Sampling Design Lecture Dr Wofford Flashcards
Sampling Bias
- •Sampling bias: occurs when individuals in a sample either underrepresent or overrepresent characteristics which are related to the phenomenon of the study
- •Can be conscious or unconscious
- •Conscious- occurs when a sample is selected for a reason which may cause bias
- •ie: select subjects who possess characteristics which make them more likely to improve with an intervention
- •Unconscious- researcher does not realize that he/she is selecting subjects which may cause a sampling bias
- •Choosing subjects based on appearance when appearance is not a criteria for the sample
- •Conscious- occurs when a sample is selected for a reason which may cause bias
- •Can be conscious or unconscious
Disproportional Sampling
A type of probability sampling
- •Select samples of adequate size from each category
- •Used when you have disproportionate sample, but still want to have an adequate sample
- •ie: accessible population has more males than females, so you take more females than males so that you have equal numbers in the sample
- •Not random since every person does not have equal probability of being selected (females have a greater chance of being selected than males)
-
•Disadvantage: may be problematic for data analysis since certain groups or characteristics will be overrepresented in the sample that is not characteristic of the population
- •Can be corrected with weighting the data
This is not random at all
Disadvantage is the MAIN THING TO KNOW, and There is a statistical correction you can do called Weighting the Data to make it work better.
We do not need to know how to weight the data yet.
Quota Sampling
nonprobability sampling technique with the least error
- •Proportionately represents each segment of the population in the sample
- •Sample adequate number of subjects in each stratum
- •Still has potential bias, but is more representative of each characteristic of the population
Population vs sample
- Population: A group of people, objects, or events that satisfy a specific set of criteria.
- Sample: A representative group from the population in order to draw conclusions about the population
Cluster Sampling
Lowest type of Probability Sampling
- •Used with large populations
- •Also known as multistage sampling
- •ie: random sample of students in one city
- •1st stage: randomly choose 10 schools
- •2nd stage: randomly choose 1 class from each school
- •3rd stage: randomly choose 4 students from each class
- •Advantage: convenience and efficiency when dealing with larger populations
- •Disadvantage: increased sampling error
Multi-stage sampling technique
More chance for sampling bias than other techniques we have gone over so far
Simple Random Sampling
The best type of Probability Sampling
- •Each selection is independent and no potential subject has any more chance of being chosen for the sample than the other potential subjects
- •Drawn from the accessible population
- •Once selected for a sample, no further chance of being selected
- •Also known as sampling without replacement
- •Methods?
- Random number generator can be used to help pick people. We can also do random group assignment.
- •Advantages?
- good representation. Sample error low
- •Disadvantages?
- Difficult to access everyone you need
- Very time intensive
- Can potentially be costly
Stratified Random Sampling
A type of Probability Sampling
(a multi-stage process, divided by relavant population characteristics)
- •Sampling from a series of strata in a sample
- •1st step: identify relevant population characteristics
- •2nd step: divide subjects into homogenous groups, called strata
- •3rd step: draw a random sample from each strata
- •Proportional stratified sample
- •Used when the formed strata are not proportional
- •ie: unequal numbers of 1st, 2nd, 3rd year students
- •Take a proportion of each strata which is relative to it’s proportion in the population
- •Still uses random sampling from each strata
- •Used when the formed strata are not proportional
- •Advantages?
- you can make it a bit more representative of the population. Make sure the groups are a bit more neutral on that defining characteristic.
Sampling from a series of strata: When the characteristic can effect the outcome, we want to have equal numbers of people with those characteristics.
Proportional stratified sample is good to use when there is a big proportion difference.
Five types of Probability Sampling (in order of excellence?)
- •Simple Random Sampling
- •Systematic Sampling
- •Stratified Random Sampling
- •Disproportional Sampling
- •Cluster Sampling
Sampling Error
•Sampling error= difference between sample averages (statistics) and population averages (parameters)
- •Random sampling says that these differences are due to chance and not bias
Purposive Sampling
the form of nonprobability sampling with the most error
- •Researcher handpicks subjects on the basis of specific criteria
- •Similar to convenience, except that specific choices are made, rather than by availability alone
- •Continues to have potential bias, but may be more representative of the spectrum of population characteristics
- •Commonly used with qualitative studies
- •ie: researcher only samples patients in the clinic who he/she feels will be compliant with an exercise program
Two main Categories of Sampling Techniques and their charecteristics
•Probability sampling: occurs through random selection
- •Every person with the specified characteristics has equal chance of being selected
- •Should be free from bias and representative of the population
- •Provides greatest confidence in the validity of the sample
•Nonprobability sampling
- •Probability of being selected is unknown
- •Subject to bias
Equal chance is a big thing for probability sampling. Sample should be relatively free from sampling bias and representative of the population.
Two big types of validity:
Internal: what we have talked about up to this (does the study measure what it is supposed to measure)
External: generalizability
Four Types of Nonprobability Sampling
(from least error to most error)
- Quota sampling
- Convenience samplingor
- Snowball sampling
- Purposive sampling
We don’t know the probability so there is more potential for sampling bias
Snowball Sampling
One of the two nonprobability sampling techniques that has the second least amount of error.
- •Most useful when the population is rare, unevenly distributed, or hard to reach
- •Performed in stages:
- •1st stage= a small sample of subjects who meet the inclusion/exclusion criteria are identified and tested
- •2nd stage= the tested sample are asked to identify others who have the requisite characteristics
- Advantage: easier to reach hard to reach people that qualify for the study
- Disadvantage: pretty large potential for bias (subjects tend to find more people like them)
*
Convenience Sampling
One of the two second best nonprobability sampling techniques
- •Most common form of nonprobability sampling
- •Subjects are chosen on basis of availability
- •Consecutive sampling: recruit all subjects who meet the inclusion/exclusion criteria as they become available
- •Advantage?
- Much more convenient
- •Disadvantage: potential bias of self-selection (threat to external validity)
- •Are those who volunteer for the study representative of the population?
What most people in our profession does. This method is EXTREMELY COMMON. It is okay to say we used this method during our thesis defense, but we must be able to recognize problems and answer questions about it.
Self-selection is a threat to External Validity
VERY IMPORTANT FOR US We should be able to map out all of the limitations of our research design from all of our methods before we do the study. We will eventually acknowledge/address the limitations in the discussion section. Be able to map out the strengths and weaknesses of our methods designs before we start data collection. THIS HAS NOT BEEN DONE MUCH BEFORE in the PT program. SHE WANTS OUR CLASS TO START THE TREND.
Example of convenience sample: Whoever responds to the advertisement
Three levels of the sampling process
- Target population: Universe of interest (who you want it to be generalizable)
- Accessible population: Population who is accessible from which to draw the sample. Should be similar to the target population.
- Sample: is only accurately representative of the accessible population (parameters of sample are set by inclusion and exclusion criteria)
- Inclusion criteria
- Exclusion criteria
- Sample: is only accurately representative of the accessible population (parameters of sample are set by inclusion and exclusion criteria)
- Accessible population: Population who is accessible from which to draw the sample. Should be similar to the target population.
