Estimation and Analysis of Variance Components for the Revised CPI Housing Sample

Owen J. Shoemaker


In 1999, the Bureau of Labor Statistics (BLS) introduced a new Housing Sample for the Rent and Rental Equivalency (REQ) estimators in the U.S. Consumer Price Index (CPI). The Housing Sample consists of, roughly, 10,000 sampled segments, composed of U.S. Census blocks, allocated in 87 Primary Sampling Units (PSUs) and collected in six panels every six months. In this paper, we model the 6-month price relative for both Rent and REQ, and analyze a random effects model that treats PSU and Segment as two random effects. We look at three years of data (1999-2001). We use the Restricted Maximum Likelihood (REML) estimation method to produce the variance components. Standard F-test procedures are applied to determine the significance of the effects in the model. Finally, the variance component results are compared to a set of variance components produced from the previous housing sample (1987-1998).