3
$\begingroup$

Probability sampling methods help reduce sampling bias. In clinical research, simple random sampling methods works well for randomised control trials, and systematic sampling methods are easy to implement in cross-sectional observational studies in the clinic setting.

However, to the best of my knowledge, systematic sampling methods assume every nth person selected will elect to take part in the study. In reality, every person sampled still has to provide informed consent, and thus the informed consent process adds a point at which selection bias can be introduced in much the same way as it is introduced through convenience sampling. That is, even if the process of selecting which patients to approach is random, the requirement for informed consent means that the final sample is a self-selected subgroup of those originally sampled (e.g., people with pain are probably more likely to volunteer for a study on chronic pain prevalence than those without pain).

I have tried to find information on this issue, but can only find standalone information on sampling methods or informed consent, and not how the two may interact in some cases. Apologies if my logic on this issue is faulty, but if it is, where is it faulty?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
3
$\begingroup$

Your logic doesn't seem faulty to me.

As soon as you give people the option not to participate you're making your sample a little more "convenient-like" than you might have intended. Even if nobody withdraws then you're still only sampling people who want to participate. Whatever results you obtain need not necessarily hold for those who might not have participated.

To some extent you may be able to reduce the reasons people might have not to participate, e.g., if participation is particularly demanding in time or energy then you may look to reduce that, but ultimately you're always going to have people who will refuse. All you can really do is to try to consider why people might withdraw consent, with reference to relevant prior literature where possible, and acknowledge this when you come to interpret your findings and draw conclusions from them.

To give an extreme example, if I'm interested in studying the attitude of students to participating in research, I would have to acknowledge that those students most opposed to participating are unlikely to appear in my sample. As a result, my findings probably don't reflect their views. it's not ideal, but as long as I acknowledge that limitation when I come to disseminate my findings then it's just a fact of life we live with.

Improve this answer
$\endgroup$
5
  • $\begingroup$ That's the way I have been approaching this issue to date, but it reassuring to hear someone else supports this approach too. Just wish the standard literature dealt with the issue. $\endgroup$
    – Peter K
    Commented Jun 30, 2016 at 7:59
  • $\begingroup$ (+1) @PeterK: Well it's just an early drop-out: see en.wikipedia.org/wiki/Intention-to-treat_analysis. $\endgroup$
    – Scortchi ♦
    Commented Feb 21, 2017 at 16:59
  • 1
    $\begingroup$ @Scortchi your suggestion makes sense, but the problem is that in obs studies you don't have data from people who 'withdraw' before consent that you can carry into your analysis. In clinical trial, if you look at the CONSORT diagram, the top level is 'assessed for eligibility' followed by 'randomized', and you account for dropouts who didn't consent as you move from the top tier to the next. Individuals who dropout at this stage (pre-randomization) are not included in the ITT population. In obs studies, 'randomization' is part of the sampling process, and precedes consent and data collection. $\endgroup$
    – Peter K
    Commented Apr 17, 2017 at 11:53
  • $\begingroup$ @PeterK: I see what you mean. Could you randomize prior to getting patients' consent, for a conservative ITT analysis (assuming you have their consent to monitor outcomes)? Another approach, often used with surveys, is to follow up a proportion of those who declined to take part & cajole/bribe them into either taking part (allowing for a stratified analysis), or providing other information that throws light on their reasons for declining. (I can of course imagine either or both of these suggestions being deemed inappropriate for clinical studies, for ethical reasons.) $\endgroup$
    – Scortchi ♦
    Commented Apr 18, 2017 at 8:45
  • 1
    $\begingroup$ I found Armstrong & Watts (1981), Biomedicine, 34, 2, "Clinical trials:randomization before consent" & Zelen (1979), N. Engl. J. Med., 300, 22, "A new design for randomized clinical trials". $\endgroup$
    – Scortchi ♦
    Commented Apr 18, 2017 at 9:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.