Modeling Monthly Birth/Death by using Sample Paradata

Jeremy Oreper


In the Current Employment Statistics (CES) survey, the monthly employment change as a result of out of sample births and deaths of establishments is modeled using an ARIMA time series. Five years of historical data, derived from the State Unemployment Insurance (UI) counts of net birth/death employment, serve as the inputs to the model. This model lacks any information current to the time that estimates are calculated, and as a result errors can increase when birth/death changes in a different pattern than the previous five years. To overcome these limitations a model was specified using paradata. The number of respondents able to report for the current month by the first deadline was combined into a modified Lotka–Volterra equation. This model has the advantage of using data contemporaneous with the production of the first release of estimates. The estimate of the birth/death using first release paradata and birth/death residuals from 2003-2012 are used in two year increments to forecast the monthly birth/death values for March 2006 through March 2012.