Quantitative Research | Week 5

Quantitative Research is typically…

homogenous

Sampling Frame:

a specific list/database/source from which you will recruit your sample (ie: a clinic list, OHIP database)

Inclusion and Exclusion Criteria (who do we include/not)

• Consider socio-demographics → ie: sex, age.

• Clinical

• Geographic → location ppl live

• Consider ability to give consent

• Criteria set boundaries for your study and relate to the research question

Probability Sampling

• Based on probability theory

• Random selection of participants (simple, stratified, cluster, etc)

• Process that assures that the different units in your population have equal probability of being chosen 

• Manual ways of random selection and computer generated random selection

Non-Probability Sampling

• Sampling technique where the samples are gathered in a process that does not give all the individuals in the population equal changes of being selected 

• Quota, convenience, purposeful, snowball, self-selection/volunteer

Sample Size Calculations and Statistical Power

• Needs to be calculated at design stage ‘a priori’ → beginning

• Statistical formula

• Statistical power: measure of how likely the study is to produce statistically significant results for a difference between groups. Ie: true difference and not difference due to change

• Consider social and clinical significance

Sampling Error

Random variation in the sampling that occurs by chance

Sampling Bias

Systematic error that leads to a non-representative sample and thus misleading results by skewing the sample into a particular direction

Common Sampling Errors

• Population specific error → researchers fail to accurately define/identify the target population (ie: young, old)

• Non-response error → signi. Proportion of the selected sample doesn’t participate in the study or provide incomplete data; resulting in bias in results if they differ signi. to those who participated in the study/responded

• Sample frame error → when a list of database from which the sample is drawn does not accurately represent the entire population (ie: incomplete lists, under representation in one group)

• Selection error: occurs when the method used to choose participants from the sampling frame i biased or flawed (ie: conducting a survey in a limited geographical or timeframe)

Impact of Sampling Errors on Research Reliability

• Generalisability → lack of representative sample means the results might not hold true for the entire group (population) you are studying

• Statistical Significance → a high sampling error might lead to missed effects or claim significance when none exists

• Replication → Transparency in sampling error documentation allows other researchers to account for these and verify-build on in future work contributing positively to the cumulative nature of scientific knowledge

Simple Random Sampling

• Subset of a statistical population in which each member of the subset has an equal probability of being chosen 

• Unbiased representation of a group

• Most purest & straightforward prob. sampling method

• Requires large samples and acquiring realistic sampling frame

Stratified Random Sampling

• Dividing population into smaller groups (strata) based on shared characteristics of members in the group, selects small and equal samples from each strata

• Process of classifying the population into groups = stratification

• Best represents the entire population being studied → offers fairer representation of pop.

• Very time consuming → requires knowledge of total population from which the sample will be drawn

Cluster Sampling

• Divide population into smaller groups (clusters), then takes a random sample from each clusters 

• Method of probability sampling

• Used w/large populations particularly with wide geographical coverage

• Requires less resources than simple or stratified

• Higher risk of bias

• Higher rates of sampling error

• Reduces risk of contamination

Systematic Sampling

• Starting point may be random, sampling involves fixed intervals between each member

• Easy + cheap 

• Need to know parameter of entire population

• Creates fractional rather than equal chance of selection

Sampling Issues to Consider

• Population specification

• Random does not mean accidental

• Sampling frame and sampling plan

• Non-response bias

• Too many hypotheses for too small sample

Most Common Sample Unit in Healthcare

Most likely human…can by documents, existing data sets (medical records), digital communication, or published reports (think systematic review)