning gender
Is it that men aren’t able to be selective?
I agree this is a more rigorous and correct model to dating than the secretary problem. Thanks for writing it up! However, I think one of its major downsides is the risk of filtering out the super interesting candidates in favor of mediocre ones (i.e., taking the "settling" part too literally). This is because (1) many criteria you have are a sliding scale instead of a binary choice, so someone could really excel on one criteria and lacking on several others but still be a better overall candidate than another person who is okay on all fronts, and (2) while many criteria correlate, they don't perfectly correlate and the exceptions to the rule could end up being a candidate with a lot of "X factor."
Two examples I've experienced in real life to illustrate my points: (1) I normally wouldn't date someone who frequently consumes alcohol and recreational drugs because this behavior strongly correlates with many other undesirable traits; however, my second longest relationship is actually with such a personal who's extremely agentic, intelligent, and creative. They were also able to keep substance use separate from our time together. (2) Someone who is ambitious and agentic usually have high earning potential, both of which would be in my normal set of criteria, but in the EA circle, many of these people end up founding non-profits which decreases their earning potential. I personally think it'd be worth giving someone who starts the next big idea in the animal welfare/AI safety space (my priority cause areas) a shot over generic tech/PE/VC/quant bros.
That said, both of my objections can be easily incorporated into your existing model if (1) the user is scoring the candidates for each criterion instead of using the criteria as a simple checklist (probability assigned to each criterion becomes a probability distribution) and (2) designate a subset of the total criteria as true "dealbreakers" (i.e., will not continue dating if score < Y) but allow the rest to trade off against each other.