Undercoverage: Used telephone books and vehicle registrations during the Great Depression.
Low response rate: Only 23% of 10 million people responded.
Large samples are not necessarily better than small samples.
Gallup, with only about 5,000 in-person interviews, got it right.
Quota samples have come under scrutiny.
Simple random sample (SRS) is now considered the gold standard.
Overview of Sampling Error
Sampling error (margin of error) is the difference between sample statistics and population statistics.
Sampling error is inevitable in every poll.
Introduction to Sampling Error
Even with efforts to avoid selection bias, sampling error still exists.
Sampling error: Difference between sample statistics and population statistics.
Inevitable, not a mistake.
Example 1: Presidential Approval Ratings
Polls ask about approval of presidents (e.g., Trump). Gallup reported a 45% approval rating in February 2025 with a margin of error of ± 4% at the 95% confidence level.
Sampling Error in Polls that Predict Election Outcomes
Pollsters study “likely voters” instead of the general population.
Identifying likely voters is challenging.
Professional pollsters disagree on methods.
So We Have Identified Likely Voters. Now What?
The population is smaller than the overall U.S. citizens (18+).
Pollsters must work harder to obtain a good sample.
Being willing to complete a poll doesn't guarantee they’ll vote.
Interpreting Margins of Error When Predicting Election Outcomes
Exercise caution when gauging multiple candidates’ support.
Multiple test statistics, each with its margin of error.
Separate, larger margin of error for the spread between candidates.
Example 2: The Texas Senate Race
August 2024 Texas Trends poll by UH Hobby School and Texas Southern Barbara Jordan – Mickey Leland Schools of Public Affairs.
1,365 likely Texas voters.
Cruz: 46.6%.
Allred: 44.5%.
Margin of error: ± 2.65%.
Based on this poll, it was hard to tell who would win.
In the end, Senator Cruz was reelected, gaining 53 percent of the vote to Allred’s 45 percent
Example 3: The National Popular Vote in the Presidential Election
Fall 2024 NPR/PBS News/Marist Poll for the popular vote in that year’s presidential election
1,294 registered voters.
Harris: 50%.
Trump: 48%.
Margin of error: ± 3.7%.
Based on this poll, it was hard to tell who would win.
In the end, Trump won 49.9 percent of the national popular vote; Harris won 48.4 percent
The Unique Challenge of Predicting American Presidential Elections
Besides identifying likely voters, there are other challenges.
Historically, state polls have been difficult and expensive.
We Need State Level Polls
Due to the Electoral College, state-level polls of likely voters are needed.
National polls don't accurately represent state-level opinions.
State polls are historically difficult and expensive to conduct.
The Relationship between Sample Sizes and Sampling Error
As sample size increases sampling error decreases.
The relationship between sample sizes and margin of error is shown on the graph.
The Trouble with State Level Polls
State-level polls require large sample sizes for a 3% margin of error.
State polls often lack resources despite their importance.
2016 state polls were of low quality.
Interpret election polls cautiously.
Example 4: The Battlegrounds
Seven battleground states with a combined 93 electoral votes.
Polling in each state was within the margin of error.
Neither candidate had a clear advantage in polls.
In the end, President Trump won all seven states.
Concerns About Question Wording
Objective of a poll is to gauge public opinion; push polls aim to change it under the guise of a poll.
What is a Push Poll?
A push poll attempts to change public opinion while pretending to be a legitimate poll.
Message Testing v. Push Polls
Message testing is used by candidates to research campaign strategies.
Push polls manipulate public opinion rather than gauge it.
Example: 2000 South Carolina Republican Primary.
What Can Be Done About Push Polls?
Unfortunately, despite push polls being considered unethical, the answer is…not a whole lot.
The American Association of Public Opinion Researchers (AAPOR) prohibits members from conducting push polls.
Issues with Question Wording
Changes in wording and answer choices can drastically affect findings.
Examples: "Welfare" vs. "aid to the poor"; death penalty and sentencing options.
Implications for campaigns: wording can influence responses.
Pre-Survey Information
General Social Survey question example: Asking about spending on various problems and whether it’s too much, too little, or about right.
General Social Survey, 2006
"Welfare" vs. "Assistance to the Poor."
Different responses based on wording.
Capital Punishment (The Death Penalty)
Historically most Americans have favored the death penalty as a punishment for convicted killers
Support/opposition is not as easy to gauge as you might think
Consider carefully the responses Gallup received to two different questions regarding the death penalty
Support for Capital Punishment - Version 1
Are you in favor of the death penalty for a person convicted of murder?
% in favor has varied from 90% to under 60% from 1936 to 2019.
% opposed has varied from 10% to around 40% from 1936 to 2019.
Support for Capital Punishment - Version 2
If you could choose between the following two approaches, which do you think is the better penalty for murder - [ROTATED: the death penalty (or) life imprisonment, with absolutely no possibility of parole]?
% death penalty has varied from 56% to around under 50% from 1986 to 2019.
% life imprisonment has varied from 34% to over 50% from 1986 to 2019.
Reflections on This Example
How does question wording influence responses?
How might pollsters use this finding for good or ill?
Tips for Writing Good Questions
Balanced arguments: "Do you favor or oppose the death penalty?" is better than "Do you favor the death penalty?".
Avoid double-barreled questions (two questions in one).
Example: "If we can now trust the Russians, do you think we should focus our attention on saving endangered species?"
Try to Avoid Negatives!
Avoid negatives due to confusion.
Example: Instead of "Do you think McDonald’s should not have been found not liable…" use "Do you think McDonald’s should have been found liable…or not?"
The Importance of Questions Order
Question order can prime people to answer later questions in a certain way.
The Classic Question Order Effect
Two versions of questions about Communist reporters: changing the order changes the responses.
Effects of Public Opinion on Public Policy
To what extent are politicians beholden to public opinion?
To what extent should politicians be beholden to public opinion?
Assessing the Impact of the Public on Policy Outcomes
Political scientists struggle with this question.
Challenges:
Congress's true intentions (reelection vs. genuine desire).
Inconsistencies in public opinion.
The public is operationally liberal yet symbolically conservative.
Studies of Public Mood
Evidence of responsiveness to public opinion.
Policy changes receive strong public support.
Researchers can predict policy changes based on public opinion.
Policy outputs move in the same direction as public mood (liberal or conservative).
Public Policy Mood
(Graph from 1952-2020 showing % Liberal Responses).
Source: James A. Stimson ( http://stimson.web.unc.edu/data/ )
Kidnapping and Murder of Polly Klaas and Three Strikes Laws
1993: Polly Klaas kidnapped and murdered.
1994: California passed a three strikes law.
Third strike leads to a mandatory sentence of 25 years to life.
Weakening Three Strikes in California
Proposition 36 (2012) thrown out with 69 percent of the vote:
Criminal conviction has to be violent to constitute Strike 3.
If sentenced under three strikes, you could appeal your sentence.
Those whose strikes include murder, rape, and/or child molestation may not do so
How Should Politicians Respond to Public Opinion?
Disagreement among scholars on the extent of correspondence between public opinion and policy.
Consider two approaches and apply them to crime and punishment.
Delegate Model
Representative should vote as their constituents would like.
Public opinion matters greatly.
Trustee Model
Public opinion is less likely to affect a representative’s votes.
Edmund Burke: Representatives should use their own judgment, not just follow public opinion.