`col_mean` using `survey::svymean`

[kinto-b]

If we use survey then we can support doing means for stratified samples. We'll want to keep the same API so something like

As a starting point, here's a rough function for doing cross tabs,

xtable <- function(.data, .row, .cols, .strata, .weight) {
  .data <- .data[!.data[[.row]] %in% NA, ] # Drop NA
  
  # Add binary variables for each level
  .rows <- c()
  for (val in unique(.data$a6)) {
    new_var <- paste0(".", .row, "__", val)
    .rows <- c(.rows, new_var)
    
    .data[[new_var]] <- .data[[.row]] %in% val
  }
  
  .design <- survey::svydesign(
    id = ~ 1, 
    strata = as.formula(paste("~", .strata)), 
    weights = as.formula(paste("~", .weight)), 
    data = .data
  )
  
  res <- survey::svyby(
    as.formula(paste("~", paste(.rows, collapse = "+"))), 
    as.formula(paste("~", paste(.cols, collapse = "+"))), 
    design = .design, 
    survey::svymean
  )
  
  # Only need TRUE cols
  res <- res[, !grepl(x=names(res), "FALSE$")]
  names(res) <- gsub(x=names(res), "TRUE$", "")
  
  t(res)
}

dat <- haven::read_sav("Z:/Reference information/CRSTabs/Example Data (SoCo ADA 16-17).sav")
xtable(dat, "a6", "year", "market", "wt_national")

[kinto-b]

It would be ideal if we could build on some way to do p-value adjustments for multiple comparisons