How to extract the chance of winning from questions in the poll?

Question

Polling is very new to me and I have a naive question:

Looking at HuffPost Pollster, I am trying to understand how numbers like "Clinton: 49%, Trump: 37%" are obtained.

First of all, in every poll a number of questions (typically 7 to 20) are asked. There are two cases:

1. direct question: one of the questions is simply "If the general election were held today, and the candidates were [...], for whom would you vote?" Like Bloomberg, NBC, etc, then the numbers are just the answer of this particular question and have nothing to do with remaining ones.

2. indirect question: For example in this poll by Morning Consult, there are 11 questions and none of them directly asks whom to vote for. Of course question like "Do you approve or disapprove of the job Barack Obama" is strongly correlated, but I have no idea how numbers like "Hillary Clinton (D) 42% Donald Trump (R) 37%" are obtained from these 11 questions.

So my question is simply the following:

a) In the first case, why beat around the bush and ask ten other questions? Isn't it a waste of time/money?

b) In the second case, do they build some magic statistical model to predict the chance of winning based on the answers to various questions and never tell you how to do it?

Answer 1

In the first case, why beat around the bush and ask ten other questions? Isn't it a waste of time/money?

More questions add trivial cost. Most of the cost is in placing the call. Some of the other questions help them calibrate the poll. Others may be interesting to them.

What I mean by calibrate, is that asking other questions can help tell them what kind of people they are polling. For example, a common question is party registration. That way if it turns out that they called all Democrats and no Republicans, they know to adjust their polling. They also may ask favorability questions to see how these voters compare to other voters. Again, if the numbers are out of line, they can poll more people.

indirect question: For example in this poll by Morning Consult, there are 11 questions and none of them directly asks whom to vote for.

When I read that poll, I see the following section:

2016 General Election: Trump vs. Clinton
Asked of 1,362 registered voters
Hillary Clinton (D) 42%
Donald Trump (R) 37%
Undecided 21%

This suggests to me that they asked a question like "For whom do plan to vote in November? Donald Trump? Hillary Clinton? Or are you undecided?" They don't actually give the actual question though, just the results.

Note that the next few sections suggest that they also asked those people about their party registration. So they use the results of those two questions to generate what are called cross-tabs. That allows them to say things about how independents, Republicans, and Democrats voted.

Winning

Polls rarely estimate chances of winning. That's usually left to other organizations. For example, 538.com builds statistical models from polling results and other information. What you should expect to see from polls are numbers like 42% of people polled said they plan to vote for Hillary Clinton.