Survey and Correlated Data
Stata’s svy: prefix now works with
- Cox proportional hazards regression (stcox)
- Parametric hazard and accelerated time survival regression (streg)
Twenty-six other commands also now support estimation with survey data.
You just declare the survey design for your data by using svyset, and then declare your data to be survival-time data by using stset. Here’s an example:
| . webuse nhefs . svyset psu2 [pw = swgt2], strata(strata2) . stset age_lung_cancer if age_lung_cancer < . [pw = swgt2], fail(lung_cancer) . svy: stcox former_smoker smoker male urban1 rural |
We could just as easily have fitted a parametric survival regression model simply by replacing svy:stcox with svy:streg.
Here’s a complete list of what’s new in statistics(survey)- Stata’s svy: prefix now works with 48 estimators, 27 more than previously. Other commands with which svy: now works include
|
bivariate probit regression | |
| clogit | conditional (fixed effects) logistic regression | |
| cloglog | complementary log-log regression | |
| cnreg | censored-normal regression | |
| cnsreg | constrained linear regression | |
| glm | generalized linear models | |
| hetprob | heteroskedastic probit regression | |
| ivregress | instrumental-variable regression | |
| ivtobit | probit model with endogenous regressors | |
| ivprobit | tobit model with endogenous regressors | |
| mprobit | multinomial probit regression | |
| nl | nonlinear least-squares estimation | |
| scobit | skewed logistic regression | |
| slogit | stereotype logistic regression | |
| stcox | Cox proportional hazards regression | |
| streg | parametric survival regression (five estimators) | |
| tobit | tobit regression | |
| treatreg | treatment-effects model | |
| truncreg | truncated regression | |
| zinb | zero-inflated negative binomial regression | |
| zip | zero-inflated Poisson regression | |
| ztnb | zero-truncated negative binomial regression | |
| ztp | zero-truncated Poisson regression |
See [SVY] svy estimation.
- svy: prefix now calculates the linearized variance estimator two to 100 times faster, the larger multiplier applying to large datasets with many sampling units; see [SVY] svy.
- svy: mean, svy: proportion, svy: ratio, and svy: total are considerably faster when the over() option identifies many subpopulations.
- svy:, svy: mean, svy: proportion, svy: ratio, and svy: total now take advantage of multiple processors in Stata/MP, making them even faster.
- Concerning svyset,
- New option singleunit(method) provides three methods for handling strata with one sampling unit. If not specified, the default in such cases is to report standard errors as missing value.
- New option fay(#) specifies that Fay’s adjustment be made to the BRR weights.
- New option singleunit(method) provides three methods for handling strata with one sampling unit. If not specified, the default in such cases is to report standard errors as missing value.
See [SVY] svyset.
- estat has two new subcommands for use with svy estimation results:
- estat sd, used after svy: mean, reports subpopulation standard deviations.
- estat strata reports the number of singleton and certainty strata within each sampling stage.
- estat sd, used after svy: mean, reports subpopulation standard deviations.
-
See [SVY] estat.
- svy: tabulate now allows string variables. See [SVY] svy: tabulate oneway and [SVY] svy: tabulate twoway.
- Existing command svydes has been renamed svydescribe; svydes continues to work.
svydescribe now puts missing values in the generate(newvar) variable for observations outside the specified estimation sample. Previously, the variable would contain a zero for observations outside the estimation sample.
See [SVY] svydescribe.
- The [SVY] manual has been reorganized. Stata’s survey estimation commands are now documented in [SVY] svy estimation. All model-specific information is now documented in the manual entry for the corresponding estimation command.