The Club for Growth has issued a press release about a poll that they commissioned on the NY-23 special election:
The poll of 300 likely voters, conducted October 24-25, 2024, shows Conservative Doug Hoffman at 31.3%, Democrat Bill Owens at 27.0%, Republican Dede Scozzafava at 19.7%, and 22% undecided. The poll’s margin of error is +/- 5.66%
Ok, this is interesting as far as it goes, but it may not go as far as some boosters of Hoffman might like to think that it does (for example, click here and pick one).
First, the sample size is remarkably small. I am not a methodologist or a pollster, but the standard minimum number for basic public opinion polling is usually thought to be 400 (and even that has limited usefulness).
Second, with an MOE of 5.66%, it is a bit misleading to declare Hoffman in the lead.
Third, with 22% undecideds, it makes it rather difficult to interpret much of anything.
Fourth, the Club for Growth is actively supporting Hoffman, making them a less than reliable source for unbiased news on the election.
October 26th, 2024 at 9:13 pm
300 likely voters? That is simply bizarre. The resulting MoE only barely establishes Hoffman as being likely ahead of Scozzafava.
More importantly, was the party label asked in the question?
October 26th, 2024 at 10:08 pm
Bizarre, indeed. I don’t think I’ve ever seen a poll with an N that small before.
And I didn’t see the actual questions, but good point about party label.
October 27th, 2024 at 5:23 am
Actually you all need to take a statistics course, the sample size will depend upon the number of likely voters. Get educated then type.
October 27th, 2024 at 5:16 pm
[...] this one and the Club for Growth poll I noted yesterday shows a similar number of undecideds which means we really have no idea who is ahead at the [...]
October 27th, 2024 at 5:42 pm
Somehow I don’t think “Mark” knows who he’s talking to.
The sample size required for valid inferences to a population has little to do with the size of the population in question. There are exceptions, but that usually only applies when the sample size is a large chunk of the population – at least 5%. Unless you think there are only going to be 6000 voters in this election (around 1-2% turnout!), the sample size needs to be in the 500-1000 range to get much that’s useful.
October 27th, 2024 at 9:46 pm
[...] Steven Taylor added this about the Club for Growth poll: First, the sample size is remarkably small. I am not a methodologist or a pollster, but the standard minimum number for basic public opinion polling is usually thought to be 400 (and even that has limited usefulness). [...]
October 27th, 2024 at 9:51 pm
[...] Steven Taylor added this about the Club for Growth poll: First, the sample size is remarkably small. I am not a methodologist or a pollster, but the standard minimum number for basic public opinion polling is usually thought to be 400 (and even that has limited usefulness). [...]
October 28th, 2024 at 7:52 am
You can find how to determine your needed sample size in any statistic book. If you think it does not depend on the population number that you are trying to study, you are just wrong.
October 28th, 2024 at 9:57 am
I will just quote Chris: “Somehow I don’t think “Mark” knows who he’s talking to.”
October 28th, 2024 at 10:04 am
To put it in a less snarky way: yes, you can find this information in a stats text. The fact of the matter is, however, if you do that you will find that the text will tell you that a sample size of 300 for this kind of survey is too small. I expect, in fact, that Chris could quote you chapter and verse on that should he choose to do so.
October 29th, 2024 at 11:37 am
I’ve read at least 5 stat books and have never seen a section titled “special rules for polital polls”. The fact of the matter is a poll is a sample and the significance of a poll, from a statistical standpoint, is only dependent upon the sample size. A sample size of 300 will give a margin of error of 5.7%. As long as the poll is unbiased and you state the margin of error all inferences drawn are valid if they are tied to it.
Since the poll was done on “likely voters”, assuming they were defined as “likely” in a valid manner, the poll would have a greater practical significance, meaning more likely to predict the outcome as opposed to who is more popular, because the population of likley voters is less than the population of people registed to vote and much less than the total population of people.
The inference you can draw from this poll, assuming it was conducted in a statistcally valid manner, is that the race appears to be between Owen and Hoffman, with Scozzafava well behind. Can the inference be wrong – yes, there is a 5.7% chance. Is the poll valid? Only God knows.