PakMediNet Discussion Forum : Biostatistics : Bias
BiasMy question is about the minimizing the bias when the researcher excludes the large number of participants from their study as they refuse to participate in the study. How can this problem of bias be minimized???
Posted by: bilal hsaPosts: 4 :: 31-03-2010 :: | Reply to this Message
Re: Bias
If you have calculated your sample size and added 10-20% drop rates and still you are getting number less than the calculated number, you need to increase your drop rate to make a higher sample size. You need atleast a minimum no. of patients required for your sample size.
For example, if you are sample size is 100 patients, and you added 20% drop rates, this would make 120 sample size. And if you get 20 drop outs, you can still make up 100 patients which would be enough for your study.
Posted by: docosamaPosts: 333 :: 03-04-2010 :: | Reply to this Message
Re: BiasThat alone will not minimize bias...you have to mention the characteristics of the people that tend to drop out...e.g more women?? More illiterate?? More elderly?? And depending on your research question this can invalidate your study.....
Posted by: aftab.iqbaPosts: 7 :: 04-04-2010 :: | Reply to this Message
Re: Bias
Bias doesn't decrease by increasing sample size. Bias is a systematic error. Increasing sample size will help to decrease random error.
An example can be of a weighing machine. If it is off by 5 kg, it will add 5 kg to everyone's weight. In this case, increasing sample size from 10 to 100 to 100 will still result in a mean that is off from true population mean by 5kg.
An example of a random error is again a weighing machine which is not very accurate and is likely to overshoot as often as it is likely to undershoot. This is a random error and in this case increasing sample size will provide a better estimate of mean and a narrower confidence interval
The question is an example of a well-known bias in epidemiology called selection bias. You can try to obtain as much data from the people who refused to participate and see if there are differences between those who agreed and those who didn't agree to participate. If there are no differences between the two groups, you can have some satisfaction that there is no selection bias.
However, the problem arises that you can gather very little information, if any, from those people who refused to participate. There are some newer statistical methods being proposed that try to account for this non-response but jury is still out there on their utility.
Posted by: rqayyumPosts: 199 :: 08-04-2010 :: | Reply to this Message