Polar. America Exam 1

Diploma divide

The diploma divide varies across groups and places. The association alone cannot

show that attending college caused it.

educational attainment now splits the parties, with college graduates moving toward Democrats and voters without degrees moving toward Republicans - roughly the reverse of mid-century class voting. Strongest among white voters; historically much weaker among Black and Hispanic voters.

ex: College graduates tend to lean Democratic while nongraduates tend to lean Republican.

Motivated reasoning

A preferred conclusion guides the search. The justification can still feel impartial: the

illusion of objectivity. we reason like lawyers rather than scientists - starting from the conclusion we want and recruiting evidence for it.

ex: A resident chooses her preferred housing policy, then searches for evidence that supports it.

Curation

Curation selects existing material. It changes what people encounter without creating the stories

ex: your social media feed

partison sorting

Sorting aligns party labels with existing views. Polarization involves growing differences in positions or feelings.

Different from polarization bcs they are still keeping policy views consistent not disdain for other party 

Polarization involves growing differences in positions or feelings

Feeling thermometer

Long-run out-party warmth fell while own-party warmth stayed roughly flat.

Respondents may picture politicians rather than voters.

rate a group from 0 (cold) to 100 (warm).

Cross-pressure

Cross-pressure comes from views that point toward different coalitions. It does not

determine the voter's party choice.

Calibration

Calibration compares probabilities with outcomes across repeated cases. One unlikely result does not invalidate a forecast.

Elite polarization

Elite polarization concerns political leaders and activists. It can follow a different pattern from public opinion.

division among officeholders, party officials, policy intellectuals, and activists.

note: Congress has become increasingly more polarized

instrument of measure: roll call votes

Hardening

Hardening is change within continuing partisans. A rising group average can also

reflect changes in who joins or leaves.

the same partisans coming to dislike each other more.

File-drawer problem

The file drawer hides whole studies. Published research can therefore be an

unrepresentative sample of completed research. results that come out flat often never get written up or published, so the published literature is a biased sample of the research actually done.

Production

Production creates new material at low cost. Whether that material persuades readers requires separate evidence

Affective polarization

partisans disliking and distrusting each other. The emotional and social dimension rather than the policy dimension. Measured by feeling thermometers, social distance questions, trait ratings, and behavioral experiments.

Dislike and distrust of the other party

Affective polarization concerns partisan hostility. Actual resource allocations measure behavior, while marriage questions measure expressed social distance

Party churn

turnover in who holds a partisan label at all. Compositional, not attitudinal. Hardening and churn both push the familiar trend line up, and only panel data separates them.

Party churn changes group membership. Hardening changes the feelings of

continuing partisans. Both can affect a trend

Conversation

Conversation responds during an exchange. A fixed message cannot adapt to each

reply.

Policy polarization

Policy polarization concerns issue positions. A single peak shifting left is different from the middle emptying out. Policy polarization is when extreme views on some matter of public policy have become more common over time

: extreme views on some specific matter of public policy have become more common over time. Issue by issue, not global - the answer can be yes for abortion and no for something else. Measured by issue distance on survey scales.

Polarity reversal

Polarity reversal separates education from income. The other two combinations face

competing political pulls.

the socioeconomic basis of partisan support inverting. Education and income are treated as two separate variables rather than one thing called "class," and the diagonal cells of the resulting two-by-two - high education with low income, low education with high income - become the most committed constituencies of left and right.

Correlation vs. causation

Correlation alone cannot establish causation. A shared cause or reverse direction can produce an association. correlation means two things move together; causation means changing one changes the other. Reverse causation and confounding are the two standard alternative explanations.

DW-NOMINATE

DW-NOMINATE shows growing party separation. Voting agreement can reflect party pressure as well as personal conviction.

: a scale built from congressional roll-call votes that scores every member of Congress from -1 (most liberal) to +1 (most conservative) by placing legislators who vote alike near each other. It runs on behavior, not rhetoric.

- ANES: a national probability-survey series that repeatedly fields many comparable core measures, making careful comparison across decades possible despite changes in wording and mode.

Panel data

Panel data track individuals. Fresh cross-sections cannot do that directly, and

selective panel dropout can distort trends.

: the same people re-interviewed over time. A fresh cross-section each year can never tell you whether a person changed or the pool changed.

Expressive responding

Expressive responding signals loyalty. A smaller factual gap does not establish an

equal change in political attitudes.

Output audit

An output audit measures responses under specified prompts. Quiz scores alone

establish neither developer intent nor persuasion.

Mass polarization

Mass polarization concerns the public. The study must still specify whether it measures policy, ideology, or feelings.

divisions among voters and citizens

for measuring: surveys for the public

Political dimension

A political dimension organizes positions. Economic agreement can coexist with social disagreement

Selective reporting

Selective reporting hides attempted analyses. The file-drawer problem hides whole

studies. : running many analyses and writing up the one that worked. Usually not fraud - ordinary incentives acting on ordinary people.

Durability

Durability is persistence after exposure ends. Population effects also depend on how many people encounter the tool.

Accuracy prompt

An accuracy prompt redirects attention toward truth. Curiosity also supports engagement with opposing evidence. simply pausing to ask "is this true?" before sharing. People share misinformation largely out of inattention, and a single accuracy nudge improves the quality of what they share afterward.

Value judgment

Evidence estimates consequences. Values determine how people weigh the tradeoffs.

Partisan bias in behavior

The allocation is an observed action. It does not directly measure policy views or

changes in party identity.

Ideological polarization

division in general orientation to politics and government, usually on one left-right dimension; beliefs become more consistent and more extreme. Measured by ideological constraint across issues and self-placed ideology.

General orientation to politics and government

• Typically think in terms of liberals and conservatives: one left-right dimension

• For elites, ideology is tightly organized and predicts nearly every vote

• For most voters, the labels are loosely held and weakly tied to issue positions

• That elite-public gap is one of the oldest findings in the field

sorting vs polarization

Polarization: positions move apart - people actually hold more extreme views

those political phenomena where the public and its leaders have become increasingly divided. Deliberately vague - it covers the public and the elites, and does not say what kind of division.

• Sorting: positions stay put, but party and ideology line up more neatly

: partisans disliking and distrusting each other. The emotional and social dimension rather than the policy dimension. Measured by feeling thermometers, social distance questions, trait ratings, and behavioral experiments.

: sorting changes WHO holds a view; polarization changes WHAT views are held

• Liberals became Democrats and conservatives became Republicans

• Same opinions, cleaner packaging - and it looks identical from a distance

partisanship vs polarization

Polarization refers to differences on policy issues, ideological orientations, or value systems

• Partisan polarization refers to these differences across members of different parties

• Partisanship can also be a team identity without a policy disagreement underneath

Dimensions in politics

factors that is measurable and predictable

an axis along which political positions can be arranged and along which parties compete, such that knowing where someone sits on it predicts a bundle of their opinions. The test is predictiveness.

liberals/ conservatives

social/economic

policy vs ideological polarization

policy is distance on a specific issue; ideological is consistency and extremity across a whole belief system.

- Affective versus everything else

affective polarization needs no policy disagreement at all. Two sides can come to loathe each other without disagreeing about a single policy.

partisanship and partisan polarization

- Partisan polarization: differences in policy positions, ideological orientations, or values lining up across party lines.

- Partisanship on its own needs no policy or attitudes underneath it. It can be inherited, social, tribal - a team allegiance. A fan needs no policy reason to hate the rival, which is exactly why partisanship explains behavior that policy differences cannot.

Mass vs Elite

- The two can move independently. Everyone agrees elites polarized; nearly every serious fight in this field is about the mass side.

Composition versus change

whenever a group average moves, ask whether the members changed or the membership changed. This earns points on all three exams.

Measurement traps

- Forced choice: binary questions manufacture division that graded scales reveal as moderate. Many "polarized" respondents are just unsure.

- Social media as a sample: a small share of users produces most political posts, so a feed samples the loudest, not the typical.

- Elite rhetoric: cable news shouting matches measure elite incentives, not voter opinion.

- Common thread for all three: mistaking a visible sample for the population.

Second dimension

civic and cultural liberties, race, and immigration - rights and belonging, rather than taxes.

- Cross-pressured voter

: someone whose economic interest points one way and whose social or cultural views point the other. A second dimension creates them, and they become the prize both parties chase.

Careful

vote share, party identification, and margin are three different things. Press coverage conflates them. Check which one a chart is plotting.

Liberal versus democratic

: liberal points to individual rights, tolerance, and limits on what any majority may do. Democratic points to majority rule and popular sovereignty. "We won, so we get to do this" is the democratic half; "you cannot do that to us, whatever the vote" is the liberal half. Both sides use both arguments depending on who holds power.

Preference falsification

misrepresenting what you want when the perceived social cost of dissent is high.

- Knowledge falsification

misrepresenting what you know, in order to make the preference falsification believable.

Common knowledge

everyone knows it, everyone knows everyone knows it, and so on recursively. Preference falsification can instead create pluralistic ignorance or the public appearance of agreement when private belief is rare; do not call the false claim itself common knowledge.

Threshold / cascade model

: individuals differ in how many others must speak before they will, thresholds are invisible from the outside, so the distribution of thresholds decides whether a spark fizzles or topples a government. Unpredictability here is a property of the system, not a failure of analysts.

herding

experts converging because they are relying on each other and on the same few studies. Counting experts is not the same as counting independent pieces of evidence.

Accuracy goals versus directional goals

sometimes we want the truth (booking a flight), sometimes we want a particular answer (our side gets accused of something). From the inside the two modes feel identical.

- Identity-protective cognition

political beliefs work as tribal badges, so being wrong with your group can be safer than being right alone. Your belief about climate policy changes nothing about the climate; dissenting from your family or friend group has costs you pay this week. That is why a factual challenge registers as a threat rather than as information.

Beliefs can protect group standing. A correction from a trusted person on the same

side can feel less threatening.

Backfire effect

the claim that a correction pushes people further into the false belief. Much rarer than you have heard.

Agent framing

the model is a mind - maybe alien, maybe superhuman, but a mind - so the interesting questions are its goals, values, and intentions, and the risks are deception, misalignment, and takeover. Note the tell: optimists and doomers share this premise and only disagree about the agent's goals.

Cultural technology framing

: a large model is a technology of cultural transmission, in the family of writing, print, markets, and bureaucracies. None of those is a mind; each let people use information other people had already accumulated, and each changed what a society could know and do.

- Why the frame matters: the agent framing sends you looking for intentions, which nobody can observe. The cultural-technology framing sends you looking for measurable things - whose voices show up in the summary, whose get compressed away, what happens to the range of what people read and believe.

This framing asks how institutions and tools shape the spread of ideas, rather than

assuming the model has independent goals.

homogenization

if millions of people consult the same summary they get the same answer; rare views get compressed away and common views amplified. Not a rogue mind - a cultural monoculture Homogenization reduces variety. Fragmentation instead gives people increasingly

separate information environments.

Fragmentation (the filter-bubble worry):

everyone gets a different, personalized reality.

- Homogenization and fragmentation are opposite predictions. Both are plausible. A claim that fits every outcome is not a claim, so a good theory has to say which one, when, and for whom.

- Three doors into politics

curation (AI chooses what you see - recommender feeds), production (AI writes what you see - generated text, audio, video at near-zero marginal cost), conversation (AI talks with you directly, adapts, remembers). The third has no historical precedent: print and cable never talked back.

- Base model versus fine-tuned model:

a base model is the raw pretrained system; fine-tuning is the later stage that turns it into a helpful chat assistant. The distinction matters because it locates where a measured political lean comes from.

Construct validity gap

a valid instrument can measure a narrow construct. A quiz score measures how a model answers quiz items in an artificial setting - not what it says in a real conversation, and not what a user takes away.

For every measure, practice saying what it cannot sho

. DW-NOMINATE says nothing about whether voters dislike each other. Thermometers say nothing about issue distance. Issue scales say nothing about identity.