[Statistics] Election by Jury

My knowledge of the science (is it a science?) of statistics is very limited, so this is why I come to you…

A friend of mine recently talked at length about his idea of replacing our current democratic system with one where the vote is held by a jury. This jury is group (smaller, much smaller than the whole population of the country) of people that is deemed to be “statistically significant” enough to stand for the population as a whole. These people are sequestered, then are given the relevant information by representatives of the various parties, a debate is held, they ask their questions, etc… and then they vote. He says this way you remove the problems of media filtration, people’s ignorance of important issues, and low voter turn-out.

Now my question is this - is this possible? Is there an objective scientific way to select a group from the population that would be representative of the entire population of a country? Is this a sound understanding of the way statistics work?

I think there are some practical problems:

  • You need to select the sample by a random process, and deciding on a process that is random and not able to be subverted is quite hard.
  • You need to make sure your sample can’t be bribed by interested parties.
  • In close elections (such as the US Presidential elections in 2000 and 2004), a sample large enough to avoid sampling error would have to be very large.

But there is also the problem that the process would exclude most of the citizens from taking part. Citizen participation is an important part of democracy, including making the process legitimate. If you didn’t take part in choosing the government, why should you treat it as legitimate? (No taxation without representation).

Something like this could be done… the problem is that everyone except the winner of such an election would claim that the jury was not representative of the whole voting population. In reality the only way to prevent this is to have a jury consisting of the whole voting population, which is what we have at the moment.

Gee, this would be a great way to resolve the issues about choosing the President that we’ve had about the last two elections. We could select a jury of prominent citizens who would pledge to make the best choice, maybe committing to one of the candidates. Each state could choose as many jurors as they have members in Congress, so the smaller states don’t feel overwhelmed by the bigger ones. Sounds like a great idea!

Oh wait…

Sure. Why wouldn’t it be?

You’d have to begin with a list of every eligible person in the country, to select from. No such list exists today. You could create one from voter registration lists, but that would introduce an element of self-selection since you’d have to choose to register.

Then, you’d have to decide what you mean by “representative”. Are you content to draw names from the list at random, or will you have requirements for a distribution within a certain percent of the population with respect to age, race, gender, state of residence, occupation, education, etc. Of course, if you want to measure those attributes, you’ll have to collect that information, either in the course of compiling your master list (privacy intrustion!) or after you make your selection. Then, if your selection doesn’t meet your criteria, you’ll need to bump and replace people (controversy!)

No. Sampling aims to draw conclusions about the parent population via attributes of the sample. Your proposal aims to change the political process via a more informed electorate. Statistics won’t help you much there.

Hey hey HEY! This isn’t my proposal here!

Isn’t the more informed electorate just one part of the proposal? The sampling would aim to discover who the population as a whole would want as their leader, no?

Polycarp, maybe it’s because I’m Canadian, but I don’t get it. Is this an allusion to the Electoral College?

I’ve just thought of another way of phrasing the question I have in mind - is statistical science capable of coming up with an objective (incontrovertible, like, say, most math solutions) judgement of what portion of a population is significant (representative) of the population as a whole? It would seem on the surface that something that’s supposed to be rooted in equations and numbers should be capable of this. Why or why not?

This reminds me of a short story, by Isaac Asimov, I believe, in which statistical sampling is so precise that only one citizen elects the president.

Well, statistics can’t prove that a sample is exactly representative of the whole. That’s impossible. What it can do is tell you exactly how LIKELY it is that the sample is representative of the population as a whole. So if you have a population of 1000 jelly beans, and you know some are red and some are green, statistics equations will tell you how much confidence to have from a sample of 1, 2, 3, 4…n jelly beans. With a sample of 1000, you have 100% confidence that your sample is statistically significant.

But note that this requires the assumption of a random sample. The killer is getting a sample that is truly random. How would you do that? OK, we take a census of everyone in the US and select N samples out of that. But how do we know our census is complete? Homeless people, transients, immigrants, and people who are paranoid about the government are notoriously difficult to census. So we know from the outset that our census cannot be 100% accurate, there will be errors in the census. Since there are errors in the census we cannot have a 100% accurate sample even with perfect random sampling of the census entries.

And if you’re going to bother conducting a census of every American every 2 years, just to get an accurate database to draw your samples from, why not just open up sampling stations every 2 years and allow anyone who feels the need to come down and record their preferences. Then we could take a tally of this self-selected sample and the candidate who got the most “votes” would be the winner. It seems a bit simpler.

Major public opinion surveys (Gallup, et al) are pretty accurate from a statistics point of view. The margin of confidence and margin of error are such that if you repeat the survey N number of times, then 95% of the time, you’ll get the same results within about 3-4%. It’s impossible to have 100% accuracy 100% of the time, so this level is pretty good.

However, this also means that any vote that falls inside the range of 52%-48% is within the margin of error. This would include not just 2000 and 2004, but also 1976, 1968, 1960, 1948 (with four candidates), 1916. . .

Perhaps you’re thinking of “Franchise” by Isaac Asimov, in Isaac Asimov & Martin H. Greenberg, eds., Election Day 2084 (Buffalo, N.Y.: Prometheus Books, 1984).

I think that’s the one. It also appears in “Earth is room enough.”

And it wasn’t just the president. That one citizen (and the computer of course, who seems to be pulling most of the strings,) decide every contest that is up for election that year, even in areas far away from where the Elector is from. (shakes his head.) I wonder if they do referenda as well??


Yep. Sorry; I don’t do sarcastic allusion well at all, and should stop trying. :o My point, such as it was, was the idea that your scheme seems akin to the FF’s idea of getting a team of wise statesmen to select the President, which is not working out in the way the FF had hoped. (Arguments about the EC’s virtues or lack thereof are food for a different, GD thread.)

Define “representative”. When it comes to presidential polling, pollsters are concerned that their sample be representative of the larger population with respect to only one attribute–choice of a presidential candidate.

Assuming random selection of the poll sample from within the population, statistical tools can establish that the sample mimics the population with a given probability and a given margin of error. For example, the preferences expressed by the sample will be within 4% of the preferences of the population, 95% of the time.

The election proposal at issue, however, seeks to do more than just mimic the results of the current election process. It seeks to establish a new process wherein the sample will have access to greater information and make its choice in a different way. For that, you will probably want a sample that’s representative of the larger population in more ways: age, race, gender, income, geography, and so forth. The math of establishing a sample which is “represntative” across multiple metrics will quickly become incomprehensible.