I was wondering how advertisers and political pollsters figure out what the nation likes and dislikes by just obtaining samples of various groups of individuals?
Ditto for the Nielsen Ratings, they survey certain amount of individuals from this particular state, income class, social class, religious class, etc to determine how much the country as a whole is watching TV.
It seems like some PHD holders have figured it out and advertisers and campaign managers at election time believe that the various samples collected are results to equate the entire nation's beliefs or habits
Polling (the political term) or consumer surveying (the business term) is a skill, like any other.
There are methods of demographic sampling that can get really good results with very small samples.
That said, nothing is perfect. And, with spam blockers and the death of the USPS, it is getting harder. Heck, to people under 40 your phone number today probably says where you lived when you were a teenager and your parents got you a phone, where as landlines were always tied to a very small area. Mistakes get made. Among those:
- Push polls. Fake polls where fake people ask you questions like "if you knew that candidate A was a child molester..." Declared unethical by the American Political Science Assoc.
- "Shy Tories". The proven fact that the more conservative candidate (Tory is the Conservative party in the UK and Canada) poll under what really happens. Some attribute this to people simply not wanting to say what they really think, others to the fear that union members or government employees are fearful to let it be know they go against their bosses.
- Public consumption polls. Polls that a commissioned by an interested party and which achieve the result they want, be that Candidate A being ahead or four out of five dentists recommend Trident.
But at the end of the day, in politics, vote how you want, even if everyone tells you that you are losing. In business, if you like Brand Z, buy it.
As to TV, I can poke holes in Nielsen's methods all day. But the simple fact is that Madison Avenue has agreed that these numbers are good enough and what everyone will use. If your show got cancelled, crying about Nielsen's methods is an exercise in tilting at windmills.
Quote from: SP Cook on April 22, 2025, 03:14:21 PMThat said, nothing is perfect.
Just watch the US presidential election coverage every four years...
Quote from: SP Cook on April 22, 2025, 03:14:21 PM- "Shy Tories". The proven fact that the more conservative candidate (Tory is the Conservative party in the UK and Canada) poll under what really happens. Some attribute this to people simply not wanting to say what they really think, others to the fear that union members or government employees are fearful to let it be know they go against their bosses.
Interesting. I always assumed it was specific to one particular candidate that has done better than polls predicted in every election they have run in.
2018 and 2022 polls were pretty spot on, and 2012 underestimated Obama. I think it's particular to one person.
Quote from: hotdogPi on April 22, 2025, 09:20:00 PM2018 and 2022 polls were pretty spot on, and 2012 underestimated Obama. I think it's particular to one person.
OTOH, there was a persistent tendency on state ballot measures on some hot-button issues (such as same-sex marriage, while that was still a live political issue) to underestimate votes in opposition.
Thought 2024 was pretty accurate.
I took a statistics class in college. Part of the class dealt with these polls and estimates, and sizes of the population and members in the sample. Interesting stuff.
Quote from: roadman65 on April 22, 2025, 02:29:41 PMI was wondering how advertisers and political pollsters figure out what the nation likes and dislikes by just obtaining samples of various groups of individuals?
Weighting.
Suppose you call a bunch of random phone numbers to ask them if they like spicy food. When you are on the phone you realize that, by pure luck, 70% of the people you called are female and 30% are male. That means that your poll is likely going to be skewed, since it's possible that women may like spicy food more or less than men do.
So you take the women's responses and multiply all of them by (50/70) and take the men's responses and multiply them by (50/30) to approximate a 50/50 male/female split. (In actuality most places don't have an exact 50/50 split, but this is just an example to explain how it works.)
You can do the same thing to correct for income brackets, political affiliation, race, education, whatever you feel like needs to be incorporated to make your results match the ideal model of what the population looks like. But what does the ideal model of the population look like? That is the question that makes some polls better than others, what makes some polls biased, etc. The poll that is the most accurate is the one who got their model of the population the most accurate to what the real world looks like.
For election results in particular, this is tricky because you're not trying to guess what the makeup of the
actual population is. (We know approximately how many, say, Native Hawaiian men there are in Las Vegas thanks to the Census.) You're trying to guess what the makeup of the
subset of people that will vote in the election will be, before it happens. (We need to know approximately how many Native Hawaiian men in Las Vegas
will actually cast a ballot as opposed to not participating in the election.) And that's the tricky part.
Quote from: Scott5114 on April 23, 2025, 04:47:11 AMQuote from: roadman65 on April 22, 2025, 02:29:41 PMI was wondering how advertisers and political pollsters figure out what the nation likes and dislikes by just obtaining samples of various groups of individuals?
Weighting.
Suppose you call a bunch of random phone numbers to ask them if they like spicy food. When you are on the phone you realize that, by pure luck, 70% of the people you called are female and 30% are male. That means that your poll is likely going to be skewed, since it's possible that women may like spicy food more or less than men do.
So you take the women's responses and multiply all of them by (50/70) and take the men's responses and multiply them by (50/30) to approximate a 50/50 male/female split. (In actuality most places don't have an exact 50/50 split, but this is just an example to explain how it works.)
You can do the same thing to correct for income brackets, political affiliation, race, education, whatever you feel like needs to be incorporated to make your results match the ideal model of what the population looks like. But what does the ideal model of the population look like? That is the question that makes some polls better than others, what makes some polls biased, etc. The poll that is the most accurate is the one who got their model of the population the most accurate to what the real world looks like.
For election results in particular, this is tricky because you're not trying to guess what the makeup of the actual population is. (We know approximately how many, say, Native Hawaiian men there are in Las Vegas thanks to the Census.) You're trying to guess what the makeup of the subset of people that will vote in the election will be, before it happens. (We need to know approximately how many Native Hawaiian men in Las Vegas will actually cast a ballot as opposed to not participating in the election.) And that's the tricky part.
Not sure why you would go 50/50 when women outnumber men in the general population.
Quote from: Rothman on April 23, 2025, 07:06:20 AMQuote from: Scott5114 on April 23, 2025, 04:47:11 AMQuote from: roadman65 on April 22, 2025, 02:29:41 PMI was wondering how advertisers and political pollsters figure out what the nation likes and dislikes by just obtaining samples of various groups of individuals?
Weighting.
Suppose you call a bunch of random phone numbers to ask them if they like spicy food. When you are on the phone you realize that, by pure luck, 70% of the people you called are female and 30% are male. That means that your poll is likely going to be skewed, since it's possible that women may like spicy food more or less than men do.
So you take the women's responses and multiply all of them by (50/70) and take the men's responses and multiply them by (50/30) to approximate a 50/50 male/female split. (In actuality most places don't have an exact 50/50 split, but this is just an example to explain how it works.)
You can do the same thing to correct for income brackets, political affiliation, race, education, whatever you feel like needs to be incorporated to make your results match the ideal model of what the population looks like. But what does the ideal model of the population look like? That is the question that makes some polls better than others, what makes some polls biased, etc. The poll that is the most accurate is the one who got their model of the population the most accurate to what the real world looks like.
For election results in particular, this is tricky because you're not trying to guess what the makeup of the actual population is. (We know approximately how many, say, Native Hawaiian men there are in Las Vegas thanks to the Census.) You're trying to guess what the makeup of the subset of people that will vote in the election will be, before it happens. (We need to know approximately how many Native Hawaiian men in Las Vegas will actually cast a ballot as opposed to not participating in the election.) And that's the tricky part.
Not sure why you would go 50/50 when women outnumber men in the general population.
🙄🙄🙄
It's 50.5% to 49.5%. (https://www.census.gov/quickfacts/fact/table/US/LFE046223)
Quote from: Rothman on April 23, 2025, 07:06:20 AMNot sure why you would go 50/50 when women outnumber men in the general population.
Quote from: Scott5114 on April 23, 2025, 04:47:11 AM(In actuality most places don't have an exact 50/50 split, but this is just an example to explain how it works.)
Actually reading the post can make you sure of many things. :D
I've often wondered how accurate phone polls can be these days when the respondents are essentially those people too stupid to screen their calls from unknown numbers.
Quote from: GaryV on April 23, 2025, 08:08:18 AMI've often wondered how accurate phone polls can be these days when the respondents are essentially those people too stupid to screen their calls from unknown numbers.
This is a major challenge to the polling industry. Another is, because of call-screening, they have to spend many more man-hours dialing numbers to get the number of responses needed for a representative sample, which in turn increases the cost of performing a poll.
Polling by Internet and text message are ways around this, but they have their own drawbacks (Internet polls in particular can be gamed since ballot-stuffing and spoofing your location are both fairly easy for anyone with technical knowledge to do).
Quote from: Scott5114 on April 23, 2025, 07:16:59 AMQuote from: Rothman on April 23, 2025, 07:06:20 AMNot sure why you would go 50/50 when women outnumber men in the general population.
Quote from: Scott5114 on April 23, 2025, 04:47:11 AM(In actuality most places don't have an exact 50/50 split, but this is just an example to explain how it works.)
Actually reading the post can make you sure of many things. :D
Pfft. You fear real math. :D
Quote from: Rothman on April 23, 2025, 08:50:33 AMQuote from: Scott5114 on April 23, 2025, 07:16:59 AMQuote from: Rothman on April 23, 2025, 07:06:20 AMNot sure why you would go 50/50 when women outnumber men in the general population.
Quote from: Scott5114 on April 23, 2025, 04:47:11 AM(In actuality most places don't have an exact 50/50 split, but this is just an example to explain how it works.)
Actually reading the post can make you sure of many things. :D
Pfft. You fear real math. :D
Real math makes describing a procedure more difficult because it introduces distracting magic numbers (https://en.wikipedia.org/wiki/Magic_number_(programming)#Unnamed_numerical_constants) which require unnecessary explanation and matter only to pedants. :D :D
Quote from: Scott5114 on April 23, 2025, 09:15:51 AM... and matter only to pedants.
Such as a nit picker of unprecedented pedantry, I suppose.