Lecture 6 - Principles of Large C Research Flashcards
A spatially and temporally delimited phenomenon of theoretical interest. For example, a country, the social media posts by a political party, or a survey respondent. Typically contains multiple observations.
Case
The lowest-level unit in an analysis at which a measured variable can only take one value ( unit of analysis). For example, a country-year, an individual post on social media, or a survey respondent. May or may not be of theoretical interest.
Observation
The set of cases or observations that are analyzed in a given piece of research. For example, all European countries during a given time period, all social media posts by the Democrats and Republicans, all individuals in the UK.
Sample
he set of cases which in combination make up the universe of cases, i.e., all cases. For example, all countries in the world during a given time period, all social media posts by political parties, all individuals in the UK
Population
Population Inference
- Empirical research typically wants to say something about populations of cases
- However we are often unable to study entire populations of cases
- Smaller sample sizes of cases are subjected to analysis instead
Inferring something we do not know (patterns in the population) from something we do know (patterns in the sample)
Population inference
Surviorship bias
Which only “surviving” cases are selected, ignoring those that fell out of view.
Selection on the dependant variable
- Dont select cases based on the values of the dependent variable
- Risk of biased causal inferences
- Generaly avoid in large C research
What are the DONTS in case selection
- Cherry-picking cases that are known to support a hypothesis
- Performing an inductive study and then using the same set of cases to test the resulting theory and hypotheses
- In large-C research: selecting cases based on the dependent variable
Small C study
An intensive study of a single case or a small number of cases, typically quant
Large C study
Less intensive study of a large number of cases using quant methods
Data matrix = Rectangular dataset
Rows = observations
Columns = variables
Cases are either groups of observations or individual observations
Data Matrix
Rows are observations and columns are variables.
Strengths of Large C
- Potentialy for generalizability, facilitates the study of random samples from a population.
- Increased ability to identify causal effects
Costs of Large C
- No intensive study of small cases, less attention to context.
- Generally less useful for inductive research, stronger focus on individual cases can make it easier to identify new patterns and potential causes.
- Limited usefulness for interpretivist research, not amenable to detailed analyses of motivation of individual actors and meaning they attach to actions.
Principles of large C case selection
- Case selection in large-C research generally has a twin goal
Large enough sample to enable robust statistical inferences
Sample that is representative of population - Samping strategies wich have the potential to meet both goals
Probability sampling
Total population sampling
-
Total population sampling
- Select all cases in a population
- Census
- High external validity
Simple random sampling
All sampling methods where cases are randomly selected from a larger population with a known, nonzero probability are referred to as probability sampling (or: random sampling)
-
Stratified Random Sampling
Populations are divided into relevant strata (i.e., subclasses, such as different people with different genders, age groups, or ethnicities)
Then cases are drawn at random from different strata
Ensures that important groups are adequately represented
Achieves more representative samples with fewer observations
Non-Probability Sampling
- Cheaper or more easily feasible
Non-Probability Sampling - Convenience sampling methods
Volunteer sampling (e.g., undergraduates, surveys on news website, participation requests on social media)
Snowball sampling, find respondends and then ask them for other similiar people.
Non-probability sampling - Quota sampling
Quota sampling requires researchers to interview a certain number (quota) of respondents with certain attributes. Researchers can fill the quotas with any respondent as long as they fulfill the quota attributes. Can be weighted, but still generalizability is uncertain. Good for experimental research.
