7.1: Systematic Error- Selection Bias & More ✅ Flashcards
Random error
Error introduced solely by chance
Is inherent in the sampling process
Systematic error
Bias
Introduced by man-made actions relating to the conduct of the study
Error in epidemiological studies
Random bias DECREASES with increase in sample size
If total population is used, random bias = 0
95% confidence interval becomes narrower increases with sample size
How is systematic error affected by sample size?
Bias always remains the same, regardless of sample size
Selection bias
Systematic error
Resulting from participants used not being representative of the source population
->leads to a bias sample, which gives rise to bias estimates
What greatly affects selection bias?
The sample method
Sampling methods
Random sampling
Systematic sampling
Non-probability sampling
Random sampling
(Also known as Probability sampling)
Sample selected by probabilistic methods
Allows strong statistical interference about the whole group
Probability sampling types
Simple random sampling
Stratified random sampling
Cluster sampling
Multi-stage sampling
Systematic sampling types
Simple systematic sampling
Proportional quota sampling
Systematic sampling
Sample selected according to simple, systematic rules
Non-probability sampling
Sample selected by convenience
Involved non-random selection based on convenience
Simple random sampling: overview
Most straight-forward
All individuals in the sampling frame have the same probability of being selected, independently of all others
Given a larger sample size, ensures individuals are representative of source population
When is SRS mostly used?
In quantitative research
SRS steps
- Identify source population
- Set up sampling frame
- Decide on sample size
- Randomly select individuals from sampling size
SRS pros
Representative sample (if sample size is large enough)
Less costly and less time-consuming
Ideal for quantitative studies and test of hypothesis
SRS cons
May be impractical if sampling frame is too large or pop is geographically diverse
If a large sample is used, could be time consuming or costly
Stratified random sampling: overview
Same as SRS but within strata of the population
Size of the random sample should be proportional to the specific stratum size in the population
Stratified random sampling: steps
- Indemnify source population
- Set up sampling frame
- Decide on sample size
- Decide on pre-defined population strata
- Based on overall proportions of the population, calculate how many people should be sampled from each subgroup
- Randomly select individuals to fill strata
Stratified random sampling pros
Allows for more precise conclusions by ensuring every subgroup is properly represented in the sample
Allows comparison of population sub-groups
Stratified random sampling cons
Time-consuming
Higher complexity might give rise to errors
Cluster sampling overview
Based on natural clusters of individuals within the population
e.g. hospitals, schools, streets, city districts etc…
Involves taking a random sample from these clusters
Sampling frame is a list of all clusters
If the clusters are large, one of the above techniques may be used to choose sample (SRS or stratified random sampling)
Difference between stratified and cluster
Stratified takes individuals from groups
Cluster takes the whole group
Cluster sampling steps
- Identify source population
- Set up sampling frame (compromised of clusters)
- Decide on sample size (number of clusters and individuals)
- Randomly select clusters form sampling frame
Cluster pros
Good for large and diverse pops
Less costly and less time consuming
Cluster cons
Substantial differences between clusters can cause errors
Difficult to ensure the sampled clusters really represent the whole pop
Representativeness may be compromised if:
-too few clusters selected
-clusters are too specific
-cluster contain too few individuals
Multi-stage sampling
Uses structure of natural clusters of individuals within the population
After randomly selecting clusters, there is a random selection of individuals within the cluster
Multi-stage sampling
- Random selection of large clusters
- Random selection of smaller clusters within large clusters
3.Random selection of individuals within smaller clusters
Multi-stage sampling pros
May improve sample representativeness
Less costly and less time consuming
Multi-stage cons
Representativeness may be compromised if:
-too few clusters are selected
-clusters are too specific
-clusters contain too few individuals
Systematic sampling
Sample selected by by simple systematic rule
Could be equivalent to simple random sample if there is no biasing pattern in selection process
Systematic sampling steps
- Identify source population
- Set up sampling frame
- Decide on sample size
- Systematically select individuals from sampling frame
Systematic sampling pros
More convenient alternative approach if random sampling isn’t possible
Faster and potentially cheaper
Proportional quota sampling
Same principal as stratified random sampling
Strata filled by non-random sampling
Proportional quota sampling
- Indemnify source population
- Set up sampling frame
- Decide on sample size
- Decide on pre-defined population strata
- Select individuals to fill strata non-randomly
Proportional quota sampling: steps
- Identify source population
- Set up sampling frame
- Decide on the sample size
- Decide on pre-defined population strata
- Select individuals to fill strata (non-randomly)
Proportional quota sampling pros
Acceptable convenient method if random sampling is not possible
Compared to systematic, could ensure original population structure (as it uses predefined population) strata
Proportional quota sampling cons
The representativeness may be compromised
-as individuals aren’t selected randomly as individuals are not selected randomly
Convenience sampling
Most frequent non-probability sampling
Based on convenience
Convenience samplings steps
- Identify source population
- Decide on sample
- Conveniently select individuals
Examples of non-probabilistic sampling methods
Convenience
Purposive
Voluntary response
Snowball
Advantages of convenience sampling
Cheap
Fast
Convenient (duh)
Cons of convenience sampling
Representativeness of the sample will DEFINITELY be compromised
->as individuals are selected in non-random fashion
How to decide which sampling method should be used?
Depends on
-aim of study
-nature of source population
-sample size
-other practical issues
Which method is best for small samples?
Stratified random sampling
Which method(s) is best to minimise selection bias?
Random sampling techniques
What do we always assume when using non-random sampling?
Selection bias is operating to some extent
Descriptive research
Prevalence of a disease in a population
Important to have perfectly representative sample as selection bias will greatly influence findings
Analytic research
Investigating exposure-outcome association
Minor selection bias may not affect findings to a large extent
What method should always be avoided?
Convenience sampling