Targeted Learning Causal Inference for Observational and Experimenta
The statistics profession is at a unique point in history. The need for valid statistical tools is greater than ever; data sets are massive, often measuring hundreds of thousands of measurements for a single subject. The field is ready to move toward
- PDF / 6,431,060 Bytes
- 678 Pages / 439.37 x 666.15 pts Page_size
- 87 Downloads / 163 Views
For other titles published in this series, go to http://www.springer.com/692
Mark J. van der Laan • Sherri Rose
Targeted Learning Causal Inference for Observational and Experimental Data
Mark J. van der Laan Division of Biostatistics University of California Berkeley Berkeley California USA [email protected]
Sherri Rose Division of Biostatistics University of California Berkeley Berkeley California USA [email protected]
ISSN 0172-7397 ISBN 978-1-4419-9781-4 e-ISBN 978-1-4419-9782-1 DOI 10.1007/978-1-4419-9782-1 Springer New York Dordrecht Heidelberg London Library of Congress Control Number: 2011930854 c Springer Science+Business Media, LLC 2011 All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. Printed on acid-free paper Springer is part of Springer Science+Business Media (www.springer.com)
To Martine, Laura, Lars, and Robin To Burke, Pop-pop, Grandpa, and Adrienne
Foreword
Targeted Learning, by Mark J. van der Laan and Sherri Rose, fills a much needed gap in statistical and causal inference. It protects us from wasting computational, analytical, and data resources on irrelevant aspects of a problem and teaches us how to focus on what is relevant – answering questions that researchers truly care about. The idea of targeted learning has its roots in the early days of econometrics, when Jacob Marschak (1953) made an insightful observation regarding policy questions and structural equation modeling (SEM). While most of his colleagues on the Cowles Commission were busy estimating each and every parameter in their economic models, some using maximum likelihood and some least squares regression, Marschak noted that the answers to many policy questions did not require such detailed knowledge – a combination of parameters is all that is necessary and, moreover, it is often possible to identify the desired combination without identifying the individual components. Heckman (2000) called this observation “Marschak’s Maxim” and has stressed its importance in the current debate between experimentalists and structural economists (Heckman 2010). Today we know that Marschak’s Maxim goes even further – the desired quantity can often be identified without ever specifying the functional or distributional forms of these economic models. Until quite recently, however, Marschak’s idea has not attracted the attention it deserves. For statisticians, the very i
Data Loading...
