Data Science Midterm 1

What are the strength of polls?

Quick, inexpensive, can represent large populations if well designed, track over time

What are the weaknesses of polls?

Sampling bias, nonresponse bias, poor question wording, bad timing

What makes a high quality poll?

Random sampling, transparent methods, weighting for demographics, reports margin of error

What makes a low-quality poll

Nonrandom or opt-in samples, small sample size, unclear methods, biased questions.

What is post-stratification weighting?

adjusting sample data to match population demographics (eg. by age, race, gender)

Why use survey weights?

To correct for over- or underrepresented groups so result better reflect the population

What”s the difference between correlation and causation?

correlation = relationship; causation = one variable directly affects another.

What conditions are needed for a casual claim?

random assignment, control group, and no confounding variables

What is a confounder?

A variable related to both the cause and effect that can distort results

What is ATE (Average Treatment Effect)?

The average difference in outcomes between the treatment and control groups

How is ATE calculated?

ATE= mean(outcome in treatment) - mean (outcome in control)

What does statistical significance mean?

The observed effect îs unlikely due to chance 

How do you calculate cost per vote?

total campaign cost / estimated number of additional votes gained

What does cost per vote tell us?

The efficiency of a political intervention in influencing votes

Why are the visualizations important to data science?

They help reveal patterns, trends, and outliers in data

What are signs of a misleading visualization?

Truncated axes, distorted scales, lack of labels, or missing content

How can visualizations be improved?

use full scales, clear labels, consistent colors, and contextual explanation

What is computational thinking?

using coding and algorithms to analyze, organize, and manipulate data

What’s data cleaning?

fixing missing or incorrect data, removing duplicates, formatting variables

What is cross tab?

A table showing the relationship between two categorial variables ( eg. party x gender)

What is inferential thinking?

using data from a sample to make conclusions about a larger population

what are the two main types of inference?

estimation( predicting values) and hypothesis testing (testing claims)

What is sampling variability?

The natural difference in results when you take different random samples from the same population

What is computational thinking in data science?

using programming to process and visualize data

What is inferential thinking in data science?

Using probability and statistics to reason about uncertainty and draw conclusions

How do computational and inferential thinking work?

code organizes and analyzes data; inference interprets what those results mean for the real world.

What are means by group

calculating the average of a variable for each subgroup (eg. mean income by gender)

What are the main goals of a political campaign?

To persuade undecided voters and turn out supporters until reaching 50% +1 of votes

What are the two main ways campaigns?

Persuasion (changing minds) and mobilization (getting supporters to votes.

What is targeting in a campaign?

identifying which voters to reach out to, using data to prioritize persuasion or turnout

What do turnout targeting and supporter targeting mean?

Turnout targeting predicts who will vote; supporter targeting who supports your candidate

Why is the mathematics of victory different across states?

Different partisan balances and voter behaviors require different strategies (eg. Alabama vs. Pennsylvania)

What are the six voter types campaigns analyze?

Supporters who will vote, supporters who might not, undecideds who will vote, undecideds who might not, opponents who will vote, opponents who might not 

What law created modern digital voter files?

The Help America Vote Act (2002) requiring statewide computerized registration lists

What does a voter file include ?

Name, address, vote history (whether you voted, not who for) and sometimes demographics

What are “commercial voter files”?

Enhanced voter lists combining public voter data with consumer info (like donations, subscriptions, property value)

Why campaigns use voter files?

They are the foundation for modeling turnout, support, and targeting.

What political factors affect voter data laws?

States pass laws shaped by partisan goals- some limit race/ party data, others expand it for campaign advantage.

What’s a major ethical concern about campaign data use?

Microtargeting can narrow political outreach and reduce exposure to diverse viewpoints

What is Random Digit Dialing (RDD) and why is it declining?

Random phone surveys once produced representative samples, but response rates have collapsed are now biased

What makes today’s polling less reliable?

Non-response bias: certain political groups (like conservatives) are less likely to participate.

Why are voter files better for sampling than phone lists?

They include verified vote histories, reducing reliance on unreliable self-reported voting.

What are “turnout models”?

Statistical models predicting how likely each voter is to vote, based on history and demographics.

What was the major reason for polling error in 2020?

Between-party and within-party nonresponse — too few Republicans and unrepresentative Democrats responded.

What is a “tipping point state”?

The state that gives a candidate the decisive electoral votes to reach 270.

Why do campaigns focus on tipping point states?

Shifting a few votes there can change the national outcome, unlike in safe states.

What is “response substitution” in polling?

When respondents answer a question about change by expressing their current opinion instead (e.g., Roy Moore example)

How do survey experiments improve over direct questions?

By randomly assigning some respondents to receive information, isolating the message’s causal effect.

Why can’t we rely on observational data to evaluate campaign strategies?

Confounding bias — people who choose to engage are already different from those who don’t.

What is selection bias in campaigns?

More politically interested voters are likelier to be contacted, making “contact increases turnout” claims misleading.

What does random assignment achieve in experiments?

It makes treatment and control groups comparable and removes selection bias.

What is the “potential outcomes framework”?

Thinking about what would happen with and without treatment for each voter to define causal effects.

What does the “fundamental problem of causal inference” mean?

You can never observe both outcomes for the same individual — only one.

What are examples of confounding in campaign studies?

Comparing contacted vs. uncontacted voters without accounting for political interest or prior turnout.

What does a high p-value (e.g., 0.8) mean?

The result could easily occur by random chance; not statistically significant.

What’s the difference between statistical and practical significance?

Statistical = unlikely by chance; Practical = meaningful for real-world strategy.

Why visualize data in campaign analysis?

To identify outliers, misreporting, and trends that affect messaging and targeting.

What is the “Lie Factor” in graphs?

The exaggeration or understatement of effects due to misleading scales or visuals.

What is the “Data-Ink Ratio”?

The proportion of ink used for data versus decoration — high ratio = more honest graph.

What is Simpson’s Paradox, and why does it matter?

Trends can reverse when groups are combined — can mislead campaign data interpretation.

Why are outliers important in campaign data?

They can either reveal new voter patterns or distort averages if not examined carefully.

What is the formula for deciding whether someone votes?

pB + D > C

In the voting formula, what do the letters stand for?

• p = probability vote is pivotal

• B = benefit of one candidate winning

• D = psychological/civic duty benefit

• C = cost of voting

Which parts of the formula can campaigns influence?

D (increase civic motivation) and C (reduce voting costs)

What is an experiment in political science?

A study where subjects are randomly assigned to treatment or control to measure a causal effect.

What’s the difference between percent and percentage point?

• Percentage point = simple subtraction of percentages

• Percent difference = relative change ((new − old)/old × 100)

What does “statistically significant” mean?

The result is unlikely due to chance (p-value < 0.05).

What is a p-value?

The probability of seeing a result as extreme as the observed one if the null hypothesis were true.

What is the null hypothesis in an experiment?

That the treatment has no effect.

What was tested in the Virginia experiment?

Whether sending text messages increased voter turnout.

What were the turnout rates?

Treatment 11.3%, Control 10.7%

How many text messages created one new voter?

167 text messages.

What was the cost per vote if each text cost $0.07?

$12 per new voter.

Was it effective?

Yes—cheap and slightly effective, but small effect.

What was the purpose of the Georgia experiment?

To test if increasing civic duty (“D”) boosted turnout.

What were the three groups?

Control, Plain Reminder, and Gratitude Letter.

Why do we randomize?

To make groups comparable and remove bias.

What is the goal of calculating ATE?

To find the average impact of the treatment on the outcome.

What does a small p-value mean?

The observed effect is unlikely due to random chance

Why is explaining your reasoning important on the midterm?

You get partial credit if the logic is clear even if math isn’t perfect.

What is the #1 rule for voter mobilization?

The more personal the contact, the more effective.

Which methods work best?

Door-to-door and volunteer phone calls.

What’s the effectiveness per contact?

 •    Door-to-door: 1 per 15

    •    Volunteer phone: 1 per 35

    •    Commercial phone: 1 per 125

    •    Mail (nonpartisan): 1 per 273

    •    Mail (social pressure): 1 per 77

Do emails or social media ads increase turnout?

No clear effect

What matters most in personal communication

Quality — authentic, unscripted, personal contact.

Do advocacy or issue-based appeals mobilize people?

Not much — making voting easy and social norms are more effective.

Is there strong evidence of synergy between tactics?

No — effects are additive or diminishing, not multiplying.