Ch. 12 - Examining Sampling Plans
Terms
undefined, object
copy deck
- Sampling
- - selecting a portion of the sample to represent the population of interest
- Sample
-
- a subset of the population
- made up of elements (ex. humans, things)
- more practical to work with a sample than a population
- usually possible to obtain good information from a sample
- in quantitative studies the 2 main sampling concerns are:
1. Representativeness
2. Size
- Otherwise, conclusions can be erroneous
- there is never a guarantee of a representative sample (some sample plans are more likely to produce representativeness than are others. Which ones?
1. Sampling Bias
2. Strata - Sampling Bias:
-
- is the product of a sample that is not representative of the population
- over-sampling or under-sampling of a characteristic that is relevant to the research - Strata:
-
- Subpopulations of a population
- Mutually exclusive segments
- ex. break it down into males & females - In what segment of a research report is the sample usually discussed?
- Methods
- Sampling Designs (QUANTITATIVE):
-
- developed in the planning phase of the study before the data collection
- goal is to achieve statistical conclusion validity and be able to GENERALIZE findings to the population of interest
- 2 major categories:
1. Non-probability
2. Probability (would use randomization) - Non-probability Sampling:
-
- elements NOT randomly selected
- every element does not have a chance of inclusion in the sample
- rarely representative of the target population
- need to be cautious about conlusions - 3 Methods of Non-Probability Sampling (similar to purposive sampling):
-
1. Convenience
(convenience of who is accessible - ex. first 10 people who walk in the cafeteria or a nursing class; risk for bias; most widely used because it is easy to do)
- accidental or chance sampling
- using the most conveniently accessible people or things
- available elements may not be typical of the population
- SNOWBALL (chain or nominated) sampling: is a type of convenient sampling, whereby subjects already in the study identify other potential subjects
- "The price of convenience is the risk of bias"
- weakest form of sampling. why
- most widely used form of sampling. why?
- Useful for new areas of research, when not much is known about the topic or when a more robust sampling method would conflict with ethics
-
3 Methods of Non-Probability Sampling:
-
1. Convenience
2. Quota Sampling
3. Purposive Sampling - Quota Sampling
-
- ex. Males vs. Females or Black vs. White
- participants are chosen from STRATA (subpopulations within the population of possible participants)
- diverse characteristics can be sampled
- still considered over straight convenience sampling - Purposive Sampling:
-
- researchers use their knowledge & judgment to hand pick people with particular characteristics; open to bias
- Subjective sampling
- can be useful when researchers want only particular characteristics (ex. expertise in a particular area) - PROBABILITY SAMPLING
-
- everyone has equal chance of getting selected; more time consuming
- elements are RANDOMLY selected from the population of interest
- do not confuse with RANDOM ASSIGNMENT within an experimental design
- all elements in a population have an equal chance of being selected
- more confidence in representativeness - eliminates conscious or unconscious bias
- labour intensive and time consuming process
- may be difficult to obtain a complete list of elements - 4 Types of Probability Sampling:
-
1. Simple Random Sampling
2. Stratified Random Sampling
3. Cluster Sampling
4. Systematic Sampling - Simple Random Sampling
-
- most basic design
- basis of more complex designs
- first step is to establish a sampling frame - list of all population elements
- elements are numbered consecutively
- a table of random numbers or a computer program is used to randomly select a sample
- NOT subject to researcher bias
- no guarantee of representativeness
- any differences between the sample and population are a function of chance
- chance of selecting a mismatched sample is LOW
- the larger the sample size, the more likely it will be representative - Stratified Random Sampling:
-
- elements are RANDOMLY selected from strata in the population
- population is divided into HOMOGENOUS subgroups
- may sample:
1. Proportionately - elements are sampled in proportion to representativeness in the population [10% of the population size of 100 males and 10,000 females - ex. 10 males and 1000 females]
2. Disproportionately - elements are sampled disproportionately to their representativeness in the population to increase the number of elements in small strata
- intended to increase sample representativeness
- may be difficult to do if information is not available on the specific characteristics that ought to be stratified - Cluster Sampling:
-
- used when it is not possible to obtain a list of all elements in the population
- used in large scale surveys (ex. national)
- involves successive random sampling of units within a population
- the first unit sampled is the LARGEST unit (ex. provinces or regions within a country), then next largest (ex. cities), the next largest unit (ex. households), etc.
- can contain more SAMPLING ERROR than simple and stratified random sampling
- more practical and economical when population is large - Systematic Sampling
-
- random but more systematic
- selection of elements at an established interval (ex. Kth number)
- similar to simple random sampling
- first step is to divide the population by the desired sample size to obtain the sampling interval - the distance between the selected elements)
- ex. a required sample of 100 from 100,000 in the population mean every 100 case would be sampled
- next step is to randomly choose the first number (ex. 256 - start 256 and select every 100th element thereafter until 100 elements are chosen) - Sampling Error:
-
- the difference between the POPULATION value on a particular variable and the SAMPLE VALUE
- probability sampling allows for estimates of the degree of error - Sample Size (QUANTITATIVE)
-
- the number of elements in a sample
- should use the LARGEST sample possible (and ethically appropriate)
- the larger the sample, the closer it will be to the population characteristics and the smaller the sampling error - the less likely it will deviate from the population
- sample size should be based on a POWER ANALYSIS: an estimate of the size of the difference on outcomes between the treatment and control groups
- estimate can be based on findings from previous research, personal clinical experience or on a pilot study of the variables
- when expected differences are large, sample size does not need to be large; when expected differences are small, a large sample is required
- for new areas of research, it is best to estimate a small difference and therefore use a large sample
- when sample size is too small, the research hypothesis may be ERRONEOUSLY rejected
- when critiquing quantitative research, need to assess both sample selection method and sample size (and basis for the size) - CRITIQUING QUANTITATIVE SAMPLING PLANS:
-
- quality of the sampling strategy
- sample size
- RESPONSE RATE: number of people participating vs. # of people SAMPLED
- RESPONSE BIAS or NON-RESPONSE BIAS: differences between those who participated and those who didn't (ex. study on teaching methods to find out how good it is - random sample - everyone gets a survey; gave out 50 but only got 25 back; 25 handed in were wonderful but the other 25 tore it up and didn't like the teaching)