Chapter 5: Selecting Research Participants (W4)
Stratified Sampling
Selection procedure dividing a population into subgroups, called strata, & a random sample is selected from each stratum. • Used to represent key subgroups in your pop.; restrictions are imposed to ensure each is represented. • Populations have many identifiable subgroups: income or SES, male/female, ethnicity • Want to make sure each is adequately represented.
Random Sample
- A sample selected from the sampling frame based on a mathematically random selection procedure. -A sample collected such that all elements of the pop have equal chances to be included. - Selection process is fair, unbiased - Most times, a proper random sample yields results that are close to the pop parameter - No guarantee that it is representative! - Other sampling techniques impose additional restrictions on this procedure to get representative samples...
Non-Probability Sampling: Examples
- Man-on-the street interviews for public opinion - Use of college students in psych research - Clinical practice: clients available for our sample - Much research: volunteers
Proportionate Stratified Sampling: Steps
1. Divide the population into subpops or strata 2. Create a sampling frame for each strata 3. Decide on a sample size & randomly select from each sampling frame Ex: - 10,000 employees --> 8% have 1 of the 4 disabilities; & 92% do not have 1 of the 4 - 800 employees who have 1 of the 4; & 9,200 employees who do not have any of the 4 - Sample of 200 employees (16 & 184)
Stratified Sampling: Steps
1. Identify key subgroups (strata). 2. Participants chosen by taking a simple random sample in each pre-identified subgroup. 3. Select equal numbers from within each subgroup. 4. Combine subgroups into overall sample.
Sampling: Reasons We Can't Include Everyone in the Population
1. Not always possible 2. Pop too large 3. Too costly 4. Time-consuming 5. Unwieldy (difficult) 6. Not necessary
Sampling Procedures
1. Probability Sampling 2. Nonprobability Sampling
Chapter 5 Topics
1. Sampling 2. Sampling Bias 3. Random Sampling Procedures 4. Non-Random Sampling Procedures 5. APA Formatting
Probability Sampling Types
1. Simple Random Sampling 2. Systematic RS 3. Stratified RS 4. Proportionate Stratified RS 5. Multistage RS 6. Cluster RS
Sampling Frame
= A list of cases in a pop, or the best approximation of it (ex: telephone directories, tax records, driver's license record) - This will still not ever be accurate, but it's as close as you will get!
Systematic Random Sampling
= A type of Simple Random Sampling. - Select one in a certain number of sample elements (participants), the number is from the sampling interval. 1. Sampling Interval • Entire population is represented & listed • Random person selected; every *n*th person after - Not as random: principle of independence is violated
Probability Sampling
= Based on random sampling: each member of a particular population has an equal & unbiased chance of being selected (good, but difficult). - Exact size of pop is known; extensive knowledge of pop needed (ex: can list all members) - Odds of selecting a certain individual is known; can be calculated - Estimating a pop size is very difficult (ex: ppl die, move, vacation), so you need the 'best estimate' of pop 1. Sampling Frame
Target Population
= Entire set of individuals sharing the characteristic of interest (ex: women with postpartum depression, older adults with diabetes, insomniacs)
Non-Probability Sampling: Purposive Sampling Pros
= Good to use for difficult to reach populations, cannot acquire list of population & random sample. - Ex: post-partum depression, suicide, addiction
Biasing Selection
= NOT random sampling [Ex: Telephone book is not the entire pop (far from it!) --> Researcher picks one name at random, then turns page, then selects another & keeps doing this until she has n] - Once one person is chosen from a given page, others on same page are biased against (0 probability of selection)
Nonprobability Sampling
= Not randomly selected: each member of a particular population does not have equal chance of being selected. - Population size is unknown; cannot list each member - Odds of selecting certain individual is unknown - Selection is biased; Greater risk of producing biased sample (Behavioural sciences)
Accessible Population
= Portion of target pop that are available to be recruited
Sampling
= Refers to the process of selecting participants for a research project - In order to decide who to recruit for your study (sample), you need to first determine who your population is. Q) Who do we want to generalize our results to? - Usually interested in findings that apply to an entire population (e.g., all university students), but we don't have access to an entire population
Cluster Sampling
= Shortcut that preserves most characteristics of random sample. - Randomly select groups rather than individuals. • Ex: Children's adjustment to daycare; randomly select among daycares in given area
Non-Probability Sampling: Snowball Sampling Pros & Cons
= Sometimes the only way to get in contact with a (hidden) population; = First participant has a strong impact on sample!
Law of Large Numbers
= The bigger the sample size, the more accurately it will represent the population - up to a certain point. - You don't need large numbers of participants to ensure representativeness: little benefit > 25-30/group - Ex: the effectiveness of Prozac in treating symptoms of depression in postpartum women - BUT a min of 10 participants per treatment condition is required for statistical purposes.
Non-Probability Sampling
= Used when the population is not completely known. - Most common in behavioural sciences. Types: 1. Convenience sampling 2. Purposive sampling 3. Quota sampling 4. Snowball sampling - Problem with these samples = no evidence that they are representative of the populations we're interested in (in many cases, they clearly are not)
Sampling Bias
= When the way we selected our participants favours the inclusion of certain people over others; no longer random - [ex: social media: younger; university clinics: more educated] - The method that we use to select our participants will affect the representativeness of our sample. Types: 1. Obvious 2. Subtle 3. More subtle
Systematic Sampling: Sampling Interval
= the size of the sample frame over the sample size. - i.e. you want 300 names out of 900, so 900/300 is 3, you will systematically count every third name until you reach 300.
Biased Sample
= when participants in your sample differ from your pop on a given characteristic - Can be the result of the manner in which participants were selected (selection or sampling bias). - [E.g., if individuals in our sample of women with postpartum depression are older or more educated than total pop with it]
Inferential Statistics
= when we infer population parameters from sample statistics - ex: we use the average heart-rate of the sample to *estimate* the average heart-rate of the pop
Population (N)
A group sharing some common characteristic(s); researcher is interested in phenomenon of entire group. = represents all the data of interest (can be narrow or general);
Sample (n)
A subset of this group; small set who participate in the study. = Anything less than all the data of interest (any subset of a population) = Individuals selected to be in your study (this # is much smaller than the target pop)
Non-Probability Sampling: Snowball Sampling
Begin with someone who meets the criteria for inclusion in your study (e.g., sex workers, those with addiction). - Then ask them to recommend others who they may know who also meet the criteria. Study ps recruit future ps from contacts. - Sample group groups like snowball until have enough data
Multistage Sampling: Example
Canadian university professors' opinions on student literacy. Stages: 1: Randomly select universities across Canada 2: Randomly select departments within universities 3: Randomly select professors within departments
Inferential Statistics Analogy
Engineer cannot make full size model of bridge; builds small scale & tests it: can infer about full sized bridge. - Larger the model, the more likely behaves in the same way.
Probability Sampling: Why is it Sometimes Unrealistic?
Especially in psychological research: 1. Too much time & effort, 2. May not be practical or possible, 3. Need a list of all members of population, 4. Representativeness is not guaranteed.
More Subtle Bias
Ex: Highly motivated subjects, large financial compensation for participation.
Obvious Bias
Ex: Low return rate of surveys
Subtle Bias
Ex: Researcher avoids higher floors of a building while doing a door to door survey.
Stratified Random Sampling: Example
If you wanted to make sure your sample had individuals from each income level, you could take 25 individuals with incomes above 100,000$, 25 with incomes between 40,000$-100,000$, and 25 with incomes below 40,000.
Principle of Independence
Once you select one person, the probability that you select another has changed (NE) - Example: Ps 5, 6, 7, 8 are biased against, Ps 9 is biased for
Non-Probability Sampling: Purposive Sampling
Purposely select participants with characteristics relevant to the research question; fit specific characteristics.
Cluster Sampling: Pros
Relatively quick & easy to get large sample; test in groups
Systematic Sampling: How to Figure out n
Steps: 1. Pop. size (N)/sample size (n) needed = n interval 500/50 = 10 2. Randomly pick a number from 1-10 3. Pick every 10th person on your list after that number
Sampling Interval
The size of the sample frame over the sample size.
Populations & Samples
We make inferences about populations based on info collected from a sample of that population - Ex: a whole pie = a population vs. a slice of that pie = a sample - We can't eat the whole pie... instead take a slice & generalize... (Flavour? Too sweet? Crust?)
Representativeness
You want to be sure that the people in your sample are a true representation of those in the population of interest. - The larger and more representative the sample, the more confidence we have that the results can be generalized to the population.
Stratified Sampling: Pros & Cons
• Ensures all subgroups are equally represented in your sample. • Useful when your goal is to describe and make comparisons among subgroups. • Does not adequately represent proportions found in pop.
Simple Random Sampling
• Entire population is represented. • Each individual has equal chance. • Each selection is random, independent. • Define population, list all members, use random process to select individuals. • Assignment of numbers to participant, random process to select #s. • Draw names out of a hat. • Use a random numbers table. • With or without replacement (more likely; small change in probability). • Is fair, unbiased, but can't guarantee it's representative!
Non-Probability Sampling: Quota Sampling
• Establishing quotas for the # ps in each subgroup • Ensure subgroups are equally represented (e.g., 50/50 men, women), but not randomly selected. • Still selected based on convenience (sample can be biased). • Can adjust quota to reflect proportions in population, but not randomly selected. • Identifies relevant categories of people (male/female, age groups, etc.); selects predetermined # of ps for each category - Ex: Pop = 60% men; 40% women; Sample = 60 men, 40 women [ - Ex: Of 32 adults & kids in the street scene, select 10 for the sample --> 4 adult males, 4 adult female, 1 male child, & 1 female child]
Non-Probability Sampling: Fixing Convenience Sampling
• Large sample with broad crosssection of individuals (e.g., diff sexes, ages, academic performance) & provide detailed description to let people draw inferences about generalizability
Multistage Sampling
• Random sampling at several stages • Start wide & narrow down at each level
Non-Probability Sampling: Convenience Sampling Pros & Cons
• Representativeness? • Bias? • Volunteer characteristics? • Easy / less costly / more timely
Non-Probability Sampling: Convenience Sampling
• Sample drawn from part of pop close by • Also called "accidental sampling" • Grab whomever you can: participants are easy to get based on availability & willingness; readily available & convenient • "Man on the street" type interviews • Setting up a booth in school cafeteria • No attempt to know population
Proportionate Stratified Sampling
• When researchers deliberately sample to ensure proportion of subgroups in sample is same as proportions found in population. • Determine correct proportions & randomly select from within that subgroup. - ex: pop = 10% men; 90% women --> sample = 10 men, 90 women - A lot of work, but guarantees the composition of the sample will be representative of composition of pop. - Problem: some subgroups may have limited representation & may not be able to say much about them.