#JSM2018 Panel on CNSTAT report on Federal Statistics, Multiple Data Sources, and Privacy Protection, with @fraukolos kicking off The discussion
#JSM2018@fraukolos Goal of panel to evaluate combining data sources to possibly replace / augment surveys. Two reports out of the panel.
#JSM2018@fraukolos Conclusions: Current Federal Statistical Agencies face threats from falling response rates, rising costs, increased desire for granularity and timelines
#JSM2018@fraukolos Great graph on types of data. Aspirational Data are social media (what we aspire to show ourselves as) and transactional Data are interactions with computerized systems for transactions (purchases, for example)
#JSM2018@fraukolos Gives example of difference of testing signal strength for mobile devices. Can send a car around to precisely measure signal strength on small strip; use crowd sourcing from mobile devices to get broad measurement
#JSM2018@fraukolos Administrative, aspirational, transactional data all going to require work. May be out there, but needs to be processed, so not fast
#JSM2018@fraukolos A new entity should be built to house data access (similar to Evidence-Based Policy Making commission)
#JSM2018@fraukolos Second report goes into detail about statistical methods for combining multiple data sources. Can be hard when multiple data sources come in at different levels of analysis.
#JSM2018@fraukolos Federal Statistical Agencies should clearly document processes that were used for combining, etc. Data - sharing code more generally could be good even without multiple data sources
#JSM2018@fraukolos As we combine multiple data sources, what makes PII is more complicated (images contain identifying information even without names). (PII = personal identifying information)
#JSM2018@fraukolos Techniques that give limits to the number of queries that can be asked of a database one thing to think about WRT differential privacy
#JSM2018@fraukolos We survey folks should be comfortable claiming our TSE framework as useful for evaluating a dataset with multiple data sources. But need to think of more general quality framework as well
#JSM2018@fraukolos A new entity should facilitate access to multiple data sources. It should be transparent in what the data sources are and what methods are for doing Linkage. Have a board of directors to guide its actions.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
#JSM2018 Tobias Schmidt Looking at interviewer experience and interview duration
#JSM2018 Schmidt In this survey, duration linked to interviewer salaries.
#JSM2018 Schmidt Looking at interviewer experience over the course of survey and respondent experience within survey and experience over repeated surveys. Looking in particular at experience within panel survey for both Iers and Rs
#JSM2018 Wuyts Interested in within-survey workload. Use call history data and interview time data. Some Measure workload by fixed measures of experience and interview order cumulated over the field period. They use actual number of cases assigned at time t in field period
#JSM2018 Wuyts Use Paradata to create new measures of interview workload, based on sample units assigned on given day
#JSM2018 Rebecca Powell from @RTI_Intl talking about an experiment on Add Health shifting from interviewer administered to self administered survey
#JSM2018 Powell moved to a 55 self-administered survey from 90 minutes interviewer administered. Worried about response burden with this length of self-admin survey. Randomized n=7600 into either full 55 minute survey or 2 modules- one 35 minutes then 20 minutes.
#JSM2018 Powell Could select to continue on the web. In paper, had to first complete module A, then sent module B. Cover letters told about modules in the incentive part, but not up front. $55 incentive total in each condition
#JSM2018 The brilliant Susan Murphy is this year’s Fisher Lecture award recipient!
#JSM2018 Murphy Lab does sequential experimentation in improving health. Some for companies.
#JSM2018 Murphy Experimentation and continual optimization is key. How do we use learning as an experiment is put into the field to improve outcomes for individuals? Mobile interventions are key here. Intervention may be either a push intervention or pull intervention
#JSM2018 Next up Hubert Hamer from NASS talking about NASS Small Area Estimation
#JSM2018 Hamer NASS has Agriculture Loss Coverage County Option program. Payments triggered based on county crop revenue falling below program guarantee. NASS surveys used to make this decision, along with other data
#JSM2018 Hamer Program paid out $3.7 billion on 2016. Small changes can affect payments
#JSM2018 Peter Miller appearing as a Northwestern University emeritus professor, providing comments on the CNSTAT reports
#JSM2018 Miller Survey paradigm vs multiple data source paradigm. Surveys may become irrelevant b/c they are slow, not granular, not nimble, costly, not sustainable
#JSM2018 Miller Multiple Data sources require new: methods, computing resources, privacy protections, training, data quality frameworks. Not cheap. What does this give us?