CMStatistics 2018: Start Registration
View Submission - CMStatistics
B1153
Title: Poststratifying on variables measured with error Authors:  Daniel Oberski - Utrecht University (Netherlands) [presenting]
Abstract: Samples, by design or by accident, often do not reproduce known population totals on average. To solve this potential problem, a common approach is reweighting, with weights calculated based on the disparity between the totals that are known and those that are measured. But when measured variables are error-prone, weights cannot be calculated. For example, when weighting a social media population based on US State, how should we account for the fact that social media users may give incorrect information about their state of residence? A novel method is introduced to postratify based on variables measured with error, the ``Mixture of States with Poststratification'' (Ms. P). The method is applied to a nonprobability sample of 73,000 Facebook users.