|Sampling and Sampling Information Protocols|
|Background Rationale of Policy Memo 98.3|
|Implementation of Item 4|
Background Rationale of Policy Memo 98.3
Discussion of Sampling and Sampling Information Protocols for the Behavioral Risk Factor Surveillance System
1. A sampling frame consisting of all telephone numbers that could possibly be assigned to households should be drawn from the Terminating Point Master--Vertical & Horizontal Coordinates Data (TPM-VHCD) database (Bellcore database) produced by Bell Communications Research. In the Bellcore database every prefix in the United States and its dependencies is assigned an NXX type number which designates the kind of service that is provided through the prefix. The NXX Types that contain household telephone numbers are
Thus, every telephone number in the NXX Types 00, 50, 51, 52, and 54 should have a non-zero probability of being selected.
Comment: This item is a more specific re-statement of the current BRFSS policy.
2. For data year 1999, all states are strongly encouraged to follow the provisions of this item. Beginning with data year 2000, this item will be standard BRFSS policy.
A disproportionate stratified sample (DSS) design is defined in this document as a design in which telephone numbers are assigned to two strata based on the presumed density of telephone household numbers. The ideal sampling ratio, that which yields the smallest standard error for a fixed cost or a given sample size, equals the square root of the ratio of the cost of completing an interview in a high and low household density stratum.
A. The ideal sampling ratio between one-plus block and zero block household density strata in a DSS design equals the square root of the ratio of the cost of completing an interview in the one-plus block stratum compared to the zero block stratum. In the absence of data on the relative cost of completing an interview in the two strata, states should use a sampling ratio of 4:1. States wanting to use a larger ratio must submit a request to do so to BSB, with documentation that the actual cost of obtaining a completed interview from low density strata is greater than 16 times the cost of obtaining a completed interview from high density strata. This request must be agreed upon in writing by the SOS Section Chief or the BSB Branch Chief.
B. States using definitions for high and low density strata other than one-plus and zero blocks must submit a proposed sampling ratio to BSB based on actual or estimated cost ratios. The sampling ratio should approximately equal the square root of the ratio of the cost of completing an interview in a high and low household density stratum. This sampling ratio must be agreed upon in writing by the SOS Section Chief or the BSB Branch Chief prior to its use.
States are strongly encouraged to manage their sample in 1999 in a way that allows them to determine the actual cost of completing an interview separately in high density and low density strata.
Comment: Although this provision has strongly counter-intuitive implications for BRFSS practice, it is based on a standard formula for optimally allocating sample among strata. The formula states that when variances across strata are equal (which seems to be a reasonable assumption for BRFSS data), the allocation of sample records that yields the smallest standard error (that is, the most precision) is proportional to the square root of the cost of completing an interview in each stratum. For example, according to Jim Lepkowski, a typical cost ratio for completed interviews between zero and one-plus blocks for telephone surveys is 10. The ratio that yields the smallest standard error in that case is /10 = 3.2. Deviations from the optimal ratio result in larger standard errors and, therefore, larger confidence intervals.
A 4:1 ratio for 0/1+ block strata was early on adopted as a BRFSS standard. (The source of the standard is unknown.) In recent years, more and more states have started using larger ratios in order to improve efficiency and obtain a larger sample size for a fixed cost. The result of that change is that they have thereby also increased their standard errors and confidence intervals. The counter-intuitive implication of this situation is that a state using a 0/1+ block sampling ratio larger than the square root of the ratio of the costs of completing an interview in each stratum can increase its precision by lowering its efficiency and its sample size.
3. The telephone numbers used to conduct a survey consist of all numbers actually called by the data collector plus, if applicable, those numbers pre-identified as non-working or business by the sample provider which were not called by the data collector but would have been called had they not been so identified.
For data year 1999, all states are strongly encouraged to follow the provisions of this item. Beginning with data year 2000, this item will be standard BRFSS policy. The monthly data files sent to BSB must consist of a record for every telephone number used to conduct the BRFSS survey, including all Waksberg stage one (prescreening) numbers used and numbers pre-identified as non-working or business which were not called but would have been called had they not been so identified. If, for unavoidable reasons, it is not possible to send a record for all numbers used in a survey, a count of numbers used, by disposition, will be sent monthly to BSB.
Comment: This provision would make outcome rates more nearly comparable across states. It would eliminate variations in outcome rates due to the exclusion of stage one numbers in Waksberg states and differences in prescreening practices among all states.
4. Commercial sample providers usually have many fields of information available about the sample telephone records that they generate. The data collection process generates many additional pieces of information for each sample record related to the data collection effort. BSB is committed to making as many of these fields available on data records as is practical given constraints of time and money.
Beginning with data year 1999, every state must, as applicable, include the following fields in each complete or incomplete data record sent to BSB. Specifications for these fields are contained in the BRFSS Data Layout:
In addition, states should provide as many of the following fields as practicable:
Comment: The fields identified in this section represent a major expansion of the design and methodological fields currently contained in BRFSS data. The inclusion of these fields will (1) provide more complete weighting information, (2) document the source of the sample, and (3)enable the creation of reports and research that can help improve the generation and use of sample. All of the fields except Number of Attempts and the code for Stage Two Waksberg numbers would be obtained from the sample provider.
This page last reviewed April 20, 2012[00_pgfooter2.htm]