SOC 220 - October 8

Overview

Jeff Brassard finishes SOC 220’s sampling unit by explaining how sample size, nonresponse, subgroup representation, and sampling method affect the strength and limits of research findings. He distinguishes probability sampling from convenience, snowball, and quota sampling; describes online polling and content analysis; and reviews the October 15 midterm format and October 13 office hours.

Key takeaways

A larger sample reduces sampling error, but the gains diminish with each increase; researchers must balance sample quality against time and cost.
Around 400 respondents is a useful rule of thumb for enabling some complex analyses, but smaller samples can still support simpler comparisons.
Representative samples may contain too few members of small populations for subgroup analysis, making oversampling and subsequent weighting useful strategies.
Convenience and snowball sampling can support qualitative interviews and pilot research, but self-selection limits confident generalization.
Quota sampling can make online polls faster and less expensive by matching categories such as region and age, but opt-in participation means it is not probability sampling.
Even a well-designed probability sample supports conclusions only about its target population and the period when data were collected.

Chapters

0:00 Midterm Schedule, Exam Location, and Review Support

Jeff Brassard says the October 15 midterm is in Education South room 129, not the regular classroom.
He plans to hold drop-in office hours on October 13 in Tory 424 instead of class, for questions about the exam or assignments.
The sampling lecture completes the content planned for the midterm; later course material will be assessed on the final.

7:50 Sample Size: Absolute Numbers Matter More Than Population Share

For large populations, sample size matters more to sampling error and confidence intervals than the sample’s percentage of the population.
Common sample sizes include 400, 900, 1,600, and 2,500; samples above 10,000 are relatively uncommon.
Increasing sample size reduces sampling error, but each successive increase produces a smaller improvement.
NORC’s General Social Survey typically samples roughly 1,500 to 3,000 people from the U.S. population, which numbers hundreds of millions.

13:24 Why 400 Respondents Can Enable More Statistical Analysis

Brassard presents 400 as a practical rule of thumb for enabling analyses such as multivariate regression and weighting.
Below 400, some complex statistical procedures become more difficult or unreliable, though simpler cross-tabulations and bivariate analyses can still be useful.
Researchers should weigh added sample quality against the time and cost of recruiting more respondents.

16:25 Oversampling Small Populations for Meaningful Subgroup Analysis

A representative sample of 3,300 people may include only about 25 Jewish respondents and 30 Muslim respondents, too few for many subgroup analyses.
Researchers focused on small populations may need to oversample them—potentially to around 400 respondents each—then weight the data to reflect population proportions.
The same principle applies to small groups within a university population; the sample needs enough members of the subgroup to support analysis.

21:15 Nonresponse and Population Diversity Shape Sample Requirements

Response rate is the share of a sample that participates; Brassard notes that some recent studies proceed with rates around 46%, rather than the once-expected 80%.
Nonresponse can bias findings when participation differs by characteristics such as age; he also notes challenges in appropriately sampling Black populations in the United States.
More heterogeneous populations generally require larger samples to include enough cases from relevant subgroups.

24:12 Convenience Sampling for Interviews and Pilot Research

Convenience sampling recruits readily available participants, such as University of Alberta students responding to a Reddit post seeking interviewees about exam-system changes.
Because participation is self-selected rather than random, findings cannot be generalized confidently to a wider population.
The method can suit qualitative research focused on detailed personal experience, or pilot studies used to test survey questions and measures.

29:56 Snowball Sampling Uses Participants’ Social Networks

Snowball sampling begins with a few participants and asks them to refer or introduce others who may be eligible.
The approach is a form of convenience sampling: a researcher might recruit through Reddit, then ask interviewees to connect them with additional contacts.
Participants’ social relationships can help researchers reach people who are difficult to recruit through broad invitations.

33:55 Quota Sampling in Online Polling: Screening to Fill Categories

Quota sampling uses screening and opt-in responses to fill specified categories rather than randomly selecting from a sampling frame.
A Canadian poll might set regional quotas, then screen by postal-code prefix and other characteristics such as gender, age, ethnicity, class, or education.
Once a category’s quota is full, further respondents in that category may be screened out.
Polling firms such as Leger, Abacus Data, and EKOS commonly use quota-based online panels; carefully configured filters matter to data quality.

43:55 Generalization Is Limited by Population and Collection Date

Probability sampling supports generalization only to the population actually sampled: a survey of University of Alberta students does not establish what University of Toronto students think.
Survey findings are time-bound; a cross-sectional survey describes the period when data were collected.
Longitudinal data may help identify trends, but later events can make earlier findings outdated.

45:57 Content Analysis Counts Patterns in Text and Media

Content for analysis can include newspaper articles, videos, television, films, and diary entries; analysis may be quantitative or qualitative.
Quantitative content analysis can use LexisNexis to track a term such as “woke” across newspaper articles and time periods.
Researchers can also sample media or years—for example, comparing how articles frame sex work in different periods.
Brassard describes quantitative content analysis as uncommon in contemporary Canadian sociology and more associated with media studies.

49:49 Online Sampling Challenges: Opt-In Panels and Coverage Gaps

Online surveys can exclude people without internet access, devices, or technical ability, and opt-in platforms do not provide a straightforward probability sample.
Survey respondents may skew younger, more educated, and more urban; unhoused people are one example of a population that can be difficult to reach online.
When a complete online sampling frame is unavailable, quota sampling is a common practical approach, though it does not match the strength of a true probability sample.

52:38 Midterm Structure: 30 Multiple Choice and Four Short Answers

The October 15 exam includes 30 multiple-choice questions, distributed roughly evenly across the major course sections.
Four short-answer questions test two content areas, such as defining terms or distinguishing concepts, and two applied skills.
Applied questions may ask students to create a research question, operationalize variables, or write survey questions with response options.
Students type their short answers on a laptop or another device; Brassard says the exam setup is detailed in the Canvas announcement.

56:38 Operationalization Questions and Closing Discussion

Brassard says students may use and cite a validated measure, such as a perceived-stress scale, when designing a survey for an assignment.
He recommends also writing some original questions to practise the skill needed for an exam scenario where no validated scale is supplied.
The remaining classroom discussion turns to university teaching, research incentives, and the differences students experience across disciplines.

Summary, takeaways, and chapters were generated by AI from the video's transcript and may contain errors. The video belongs to its creator, Jeff Brassard.