Week 6: Sampling

Sampling plan is...

- key for credibility and generalizability

- the better the sample represents a population, the more valid the results

population

all the people in a particular group

sample

- a subset of the people in a group that we select for a study

- should reflect the composition of the population selected

Hierarchy of populations

- population

- sampling frame (accessible population)

- targeted sample

- actual sample

Sampling error

- differences in the sample compared to the population

- cannot be completely avoided

selection bias

- sample is selected in a way that is not impartial or participants are assigned to groups in a way that introduces bias

- types: under or over representation of a characteristic in the population, homogeneity, low response rates or increased attrition

ways to reduce selection bias

- understanding the population first

- considering characteristics that can influence results

- increasing variability in characteristics or heterogeneity

- increasing the number of participants

- analyzing characteristics of who decides not to participate and who decides to leave the study before completion

inclusion criteria

- "who are we looking for"

- characteristics: clinical, demographic, geographic, temporal criteria

exclusion criteria

- "what characteristics might influence the outcome in ways we don't want"

- "what are rival explanations"

- characteristics: comorbid conditions, language

- must have a good reason

What are types of random sampling

- simple

- systematic

- stratified

- cluster

simple random sampling

- sampling frame includes entire population - hardest to do

- must have list of the entire population

systematic random sampling

- not a true probability sample, used for prospective groups

- first participant chosen randomly then every 10th afterwards

stratified random sampling

- reduces the risk of under representation of a subgroup

- must divide the population into groups first based on characteristic like gender or ethnicity

- then randomly selects a number from each group

cluster random sampling

- randomly selecting groups first, then randomly selecting individuals, chosen for feasibility

- Ex: hospital units or magnet facilities

random sampling is...

not the same as randomizing to treatment conditions

independence

- necessary for a true probability sample

- the selection of one subject does not influence the selection of another subject

- violated if selected subjects are related to another in any way or if more than one score is collected from the same subjects

- example: pre-test post-test designs are not considered independent samples

- systematic sampling are not independent samples

What are types of non-probability sampling

- convenience sampling

- snowball sampling

convenience sampling

- the most common type

- subjects who are accessible to the researcher, much more feasible to sue

- can introduce sampling bias, especially with more involvement by the researcher

convenience sampling example

Nursing students in New England or Nursing students at Saint Anselm college

snowball sampling

- referral sampling

- "word of mouth"

- helpful when getting participants who may not otherwise participate

- can then randomly select from referrals or increase sample size to reduce bias

What are hard to reach samples

- vulnerable or marginalized populations with reduced trust in healthcare system

- shame about topic being researched

What are strategies to reach the vulnerable populations

- where and who (have an insider do the recruiting and face to face study measures at a location they are more likely to feel comfortable)

- stratified random sampling, snowball sampling

- incentives (but always consider ethics!)

- service based sampling

size of quantitative studies

- size matters

- statistical power

- if effect size isn't known, estimate by 50 x each independent variable then add 8

- less than 30 participants is unlikely to be enough to generalize findings

- insufficient size increases the risk for a type 2 error

statistical power

- need enough subjects to detect a difference in the outcome variable

- calculation: effect size, significance level, number of variables to be measured, independence

type 2 error

there was a significant difference, but it was not detectable

effect size

- the magnitude of the difference between groups

- cohen's effect sizes "d" = small (0.2) medium (0.5) large (0.8) d = m1-m2/s

- "odds ratio"

- "relative risk"

- the larger the effect size the smaller the sample needs to be to find statistical significance

external validity

the ability to generalize findings from a study to other populations, places, and situations

what are the two types of external validity

- ecological

- population

ecological external validity

how similar is the study setting to where we would like to apply it

population external validity

how similar was the study sample to the population in which we want to apply it

surveys

- response rate will greatly affect your sample and can increase sampling bias

- total rate of distributed surveys/returned surveys

- how many surveys were completed? What to do about missing data?

- nonresponse error

nonresponse error

when respondents differ in characteristics than the population from which they are drawn

How to create a sampling plan

1. define the population

2. create inclusion criteria

3. create exclusion criteria

4. design a recruitment plan

5. determine the number of subjects needed

6. apply the selection methodology

7. implement strategies to maximize retention and increase response rate