Are Pot-Tolerant Californians More Candid When They Talk to Robots?

Six surveys have asked California voters about Proposition 19 since the marijuana legalization initiative officially qualified for the ballot in March. Three of them show the measure losing narrowly, while the other three (including the most recent one) show it winning by at least 10 points. Poll maven Nate Silver notes that the former three surveys were conducted by humans, while the latter three were conducted by "robopoll" machines. Two recent national surveys show a similar pattern. Silver speculates that reluctance to endorse a controversial position in a conversation with a live person, especially when that position may be seen as reflecting one's own stigmatized habits, may explain the discrepancy between the two types of polls. If so, that's good news for the proposition's supporters, since voters will be dealing with machines at the polls this fall. 

I discussed Prop. 19 in my column this week.

[via the Drug War Chronicle]

Comments

[?]

I'm way more freaky in bed with robot chicks. Especially the dual-exhaust models.

[Episiarch]

The path to robot hell is paved with human flesh.

[AA]

+1

[jasno]

Or these polls are complete bullshit. I remember the first one that came out showing less than 50% support - I think they polled 1,000 or so people. California has 22 million eligible and 16 million registered voters. Please explain to me how 1,000 samples, or 1/16,000 of the voting population, can give you a meaningful idea of what people will vote for.

Besides, who the hell takes polls anyway? People with nothing better to do. So the elderly and unemployed are going to be unfairly represented.

[toxic]

Statistical analysis, Jasno. It's the art of deriving valid generalizations about a larger without having to check on every single individual person/thing in the group by randomly selecting a smaller sample group that reflects the traits of the population as a whole.

Or something like that.

[John Thacker]

Please explain to me how 1,000 samples, or 1/16,000 of the voting population, can give you a meaningful idea of what people will vote for.

Without defending the rest of it, once the voting population is big relative to the size of the sample, it doesn't really matter exactly how big it is.

It's roughly just about as unlikely to get a strange result from polling 1,000 people out of 1 million as out of 16 million.

[Jason]

You only need 384 people out of California 2009 population of (total) 36,961,664 for a confidence level of 95% and a confidence interval of 5%.

1000 people gives you a confidence level of 99% and an interval of 4.08%.

Play with this sample size calculator for a bit and see what size sample you need for various confidence levels and intervals.

[jasno]

You only need 384 people out of California 2009 population of (total) 36,961,664 for a confidence level of 95% and a confidence interval of 5%.

You must be a pollster 🙂

Do you seriously believe that or is that what someone's simplified mathematical model told you? I guess it's time for me to crack open some statistics books, because that sounds like complete and utter bullshit. Maybe it would be true if you took a truely random sample from a population following a perfectly normal distribution.

[EscapedWestOfTheBigMuddy]

There are some assumptions about correctly representative sampling that go into those figures---and they are surprisingly tough to fulfill when you're talking about people and not, say, coherent excitation of the delta resonance---but yes it is the result of a theory, and yes, you can believe it (within the limits of sampling technique).

The matter of sampling technique being more important than sheer numbers played a large role in the Dewey/Truman election campaign. That story is worth reading.

[John Thacker]

Yes, I highly agree on all points.

The mathematics behind it in an idealized situation is fairly simply and easy to comprehend, and airtight. Complaints about sampling technique are of real concern and a much better place to direct questions than concerns about the number of people polled.

If your sampling technique is good and your sample representative, 1000 is certainly enough. If your technique is bad and your sample skewed, 100,000 would not be enough, nor would any amount short of a significant fraction of the total.

[jasno]

I'll contend that it is impossible - not nearly impossible or highly unlikely - but straight out fucking impossible - to come up with an unbiased sampling technique.

How would you even do it? Would you make sure your sample had a racial and economic breakdown that matched the overall population? Ok, that's a start, but I would argue that there is an inherent bias towards assuming people from the same racial and economic background agree on a particular issue.

Going back to the 384 out of 36 million - does anyone seriously think that you could call 384 people and get a confident projection?

It also ignores the fact that many people just don't like to take polls. It is incorrectly assumed that those people's beliefs can be divided up in the same ratio as those who do take surveys. That is simply absurd.

[EscapedWestOfTheBigMuddy]

Remember that you don't have to come up with an unbiased group by setting up a representative sample (so many black people, so many whites... so many baptists, so many catholics...): you have to come up with a random technique (well, strongly pseudo-random) with uniform coverage.

That is still hard, but doesn't require you to have an near-infinite amount of data and data processing capacity. The trouble is in the uniform coverage, BTW: if you conduct a telephone poll and the non-phone owning population is (1) significant in size and (2) has different views from the phone owning population you've just set yourself up for trouble. That's why the headline read "Dewey Defeats Truman".

They are ways to mitigate the risks, but they take resources (i.e. money).

[skr]

I occasionally take polls, but only if they get me after a big bongload.

[JoeF]

It's probably a different sample of people who are willing to talk to robots. If you're unemployed, bored, or just generally have a lot of time on your hands, you might not hang up immediately upon recognizing a robot call.

As much as I hate to say it, the demographic that is employed, gainfully busy, (i.e. not employed by the government) and generally doesn't want to put up with that crap is also probably less likely to favor legalizing pot. They don't have time to smoke up, either.

[EscapedWestOfTheBigMuddy]

This is the way I was thinking about it too, though I was wondering if "conservative" type were less willing to be rude to a real live human than to a robot.

[JoeF]

You pretty much have to put quotes around that word these days, don't you?

I don't consider myself "conservative" except in social manners, but I am employed, and fwiw, I'm willing to give human telemarketers the cold click if I feel like I have something better to do.

Survey takers get more of a polite brush-off, and robots/recordings are cut off as soon as they are recognized, so I generally don't know which category they fall into.

Anyone else care to share their policy?

[Darkeyes]

What, no alt-text? You fucking racists...

[AA]

If R2D2 is pro-Prop 19, shouldn't you be?

[silent v]

Either way, the issue is close enough that it is going to come down to voter turnout. Unfortunately, the stoners' likely gubernatorial candidate, Jerry Brown, isn't likely to get anyone fired (get it) up. Fortunately, the law and order types probably aren't any more enthusiastic for Meg Whitman.

[skr]

i know that people give stoners a bad rap, but if anything can get them out it's this. I figure that a lot of them actually registered for the Obama vote, and those actually growing weed for which this prop could be a problem will never register. So low voter turnout could be a good thing percentagewise.

[joshua corning]

R2D2 with a background of pot leaves.

So did you search for this from a stock photo archive or did you put it togethr yourself?

[Jason]

You missed the joint sticking out of R2-D2's "mouth".

[Publis Cato]

I thought it was a blunt

[fishfry]

It all comes down to turnout. Since the top three California races all come down to none of the above, I expect low turnout and a defeat for prop 19. The races I refer to are Brown/Whitman for gov, Fiorina/Boxer for Senator, and Newsom/Maldonado for Lt. Gov. A more useless collection of candidates would be hard to find. I don't see much enthusiasm for this election, hence the low turnout, hence the defeat for prop 19. The youth vote will stay home.

[Jason]

Wouldn't Prop 19 make the youth vote turn out?

[jasno]

I disagree. In fact, I think the only thing that will get Jerry Brown elected is the turnout for prop 19. This will also be a good year for fringe candidates(I guess that would include Brown).

[Azathoth]

Hint: it's not a 'fringe' candidate if it's already been governor

[wayne]

I am a Californian. The legalize MJ proposition is the single thing that will get me to "turn out".

[Suki]

Much more open with aliens than with any robot.

[JOhnny MAckson]

LOL. I would've conducted this poll, but I got high.

Jess
http://www.anonymous-surfing.es.tc

[AA]

Thats what I'm talking about bot!

[BeesInTheBran]

Hmmmm, pot heads receiving calls from blocked numbers asking their opinion on drugs. Kinda wondering if the numbers aren't skewed even further than mentioned above.

[juanita]

The addicts and apologists for Big Pot shouldn't think they can cowardly hide in the anonymity of the secret ballot.

[Publis Cato]

I'd like some of what you're smoking...

[JOhnny MAckson]

You must be an apologist for Taco Grande. Never apologize for having a huge, meat-filled, sour cream covered taco! LOL . . . juanita

Jess
http://www.anonymous-surfing.es.tc

[juanita]

Paranoia and the"munchies" are just the tip of the iceberg of marijuana's deleterious side-effects.

[James Henry]

Anybody is going to be more open with a bot. If it does become legal this year the Landscaping of Cali with pot plants will be something to see.

[Geotpf]

It's California. Trust the Field Poll above all others. They are scary-accurate (best pollster overall according to Nate, for instance). I believe they use human dialers, but I'm not sure.