Premium
Calibrating non‐probability surveys to estimated control totals using LASSO, with an application to political polling
Author(s) -
Chen Jack Kuang Tsung,
Valliant Richard L.,
Elliott Michael R.
Publication year - 2019
Publication title -
journal of the royal statistical society: series c (applied statistics)
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.205
H-Index - 72
eISSN - 1467-9876
pISSN - 0035-9254
DOI - 10.1111/rssc.12327
Subject(s) - polling , statistics , lasso (programming language) , coverage probability , econometrics , calibration , voting , selection bias , computer science , mathematics , confidence interval , political science , politics , world wide web , law , operating system
Summary Declining response rates and increasing costs have led to greater use of non‐probability samples in election polling. But non‐probability samples may suffer from selection bias due to differential access, degrees of interest and other factors. Here we estimate voting preference for 19 elections in the US 2014 midterm elections by using large non‐probability surveys obtained from SurveyMonkey users, calibrated to estimated control totals using model‐assisted calibration combined with adaptive LASSO regression, or the estimated controlled LASSO, ECLASSO. Comparing the bias and root‐mean‐square error of ECLASSO with traditional calibration methods shows that ECLASSO can be a powerful method for adjusting non‐probability surveys even when only a small sample is available from a probability survey. The methodology proposed has potentially broad application across social science and health research, as response rates for probability samples decline and access to non‐probability samples increases.