Subjects are recruited by mailing the questionnaire to randomly selected adults who belong to a Lombardo health maintenance organization. Subjects who return the questionnaire will then be asked further questions about their health at later dates (1 year, 3 years).
What is the study design and what are the potential problems with this design? What alternative strategies for recruiting subjects would you consider?
This study is a random stratified quantitative survey. To better understand we'll break it down to the components. The random portion is given in the study problem: it is sent to random people. The stratified group is a portion of the population separated from the remainder based on some characteristic defined by the researcher. In this case belonging to the medical group is the stratified population. It is quantitative in nature because the answers will be put into a numerical or data centric device to gauge response. A qualitative survey allows subjects to input their own data, whereas in a quantitative survey, it is provided for by the researcher.
There are a number of problems with this style. The first is going to be initial response percentage and it will be further compounded by tertiary response. The better way to understand this is the response rate over time. After the first random mailing, only X percent of the population will return the survey. This is your initial response rate. In this design, follow ups will be conducted at later times; this is your Y and Z response rates. Some people will provide X and Y responses, but not Z or only X responses. The drop in response rate over time will dramatically reduce the useable data.
Another problem is incomplete or erroneous data provided. Some X responders won't provide the entire survey or will double mark some entries corrupting the data. To tie in with this is the misunderstanding of the question or miscalculation of the data. Who is to say what "good nutrition" is?
A better way to address this sampling problem is to do a regression model on the strata. In this model, researchers review medical histories for current patients and compare their histories to their current health. Another way to prevent a reduction in Y and Z response is constant monitoring through the medical group and reminding the respondents of their participation. One great way to reduce tertiary response rate drop is to offer a prize at the completion of the survey.
Corrupt data can be reduced by encouraging participants to complete the entire survey in exchange for a coupon or prize entry. Additionally, explicit definitions should be included so a person understand what the researcher means by certain ambiguous phrases.