Description

This help page documents the options that control the behaviour of the survey package.

Details

All the options for the survey package have names beginning with "survey". Four of them control standard error estimation.

options("survey.replicates.mse") controls the default in svrepdesign and as.svrepdesign for computing variances. When options("survey.replicates.mse") is TRUE, the default is to create replicate weight designs that compute variances centered at the point estimate, rather than at the mean of the replicates. The option can be overridden by specifying the mse argument explicitly in svrepdesign and as.svrepdesign. The default is FALSE.

When options("survey.ultimate.cluster") is TRUE, standard error estimation is based on independence of PSUs at the first stage of sampling, without using any information about subsequent stages. When FALSE, finite population corrections and variances are estimated recursively. See svyrecvar for more information. This option makes no difference unless first-stage finite population corrections are specified, in which case setting the option to TRUE gives the wrong answer for a multistage study. The only reason to use TRUE is for compatibility with other software that gives the wrong answer.

Handling of strata with a single PSU that are not certainty PSUs is controlled by options("survey.lonely.psu"). The default setting is "fail", which gives an error. Use "remove" to ignore that PSU for variance computation, "adjust" to center the stratum at the population mean rather than the stratum mean, and "average" to replace the variance contribution of the stratum by the average variance contribution across strata. As of version 3.4-2 as.svrepdesign also uses this option.

The variance formulas for domain estimation give well-defined, positive results when a stratum contains only one PSU with observations in the domain, but are not unbiased. If options("survey.adjust.domain.lonely") is TRUE and options("survey.lonely.psu") is "average" or "adjust" the same adjustment for lonely PSUs will be used within a domain. Note that this adjustment is not available for replicate-weight designs, nor (currently) for raked, post-stratified, or calibrated designs.

The fourth option is options("survey.want.obsolete"). This controls the warnings about using the deprecated pre-2.9.0 survey design objects.

The behaviour of replicate-weight designs for self-representing strata is controlled by options("survey.drop.replicates"). When TRUE, various optimizations are used that take advantage of the fact that these strata do not contribute to the variance. The only reason ever to use FALSE is if there is a bug in the code for these optimizations.

The fifth option controls the use of multiple processors with the multicore package. This option should not affect the values computed by any of the survey functions. If TRUE, all functions that are able to use multiple processors will do so by default. Using multiple processors may speed up calculations, but need not, especially if the computer is short on memory. The best strategy is probably to experiment with explicitly requesting multicore=TRUE in functions that support it, to see if there is an increase in speed before setting the global option.