F4 Data input and opinion polls

Question 1

What does polling mode mean? Give an example

Answer 1

How a poll is collected: Live calling, internet poll etc.

Question 2

What is weighting and how can it be problematic?

Answer 2

Weighting is accounting for nonresponse by weighting underrepresented sociodemographic groups in the poll.

If the units in the sample are not representative of their demographic group, then weighting only make things worse.

Question 3

What is a likely voter model?

Answer 3

A model that tries to estimate who is going to vote. They predict turnout and vote choice.

Population: Citizens who can expected to vote

Question 4

What is probability based polling? What is the opposite?

Answer 4

Random sampling. The gold standard earlier: Random digit calling.

Now the opposite: Internet polling, online panel etc. (quality is low)

Question 5

What is a sample frame?

Answer 5

A list or representation of all the elements (individuals, items, or units) in the population from which a sample is drawn for a study or survey.

Question 6

What is the difference between a probabilistic and deterministic likely voter model? What text describes this.

Answer 6

Deterministic: A specific cutoff for inclusion (PG index).

Probabilistic: Inclusion based of likelihood of voting. Include both PG and demographics.

Rentsch et al. (2019)

Question 7

What is polling essentially?

Answer 7

A snapshot of current voter preferences at a given point intime. Nothing more.

Question 8

What is differential nonresponse

Answer 8

The likelihood of individuals responding to a survey differs systematically across subgroups of the population (e.g. GOP and dem).

This can create biases in survey results if certain groups are overrepresented or underrepresented due to their differing response rates.

Question 9

What can a bump in the polls be due to?

Answer 9

Motivation/excitement about a candidate e.g. after a convention.

Question 10

What do we suspect the next polling biases will be?

Answer 10

Hispanics
Non-college educated
White Trump supports
Young men

Question 11

What is house effects?

Answer 11

Some pollsters have a slight bias towards either GOP or dem (due to polling mode, adjustments, the secret sauce etc.).

Can be both intentional (Fox News) and non-intentional (Pew Research).

Question 12

Explain herding

Answer 12

Manipulation of data. Pollsters try to look alike or deliver a surprising result.

Everyone would rather be part of the skewed result than put themselves out there (just like our own models)W

Question 13

Why is turnout important to consider?

Answer 13

There is a big difference between voting intention and turnout in the US compared to DK. Barriers to voting.

Important if turnout is systematic and not random.

Question 14

What was turnout in 2024?

Answer 14

63.9 percent.

That is below the 66.6 percent voter turnout recorded in 2020, which was the highest voter turnout rate in a U.S. presidential election since 1900. Nonetheless, turnout in 2024 was still high by modern standards

Question 15

How does turnout differ across demographic groups?

Answer 15

Race matter (white turnout higher than Hispanic)
Age matters a lot (older turnout)
Education matters a lot (higher educated turnout)

Question 16

What is nonresponse bias and when is it problematic?

Answer 16

When people in a sample does not respond.

Ignorable (coverage): Lack of a specific demographic group. Can be accounted for through weighting.

Non-ignorable (missingness): Partisan response bias. Difficulties getting republicans to participate in polls.

Question 17

What was the polling error in 2016/2020 and why?

Answer 17

2016: 3,0+ dem (lack of lesser educated)

2020: 4,3+ dem (polling mode + more error in areas with more white and rural state)

Question 18

How can error be adjusted in the model?

Answer 18

Higher uncertainty in prior: Allow for more error - we don’t trust polls.

Low uncertainty in prior: We trust polls and do not expect a lot of error.

Question 19

How can you adjust for nonresponse bias?

Answer 19

Impossible to know how much nonresponse there is.

Question 20

How are weights decided?

Answer 20

Rentsch et al. (2019) point to how past elections are used to estimate weights.

Question 21

What happens if you use registered votes instead of likely voters?

Answer 21

You typically overestimate Democrats.

Question 22

Does likely voter models account for nonresponse bias?

Answer 22

No it’s still vulnerable to non-ignorable nonresponse

Question 23

What is the Perry-Gallup index?

Answer 23

7 questions that try to estimate intention to vote. Used in both probabilistic and deterministic models.

Questions: Interest in election, past voting behaviour, motivation, registration, intention etc.

Question 24

What demographic variables could be included in a probabilistic model?

Answer 24

Race, age, education, gender etc.

Question 25

What are shy Trump-voters?

Answer 25

Voters that due to social desirability bias say the voted dem in 2016 but actually voted Trump