Publication | Closed Access
Regression Composite Estimation for the Canadian Labour Force Survey with a Rotating Panel Design
21
Citations
8
References
2001
Year
Unknown Venue
Regression CompositeApplied EconometricsRegression AnalysisPanel DataRegression Composite EstimationGeneralized RegressionStatisticsLatent Variable MethodsEconomicsEstimation StatisticLabor Force TrendEconometric MethodMarginal Structural ModelsRotating Panel DesignBusinessEconometricsTime-varying ConfoundingStatistical InferenceMedicineUnemploymentSurvey Methodology
We consider the regression composite estimation introduced by Singh (1994, 1996; termed earlier as “modified regression composite” estimation), a version of which (suggested by Fuller 1999) has been implemented for the Canadian Labour Force Survey (CLFS) beginning in January 2000. The regression composite (rc) estimator enhances the generalized regression (gr) estimator used earlier for the CLFS and the well known GurneyDaly akcomposite estimator in several ways. The main features of the rcestimator are: (a) it considerably improves the efficiency of level and change estimates for key study variables resulting into less volatile estimate series; (b) it is calculated like the grestimator as a calibration estimator such that all the usual poststratification controls used in gr as well as the new controls corresponding to correlated variables from the previous time point are met; and (c) it respects the internal consistency of estimators without having to calculate part estimates differently as residuals. The main innovations used in rcclass of estimators entail: (a) using the idea of working covariance matrix in estimating functions as an alternative to superpopulation modeling for defining regression coefficients for the predictors in the grestimator, (b) treating random controls (the ones based on the key correlated variables from past) as fixed, while computing the regression coefficients, similar to twophase estimation, and motivated from the working covariance idea, and (c) that of the use of micromatching to obtain previous time point’s microlevel auxiliary information for realizing higher correlation with the present time point’s study variables. As a by product, a new version of the akestimator which uses the micromatching based predictors from past rather than the traditional macrolevel is recommended in the interest of higher efficiency gains. The paper also presents an interesting heuristic justification of the smoothness feature of composite estimates using the amortization idea. Empirical results based on the Ontario 1996 CLFS data are presented for comparison of various estimators.
| Year | Citations | |
|---|---|---|
Page 1
Page 1