Hi,
We plan to do a KAP survey on risk factors for certain NCDs and wish to do this survey online in educational institutions (targeting students) that regularly use emails to send internal circulars. All educational institutions having this facility will be considered (We are aware of the selection bias here).
Additionally, we do not want to randomly select students but would like to invite all willing to participate.
1. Is this method justifiable?
2. Should we still do a sample size calculation to know what the minimum sample is?
3. If sample size calculation is required, then how to go about it? Any resources to refer to? Majority of the questions are closed-ended.
Thank you in advance
Online solicitations almost always produce convenience samples, so I would start by accepting that limitation. The main issue with regard to sample size would then be whether you have the "power" to detect significant relationships. The standard for addressing this issue is the g*power program.
Prof Morgan power is a meaningless concept in nonprobability samples . I refer you to any reputable mathematical statistics text. Devore and Berk Mathematical statistics, springer is one of my favorites. Best wishes David Booth
David Eugene Booth So what is your recommendation?
FYI I checked the g*power manual and the program does have an option for "non-probability" samples (with the obvious warning against selection bias in convenience samples).
Thank you both for the valuable inputs.
David L Morgan, David Eugene Booth
Prof Morgan as Abraham Lincoln once said calling a horse's tail a leg does not make it so. If you check the math stat text you find power is a particular probability of an event. This means no probability sample no accurate estimate for power. Best wishes David Booth
A lot of work has recently been done to use covariates for either modeling or to arrive at pseudo-random sampling approaches to ameliorate nonprobability sampling shortcomings, but there is no attempt at this here. The volunteer surveys described would be potentially very biased. You can't use random probability-of-selection-based theory for something that is anything but random, and expect accurate results, as David Booth has already noted.
- Badges
- Science topic
- Similar topics
- Biological Science
- Physiology
- Physiological Processes
- Stress