sampling method Flashcards
Cohort studies
Also called longitudinal or epidemiological studies, Cohort studies follow a large group of people over an extended period of time to see how their exposures affect their outcomes. This type of study is normally used to look at the effect of suspected risk factors that cannot be controlled experimentally – for example, the effect of smoking on lung cancer. These studies are frequently used to determine long term effects of a lifestyle, diet, or other interventions. Cohort studies may include a second group that did not engage in the same intervention as a control comparison. Although these studies are a step up in reliability and generalizability, they can be difficult to blind, can’t be controlled for outside variables, and are usually not randomized.
The Randomized Control Trial (RCT)
Now we have reached a major point in the Pyramid – the Randomized Control Trial, the true experimental design. In this study design, individuals are assigned by special randomization techniques into two or more groups, where one group receives the intervention under investigation and the other(s) receives no treatment, a placebo, or a standard intervention.
The systematic reviews
Now we have finally arrived at the pinnacle of the Evidence-Based Medicine Pyramid.
From this dizzying height, we take in a panoramic view of all of the evidence about an intervention. Systematic reviews take a bird’s eye view by comparing the results of studies side by side, typically on a forest plot. Systematic reviews are considered the strongest and highest quality of evidence.
We could also include meta-analysis here as well. This is where multiple studies are reviewed and a statistical summary is made that represents the effect of the intervention across multiple studies.
The Cochrane Collaboration takes systematic reviews to the next level. They are the experts of the systematic review and have an added a level of rigor as an independent voice, as well as developing special techniques to identify bias in studies.
Case control studies or case series reports
This level represents the first stage of testing an observation. Case Series reports usually include only a few participants who are given a similar intervention and follow-up. Case Control Studies are similar to case series, except it looks retrospectively at individuals and compares with a similar group who did not have the intervention. These studies are conducted in the early stages of research to help identify variables that might predict a condition. One of the weaknesses in these designs is that there are small numbers of participants and they are frequently not randomized or controlled for confounding variables.
SAMPLING METHODS AND EXTERNAL VALIDITY
impossible or extremely costly to study complete populations
therefore collect data on sample of cases
ie. Study subset (small group) and generalize to population
Examine ways of sampling to allow generalization
advantages and disadvantages of different methods
relationship between sampling error and sample size
“external validity” – generalizing findings to population or other settings (related to induction)
WHAT IS SAMPLED IN A STUDY
Select subjects/participants, but also: info to be collected procedures to collect info research location clinicians and researchers involved
BASIC ISSUES IN SAMPLING
Not only beyond resources to study whole population, but often would be wasteful
if sample representative can generalize to population
Researcher defines population to which wishes to generalize
sample = subset
eg. % of people in Sask who take nutritional supplements?
population = people in Sask
sample = 500 people from different cities/regions
eg. amount of resveratrol in different wines?
population = all wines
sample= 5 from each variety, red, white, rose
Representative Samples
If sample is representative can confidently generalize to rest of population
If sample biased (not representative), generalize less validly and may lead to incorrect conclusions
eg. population ○○○○●●□□
representative sample ○○●□ = precise miniaturized representation
unrepresentative sample ●● = biased/incorrect proportion
Selection of sampling method depends on aims and resources
eg. if designing expensive dietary program based on survey of client’s needs want good representative sample
but good sampling methods more expensive and difficult to implement
Scientific and clinical research – generally use incidental or random sampling
Incidental Samples
incidental sampling
quota sampling
Incidental sampling – cheapest, easiest
selection of most readily accessible members of target population
eg. stand in middle of street and ask people about food choices
Quota sampling – if know in advance have important subgroups
eg. males/females – know general population has proportion 49:51
stand in street and sample 49 males and 51 females
Quota sampling (cont.)
need to know which popula tion groups important to a question, and proportion of different groups in the population
ex. male, femal, white collar, blue collar, unemployed
Random and Systematic Samples
Random sampling – one of best but difficult method
construct list of all members of population
use method eg. dice, coins, random number method to select randomly from list
all members of population have equal chance of selection
eg. lottery (for those who bought tickets)
computer can do efficiently, even with large populations
if need to select replacements (eg. refusal to participate, mortality) may introduce bias
eg. survey on food choices – those who respond may be different than those who don’t
d
Stratified random sampling
like quota sampling but with random sampling for each subgroup
eg. prepare list of all males and female in a population and sample randomly in each list to the correct proportion (eg. NUTR 305 class)
Advantages
Stratified random sampling
important groups proportionately represented
particularly important with key subgroups in low proportion
know exact representativeness – important for statistics
Disadvantages
need to list all of a population, characteristic(s), and proportion
cost
gain in accuracy small relative to simple random sampling
Area sampling
sample on basis of location with known characteristics
eg. areas in a city with low or high unemployment
divide areas into streets and contact occupants of eg. every 3rd house
ie. random selection within location and don’t need list of individual members
Effective, cheap method for social surveys
Systematic sampling
choose every eg. 10th case from a list of the population
ie. not truly random but usually gives representative sample
based on supposition that the list is not made in a systematic way that coincides with the sampling system
easy, convenient
eg. clinical practice – measure temperature and blood pressure every hour
SAMPLE SIZE
Optimal numbers of cases to include in a sample (often poorly understood)
Balance number with costs of data collection (or danger/pain to patients or lab animals)
Are some principles to guide researcher
in general optimal sample size is one adequate to make correct inferences to population
Sampling Error
Discrepancy between the true population parameter (eg. average age) and the sample statistic (eg. average age of sample)
eg. average age in a district = 35 years
sample average age = 30 years
sampling error = 5 years
Relationship: sampling error 1/n
ie. greater the n, smaller the sampling error (in fact to square root of n)
eg. 9-fold n only 3-fold sampling error
Therefore benefit/cost less at larger sample sizes
Selection of n depends on research situation
eg. election poll, sampling error estimated 10% with 100 people
if 25% A, 75% B – error small relative to effect (difference) – predict B in landslide
no need (too costly) to poll more people
if 48% A, 52% B – error too large relative to effect to predict outcome
need to poll more people (increasing cost)
Sometimes use pilot study to estimate size of effect and thereby estimate sample size needed to show it statistically
EXTERNAL VALIDITY AND SAMPLING
External validity – extent results can be generalized to other samples or situations
two types population and ecological
Population validity
Population validity – generalization to sample’s population
depends how representative sample
may not always have access to whole target population
eg. develop diet to control postnatal depression (for all women having babies)
test at local maternity ward (accessible sample)
lacks some population validity if sample different in different wards
Ecological validity
Ecological validity – extent results can be generalized to other situation
eg. coronary arteriography (insert catheter to inject dye for visualization)
mortality from procedure initially reported 0.1% (in first- class medical institution)
mortality found as high as 8% in other institutions
eg. find flavonoid can protect cultured cardiomyocytes from oxidative damage
may not protect heart from ischemia-reperfusion injury in vivo
-need caution to generalize to other situations
