Whats the likelihood of a Palin "nu-cu-lar" Presidency?

October 4, 2008 12:56 PM

"Resolve an argument" filter: What to use to calculate the life expectancy of a 72yr old male? Yes, we are talking about John McCain.

So my brother (currently an associate, but almost a "fellowshipped" actuary by exams) and I (last took calc as a college frosh, and stats as a jr) have been having an argument regarding calculating life expectancies.

When I asked him how I could calculate the life expectancy, or the likelihood of death in 1 year, 2 years, 3 years, 4 years...from today, he came back with a life table from the SSA site (duh), He told me that there's about a 14.4% chance of death within 4 years.

Yeah, I can do the same thing. I also know that life tables are awesome when looking at the aggregate, but horrible when looking at individuals. If I wanted to know how many 24 year olds out of 10,000 would die this year, that life table would be spot on. It won't really tell how likely one 24 year old is likely to die accurately.

He told me that the SSA table is the tool actuaries use. Ummm no. I don't buy it. Sure it may be an initial reference point, but there is no way that one studies math for decades only to rely on a chart to make all the decisions. To be fair, he doesn't do THIS kind of actuarial work.

I KNOW that I won't understand the *why* of how the math was arrived at, but how is life expectancy arrived at? What is the likelihood of death for John McCain while in office for the first term.

Although life insurance applications may seem HUGE, they don't ask that many questions (to get the variables that need to be plugged into a formula). They take a simple medical exam (I just went through one: questions, height/weight, blood pressure, and some blood).

Without having ALL the specifics of McCain's medical records, but being able to google a good portion of it, can someone calculate the likelihood of a Palin presidency contingent upon the Republicans winning the 2008 Presidential election?

I've seen some articles all over the place using basic math to arrive at a number (some as high as 40%[1 out of 2.5], and some as low as 0.3%[1 out of 333]), and I don't trust the math behind it. Sure this question has some political implications, but it is a STRICT MATHEMATICAL and SCIENTIFIC question. I don't want Obama supporters telling me that their pappy died at 71, so there's a 95% chance that he will die within the first year, and I don't want McCain supporters telling me that "71 is the new 30". I just want a straight answer.

posted by hal_c_on to science & nature (34 comments total)

Addendum: If you need ANY additional information to do the calculation, just ask it here. I'm sure I, or some black-belted google-fu solitary monks will be able to find it. I betcha they will!
posted by hal_c_on at 12:58 PM on October 4, 2008

It will be very difficult to factor in the effect of top-notch medical help always at hand.
posted by Idcoytco at 1:02 PM on October 4, 2008

It will be very difficult to factor in the effect of top-notch medical help always at hand.

Actually not at all. Thats why they ask for income/savings on insurance applications. The wealthier one is, the better access to health care one has. If McCain does become president, he will have better doctors than say...the governor of a state with less than a million people. But I'd put money on the idea that he wouldn't have better health care than a health-conscious billionaire, or even a hundred-millionaire.
posted by hal_c_on at 1:09 PM on October 4, 2008

Here's one article about this, that tries to take everything into account.
posted by aubilenon at 1:10 PM on October 4, 2008

You've got a continuum here. On one end is the actuarial tables which are true for large populations, on the other end you have a super sci-fi machine that looks at every atom in his body and figure out exactly what might eventually go wrong. We don't have that machine. Anything other than that machine would need to take into account the effects of the stress of having the most important job in the world while simultaneously having the top health care resources in the country dedicated on the job. No president in this century has died of medical reasons while in office, so we don't have any data on this. The best you can do here is just fudge the numbers (+: John McCain is more active than most people his age; -: John McCain has has cancer)
posted by 0xFCAF at 1:11 PM on October 4, 2008

0xFCAF,

Your answer would be totally correct if life insurance companies would say "nope, we can't calculate premiums for individuals". Thanks for the input, though.
posted by hal_c_on at 1:15 PM on October 4, 2008

Insurance companies don't calculate premiums for individuals. They calculate them for classes of individuals: Obese smokers age 65-68, thin non-smokers age 70-72, etc. John McCain is in one of those classes, but you're asking us to take in to account much more information than those classes bother to make distinction for.
posted by 0xFCAF at 1:19 PM on October 4, 2008

aubilenon,

I thought the link was totally awesome until I actually read it. A firm in atlanta is claiming that John Mccain has 8 years of life left (80 years), but Barrack Obama has only 21 years left (68 years)? It seems odd, but could totally be right.

It just seems odd to me, that the SSA tables make it seem as if McCain would have a 30%+ chance of dying before the end of his second term, but this atlanta firm puts it at a quarter.

Cool, though. Kinda surprising, but it seems like the most accurate answer so far...but I'm sure mefi has some actuaries who have nothing to do on a gloomy saturday night...

Best answer....so far.
posted by hal_c_on at 1:22 PM on October 4, 2008

0xFCAF,

You're right about the classes. But what other classes must be taken into account to do this calculation?
posted by hal_c_on at 1:24 PM on October 4, 2008

It won't really tell how likely one 24 year old is likely to die accurately.

False. That's precisely what that table tells you.

It sounds to me like what you REALLY want to know is whether or not McCain will die in office. Unfortunately that's not an answerable question.

It also sounds like you just don't have a very firm grasp on the nature of statistics. Predicting the future for specific individuals is not really the goal, nor is it feasible. The goal is to predict in the aggregate... which is why percentages are used.

At the end of his term, McCain will not be 14.3% dead. He will either be 0% dead or 100% dead. Now, is there a table for presidential life expectancy? No, there's not, but the people who make the tables don't do it by sitting around a crystal ball and guessing... they do it by looking at all of the data they can find... aggregate data. If there's no data, there's no table.
posted by toomuchpete at 1:49 PM on October 4, 2008 [2 favorites]

I was just about to post what 0xFCAF said.

hal_c_on, you seem to be making an assumption that the insurance company needs to arrive at an accurate figure for the individual. But they don't. They insure sufficiently many people that their errors even out.

The answer to this:

"Without having ALL the specifics of McCain's medical records, but being able to google a good portion of it, can someone calculate the likelihood of a Palin presidency contingent upon the Republicans winning the 2008 Presidential election?"

is no. What they can calculate is the risk for an arbitrary person in a pool of people who share those characteristics. When people talk about the risk of McCain dying during his term, they're glossing over those distinctions.

"there is no way that one studies math for decades only to rely on a chart to make all the decisions. "

Au contraire, it could well be that having studied the problem for years, I realise that there is only a handful of factors required for a model with useful predictive power, and that adding more factors makes predictions worse, or at least no better. And you know what? Here's a handy chart summarising my work.
posted by i_am_joe's_spleen at 1:52 PM on October 4, 2008 [2 favorites]

toomuchpete is correct. You sound like you want the results of an assessment of McCain: a doctor looks at his health and habits and determines exactly how likely he is to die. There's no one who goes around and does that for each individual person... that's why we go by statistics and rely on likelihoods. They're not always completely accurate because that would be like accurately predicting the future.
posted by Solon and Thanks at 2:00 PM on October 4, 2008

toomuchpete, and i_am_joe's spleet,

thanks. that makes it a BIT more clear to me.INS companies can't predict for an individual, but for an aggregate. Got it.

I am not asking for CERTAINTY, but a percentage (what is the chance, in percentage, that mccain will be 100% dead before he finishes his term?).

Let me rephrase my question:
The SSA table uses 3 factors (american, age, sex) to figure out the table. What would the table look like if other stats were known (google-able health records, personal history, work history).

I'm not looking for a 100% percentage saying "john mccain will die on march 3. 2011", what I am looking for is a more accurate percentage because more variables are known than the 3 represented on the SSA chart. The more data there is, the more validity there is.

Sure there is no life table for "presidents" (that i know of), but there were also no life tables for "webmasters" in 1990. That doesn't mean that one can't make an even MORE accurate assessment based on MORE information.

I appreciate your comments, that definitely let me clarify the question.
posted by hal_c_on at 2:12 PM on October 4, 2008

You could probably argue that anyone who becomes President takes on a (significantly?) higher risk of death by assassination. That should probably be factored in, even though it's not an age-related issue.

Another (incredibly vague and hard to pin down) factor is the fact that his mother is alive and seemingly in good health - I would guess that her advanced years make him more likely to reach the same age.

I have no idea how you would actually use these observations though. This post is essentially useless.
posted by Kirn at 2:24 PM on October 4, 2008

For your average 72 year old, there is a negligible risk of death due to assassination.
However, for a 72 year old who happens to be president of a country, this is not the case -- in fact, I think that assassination or injuries from failed assassination might be the leading cause of death of sitting US presidents (but I don´t have time to look that up at the moment). This needs to be taken into account in any statistical analysis.
posted by yohko at 2:32 PM on October 4, 2008

The way you would do this works like this:

(1) Get individual level data for a whole bunch of individuals. Tens of thousands at least. For as many potentially causal or even correlative variables as you can get. This will be hard for you.

(2) Put together a duration model; I can't think offhand which of several varieties would be most appropriate. It'll be a big complicated model with lots of interactive effects and so on. This is hard, but you could do this if you really wanted to.

(3) Get the same variables for McCain. This is impossible. But if you could, then you could input the relevant data, find his hazard rate, and get a percentage chance of death with a 95% prediction interval around it. Note that prediction intervals (for a single observation) are wider than confidence intervals.
posted by ROU_Xenophobe at 2:37 PM on October 4, 2008

Sure there is no life table for "presidents" (that i know of), but there were also no life tables for "webmasters" in 1990. That doesn't mean that one can't make an even MORE accurate assessment based on MORE information.

You´ve got kind of a small sample size for ¨presidents¨, so it´s going to be pretty hard to come up with a table.

If you go to see a doctor and ask them if you will live for another four years, they won´t be able to tell you. This uncertainty is part of the human condition.
posted by yohko at 2:37 PM on October 4, 2008

According to my actuary father:
Strictly using the general population table, a 72 year old male has a 86% probability of living 4 years and a 69% probability of living 8 years.
I realize that you're looking to include more variables than what's in the table, but those are some concrete numbers for you and others curious in the thread.
And of course, what is "living"? Someone could be living but still be in no shape to be president.
posted by operating thetan at 2:40 PM on October 4, 2008

It may be a bigger issue (but not as newsworthy and shocking on face value), is that if he has medical problems that impair him she will have to take higher percentage of responsibility, so eventually it might get to the point where she has to make nearly all decisions, but at the same time nobody would be able to say, oh my, Palin is now president, that's crazy.
posted by rainy at 3:02 PM on October 4, 2008

"The more data there is, the more validity there is."

Nope, not necessarily. Only if you have a model that can use that data.

Studies that look at risk factors try to look at just one at a time.

Suppose that a couple of drinks a day lowers your risk of heart attack, raises your risk of liver disease. Which has more predictive power, modelling heart attach risk and liver disease separately, or just in aggregate looking at death rates for drinkers?

One of my hobby interests is statistical modelling for share valuation and price. Models with fewer variables tend to outperform models that try to account for lots of variables.
posted by i_am_joe's_spleen at 3:06 PM on October 4, 2008

No president in this century has died of medical reasons while in office, so we don't have any data on this.

Well, THIS century, no, since we've only had Clinton & Bush. But FDR died of medical reasons (cerebral hemorrhage) while in office in 1945, and he was only 63.
posted by lisa g at 3:15 PM on October 4, 2008

And, of course, William Henry Harrison died one month into office of pneumonia.
posted by adamrice at 3:37 PM on October 4, 2008

And factor this in: McCain smoked two packs of cigarettes a day for 25 years.
posted by Jaltcoh at 4:14 PM on October 4, 2008

Other than FDR, who had significant prior health problems, the life expectancy of U.S. Presidents, current and former, is pretty good. Barring assassination, which affects every President, the access to health care of your average rich, white (up to now) guy in the Oval Office has rendered some pretty impressive results in the last several decades.

Hoover died at age 90.
FDR died at 63.
Truman died at 88.
Eisenhower died at 79.
JFK was shot at the age of 46.
LBJ died at 65, after three heart attacks.
Nixon lasted until 81.
Ford was 93.
Carter is still with us, at age 84.
Reagan died at age 93.
George H.W. Bush, still alive, is 84.
Clinton is 62.
Our current President is also 62.

So, but for FDR, JFK, and LBJ, no President or former President has died before the age of 79 in almost 80 years. A small sample, true, but it's useful to keep in mind.
posted by mikewas at 4:38 PM on October 4, 2008

operating thetan,

That is EXACTLY what my brother says also (off by a few tenths of a percentage point). Besides the post of aubilenon. You have the most relevant posting here. Thanks for consulting a real source!

So here are the posts I see on here:

1. "There's no way to tell. EVER. We need more presidential data."

2. Well there's always the risk of "X". So you gotta factor that.

ummm...thanks, I guess to everyone else. I just figured mefi would have someone who could have actually done the math.

I'll pick the best answer 24 hours after this was originally posted. So far, thetan, and aubilenon are winning.
posted by hal_c_on at 5:44 PM on October 4, 2008

But we do have a life table for the presidents. There have been 42 presidents (if we count Grover Cleveland's two term as one pres). Four (Lincoln, Garfield, McKinley, Kennedy) were assassinated and four (W Harrison, Taylor, Harding, FDR) died of an illness. A bit of basic math that even I can do therefore gives a 19% chance of a sitting president dying in office. It could be argued that only FDR died of an age-related disease (and even that can be argued). It could also be argued that, for obvious reasons, Obama is more likely to be assassinated than McCain. Insurance companies, as others have pointed out, are looking for much larger statistical cohorts (e.g. all smokers or all males between 65 and 75) so all the presidents is probably a statistically unacceptable cohort but it's all we've got. Now if we add in presidents/prime ministers of other countries... So there is quite a good chance of President Palin being even scarier than the idea of President Cheney. But, of course, we have no life table for female presidents.
posted by TheRaven at 6:14 PM on October 4, 2008

Article on the subject here. They estimate a 1 in 3 chance of dying in office in an eight year term.

Voters should also consider that dying isnt the only reason he would have to step down. He may suffer a stroke or take an extended leave for chemotherapy. The chances of a Palin presidency are much higher than the mortality statistics suggest. These people suggest 40%. I would think these numbers are as good as any.
posted by damn dirty ape at 6:43 PM on October 4, 2008

OK, I can pretty much guarantee I have the best answer to this question.

I have a masters in statistics and PhD in demography, both from UC Berkeley. I did my PhD dissertation on this topic. Using the longitudinal EPESE survey questions, I developed a very high-powered predictive model to gauge the probability an elderly person would die within three years. It was titled "A Method For Predicting Death And Illness In An Elderly Person".

Without going into too much detail: The EPESE survey, a gold-standard epidemiological survey of the elderly, asked hundreds of health-related questions, and tracked respondents over time. I used sophisticated computer methods (a genetic algorithm search for classification models designed to optimize the bias/variance tradeoff) to select the best questions out of these hundreds.

Furthermore -- and this is important -- I actually validated my model by testing it on a sample of respondents I had not seen until AFTER I constructed the model. The predictions were spot on.

I'll have to scan in the test and post it online for you, but for now I can do the calculations, without making any too-strong assumptions about how McCain would answer the questions.

Give me a few minutes.
posted by mikeand1 at 9:19 AM on October 5, 2008 [1 favorite]

OK, based on some of the answers given here, and making a few mild assumptions about how McCain would answer a few of the other questions, my model predicts that he has about a 15% chance of dying within the next four years.

That's very close to what the SSA life tables tell you, but that's probably because his health is pretty average for a 72-yr-old male.

I'll try to post the exact form of the model tomorrow morning, when I get a chance to scan it in. But basically, it has 15 questions, many of which are obvious (age, sex, weight, smoker, cancer, heart disease) and some of which are less obvious.

The model was developed to optimize classification error: That is, trying to predict whether someone would be alive or dead in three years. Error was based on the proportion of false positives and false negatives. The hard parts are (1) trying to figure out which of the hundreds of possible questions -- and countless combinations of them -- have the most predictive power; and (2) trying to figure out how many questions are optimal (too many, and you build noise into the model; too few, and you lose predictive power; this is known as the bias/variance tradeoff problem).

A hard copy of my dissertation is at the UC Berkeley Department of Demography. One of these days I'm going to put the whole thing online.
posted by mikeand1 at 9:39 AM on October 5, 2008 [2 favorites]

BTW - I should qualify my answer by stating the obvious: It does not take into account the chance he could be assassinated as the result of being President!
posted by mikeand1 at 9:47 AM on October 5, 2008

"But we do have a life table for the presidents. There have been 42 presidents..."

I think there's way too many deficiencies to this approach. First, 42 is not a large-enough number. Second, you can't compare McCain with someone who lived two centuries ago, when life expectancies were much, much lower.
posted by mikeand1 at 9:53 AM on October 5, 2008

And factor this in: McCain smoked two packs of cigarettes a day for 25 years.

Interesting link- some might say TMI but I found it fascinating.

I hope somebody can answer this: McCain smoked two packs a day for 25 years and quit in 1980. But he spent 5.5 years as a POW enduring torture from the VC... from 1967 to 1972- and they gave him two packs of smokes a day? Either he "quit" for 5 out of those 25 years or his captors were pretty humane (at least as far as feeding his tobacco addiction- and seriously, 2 packs a days is hardcore- goes).
posted by ethnomethodologist at 11:21 AM on October 5, 2008

Sounds somewhat similar to what Mikeand1 is suggesting, but you could use something like this that calculates a person's life expectancy using some general questions about their health and family history. Background on that calculator here.

Just based on my layman's observations, the quality (and actual result) you get out of these type of calculations seems to vary quite a bit, but still it is generally what you are asking for--a more personalized version of actuarial tables that takes into account the actual health & other factors that might affect a particular person.
posted by flug at 3:51 PM on October 5, 2008

"Just based on my layman's observations, the quality (and actual result) you get out of these type of calculations seems to vary quite a bit..."

That's why it is important to know that I actually validated my model by applying it to an independent sample of individuals after the model was built.

As far as I have seen, I'm the only one who has done that.
posted by mikeand1 at 6:32 PM on October 5, 2008