The research question that is interesting me is how a formal mentorship program can improve orthopaedic surgical education. I’ve talked about this in a previous blog, but, as a brief recap, I would like to study if a standardized mentorship program improves resident quality of life, happiness, burnout, education, etc. I have a nice, simple, captive sample of participants for this study, i’e., my residents. They would qualify as a convenient and purposeful non probability sample. The pros to this sample include ease of study (they are right at my disposal), low cost (don’t have to travel or use any sort of long distance devide, e.g an internet based survey), and high participation (they pretty much have to do what I ask them to, after all). It would be a very convenient study sample, literally. The cons would be a small sample (big “n’s” impress medical researchers), the results may not be representative of the vast population of orthopaedic residents (there are over 100 orthopaedic residencies in america with up to 45 residents in a program), and the results could be biased by the relationship I have with the residents.
To address the problem of a small sample and the iablity to generalize the results I have the option to track down the contact information of every orthopaedic resident in america, and use that as my sample. I could still use every single particpant in my study (another nonprobability scenario), but if I had enough responses and participants I could begin to include different types of probablity sampling. I could use a simple random sample where a certain number of participants are selected at random. The pros would be ease of selection, analysis, and understanding, but it may not be representative of smaller sub groups of the population. Similar pros exist for systematic sampling (when every nth participant is chosen). The only other problem with systematic sampling is that a pattern or periodicity to the list being chosen from would lead to a non-representative sample. If there were enough responses from orthopaedic residents then stratified sampling could be used. This sampling occurs after the larger population is broken down into sub-classes (residents from programs with more or less than 6 residents per year or residencies from different parts of the US, e.g.). This type of sampling can ensure better representation of the larger population (by ensuring adequate representaion from all sub-groups) and provide more information by comparing sub-populations. It can be done in a proportionate manner (the number of participants selected from each sub-group are proportional to the size of the sub-group in comparison to the whole) or disproportionate (for whatever reason more or less subjects from each sub-group are chosen despite the size of the sub-group). Both of these probabilty types of sampling require more work: you have to accurately identify sub-groups and their characteristocs and then reproducibly place participants in the appropriate sub-group and you have to decide whether to use proportionate (results represent the population without having to weight the numbers) or disproprotionate (requires weighting of the data to extrapolate it to the larger population) methods. I could also use a cluster sample. This method uses a more convenient sample (e.g., the orthopaedic residents in the mid-atlantic) to try and collect representatice data. I would assume a cluster of residents from the mid-atlantic would represent the entire US, and my research would be easier, quicker, and cheaper, but it would be less accurate and possibly non-generalizable.
At first, I would start with the simple, convenient nonprobablity study for the reasons above. If the results were interesting and meaningful then I could expand the study to include multiple other residents and change my sampling techniques. However, it is stil likely I would use all the results obtained in a nonprobable fashion so that all the data was analyzed.