Measuring income inequality using the R survey package

Question

I'm working with American Community Survey microdata using the survey package, and am hoping to calculate some basic income inequality statistics. I've set up the following as my design:

testsurv <- svrepdesign(data=test, repweights = test[,8:87], weights = test$HHWT, 
                   combined.weights=TRUE, type = "Fay", rho = 0.5,scale=4/80, 
                   rscales = rep(1, 80), mse=TRUE)

From that, I'd like to calculate gini coefficients by year, as well as quantile ratios of income, also by year. Generating the quantiles and the related errors is straightforward using svyby and svyquantile:

quants <- svyby(~INCOME, ~YEAR, testsurvey, svyquantile, 
              quantiles=c(0.9, 0.75, 0.5, 0.25, 0.1), keep.var=TRUE)

That brings me to my first question: How do I calculate the the standard errors for ratios of income quantiles (e.g. 90/10) if I have the replicate-weight-based errors for each quantile? I tried using svyratio but that's for the ratios of entire variables, not for selected observations within variables.

Second question: Is there a way to calculate the gini coefficient (with replicate-based errors) within survey using existing functions like gini from reldist? I tried using withReplicates but it didn't work well, maybe because gini orders its arguments as variable, then weights, but the instructions for withReplicates specify the opposite order. I tried both ways but neither worked. For example, this, where HHWT is the sample weights:

> withReplicates(testsurv, gini(~HHWT, ~INCOME))

That yields the following error message:

Error in sum(weights) : invalid 'type' (language) of argument
In addition: Warning message:
In is.na(x) : is.na() applied to non-(list or vector) of type 'language'

Answer 1

use the R convey package. this is not yet available on CRAN but you can install it quickly with

devtools::install_github("djalmapessoa/convey")

for the ratio of 90th to 10th, use the ?svyqsr function and set alpha= to 0.1 because it defaults to 80th and 20th

for the gini coefficient, use the ?svygini function

these should both be straightforward computations so long as you have the acs replicate-weighted survey design. be sure to use the convey_prep function immediately after the svrepdesign call!