BACKGROUND: Experimental research in nursing has increased considerably in recent years. To improve the quality of such research, it is critical to reduce threats to internal validity. One threat that has received inadequate attention in the nursing literature is Simpson's paradox--a case of extreme confounding that can lead to erroneous conclusions about the effects of an experimental intervention. In fact, it can lead to a conclusion about an intervention effect that is the opposite of the correct inference. APPROACH: The aims of this study were to describe Simpson's paradox, provide a hypothetical example, and discuss approaches to avoiding the paradox. RESULTS: The paradox is due to the combination of an overlooked confounding variable and a disproportionate allocation of that variable among experimental groups. Different designs and analysis approaches that can be used to avoid the paradox are presented. DISCUSSION: Simpson's paradox can be avoided by selecting an appropriate experimental design and analysis that incorporates the confounding variable in such a way as to obtain unconfounded estimates of treatment effects, thus more accurately answering the research question.
BACKGROUND: Experimental research in nursing has increased considerably in recent years. To improve the quality of such research, it is critical to reduce threats to internal validity. One threat that has received inadequate attention in the nursing literature is Simpson's paradox--a case of extreme confounding that can lead to erroneous conclusions about the effects of an experimental intervention. In fact, it can lead to a conclusion about an intervention effect that is the opposite of the correct inference. APPROACH: The aims of this study were to describe Simpson's paradox, provide a hypothetical example, and discuss approaches to avoiding the paradox. RESULTS: The paradox is due to the combination of an overlooked confounding variable and a disproportionate allocation of that variable among experimental groups. Different designs and analysis approaches that can be used to avoid the paradox are presented. DISCUSSION: Simpson's paradox can be avoided by selecting an appropriate experimental design and analysis that incorporates the confounding variable in such a way as to obtain unconfounded estimates of treatment effects, thus more accurately answering the research question.
Authors: Bernard L Marini; Shannon M Hough; Kevin S Gregg; Haya Abu-Seir; Jerod L Nagel Journal: Support Care Cancer Date: 2015-01-10 Impact factor: 3.603
Authors: Patrick Nadol; Siobhan O'connor; Hao Duong; Linh-Vi N Le; Pham Hong Thang; Tran Hong Tram; Hoang Thi Thanh Ha; Michelle S Mcconnell; Jeff Partridge; John Kaldor; Matthew Law; Tuan Anh Nguyen Journal: PLoS One Date: 2015-02-18 Impact factor: 3.240
Authors: Kin Y Mok; Una Sheerin; Javier Simón-Sánchez; Afnan Salaka; Lucy Chester; Valentina Escott-Price; Kiran Mantripragada; Karen M Doherty; Alastair J Noyce; Niccolo E Mencacci; Steven J Lubbe; Caroline H Williams-Gray; Roger A Barker; Karin D van Dijk; Henk W Berendse; Peter Heutink; Jean-Christophe Corvol; Florence Cormier; Suzanne Lesage; Alexis Brice; Kathrin Brockmann; Claudia Schulte; Thomas Gasser; Thomas Foltynie; Patricia Limousin; Karen E Morrison; Carl E Clarke; Stephen Sawcer; Tom T Warner; Andrew J Lees; Huw R Morris; Mike A Nalls; Andrew B Singleton; John Hardy; Andrey Y Abramov; Vincent Plagnol; Nigel M Williams; Nicholas W Wood Journal: Lancet Neurol Date: 2016-03-24 Impact factor: 59.935