POSC318 Exam 1

electoral connection

elections are mechanisms for holding elected officials accountable

• Not totally straightforward in practice bc of gerrymandering, polarization, etc

• Representation between representation and reelection incentives…elected officials must eventually face voters again

classical democratic theory

Full participation (40%), high expectations about citizen competence, general will or common good among the people

• Full engagement only works in small, homogenous societies

  • Presence of actionable “general will” depends on this

democratic elitism

power lies within elites, who are entrusted with determining the common good

• Elites are the trustees bc they use their own judgement to decide what is best for the public, and better situated to govern than the masses

  • As opposed to delegates, which act according to what their constituents want

democratic pluralism

• Interest groups serve as intermediaries between public and elected officials

• Groups collect and direct the public will

  • Intense minorities, outsourcing effort to groups

participatory democracy

participation is higher and more consistent amongst groups with more resources

• Public opinion exists in people participating in voting, but there are barriers to political engagement

• Emphasizes citizens actively and directly participating in political decision-making beyond just voting

census

seeks to measure the whole population

survey weights

adjust the sample so it more closely matches known population characteristics; used to adjust influence of over/underrepresented groups

stratified sampling

divide the population into subgroups/strata, then sample within each

cluster sampling

select cluster and survey everyone within them

• Sometimes used when you can’t access a list of everyone individual but you can access groups/clusters

multistage sampling

select clusters then randomly select people within them

systematic sampling

selecting every nth person

quota sampling

divide population into specific subgroups and recruit a predetermined quota of participants from each group without random selection

convenience sampling

survey whoever is easiest to reach

• fast but not usually representative

snowball sampling

participants recruit other participants

• useful for hard to reach groups

• can produce homogeneous, non-representative sample

purposive sampling

researcher deliberately targets people with specific characteristics relevant to study

• useful for narrow populations, limited generalizability

respondent-driven sampling (RDS)

starts with “seed” participants that recruit others; is more meticulously controlled than snowball sampling

• Designed for hard-to-reach populations

• Can reduce some biases of ordinary snowball sampling

random errors

idiosyncratic errors that tend to cancel each other out in the aggregate and have only modest effects on our estimates of population parameters

systematic errors

consistent errors that lead to significant exclusions when sampling with significant effects on our estimates of population parameters

• Margin of error reported with surveys is ONLY telling us about random sampling error…there could be other sources of error that aren’t reported in an obvious way but have a big effect on data quality

coverage error

this is the gap between the population and sampling frame; people should be in the sample, but they have a 0% probability of being selected for it

• Problems arise when groups are systematically excluded from sample

sampling error

Difference between sample and true population, caused by only studying a sample and not everyone

nonresponse error

People who don’t answer the phone for example, and are systematically less likely to be included

refusal conversions

trying to convince people to take the survey who are refusing

• Persuading arguments, financial compensation, etc

open-ended questions

respondents give their own answers

• Advantages

  • Flexible, get more detailed responses

• Disadvantages

  • Harder to code and analyze

  • Very labor intensive, more difficult to provide aggregated result

closed-ended questions

predefined set of responses and participants must select one

• Advantages

  • Easier to code and aggregate

  • Quick to administer 

• Disadvantages

  • Opinion may not be represented among the available options

question wording error

errors in survey results caused by how question is phrased or worded

question order effects

when the order of questions influences how people answer later questions

social desirability

Desire to withhold true opinions bc of concerns about social judgement and legal considerations

• need to make it unobtrusive and nonreactive, so people don’t modify their response

leading questions

wording or framing that encourages a particular answer

loaded questions

uses biased or emotionally charged wording that influences how you respond, or that you might not agree with

double-barelled questions

asks about multiple things at once

confusing questions

wording is unclear and hard to understand

technical questions

uses technical terms or jargon that respondents may not understand

odd response options

answer choices that are unclear, unbalanced, or do not adequately cover the possible answers

priming/framing

earlier questions can influence responses to later questions

answer option effects

position of options can effect which one is chosen

salient events

people’s answers may be influenced by what is happening around them at the time of the survey

expressive responding

Occurs when people give answers that exaggerate or misrepresent their actual beliefs to signal loyalty to a political or social group

• Group signaling, psychological comfort, partisan inflation 

congenial inference (aka motivated reasoning)

allowing political bias to subconsciously influence memory and reasoning, leading to skewed answers that still feel true to the respondent

nonattitudes

people providing an opinion when they don’t really have one

multi-item scales

When your goal is to measure concept and you use multiple different statements

cognitive interviewing

check questions before going large scale: work through the questions conversationally to assess understanding and avenues for unconventional responding

think-aloud technique

Respondents explain their thinking while answering and narrate thought process without interruptions

probing technique

interviewer asks follow-up questions about the questions

piloting

test of how the survey actually works in practice before full survey

• Researchers look for non-response, don’t know answers, dropouts, and confusing questions that point to formatting errors

• check for missing data and technical problems

cleaning

researchers check data for anything that needs to be removed

• Bots, malicious responding, coding errors, missing data, unusual response patterns, careless answers, fabricated data

interviewer training

Helps them all follow same procedures and make responses more comparable

back translation

translate to another language, then back to original to compare meaning

parallel translation

multiple translators translate independently and resolve differences

sensitivity analysis

researchers compare their results with other studies and/or real-world population benchmarks

• Differences don’t mean survey is automatically wrong…could result from sampling differences, weighting, or genuine differences, etc

margin of sampling error

quantifies uncertainty around a sample statistic based on frequentist inference for probability samples

• Only covers random error, NOT systematic error

• Only applies to statistics based on full sample

• Increases the more you look at subgroups

relationship between sample size and MOSE

diminishing returns

• MOSE exponentially decreases at first with more people, but beyond 1000 doesn’t make much of a difference 

poll aggregation

a collection of polls from many sources that are averaged

• Capitalize on cancelling out random error, but doesn’t fix systematic errors that can become compounded

herding

pollsters adjusting their findings to match or closely approximate the results of other polls

• polls are weighted to try to match census, and if pollsters are making same mistakes in weighting data, it can mess up the aggregate

issue in horserace polling

if candidate is polling ahead of another, but confidence intervals/margins of error overlap, it is essentially a toss up

sampling frame

list/source used to identify potential respondents

exit polling

a survey of voters leaving their polling place on election day

• Used to get more accurate results ensuring people actually voted (people tend to overreport voting) 

• two stage process: cluster sampling

  • Voting locations are selected at random, then interviewers seek to interview voters according to a systematic rule (every sixth person)

issues in exit polling

changes in the ways people vote, selective refusal, election regulations, early leaks

public comment periods

where gov accountability office posts comment board and people can give opinions on federal legislation (self-selection)

• cheaper to invite comment, but not generalizable

• opt-in internet survey

flattening

tendency to force a unidimensional conception and operationalization of key variables

non-probability sampling

people are not selected through known probabilities

Advantage

• often cheaper and faster

• allows selection of sample based on expert knowledge

Disadvantage

• people in the sample are systematically different from those not in the sample

• limited generalizability, can’t quantify uncertainty

probability sampling

• Advantages 

  • Able to quantify uncertainty around estimates

  • Increased confidence in generalizability of results

• Disadvantages

  • Cost, feasibility

  • Superficial responding

best practices for journalists writing about survey results

• Always be clear about who was surveyed

• Let readers know when survey was conducted

• Stay faithful to the way survey questions were worded

• Pay attention to margins of error

• Good to provide context, but dangerous to ascribe causality 

  • Esp with single snapshot of time

• As a consumer, if this info isn't available, you should be skeptical about the data and/or reporting