@techreport {handle:1813:43895, title = {NCRN Meeting Spring 2016: Developing job linkages for the Health and Retirement Study}, number = {1813:43895}, year = {2016}, institution = {University of Michigan}, type = {Preprint}, abstract = {NCRN Meeting Spring 2016: Developing job linkages for the Health and Retirement Study McCue, Kristin; Abowd, John; Levenstein, Margaret; Patki, Dhiren; Rodgers, Ann; Shapiro, Matthew; Wasi, Nada This paper documents work using probabilistic record linkage to create a crosswalk between jobs reported in the Health and Retirement Study (HRS) and the list of workplaces on Census Bureau{\textquoteright}s Business Register. Matching job records provides an opportunity to join variables that occur uniquely in separate datasets, to validate responses, and to develop missing data imputation models. Identifying the respondent{\textquoteright}s workplace ({\textquotedblleft}establishment{\textquotedblright}) is valuable for HRS because it allows researchers to incorporate the effects of particular social, economic, and geospatial work environments in studies of respondent health and retirement behavior. The linkage makes use of name and address standardizing techniques tailored to business data that were recently developed in a collaboration between researchers at Census, Cornell, and the University of Michigan. The matching protocol makes no use of the identity of the HRS respondent and strictly protects the confidentiality of information about the respondent{\textquoteright}s employer. The paper first describes the clerical review process used to create a set of human-reviewed candidate pairs, and use of that set to train matching models. It then describes and compares several linking strategies that make use of employer name, address, and phone number. Finally it discusses alternative ways of incorporating information on match uncertainty into estimates based on the linked data, and illustrates their use with a preliminary sample of matched HRS jobs. Presented at the NCRN Meeting Spring 2016 in Washington DC on May 9-10, 2016; see http://www.ncrn.info/event/ncrn-spring-2016-meeting}, url = {http://hdl.handle.net/1813/43895}, author = {Mccue, Kristin and Abowd, John and Levenstein, Margaret and Patki, Dhiren and Rodgers, Ann and Shapiro, Matthew and Wasi, Nada} } @techreport {2410, title = {Using Social Media to Measure Labor Market Flows}, year = {2014}, type = {Mimeo}, url = {http://www-personal.umich.edu/~shapiro/papers/LaborFlowsSocialMedia.pdf}, author = {Antenucci, Dolan and Cafarella, Michael J and Levenstein, Margaret C. and R{\'e}, Christopher and Shapiro, Matthew} } @techreport {2413, title = {Reconsidering the Consequences of Worker Displacements: Survey versus Administrative Measurements}, year = {2013}, institution = {University of Michigan}, type = {mimeo}, abstract = {Displaced workers suffer persistent earnings losses. This stark finding has been established by following workers in administrative data after mass layoffs under the presumption that these are involuntary job losses owing to economic distress. Using linked survey and administrative data, this paper examines this presumption by matching worker-supplied reasons for separations with what is happening at the firm. The paper documents substantially different earnings dynamics in mass layoffs depending on the reason the worker gives for the separation. Using a new methodology for accounting for the increase in the probability of separation among all types of survey response during in a mass layoff, the paper finds earnings loss estimates that are surprisingly close to those using only administrative data. Finally, the survey-administrative link allows the decomposition of earnings losses due to subsequent nonemployment into non-participation and unemployment. Including the zero earnings of those identified as being unemployed substantially increases the estimate of earnings losses.}, url = {http://www-personal.umich.edu/~shapiro/papers/ReconsideringDisplacements.pdf}, author = {Flaaen, Aaron and Shapiro, Matthew and Isaac Sorkin} } @article {2262, title = {Ringtail: Feature Selection for Easier Nowcasting.}, journal = {WebDB}, year = {2013}, pages = {49-54}, chapter = {49}, abstract = {In recent years, social media {\textquotedblleft}nowcasting{\textquotedblright}{\textemdash}the use of on- line user activity to predict various ongoing real-world social phenomena{\textemdash}has become a popular research topic; yet, this popularity has not led to widespread actual practice. We be- lieve a major obstacle to widespread adoption is the feature selection problem. Typical nowcasting systems require the user to choose a set of relevant social media objects, which is difficult, time-consuming, and can imply a statistical back- ground that users may not have. We propose Ringtail, which helps the user choose rele- vant social media signals. It takes a single user input string (e.g., unemployment) and yields a number of relevant signals the user can use to build a nowcasting model. We evaluate Ringtail on six different topics using a corpus of almost 6 billion tweets, showing that features chosen by Ringtail in a wholly-automated way are better or as good as those from a human and substantially better if Ringtail receives some human assistance. In all cases, Ringtail reduces the burden on the user.}, url = {http://www.cs.stanford.edu/people/chrismre/papers/webdb_ringtail.pdf}, author = {Antenucci, Dolan and Cafarella, Michael J and Levenstein, Margaret C. and R{\'e}, Christopher and Shapiro, Matthew} }