This is an example of response bias (specifically leading question bias or wording bias).

What response bias means: Response bias occurs when question wording, question order, or surveyor behavior influences responses, causing answers to differ from true opinions.

Why this question is biased: The phrasing 'Don't you agree...' is leading — it suggests the desired answer is 'yes.' Most respondents will comply rather than disagree with a leading question, inflating support for school funding.

How to fix it:

Reword as neutral and balanced alternatives:

1. 'How should school funding change?' (open-ended)
2. 'Do you support increasing, maintaining, or decreasing school funding?' (balanced options)
3. 'School funding should be increased' vs. 'School funding should be maintained' (state both sides)

Key principle: Survey questions must be neutral and not suggest a particular answer. Both sides of an issue should be presented equally.

Other response bias types: Acquiescence bias (always agreeing), social desirability bias (answering what sounds good, not true), and question order effects.

Biases introduced:

1. Nonresponse bias:

People who don't respond differ from respondents. Who doesn't respond?

• Busy people (less time to fill out surveys)
• Less interested in voting/politics
• Elderly or tech-unfamiliar people (online survey may exclude them)
• People with contrary opinions (may ignore it)

Result: Sample overrepresents voters and politically engaged people; underrepresents non-voters.

2. Voluntary response bias (self-selection):

Online polls are voluntary. Who volunteers?

• People with strong opinions (both for and against)
• More motivated individuals
• Those with internet access

Result: Response represents passionate people, not average population.

3. Undercoverage (selection bias):

Only people with internet access can take online survey. Missing:

• Elderly, rural poor (limited internet)
• Low-income households
• Tech-averse individuals

These groups have different voting patterns than internet users.

4. If the question assumes voting:

'Have you voted...?' assumes respondents are voters. People who think they shouldn't answer (non-voters, non-citizens) might skip the survey, creating further nonresponse bias.

Consequences:

Estimated voter turnout would be inflated — the poll would overestimate how many people actually voted.

How to reduce bias:

1. Use random sampling (not voluntary signup)
2. Use multiple data-collection methods (phone, mail, in-person) to reach diverse groups
3. Offer incentives for response
4. Use careful follow-up of non-respondents
5. For online surveys, weight results to match known population characteristics

Sampling Error:

What it is: Natural variability in sample statistics due to random sampling. Even with a perfectly unbiased sample, different samples give different results.

Example: Two SRS of 500 voters each might give 52% and 48% support for a candidate just by chance.

Is it a bias?: NO — it's unavoidable randomness, not systematic error. Reduced by larger sample sizes.

Undercoverage:

What it is: Some population groups are systematically excluded or harder to reach.

Example: Phone survey doesn't reach people without phones (young, mobile-only) or language minorities.

Result: Sample doesn't reflect population structure; biased estimates.

Is it a bias?: YES — systematic, not random.

Non-response:

What it is: People who don't respond differ systematically from those who do.

Example: Survey of job satisfaction has 30% response rate. Non-responders might be very satisfied (don't bother responding) or very dissatisfied (don't want to engage).

Result: Biased estimate depending on who non-responders are.

Is it a bias?: YES — systematic, not random.

Example: All three in one survey

A company conducts phone survey about workplace satisfaction:

• Sampling error: They randomly select 300 employees, but by chance, younger workers (more critical) are overrepresented → slight downward bias in satisfaction estimate
• Undercoverage: Survey doesn't reach part-time remote workers or recently hired employees without company phones → missing diverse perspectives
• Non-response: 40% of called employees don't answer or decline; non-responders are either very satisfied (busy) or very dissatisfied (disengaged) → middle ground overrepresented

Result: Satisfaction estimate is biased AND has random variability.

Which biases are most problematic?

Ranking by severity:

1. Non-response bias (most problematic) — directly distorts results; unknown magnitude
2. Undercoverage (very problematic) — missing entire subgroups; results don't apply to all
3. Sampling error (least problematic) — unavoidable but understood; reduced with larger samples; we can quantify it with confidence intervals

Why?: Sampling error is predictable and decreases with n. Biases are systematic distortions we can't easily quantify or correct without knowing the truth.

Lesson for decision-making: Always ask: "Who is missing?" (undercoverage) and "Who didn't respond?" (non-response). These are far more problematic than sampling error.