Sampling Methods in Research PDF

Sampling -​ Target population: the general population that the study seeks to understand -​ Source population/sampling frame: the specific individuals from which a representative sample will be drawn from -​ Sample: individuals asked to participate -​ Study population: eligible pa...

Sampling -​ Target population: the general population that the study seeks to understand -​ Source population/sampling frame: the specific individuals from which a representative sample will be drawn from -​ Sample: individuals asked to participate -​ Study population: eligible participants Probability Sampling -​ Techniques wherein the probability of selecting each sampling unit is known -​ Equal likelihood of being invited to participate -​ Prob sampled = n(sample)/N (sampling frame) -​ Simple random sampling -​ No selection bias -​ Randomly chosen -​ Stratified random sampling -​ Random sampling from distinct groups and simple randoms are selected from those strata -​ Geography, sex -​ May result people from larger demographics to be represented more -​ Cluster sampling -​ Natural clusters rather than individual units are selected -​ Need to be from same categories -​ Randomly select clusters and all people from chosen clusters are interviewed as opposed to choosing people randomly -​ Multistage sampling -​ Primary, secondary, tertiary sampling units -​ Cities → only some cities → random selecting neighbourhoods → houses → people Nonprobability sampling -​ Convenience sampling: selection from a non probability based source population due to the ease of access to those individuals, schools, workplaces, organizations and communities -​ Not randomly selected -​ Selected based on needs and wants -​ But can be systematically different from the target population they are supposed to represent -​ Purposive sampling: recruitment of participants of a qualitative study because of the special insights they provide -​ Key informants: people selected to participate because they have expertise relevant to the study question Errors with sampling -​ Does the sample actually represent the target population, external validity and sampling bias (end up missing some people and have a lack of representation) -​ No matter who was selected the goal is to get a sample population that is representative of the source population and ideally the target population -​ Source population small = everyone should be invited -​ Source population too large = need smaller size  -​ Bias: a systematic problem in the design, conduct, analysis of a study that can cause the results to be erroneous -​ Sampling bias: when the individuals sampled for a study systematically are not representative of the source population as a whole -​ Nonrandom sampling bias: each individual in the source population does not have an equal chance of being selected for the source population -​ Berkson's bias: when cases and controls for a study are recruited from hospitals and therefore are more likely than the general population to have comorbid conditions -​ Healthy worker bias: participants recruited from occupational population and therefore are systematically healthier than general population -​ Exclusion bias: different eligibility bias are applied to cases and controls (when control with health condition related to an exposure are excluded but cases with those comorbidities are not excluded) -​ Selection bias: members of the target population are not representative of the source population from which they are drawn -​ Nonresponse bias: if members of a sample population who agree to participate in study are systematically different from non participants Ethics with sampling in vulnerable population -​ Definition: those who might have limited ability to make autonomous decision about volunteering to participate in a research study -​ Young children -​ People with health issue -​ People in prison -​ Socially marginalized populations -​ Limited ability to make an independent decision about volunteering because of other reasons -​ Should not be selected unless needed -​ But at the same time, exclusion is unethical -​ Their health issues should be studied -​ Extra consideration of the potential risks of research to participants Importace of sample size -​ Need right amount of people -​ Too many = wastes resource -​ Too less = study is invalid -​ Sample size: the number of observations in data se, # of individual humans in the study population -​ The desired sample size for quantitative study is based on statistical estimations about how many data are required in order to answer the study question with a specified level of certainty -​ The effect estimate of exposure on the outcome: need smaller sample for larger effect -​ The amount of errors acceptable: smaller errors need larger sample -​ Variation of the outcome in the population: complex, largest when prevalence is 50% -​ Confidence interval (CI): statistical estimate of the range of likely values of a parameter in a source population based on the value of that statistic in a study population -​ Narrow CI - more certainty about the value of the statistic than a wide CI  -​ However when sample size is small the sample mean might be far from the actual mean of the population -​ When a greater proportion of the total population is sampled for inclusion in the estimation of the mean age, teh CI for mean age will be narrow because there is greater certainty about the sample mean being close to the population mean -​ Larger sample sizes result in sample means that are closer to the population mean Type 1 (a) and 2 (B) errors -​ Null means when we hypothesize there is no outcome/association -​ Left column: research about difference with the study population, top row: actual difference in source population -​ Want null+null or alt+alt -​ Type 1 error (false positive): study population yields a statistically significant test result even though a significant difference/association does not actually exist in population -​ Type 2 error (false negative): when statistical test of data find no significant result even though a significant difference/association does actually exist in the source population -​ Power: is the ability of a test to detect significant difference in a population when difference really do exist, detects a true relationship -​ 1-B is referred as the power of a study -​ 20% likelihood of 1-B error, the power is then 80% (should be 80 or greater) -​ Power increase with sample size -​ Too few people lack power -​ When incidence rate ratio or odds ratio has point estimate close to 1, a very large sample size will be required to yield a CI that is statistically significant and does not overlap Data collection -​ Often by interviewing participants by questionnaires or survey instruments -​ Direct measurement of physical functioning -​ Blood work -​ Sampling tissues -​ Imaging -​ Questionnaire/survey instrument: a series of questions used as a tool for systematically gathering data from study participants 1.​ List topics that the survey instrument must cover 2.​ First set of questions are used to enable the researcher to confirm that participants meet the eligibility criteria for the study, exclude ineligible people 3.​ Several question may be required in order to accurately assign participants to key exposure and disease categories -​ Self administered survey: questionnaire form that participants complete by themselves using either paper or online version -​ Pros: cost and time efficient, can approach a large number of people, can get more honest answers to sensitive questions  -​ Cons: problematic for low literacy populations, those who have limited internet access and unfamiliar with computers -​ Interview: process of verbally asking participant questions and recording their responses -​ Pros: can train interviewers to ensure accuracy and completeness of each questionnaire, record nonverbal communication -​ Cons: major time commitment, expensive -​ Semi-structured interview: start with list of open ended questions, acts as a starting point for eliciting response from participants -​ Probing: interviewing technique the prompts interview to provide a more complete and specific response -​ Training interviewers -​ Standard protocols to make sure the interview process is the same for all participants -​ Need to undergo role specific training and have an opportunity to practice their interview skills, pilot testing, quality control to avoid interview bias -​ Interview bias: when interviewers systematically question cases and controls or exposed or not exposed member of a study population differently, such as only probing some individuals because they believe to have the most information -​ Close ended questions: limited number of response -​ Date and time -​ Numerica questions -​ Paired comparison -​ Categorical -​ Ranked categorical (ordinal) answers have inherent order -​ Unordered categorical (nominal) answers do not have the built order -​ Open ended questions: free response questions, allows unlimited number of possible response Practical considerations 1)​ Order of questions -​ Start with easy or less general questions before moving to more difficult or sensitive questions -​ Best to group similar questions with similar response types, so that they are asked consecutively 2)​ Getting the correct answer -​ Prevent habituation: participants become too accustomed to giving the same response, and they continue to do so even if it doesn’t match their true perspectives 3)​ Data recording methods -​ Option 1: record response on paper and enter into computer later -​ Better for low resource environments -​ Allow for quality control/filtration -​ time consuming -​ Costly -​ Option 2: have interviewees or participants enter response directly into database -​ Eliminates need for later data entry -​ Some populations are uncomfy with technology 4)​ Layout and formatting -​ Grammar errors, misspellings, gaps in logic, unclear instruction, readability  -​ Shouldn't force an answer to selected questions, may cause them to leave survey -​ Skip logic code; can automatically hide questions from participants based on their responses to filter questions -​ If unemployed remove all questions about jobs 5)​ Back translation -​ Crucial to have harmonizes surveys -​ Translate survey from original to new language with one person and second person translates it back to original -​ Comparison of the original and back translate version will reveal where the second language translation does not match the intended meaning of the original one 6)​ Pilot testing -​ Small scale preliminary study conduct to evaluate the feasibility of a full scale research project -​ Survey instrument is revised based on these observations