National Survey of Family Growth (NSFG)

The principal survey to measure reproductive behavior in the United States population.

Multiple tables with one row per respondent for the female and male tables, then a separate table with one row per pregnancy.
A complex sample survey designed to generalize to the 15-49 year old population of the United States, by gender.
Released every couple of years since 1973.
Administered by the Centers for Disease Control and Prevention.

Please skim before you begin:

Sample Design Documentation
Wikipedia Entry
This human-composed haiku or a bouquet of artificial intelligence-generated limericks

# family structure
# questions cuz radar fails at
# storks with bassinets

Download, Import, Preparation

library(SAScii)
library(readr)

dat_url <-
    "https://ftp.cdc.gov/pub/Health_Statistics/NCHS/Datasets/NSFG/2017_2019_FemRespData.dat"

sas_url <-
    file.path( dirname( dat_url ) , "sas/2017_2019_FemRespSetup.sas" )
    
sas_positions <-
    parse.SAScii( sas_url )

sas_positions[ , 'varname' ] <-
    tolower( sas_positions[ , 'varname' ] )

sas_positions[ , 'column_types' ] <-
        ifelse( sas_positions[ , 'char' ] , "c" , "d" )

nsfg_tbl <-
    read_fwf(
        dat_url ,
        fwf_widths( 
            abs( sas_positions[ , 'width' ] ) , 
            col_names = sas_positions[ , 'varname' ] 
        ) ,
        col_types = paste0( sas_positions[ , 'column_types' ] , collapse = "" ) ,
        na = c( "" , "." )
    )
    
nsfg_df <- data.frame( nsfg_tbl )

Save locally

Save the object at any point:

# nsfg_fn <- file.path( path.expand( "~" ) , "NSFG" , "this_file.rds" )
# saveRDS( nsfg_df , file = nsfg_fn , compress = FALSE )

Load the same object:

# nsfg_df <- readRDS( nsfg_fn )

Survey Design Definition

Construct a complex sample survey design:

library(survey)

options( survey.lonely.psu = "adjust" )

nsfg_design <- 
    svydesign( 
        id = ~ secu , 
        strata = ~ sest , 
        data = nsfg_df , 
        weights = ~ wgt2017_2019 , 
        nest = TRUE 
    )

Variable Recoding

Add new columns to the data set:

nsfg_design <- 
    update( 
        nsfg_design , 

        one = 1 ,
        
        birth_control_pill = as.numeric( constat1 == 6 ) ,
        
        age_categories = 
            factor( findInterval( ager , c( 15 , 20 , 25 , 30 , 35 , 40 ) ) ,
                labels = c( '15-19' , '20-24' , '25-29' , '30-34' , '35-39' , '40-49' ) ) ,
        
        marstat =
            factor( marstat , levels = c( 1:6 , 8:9 ) ,
                labels = c(
                    "Married to a person of the opposite sex" ,
                    "Not married but living together with a partner of the opposite sex" ,
                    "Widowed" ,
                    "Divorced or annulled" ,
                    "Separated, because you and your spouse are not getting along" ,
                    "Never been married" ,
                    "Refused" ,
                    "Don't know" )
            )
    )

Analysis Examples with the `survey` library

Unweighted Counts

Count the unweighted number of records in the survey sample, overall and by groups:

sum( weights( nsfg_design , "sampling" ) != 0 )

svyby( ~ one , ~ age_categories , nsfg_design , unwtd.count )

Weighted Counts

Count the weighted size of the generalizable population, overall and by groups:

svytotal( ~ one , nsfg_design )

svyby( ~ one , ~ age_categories , nsfg_design , svytotal )

Descriptive Statistics

Calculate the mean (average) of a linear variable, overall and by groups:

svymean( ~ pregnum , nsfg_design , na.rm = TRUE )

svyby( ~ pregnum , ~ age_categories , nsfg_design , svymean , na.rm = TRUE )

Calculate the distribution of a categorical variable, overall and by groups:

svymean( ~ marstat , nsfg_design )

svyby( ~ marstat , ~ age_categories , nsfg_design , svymean )

Calculate the sum of a linear variable, overall and by groups:

svytotal( ~ pregnum , nsfg_design , na.rm = TRUE )

svyby( ~ pregnum , ~ age_categories , nsfg_design , svytotal , na.rm = TRUE )

Calculate the weighted sum of a categorical variable, overall and by groups:

svytotal( ~ marstat , nsfg_design )

svyby( ~ marstat , ~ age_categories , nsfg_design , svytotal )

Calculate the median (50th percentile) of a linear variable, overall and by groups:

svyquantile( ~ pregnum , nsfg_design , 0.5 , na.rm = TRUE )

svyby( 
    ~ pregnum , 
    ~ age_categories , 
    nsfg_design , 
    svyquantile , 
    0.5 ,
    ci = TRUE , na.rm = TRUE
)

Estimate a ratio:

svyratio( 
    numerator = ~ pregnum , 
    denominator = ~ lbpregs , 
    nsfg_design ,
    na.rm = TRUE
)

Subsetting

Restrict the survey design to ever cohabited:

sub_nsfg_design <- subset( nsfg_design , timescoh > 0 )

Calculate the mean (average) of this subset:

svymean( ~ pregnum , sub_nsfg_design , na.rm = TRUE )

Measures of Uncertainty

Extract the coefficient, standard error, confidence interval, and coefficient of variation from any descriptive statistics function result, overall and by groups:

this_result <- svymean( ~ pregnum , nsfg_design , na.rm = TRUE )

coef( this_result )
SE( this_result )
confint( this_result )
cv( this_result )

grouped_result <-
    svyby( 
        ~ pregnum , 
        ~ age_categories , 
        nsfg_design , 
        svymean ,
        na.rm = TRUE 
    )
    
coef( grouped_result )
SE( grouped_result )
confint( grouped_result )
cv( grouped_result )

Calculate the degrees of freedom of any survey design object:

degf( nsfg_design )

Calculate the complex sample survey-adjusted variance of any statistic:

svyvar( ~ pregnum , nsfg_design , na.rm = TRUE )

Include the complex sample design effect in the result for a specific statistic:

# SRS without replacement
svymean( ~ pregnum , nsfg_design , na.rm = TRUE , deff = TRUE )

# SRS with replacement
svymean( ~ pregnum , nsfg_design , na.rm = TRUE , deff = "replace" )

Compute confidence intervals for proportions using methods that may be more accurate near 0 and 1. See ?svyciprop for alternatives:

svyciprop( ~ birth_control_pill , nsfg_design ,
    method = "likelihood" , na.rm = TRUE )

Regression Models and Tests of Association

Perform a design-based t-test:

svyttest( pregnum ~ birth_control_pill , nsfg_design )

Perform a chi-squared test of association for survey data:

svychisq( 
    ~ birth_control_pill + marstat , 
    nsfg_design 
)

Perform a survey-weighted generalized linear model:

glm_result <- 
    svyglm( 
        pregnum ~ birth_control_pill + marstat , 
        nsfg_design 
    )

summary( glm_result )

Replication Example

This example matches the Variance Estimates for Percentages using SAS (9.4) and STATA (14):

Match the sum of the weights:

result <- svytotal( ~ one , nsfg_design )

stopifnot( round( coef( result ) , 0 ) == 72671926 )

stopifnot( round( SE( result ) , 0 ) == 3521465 )

Match row percentages of women currently using the pill by age:

row_percents <- c( 19.5112 , 23.7833 , 19.6916 , 15.2800 , 6.4965 , 6.5215 )

std_err_row_percents <- c( 1.8670 , 2.1713 , 2.2773 , 1.7551 , 0.9895 , 1.0029 )

results <- svyby( ~ birth_control_pill , ~ age_categories , nsfg_design , svymean )

stopifnot( all( round( coef( results ) * 100 , 4 ) == row_percents ) )

stopifnot( all( round( SE( results ) * 100 , 4 ) == std_err_row_percents ) )

Analysis Examples with `srvyr`

The R srvyr library calculates summary statistics from survey data, such as the mean, total or quantile using dplyr-like syntax. srvyr allows for the use of many verbs, such as summarize, group_by, and mutate, the convenience of pipe-able functions, the tidyverse style of non-standard evaluation and more consistent return types than the survey package. This vignette details the available features. As a starting point for NSFG users, this code replicates previously-presented examples:

library(srvyr)
nsfg_srvyr_design <- as_survey( nsfg_design )