For example, suppose your target population is college students. However, you advertised for volunteers off-campus and attracted some locals as well. Locals who don’t attend the college may not have the same characteristics as your target population and including them could lead to selection bias. The number of participants in your study must also be of adequate size for you to be able to apply the results of your study to the population at large. The necessary sample size will vary depending on a variety of factors, such as the magnitude of the effect you’re studying and its variability within the population. You might also get some help from an online calculator that helps you determine your sample size, such as the one available at https://clincalc. com/stats/samplesize. aspx.
Create a questionnaire with inclusion and exclusion criteria. For example, if you’re studying the effect of sleep on college student grades, you might want to ensure you have a balance of students with a lot of early morning courses and night students. In that case, you would include a question about the volunteer’s class schedule. If you only wanted to include full-time students, you would ask how many course hours the volunteer was taking. Once you have approximately 2-3 times the number of potential participants you need for the study, assign them each a random number. Then choose your study participants randomly based on those numbers. This helps reduce selection bias as well as volunteer bias.
Since it’s not the real thing, your sample size doesn’t have to be as large as it would be for the full study, which helps reduce costs. Pilot studies also give you an idea of how quickly you’ll be able to recruit participants for your study and which methods of recruitment seem to work the best.
For example, if your investigators were asking the participants a series of questions, your operations manual would include the exact questions asked. Then, you could coach your investigators on their tone of voice and other factors that might skew the participants’ responses. If you have multiple people involved in the study, train them on the methods you want them to use during the study and test them to make sure they’re all doing everything the same. If your study is going to take place over the course of months or years, it might be necessary to have “refresher” courses to keep investigators up to speed on your protocol, especially if they are away from the study for a while.
Most universities have research support units to assist with randomization. There are also computer programs that will do randomization for you. If you don’t have access to research support, use a free random number generator, such as the one at https://www. random. org/. Larger studies typically use a remote randomization facility to ensure that there’s no way anyone involved with the study could know which group any given participant was in.
For example, if your study included surgery, it would be impossible for your participants not to know if surgery was being performed on them. In that case, your investigators could be blind as to a particular subject’s group while taking their measurements and compiling data, but the participant could not because they would have to consent to the surgical procedure. Even if you have double-blinding in place, it might break down. For example, if you’re studying a drug that turns out to have dangerous side effects, you might need to know which participants were taking the drug so you could monitor them or warn them of the side effects.
For example, if you’re studying a population’s likelihood of contracting a disease after exposure to the virus that causes it, you would want a sample that was similar in age, socio-economic status, and access to healthcare. Maintaining these similarities reduces the possibility that some participants’ outcome was affected by their health or medical treatment.
For example, if your case population comes from patients referred to a particular hospital for treatment, you might seek out your controls from the healthcare providers who made those referrals.
For example, if you’re doing a study on smoking and chronic heart disease, having hospitalized controls would weaken the association because smoking is a factor that leads to many health problems that could also result in hospitalization.
For example, suppose a local restaurant is responsible for a viral outbreak, but you don’t know which one. The local population who contracted the virus are your cases. To identify which restaurant is responsible, you could enroll people from the local area who matched your cases in terms of neighborhood, age, and gender, but didn’t contract the virus, as your controls.
Choose a population dataset for your control that matches the population of the cases you’re studying. For example, if all of your cases are located in the state of California, you might use a state database to get your population data. However, you wouldn’t want to use a national database.
For example, suppose you are studying the connection between coffee and migraines. You sent out postal surveys to households in the state of California. However, you’re aware of previous studies that have shown older people are typically more interested in participating in postal surveys than younger people, so this could bias your study by age. To adjust for bias in the study of the connection between coffee and migraines, you could separate your data so that it measured the connection in different age groups separately (stratification). This would reduce the selection bias that would occur by having too many older people in your sample.
For example, suppose you were studying the effect of sleep on grades among college students. The student population at the school you’re studying is 40% male and 60% female. However, your sample is only 20% male. To weight the male responses, divide the population percentage by your sample percentage (40% divided by 20%). The result is 2, so each male’s response counts double.
For example, suppose you want to evaluate the association between working the night shift and having a particular health problem by comparing people who work at the same factory doing the same work, with the only difference being that some work during the day and some work at night. However, there are likely to be many other differences between these groups that you can’t possibly account for, such as their socio-economic status or access to health care. In the report of your study, acknowledge that there are a lot of other differences that your study didn’t take into account. You might also mention what some of those differences could be and include references to other studies that have analyzed those variables in depth. To conduct effective scientific research, you should have a clear idea of your objectives before starting the project. Knowing what specific question you are asking will make it easier to target your research. Include experiments with the hypothesis to provide an answer.
