?

Log in

No account? Create an account
Over in wellinghall's LJ... what an interesting question!* I'm not… - Sally's Journal
June 16th, 2008
09:57 am

[Link]

Previous Entry Share Next Entry
Over in wellinghall's LJ... what an interesting question!* I'm not sure I know the answer.

*Is it reasonable for women to be charged less than men for life insurance? Is it reasonable for women to be charged more than men for an annuity or pension? Yes, if it's backed up by statistics, or no, even if it's backed up by statistics...

(41 comments | Leave a comment)

Comments
 
[User Picture]
From:simont
Date:June 16th, 2008 09:27 am (UTC)
(Link)
This cuts right to one of my fundamental concerns with statistics that make me worry that it's all arbitrary guesswork rather than principled science. (The other is "how do Bayesians work out the correct prior?".)

Given a division of a population into subsections, you can do statistical analysis to determine whether some property of that population is correlated with the division. In non-Bayesian terms (I still haven't quite managed to throw off my A-level training), you do this by determining that there is a very small chance of the property being this unbalanced between divisions by pure chance, and that therefore you have to believe that either a one-in-(say)-100 statistical anomaly has taken place or that there really is a correlation.

This is fine if you only have one division and it's been picked for you by some external constraint: in that situation the logic makes sense. But if you have to start by thinking up your division in the first place, it's iffier. If you statistically analyse your population by race, sex, social class, inside leg measurement, star sign, first three bits of MD5 hash of their DNA profile and so on, then after you've tried a couple of hundred different divisions you're almost sure to have found one which yields what looks like a statistically significant correlation to your property. Does that mean it really is correlated, in the sense that there's any above-average chance of the correlation persisting in future additions to the same population? Not necessarily: it just proves that if you stare hard enough at a dataset you're sure to eventually find a vegetable that looks like Jesus. As it were.

Then the other question is that of justice: if you're advantaging people differently depending on the apparent correlation you've found, is that just? And this also comes back to the question of what your divisions ought to be. Should I be lumped in with the category "men" for my car insurance premium calculation, or should I be lumped in with the category "people who drive small pootly cars and don't go roaring down motorways at 100mph all the time"? My intuition says that the question of which is more just might be largely orthogonal to the question of which has a statistically stronger correlation to road safety.
[User Picture]
From:filecoreinuse
Date:June 16th, 2008 11:57 am (UTC)
(Link)
> The other is "how do Bayesians work out the correct prior?"

They don't. They choose a conjugate prior* and, in many case, guess the distribution parameters or fit them to a training set.

* i.e. a distribution where P(A|B)P(B) is integrable analytically.
[User Picture]
From:naath
Date:June 16th, 2008 07:01 pm (UTC)
(Link)
correct prior> they guess, silly (educated guess, but still). I think the point of argument is that using Bayesian methods you are forced to own up to the guessing part Very Clearly. Whereas with the methods commonly taught at A level you get to, er, gloss over rather the assumptions (which are built into the methods, and often not examined). Although when it comes to "do my life insurance tables" I get rather lost - and refer you to Dr MacKays excellent (and online) textbook (http://www.inference.phy.cam.ac.uk/mackay/itprnn/book.html); he does at least try to explain how it is done (although I admit to tripping over somewhere between the trivial questions about balls drawn from a jar and the more interesting ones about population distributions - since we were using the methods for looking at things like error correcting codes rather than actuarial work this wasn't *too* bad).

Most statistics is 'orrible guess work. Nasssssty stuff.
[User Picture]
From:robert_jones
Date:June 20th, 2008 01:38 am (UTC)
(Link)
I'm confused. I can see how that might be a problem in general, but I don't see how it can be a problem in this context. I don't think there's any doubt that there's a correlation between femaleness and increased life expectancy.
[User Picture]
From:lavendersparkle
Date:June 16th, 2008 09:35 am (UTC)
(Link)
I suppose the question boils down to whether we want these things to act as individual insurance or as redistribution between individuals? If we just want it to act as individual insurance then it's reasonable to charge a different price based upon anything which might be correlated with the risk being insured. If we want it to redistribute then it's reasonable to force insurers to charge the same fee to everyone to redistribute fro the low risk to the high risk.
[User Picture]
From:wellinghall
Date:June 16th, 2008 09:38 am (UTC)
(Link)
But that would lead to the same price for life insurance being charged to both a centenarian and a fit 20 year old. Which may be what you want; but you've got to be sure it's what you want.
From:hsenag
Date:June 16th, 2008 09:45 am (UTC)
(Link)
Except then most 20 year olds just wouldn't buy it. You can't do much of this kind of tinkering on prices of an optional product without causing self-selection and driving the prices up for the high-risk people anyway.
[User Picture]
From:atreic
Date:June 16th, 2008 09:46 am (UTC)
(Link)
So if that happened, everyone who was 100 would suddenly take out life insurance, and everyone who was 20 would think 'blow this for a game of soldiers' and the price would rocket, the 100 year-olds would be paying pretty much what they were paying in the first place, and the 20 year olds wouldn't have life insurance at all... yes, I'm not sure that's a good outcome either.

I guess there's something in the depths of all this about the distinction between what the state mandates and what people choose to buy privately, too.
[User Picture]
From:rmc28
Date:June 16th, 2008 01:25 pm (UTC)
(Link)
This is the model for what happens with health insurance, and is discussed at length in The Undercover Economist, who also discusses a model of compulsory health insurance used in Singapore that seems to avoid the horrors of the USA system while apparently being an improvement on the UK's NHS.

I know I keep going on about this book to you, but you keep bringing up things that remind me of it :)
[User Picture]
From:wellinghall
Date:June 16th, 2008 01:46 pm (UTC)
(Link)
I guess there's something in the depths of all this about the distinction between what the state mandates and what people choose to buy privately, too.

I will bring up the example of third party car insurance. The state says you must have it. But the price you pay depends on a range of factors, including some things you have no control over, like age (and, in some cases, sex - although sex isn't as universally used in car insurance as it is in life insurance).
[User Picture]
From:robert_jones
Date:June 20th, 2008 01:42 am (UTC)
(Link)
Well quite. There's no way to stop the purchasers of life assurance from using information about their likely life expectancy in deciding whether or not to buy the product. Therefore the insurers realistically have to take the same information into account. Otherwise they create an arbitrage opportunity. For example, terminally ill people would all rush out and buy life assurance. The difference between men and women is less marked than that, of course, but the effect would still be there.
[User Picture]
From:atreic
Date:June 16th, 2008 09:39 am (UTC)
(Link)
Yes, I think that's what I was trying to say on Wellinghall's post, but that seems clearer.
[User Picture]
From:neoanjou
Date:June 16th, 2008 10:09 am (UTC)
(Link)
Considering substituting man/woman with white/black people*, it suggests to me that its probably unfair. That is assuming we want to treat people of sexes as equally as we treat people of different races.

* American statistics suggest that black people live significantly less long than white people.
[User Picture]
From:theinquisitor
Date:June 16th, 2008 10:39 am (UTC)
(Link)
My thoughts on this one always lead me to coming down on the side of life insurance companies being able to make as accurate an assessment of someone's chances as possible. Forcing them to disregard information obviously works counter to this.

If you are less accurate than you could be, people come in two varieties: Those paying more than they ought, who may well realise this, and stop - and those paying less than they ought, who will eventually stop being subsidised, unless we rely on some sort of inertia.

Attempting to make life fair is an entirely reasonable goal, but forcing insurance companies to jump through hoops in order to use information they already have is not the way to go about it.
[User Picture]
From:lavendersparkle
Date:June 16th, 2008 11:53 am (UTC)
(Link)
But then I sometimes think it is reasonable to force insurance companies to not use some information if insurance sompanies using some information has very negative unintended consequences.

For example, for a while I think some life insurance companies would ask applicants whether they'd ever been tested for HIV and would charge them more/not insure them if they had. The reason was that statistically people who get tested for HIV tend to be people who are more likely to get infected with HIV at some point because most people would only get tested if they'd done something that involved a risk of infection* and are more likely to do it again. This approach would sort of be fair enough if it weren't for the fact that knowing that insurance companies took this into account made people less likely to get tested and this was a Bad Thing because people who know that they have HIV can start treatment earlier and are less likely to infect other people. So in that case it was a lot better all around to stop the insurace companies asking the question.


*Yes there are lots of 'good' 'responsible' people who go who regular tests despite practising safer sex, just in case, but they're outnumbered by the people who get tested because they engage in risky behaviour.
[User Picture]
From:aiwendel
Date:June 16th, 2008 11:08 am (UTC)
(Link)
it is reasonable for an insurance company etc to decides its own terms for what it thinks it can afford to offer. If they didn't want to insure women at all, that would be their perogative.
It's a free world, and I don't think we have a 'right' to either insurance or pensions.

So yes, it's reasonable to follow the stats to deduce what they can afford to offer to whom.

Having said that I don't think it would be unreasonable to spread the load across both.
[User Picture]
From:atreic
Date:June 16th, 2008 11:37 am (UTC)
(Link)
If they didn't want to insure women at all, that would be their perogative

That's the old "it's OK for a sausage seller to decide not to sell sausages to people in pink hats" argument. Which, when I agonised about it for ages, I decided that if one sausage seller has oddities then it's not worth getting the government involved, but if every sausage seller is refusing to sell sausages to people in pink hats and the people in pink hats need sausages, then Something Must Be Done TM. We might not have a 'right' to pensions, but if only men could get pensions that would be Evil Bad and Wrong, in my book.
[User Picture]
From:cartesiandaemon
Date:June 16th, 2008 01:04 pm (UTC)
(Link)
I have oft repeated this analogy. People often find it incongruous that only some sorts of discrimination are prevented by law, but after some thought, I decided there's actually good reason to do it that way.
[User Picture]
From:atreic
Date:June 16th, 2008 01:07 pm (UTC)
(Link)
I quickly forget why it shouldn't be illegal for sausage sellers not to sell sausages to men in pink hats - ie why the obscure individual discriminations shouldn't be cared about. Or is it just that they're bad, just not bad enough to worry about because they don't cause that much trouble?

I can definitely see a personal / business line in there - I think it's fundamentally impossible for the nation not to be sexist when choosing who to have sex with, or to only want to live with people who like board games etc.
[User Picture]
From:cartesiandaemon
Date:June 16th, 2008 01:14 pm (UTC)
(Link)
Or is it just that they're bad, just not bad enough to worry about because they don't cause that much trouble?

That was my conclusion. That if it's just one sausage seller, his inconvenience in having to serve people he doesn't like may outweigh the inconvenience of the pink-hat brigade of having to go to the second-nearest sausage shop.

And possibly even that there may be good reasons -- where is the line between "I don't sell to Mr Pinkhat" and "I don't sell to Doesn't Pay Their Bills Inc"? A shop is obviously set up to serve everyone but a window-cleaner may not be, do they always have to justify who they choose?

OTOH, the personal/business line is an interesting thought. Could you enforce no non-personal discrimination at all?
[User Picture]
From:atreic
Date:June 16th, 2008 01:19 pm (UTC)
(Link)
What would you count as personal though? One job that a lot of muslim indian women I met in Calcutta have started doing is going round to other muslim indian women's houses and doing hair / make-up / beauty stuff. For them, it's very important that they're working with women - I'm not sure they'd be culturally 'allowed' to go and do men's hair etc. So if a man complained they wouldn't cut his hair, that seems a harsh reason to shut them down for being -ismist.

[Actually, I think that argument is flawed, because the situation is only as it is because of a huge sexual inequality, and the answer is actually 'no, this is bad, but it has to stay bad until we've fixed 5 million other things first']

Another interesting personal / business edge case is that of the live-in lodger. In fact, just renting your house to anyone even if you're not living there, you're allowed a whole list of 'no students, no pets, no kids' etc.
[User Picture]
From:cartesiandaemon
Date:June 16th, 2008 11:19 pm (UTC)
(Link)
Quite. I think I meant something like "My natural assumption is that there are many cases where people have to pick and choose their clients, and so it's only possible for discrimination to apply to particular sorts of common discrimination, however, that would be perpetuating the status quo, and obviously it's hard to tell if I think it because it's sensible, or just because it's what I used to, so now you suggest hypothetically some other division between allowed and disallowed discrimination, I wonder if that might possibly be better, despite it seeming implausible to my first gut reaction."
[User Picture]
From:cartesiandaemon
Date:June 16th, 2008 11:24 pm (UTC)
(Link)
Whether there is actually any better compromise than the current, I've no idea. Ironically the example you suggest might actually be illegal because of existing sex-discrimination cases, although I hope there is an exception which fits.

AIUI current law typically says something like "no discrimination against X, Y or Z people, unless you actually have to be X, Y or Z to do the job." (And maybe some exceptions for size of company?) Which mostly makes sense.

Improvements would have to be expanding to V and W people (which does happen), or saying when discrimination is allowed and when it isn't -- it feels this would have to come down to saying "no discrimination unless you need a P person for a 'good' reason", but you'd constantly get lawsuits to decide if something was a 'good' reason or not, if the person providing the service reasonable wanted to pick and choose clients. So it sounds unworkable at first glance, but I don't know if it could be made to work. (Eg. if any country has more anti-discrimination legislation.)
[User Picture]
From:wellinghall
Date:June 16th, 2008 01:51 pm (UTC)
(Link)
A shop is obviously set up to serve everyone

A shop doesn't have to serve everyone, though. It can choose to bar certain customers (eg someone who caused a row last time), or even certain classes of customers (those wearing pink hats), as long as this isn't based on race or sex.

Some shops discriminate, quite legally, against whole classes of people - eg tobacconists, off-licences, firework shops and gunsmiths.
[User Picture]
From:cartesiandaemon
Date:June 16th, 2008 11:27 pm (UTC)
(Link)
Sorry, see my response to Sally: I was trying to say something like "it might just be possible to ban any discrimination by shops, but it seems untenable for many other businesses (I wonder if that's right)". So I didn't explore the case of shops as much as I might have.

I don't know whether or not it would be plausible to disallow shops from discriminating at all: it would certainly be difficult for all the reasons you name, and it would certainly be much more difficult for other sorts of companies.
[User Picture]
From:aiwendel
Date:June 16th, 2008 04:01 pm (UTC)
(Link)
I guess I'm coming at it from the companys perspective.
As a small business I can and must choose jobs which will earn me a living and not bankrupt my company. Someone's right to a nice garden etc doesn't come into it.
Maybe when they get big enough to be their own buffer they can choose to take on riskier people etc, but if you force a private company into doing unprofitable things from the start then you sink it and you have no pensions/insurance etc at all, which is worse than just some people having to pay more or less.

I don't think it's quite the sausage seller thing. As it's more 'people in pink hats never pay so I won't sell to them' or 'people in pink hats will eat all of my sausages at half price, preventing other people from eating sausages at all, so either I need to charge the innocent people more to subsidise the pink hatters, or I need to avoid pink hatters, or I need to charge them more...'

[User Picture]
From:cartesiandaemon
Date:June 16th, 2008 01:09 pm (UTC)
(Link)
As several people said, we seem to be approaching the divergence of two overlapping goals. For a while, insurance was a profitable business with only the most broad categorising of subjects, which was very convenient as the free market fills a need. But now we're discovering it doesn't quite work, and what's most profitable for an insurance company can be horribly unfair for people needing insurance. And we have to decide which goal we want.
From:hsenag
Date:June 16th, 2008 01:46 pm (UTC)
(Link)
In the extreme, "most profitable" = "viable" - the alternative to unfairness won't be reducing the profits of insurance companies, it'll be state provision.
[User Picture]
From:wellinghall
Date:June 16th, 2008 01:47 pm (UTC)
(Link)
Should, then, all life insurance and pension provision be via the state?
(Deleted comment)
[User Picture]
From:wellinghall
Date:June 16th, 2008 02:10 pm (UTC)
(Link)
Not at all - this is a false distinction. "Assurance" can only be used of life, but "insurance" can be used of both life and non-life.

If you really want to restrict life cover to being described as assurance, you shouldn't use it of term / temporary insurance - only of whole life or endowment assurances.
(Deleted comment)
[User Picture]
From:wellinghall
Date:June 16th, 2008 06:19 pm (UTC)
(Link)
Insurance can be applied to either; assurance only to the latter.

There is another point. Some life in/assurance does what you say - this covers both whole life assurance, which pays out when you die, whenever that is; and endowment assurance, which pays out if you die within a particular term, or at the end of that term if you are still alive then.

But some life insurance is term / temporary insurance - if you die within a particular term it pays out, but if you don't die in that period, you get nothing. A life as/insurance company will typically (indeed, almost always) write both sorts.
From:hsenag
Date:June 16th, 2008 02:45 pm (UTC)
(Link)
People would also have to be denied the right to choose how much provision to get, otherwise they could self-select there too.

Again, this is in the extreme. And no, I would not like for it to happen.
[User Picture]
From:randomchris
Date:June 16th, 2008 02:07 pm (UTC)
(Link)
Actually, it's often quite reasonable for women to be charged less for life insurance (same lump sum payout, but a statistical probability that more monthly payments will be made), but the same amount for a pension, because of how pensions work.

Quite frequently, a pension paid to a married couple continues to be paid until the death of both partners - so that would be the same whether the pension-payer was male or female.

Also, pensions are funded partly or wholly by the interest earned on the stock market from the amount paid into the pension. Some of them will return the leftover starting pot as a death benefit, decreased if necessary depending on the amount paid out - in this type of pension, the total amount paid out would be identical no matter how long the pensioner lived. In either case, the amount paid out by the company is similar to (starting amount of pension + interest earned on it), no matter how long they've been paying out on it for.

So quite often lifespan is not actually relevant to (the amount of pension paid out - the amount of pension paid in), therefore women should not be charged extra for pensions. Does that make sense?
[User Picture]
From:wellinghall
Date:June 16th, 2008 02:13 pm (UTC)
(Link)
Only if the same benefit is paid to the widow/er as was paid to the original life. Often, only a lower level of benefit (eg 50% or 2/3) is so paid.

And "returning the leftover pot as death benefit" is pretty rare, and in any event the age of death still affects the present value of moneys to be paid out.
[User Picture]
From:naath
Date:June 16th, 2008 07:18 pm (UTC)
(Link)
Hmmmm. I think it's not reasonable to charge women more NI for state pensions (I do however think raising the retirement age is fair) - this is because we don't choose to contribute to the scheme, and payout is not really dependent on amount paid in (only on number-of-years paid, which is not quite the same), and anyway it's clearly a Socialist Scheme to Redistribute Money (and should do it better dammit!). And besides women do lots of unpayed work that benefits the state - like having kids; maybe the primary carer of a minor child (the person getting the child benefit) ought to be deemed to have paid their NI if they aren't working and paying it? That would be nice. (also persons receiving the whatever-it-is allowance you get if you look after your elderly parents/siblings/spice/etc).

On the other hand I think it is fairly reasonable to charge people more for insurance if they are more likely to need it (and, um, less per month if more months will be paid before it is needed). I think that if we want Socialised Insurance (take money from people as have it and give it those as don't but who need it) we should damn well have Socialised Insurance and not pussy foot about telling private companies what to do. I don't think it's especially unfair that Rich People who put more money into their pension get more pension; I don't think it's especially unfair that a person who is very ill gets more pension per month for their pension-savings than a person who is likely to live a long time (for one thing it might help the ill person get nicer care whilst they are alive)... I just think it's Not Nice that this means some people end up with very little pension.

Plus I think life insurance is a complete waste of time. But then I don't have any dependents, so maybe I'm biased.
[User Picture]
From:wellinghall
Date:June 16th, 2008 08:19 pm (UTC)
(Link)
For people with no dependents, life insurance (in the traditional sense) is a waste of time. Once you get dependents, it becomes really, really important.
[User Picture]
From:naath
Date:June 16th, 2008 08:44 pm (UTC)
(Link)
In a state where being a dependent without the depended-upon person is enormously awful then, yes, life insurance clearly serves a purpose.

I would like to see the structures of society move in to *take care of* your dependents when you die - surely orphans should be fostered (and adopted) and elderly or sick people provided with care adequate to their needs?
[User Picture]
From:pm215
Date:June 16th, 2008 10:29 pm (UTC)
(Link)
I think even if the safety net provided by society is improved, life assurance is still worthwhile for anybody with dependents and a higher-than-average income. If you have a salary which permits a better-than-average lifestyle you're probably also happy to buy insurance so your children to continue to have a better-than-average lifestyle if you happen to die early.
[User Picture]
From:passage
Date:June 17th, 2008 07:01 am (UTC)

Reasonable requires consistent

(Link)
I'd like law to be consistent. If a company is allowed to charge me more for something, for being a man, I think it should be allowed to pay me more for something, for being a man (either assuming statistical evidence for both or for neither).

If a company isn't allowed to pay me more for something then I don't think it should be allowed to charge me more for something.

Incidentally who else suspects that if the statistics had said men were safer drivers then this pricing practice would have been outlawed decades ago?
Powered by LiveJournal.com