Please enable JavaScript.

Coggle requires JavaScript to display documents.

PS906 Sampling (Week 5) - Coggle Diagram

- - - - questions
      - who it is given to
        
        randomization
        
        e.g. random digit dialing of telephone
        
        non-response bias
        
        not representative
        
        voluntary response bias
        
        e.g. customer service surveys
        
        only respond with positive or negative
        
        under-representativeness
    - - social desirability bias
        
        anonymity
        
        wording in neutral way
- - - - Random assignment to conditions
    - - Random sampling (and representative)
  - - - Representative Samples
        
        Sampling error is random
        
        Allows for generalising from specific sample to population
        
        (external validity)
      - Simple
        
        Equal Probability of Selection Method (EPSEM)
      - Stratified (Random)
        
        Division of population into mutually exclusive groups
        
        select random sample from each group
        
        set of groups make up levels of stratification variable
        
        Proportional
        
        No. of people selected from groups are proportional to sizes in population
        
        Strong type of sampling -> allows generalization
        
        More efficient than random sampling (cost-wise)
        
        Disproportional
        
        No, of people selected from groups are not proportional to sizes in population
      - Cluster (random samplling)
        
        Clusters rather than individual elements are chosen
        
        e.g. schools, hospitals
        
        One-Stage
        
        Randomly select a sample of clusters
        
        All elements in the cluster constitute the final sample
        
        Two-Stage
        
        Randomly select a sample of clusters
        
        Randomly select a sample of elements from these clusters
      - Systematic Sampling
        
        Determine sampling interval (k)
        
        population / desired sample size
        
        randomly select a number between 1 and k
        
        include each kth element in your sample
        
        periodicity
        
        problem when there is a cyclical pattern in the sampling frame
        
        Remember!
        
        Random starting point
        
        Regular interval
        
        Population list must be in random order to avoid systematic bias
        
        Good for
        
        Large population
        
        Randomized population
        
        Representative sample (if population is homogenuous)
        
        Examples
        
        Surveying 100 students from 1,000 students in school
        
        Quality checks of 100 products in manufacturing factory of 10k
    - - Not Representative Samples
        
        Biased samples
        
        weaker in external validity
      - Types
        
        Convenience
        
        include people who are readily available / volunteer / are easily recruited
        
        Quota
        
        setting quotas for certain groups
        
        convenience sampling until quota are filled
        
        Purposive
        
        specify characteristics of population of interest
        
        locate individuals who match the required statistics
        
        Snowball
        
        identify other potential participants with specific inclusion criteria
        
        useful when sampling from hard-to-find population
  - - - population is heterogeneous
      - I want to break data in multiple subcategories
      - I want study to be more precise (having a relatively narrow confidence interval)
      - I expect a small effect
        
        because small effect size means subtle difference
      - I use less efficient sampling techniques
    - - Experimental Design
        
        a priori Power analysis
        
        determine N needed to reach reasonable power for desired effect size
        
        Reasonable power is > 0.8
        
        Desired effect size
        
        (H1 - H0) / standard error (e.g. z-score)
        
        need to know anticipated standard error
- - - - e.g. therapy clients at a college clinic who have high, medium and low self-esteem
    - - e.g. therapy clients at a college clinic who have very high and very low self-esteem
    - - e.g. select adolescent girls for a FGD on diet and ideal body images
    - - e.g. select and in-depth interviews with several college students without healthcare coverage
    - - select cases that are known to be very important
        
        e.g. identify a department with a highly effective remote work policy that has seen significant improvements in employee morale and productivity when we want to study the link between RWA and employee morale and productivity in a large tech company
    - - select cases that will disconfirm generalisations
        
        making sure that I am not just selectively finding cases to support my personal theory
        
        e.g. I want to study employee burnout in a high stress industry
        
        Find employees who have been working in high-stress environment for long time but have not experienced burnout
        
        By studying these cases, I can identify protective factors that may mitigate stress and prevent burnout
    - - e.g. when a major org change (merger / acquisition) happen unexpectedly
        
        take the chance to study how employees respond to and adapt to org change in a manufacturing company