4 ms·
People will likely freak out about this but I don't see anything wrong with it. Making more accurate predictions is a good thing. Why does it matter what the da
by Houshalter 13y ago
People will likely freak out about this but I don't see anything wrong with it. Making more accurate predictions is a good thing. Why does it matter what the data used was?
Unfortunately I expect this to become another "illegal datapoint" that is illegal to take into account. People hate the idea of being judged for seemingly arbitrary things. Which makes sense. But if it actually does correlate with the truth, if it actually makes predictions more accurate on average, why shouldn't it be used?
- DougWebb 13y agoBut if it actually does correlate with the truth, if it actually makes predictions more accurate on average, why shouldn't it be used? Because statistics. Correlation applies to populations of data points, not individual data points. Even if 99% of the data points have a certain feature (an extremely strong correlation), the 1% of outliers should not be penalized for being similar to the rest of the group in other ways.
- Houshalter 13y agoBut this is an argument against making any kind of predictions at all, as it could negatively affect some individuals. I don't see why that should be so, especially if it positively benefits everyone else a lot more and is a net benefit. For every person that loses one dollar, 500 people gain 50 cents. To look at it another way, by not taking that data into account, you are making those 500 worse off. I don't see how that is preferable. If we had a time machine to go into the future and see what loans would be paid back and which ones wouldn't, that would be ideal. But we can only guess. If our guesses aren't 100% accurate, are we saying we shouldn't try at all?
- fnayr 13y agoTaking a millionaire's wealth and distributing it to the starving would have a positive net benefit as well.
- Houshalter 13y agoYes, it definitely would. What is your point?
- sliverstorm 13y agothis is an argument against making any kind of predictions at all, as it could negatively affect some individuals. The question we ask is, what is this prediction based on? Some things are sacred; we will not allow predictions to be made based on race or gender, but we will allow predictions to be made based on billing history. The latter is much more within an individual's control than the former!
- Houshalter 13y agoIt's not always under an individuals control. Things happen outside their control so they can't pay back their loan. In insurance, getting into an accident isn't always the persons fault and we still use it to determine their rates. Why should some predictors should be considered "sacred"? The point is to make the best predictions possible. It doesn't really matter what information is used to determine that.
- sliverstorm 13y agoWe aren't talking about vehicle insurance, which does allow the use of gender, age, and happenstance.
- Houshalter 13y agoIt's basically the same thing though.
- sliverstorm 13y agoIf we assume they are the same, then one is wrong. How do we know which one is wrong?
- Houshalter 13y agoI meant that risk prediction is insurance is very similar to risk prediction in loans. I'm not sure what you mean. What one is wrong?
- kintamanimatt 13y agoHere's why: "Hi, I can't be friends with you on Facebook/Twitter/Google+ because although you've been a really great friend to me, you lost your job last year and it's had a bit of a dent on your credit history." -- or -- "Hey, I don't want to add you to Facebook because I don't know how well you pay your bills." (This isn't a real issue for me because, like the nouveau cool, I deleted my Facebook.) Yes, you could argue the person doing the digital ostracizing is a craptastic piece of shit, but the criteria for friendship shouldn't be creditworthiness. If I had to eliminate otherwise good friends (or funny acquaintances) because I know they've had problems paying their bills, I'd not be a happy bunny. It's trite to highlight this, but credit scores impact all kinds of things, from interest rates, availability of financial products, ease of getting a merchant account, and apparently, how easy it is to get certain jobs. Where does it end? Phone records? Emails? Chat logs? Who you Tweet with most? CCTV hooked up to facial recognition software to figure out who you're going to lunch with? There's nothing good about this and I hope their business fails.
- Houshalter 13y agoYou are right, it could create bad incentives. If that is the case then I will agree with you, but many of the people are arguing that it is inherently unjust to use certain things to make predictions, which makes no sense at all. I don't think in practice many people will change their friends over this, and this is only one piece of information being used, so it probably wouldn't have a huge effect if they did so.
- kintamanimatt 13y ago> many of the people are arguing that it is inherently unjust to use certain things to make predictions, which makes no sense at all Would it be wrong to judge my creditworthiness by the fact I'm white and gay if you could find some correlation between those traits and bill paying habits? What if white gays were typically irresponsible with money† (because omg shoes) and as a consequence my credit score is forever (or at least initially) tarnished because I'm a white homo, irrespective of my bill-paying ability. I can see many ethical issues with this. > I don't think in practice many people will change their friends over this, and this is only one piece of information being used, so it probably wouldn't have a huge effect if they did so. LOL. People will do a lot in the name of money. If Ms. Sally wants a mortgage, like really, really, really bad, Ms. Sally is gonna drop her deadbeat Facebook friends and game the fuck out the system. If she's successful, not only is she more likely to be approved for that mortgage, but she'll get a better rate. I think it's naive to believe that any other behavior is being encouraged by this kind of predictive credit scoring. When people figure out how to game a system, they tend to do so. Such signals would quickly be rendered useless, unless you could effectively figure out who is gaming the system, and how hard. †I'm not advocating this idea.