convert weights to psw class via vctrs

[malcolmbarrett]

This PR adds the psw object type for managing weights. A key benefit here is tracking the estimand and transformations of the weights.

wt_ate() and friends do these behind the scenes so most users won't need these but they are exported for expansion purposes

library(propensity)
library(tibble)

# Example 1: Creating and inspecting psw objects
wts <- psw(c(0.5, 1.0, 1.5, 2.0), estimand = "ate")
wts
#> <psw{estimand = ate}[4]>
#> [1] 0.5 1.0 1.5 2.0
is_psw(wts)
#> [1] TRUE
estimand(wts)
#> [1] "ate"

# Example 2: Arithmetic operations with psw
result <- wts * 2
result
#> <psw{estimand = ate}[4]>
#> [1] 1 2 3 4

wts2 <- psw(c(0.3, 0.7, 1.1, 1.5), estimand = "cens")
combined <- wts * wts2
combined
#> <psw{estimand = ate * cens}[4]>
#> [1] 0.15 0.70 1.65 3.00

# Example 3: Math operations with psw
sqrt_wts <- sqrt(wts)
sqrt_wts
#> [1] 0.7071068 1.0000000 1.2247449 1.4142136

# Example 4: Using psw with tibble
data <- tibble(
  y = c(2, 4, 6, 8),
  x = c(0, 1, 0, 1),
  wts = wts
)

data$weighted_y <- data$y * as.double(data$wts)
data$weighted_mean <- sum(data$weighted_y) / sum(data$wts)
data
#> # A tibble: 4 × 5
#>       y     x        wts weighted_y weighted_mean
#>   <dbl> <dbl> <psw{ate}>      <dbl>         <dbl>
#> 1     2     0        0.5          1             6
#> 2     4     1        1.0          4             6
#> 3     6     0        1.5          9             6
#> 4     8     1        2.0         16             6

# Example 5: Complex expression with psw
y <- c(2, 4, 6, 8)
x <- c(0, 1, 0, 1)

term1 <- sum(y * x * wts) / sum(x * wts)
term2 <- sum(y * (1 - x) * wts) / sum((1 - x) * wts)

term1
#> [1] 6.666667
term2
#> [1] 5

^{Created on 2025-01-11 with reprex v2.1.1}

If a model is fit with psw weights, we can tell what type of weights they are

# model code omitted 
model.frame(outcome_mod) |>
   model.weights() |>
   estimand()
#> [1] "ate"

[malcolmbarrett]

I rendered the book with this version of propensity to make sure it all still worked