Hey everyone!
Because the names are so hard to remember, I make silly associations out of them and I hope it helps you all too =)
Berksonian bias: Selection bias that arises from evaluating data on biased patients and hospital records only.
When it comes to Berksonian bias I think of Preston Berke from Grey's anatomy and I know that it's associated with the hospital and patients!
Pygmalion effect: The observer-expectancy effect is a form of reactivity in which a researcher's cognitive bias causes them to unconsciously influence the participants of an experiment.
Researchers are pigs. They think their research is always right. Pygmalion effect! xD
Hawthorne effect: When behavior of the subjects of the study change because they know they're being observed.
Haww I am being studied. Hawthorne effect!
That's all!
Pray that I do well in my exams.
-IkaN
This is awesome. A Tumblr post with gifs and tables. Yaay!
Because the names are so hard to remember, I make silly associations out of them and I hope it helps you all too =)
Berksonian bias: Selection bias that arises from evaluating data on biased patients and hospital records only.
When it comes to Berksonian bias I think of Preston Berke from Grey's anatomy and I know that it's associated with the hospital and patients!
Pygmalion effect: The observer-expectancy effect is a form of reactivity in which a researcher's cognitive bias causes them to unconsciously influence the participants of an experiment.
Researchers are pigs. They think their research is always right. Pygmalion effect! xD
Hawthorne effect: When behavior of the subjects of the study change because they know they're being observed.
Haww I am being studied. Hawthorne effect!
That's all!
Pray that I do well in my exams.
-IkaN
This is awesome. A Tumblr post with gifs and tables. Yaay!
Yeah it's good. . But I couldn't understand berksonian bias. . Plz explain by example
ReplyDeleteBerkson bias is using only hospital records to estimate population prevalence.
DeleteThere are a lot of cases of hypertension in the community. If you are counting only those who were admitted to the hospital, you are largely underestimating the actual number of hypertensives in the community!
(This is a selection bias. Selecting only those cases who were severe enough to reach the hospital or something.)
Similarly, if you do a research on hospital patients only, there are a lot of factors that differ in hospital patients than in those who are not in a hospital. Solution is, your sample must be randomly selected so you don't just talk about fatal cases in your study.
^ I have a poor choice of words in the above comment and thanks to Jewel, here's the same but better stated example of Berksonian bias -
DeleteBerksonian bias is using only hospital records to select cases and controls.
There are a lot of cases of hypertension in the community. If you are counting only those who were admitted to the hospital, you are largely neglecting other factors that may or may not be present in hospitalized patients.
Solution is, cases and controls must be randomly selected. So you talk about both - hospitalized people as well as people in the community who are not hospitalized.
Updated on 12th June, 2014.
Ikan's explanation is not completely correct. Berkson bias usually occurs when cases and controls are selected from hospital inpatients.More specifically, when both the exposure and outcome affect the selection and leads to a false negative association.It looks confusing but just look at this example:
ReplyDelete1.Consider an investigator studying a relation between diabetes and CHD.He goes to a hospital and gets a list of people admitted with CHD and he selects equal number of controls(inpatients not having CHD).
Let us create a 2x2 table here
Disease+ Disease-
Exposure+ a b
Exposure- c d
Both diabetes and CHD are causes of hospitalisation.And we select our cases and controls from these hospitalised patients.The problem of choosing the control group(those who don't have CHD) from inpatients is that when compared to general public,the number of diabetics among the controls are much more(diabetes is a cause of hospitalisation). So that increases the value of b(exposed controls) with a corresponding decrease in d(not exposed controls) when compared to general population.
Odds ratio=ad/bc
So decrease in d and increase in b will cause a decreased odds ratio(decreased association) when compared to the value that would have been obtained if the study has been conducted I'm general population.Thus,here when exposure and result both affects selection,we get a decreased association.
Hope it is useful.
All the best for your exams Ikan!
That's another way to look at it.. There are a lot of factors that differ in hospitalized cases or controls.
DeleteThank you for the example though! :)
And berksonian bias is not used to find out prevalence.It is defined in case control studies.
ReplyDeleteI never said you use it to find out prevalence.
DeleteIf you are assuming that all you cases are hospitalized patients, you are indirectly assuming the prevalence is 100% in hospitalized patients and 0% in the community.
Berksonian bias is a selection bias. You select cases in a case control study from the people affected by the disease (or simply, the prevalent cases).
Hope you understand :)
Actually you are never assuming that hospital patients represent all your cases!If you assume that it would rather be a cross sectional study where you can just go to the hospital,count the number of patients and find out the prevalavnce in the general population. There are two mistakes in your reply
ReplyDelete1.Case control study has nothing to do with prevalence. Even if an investigator knows that prevalence of a disease in inpatients and general public is grossly different,he can select cases from inpatients.Because asfar as case control study is concerned,you are only worried about getting cases and where you get them from is irrelevant.
2.Case control studies are done in hospitals because,it is easy to get cases there.And not because anyone assumes that all your cases are admitted or treated.
Your second example is correct.
But in the first example
There are a lot of cases of hypertension in the community. If you are counting only those who were admitted to the hospital, you are largely underestimating the actual number of hypertensives in the community!
Obviously,this is a selection bias,but this has nothing to do with a case control study.What you said here is an example of a selection bias for a cross sectional study.Berkson bias has nothing to do with it.
All selection biases including hospitals are not berkson bias,it is defined only when a case control study is tested
Oh I see. I had a poor selection of words.
DeleteI would re-frame it as, "Berkson bias is using only hospital records to select cases."
There are a lot of cases of hypertension in the community. "If you are counting only those who were admitted to the hospital, you are largely neglecting other factors that may or may not be present in hospitalized patients."
Solution is, cases must be randomly selected. "So you talk about both - hospitalized cases as well as cases in the community who are not hospitalized."
Does it sound accurate now?
What I meant is berkson bias is NOT due to a false assumption that hospitalised patients represents 100% of your cases(as mentioned in your first example). Obviously,your example shows a selection bias,but it is not burksonian bias.
ReplyDeleteYes,there are a lot of factors different in hospitalised patients.But burkson bias only refers to the factor that you study.Other factors that you don't study are rather confounding factors than selection bias.There may be 100 different factors in patients,but it becomes Burkson bias,only if they affect your selection
ReplyDeleteFirst sentence in your explanation of Berkson bias:
ReplyDelete"Berkson bias is using only hospital records to estimate population prevalence."
That is why I told it has nothing to do with estimating prevalence.
"If you are assuming that all you cases are hospitalized patients, you are indirectly assuming the prevalence is 100% in hospitalized patients and 0% in the community"
ReplyDeleteIn the scenario of case control study(and so Berkson bias),you are not at all assuming that 100% of your cases are admitted there.
Because it doesn't matter if 100% or 50% or 10% percent are recorded in hospitals,you just need to find cases so that you can study the risk factors!
ReplyDeleteEven if 10% percent cases are admitted you get them in one place rather than the 90% scattered in population.That's why you go there.And you don't need to find all the cases in a case control study.
ReplyDeleteAbsolutely!:)Sorry for the trouble I caused!Really appreciate you for your wonderful blog!Keep up the good work!:)
ReplyDeleteBut just one thin. I would rather say problem in selecting controls from hospitals:)Again sorry for the trouble!:)
ReplyDeleteOh it's not an issue Jewel as long as we are learning.. Thank you for the correction <3
ReplyDeleteI'll update my mistake ^__^
You must be an MD now 🤔🤔🤔🤔
ReplyDeleteThank you for the nice mnemonics, they have been sooo supportive and helpfull.
ReplyDeleteI'v been searching for visual mnemonics... But I think yours are more engaging since they have a fun humour.
This comment has been removed by the author.
ReplyDeleteGood afternoon . my first time here.
ReplyDeleteplease I need to be educated on the issue of sample size calculation. I am doing a population study and my colleagues are saying I need to calculate sample size using a formular for the study. I totally disagree with them. I may be wrong so please help to educate me on the use of sample size when all the subjects are being recruited for the retrospective study. Thanks for I await your response. omoregbe Isaac