Sample Selection by Powers of Size when Needing Estimates at Multiple Levels

• Activity Number: 278
• Type: Contributed
• Date/Time: Tuesday, August 10, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Harley K. Heimovitz*+ and Pedro J. Saavedra
• Companies: ORC Macro International Inc. and ORC Macro International Inc.
• Keywords: sampling ; PPS ; institutions ; variance ; simulations ; establishments

Abstract

Institution and establishment surveys often require two kinds of estimates: one may be the percentage of units meeting a particular characteristic, and the other may be the percentage of clients served or volume sold associated with units meeting the characteristic. For example, a survey of schools may need to estimate the percent of schools that implement a given program, and the percentage of students attending such schools. For the first estimate, one should assign the same probability of selection to every school, whereas for the second one should sample with probabilities proportional to the enrollment. When both are needed, some researchers have sampled with probabilities proportional to the square root of enrollment. The question is what power to use to minimize the larger of the two coefficients of variation. Using a national database of schools, five powers of the enrollment were used and estimates were obtained for several simulated and actual categorical variables varying in magnitude and in their correlation to the enrollment. The coefficients of variation were obtained through multiple simulations and the maximum of the two CVs compared.