Terms: Respondents (people who respond to survey),
Answers the following:
-What are surveys used for?
Lots of stuff. Predicting elections. Evidence in courts of law. Misleading advertising. Unemployment rate.
-What can they measure?
Attitudes (likes and dislikes). Preferences (comparisons of attitudes toward different objects). Beliefs (opinions about the objective state of world). Predictions. Facts (how many years people attended school). Past behavioral experiences.
-What kinds of questions can they be used to answer?
Four broad classes of questions: (a) prevalence of attitudes, beliefs, and behavior (b) changes in them over time (c) differences between groups of people in their attitude,s believes, and behavior; and (d) causal propositions about these attitudes, beliefs, and behavior.
Explores the following:
-Deciding what are the objectives of the study?
Deductive approach: research should derive hypotheses from theories that it tests. Inductive approach: don’t create hypotheses from theories, just learn relationships from data.
-Deciding the design of the survey?
What population should be studied? How should the data be collected? Are follow-up surveys necessary?
Sampling Frame: the list of units from which the sample will be drawn.
Haphazard sampling: sampling people who are easy to contact.
Quota sampling: based on what you believe is the population’s stats, you only ask that percent of people. So if the census says 50% women, you only admit 50% of women to your study.
Stratifying: dividing the population into small chunks, and then randomly sampling from each chunk.
Cluster Sample: taking clusters of samples in certain areas.
Multistage area sampling: sort of percolate down within your choices, so first choose cities, then neighborhoods, etc.
Parallel samples: samples taken of the same population to compare them.
PSU (primary sampling unit): largest regions to look at. So they could be a city, or a group of smaller counties.
Answers: what to ask? What forms should the questions take? How should questions be worded? What response choices should be offered? In what sequence should the questions be asked?
Social desirability bias: respondents not answering honestly because the answer is not socially acceptable.
Double barreled questions: asking two questions at once.
Consistency bias: tendency of respondents to answer similar questions similarly.
Primary ways of collecting data: face-to-face interviewing, telephone interviewing, self-administered.
Chart of page 121 is good.
Manifest coding focuses on the content of the answer.
latent coding focuses on the style of the answer.
Coding: taking responses and turning them into numbers, which correspond to mutually exclusive and exhaustive categories of answers.
How to design a survey.
Forward telescoping: remembering event having occurred more recently than it actually did.
Backward telescoping; the opposite (longer ago).
Randomized response technique: conceal the information from the interviewer so the respondent feels save by randomizing the question answer.
Non attitude: a respondent not having an answer to the question.
Panel: repeated interviewing of the same people over time.
Double-sample: used when you have a small subgroup of people (like Blacks), you interview twice as many and half their weight.
Answers steps of analyzing survey data: state hypothesis. Operationalizing concepts. Testing hypothesis.
Reciprocal Causation: When two variables “cause” each other. They are tied, essentially.
Nominal Variable: distinct categories, not related or numbered. Exhaustive.
Ordinal Variable: Inherits nominal, but is also ordered. “strongly agree”, “agree”, etc.
Interval/ratio Variable: Inherits Ordinal, but also has numbers.
Null Hypothesis: the opposite of the hypothesis.
Type 1 Error: Falsely rejecting a true null hypothesis.
Type 2 Error: accepting a false null hypothesis.
Sampling Error: Essentially, the margin of error.
Central Tendency: ways of measuring the typical responded. Mean, median, mode.
Bivariate recoding: Creating another variable based on two existing ones, essentially finding correlations.
Additive indices: Splitting up people based on their answer to a single question, and looking at how those groups compare.
Exam 1 Review:
Random things to remember: equivalent wording when analyzing over time, same target population.
“panel studies” are longitudinal studies.
Dichotomous (yes or no) are interval.
You do need to know the difference between “reliability” and “validity.” Validity is the distance between what’s intending to be measured and what’s being measured. Reliability is making sure that when they are asked repeatedly, the same result is found.
Margin of Error: how does it go up as sample size goes down? When you quadruple the sample size, the margin of error goes down by half. Doubling sample size, I think the margin of error would go down by around 1/4th.
Look over Sample Test 1. Done.
Go through all the terms in Chapters 1-9. <— do this.
Cross-tabulation: Essentially comparing two variables by splitting them up into categories.
You should put the independent variable spanning the columns.
Should probably do column percentages (adding up to 100%).
Measures of association: how strongly two variables are related. Good for when you have a lot of categories.
Somer’s d: ordinal measure of association. Based on percentage point difference. 1 is high association.
Monotonic relationship. Consistently increasing or decreasing going along columns.
Dichotomous variables are interval, so you can use ordinal measures of association (like Somer’s d).
Significance: Essentially, how likely was it that these results were obtained from sampling error. Significance is a function of sample size and the size of the significance observed.
Exam 2 Review:
Nominal style hypothesis: a ____ person is more likely to believe ____ than a ____ person.
Ordinal style hypothesis: the more ___ a person is, the more likely they think or do ____.
A causal hypothesis: States an explicit relationship between an independent and dependent variable. It tells us the independent variable causes the dependent variable in some way.
The independent variable columns should add up to 100%,.
Focus groups don’t represent and aren’t random samples.
Things that are probability sample: stratified, cluster.
Margin of error should not be reported for: mall intercept, small samples.