How must sample size calculations be adjusted, if a hawthorne effect might be occuring during study conduction? Is there a publication concerning this problem?
Hello Christian,
I am not absolutely sure how to account for the hawthorne effect at sampling level but what I can suggest is that if your experimental group and control group are observed similarly that effect could be accounted for. Another way could be trying to run a pilot study where the study groups should be giving anonymous responses....then you can expand the study if your responses are great. What has worked very way in medical/ veterinary studies is blinding,,,,that is either, single blinding, double and also maybe triple blinding,,,,so that groups collecting data does not know which is the treatment group so are study subjects that way, you could do your sampling as in the usual way.
"One way to deal with the Hawthorne effect (and demand characteristics) is to observe the participants unobtrusively. This can be done using the naturalistic observation technique. However, this is not always possible for all behaviors. Another way to deal with the Hawthorne effect is to make the participants' responses in a study anonymous (or confidential). This may eliminate some of the effects of this source bias."
(McBride, 2013)
Kind regards.
Christian.... This is a faulty design issue and canot be "adjusted" per se. The recycling bin is the best option, followed by a fresh start with some experienced help in the design phase. I suppose if you were fortunate to have differerent groups, some of whom were 'aware' and others were 'not aware', you would be able to actually measure the Hawthorne effect in your study.
Does your question concern a randomized trial where the Hawthorne effect would be comparable to a placebo effect in a pharmacological trial?
Then I think you should simply adjust your expected trial results relative to what is observed before the trial based on how much you expect the Hawthorne effect will influence the results in the control arm; apply the expected intervention effect to that and calculate sample size/power based on those assumptions.
- Badges
- Science topic
- Similar topics
- Medicine
- Epidemiology