Possible Plot Ideas for Optmatch Objects

[jwbowers]

Hi Y'all,

This is just a prototype of what graphs of optmatch objects could look like (obviously nicer if using ggplot etc..)

## perhaps try this https://briatte.github.io/ggnet/#example-2-bipartite-network next time
library(igraph)
blah0 <- outer(fm0, fm0, FUN = function(x, y) {
  as.numeric(x == y)
})
blah1 <- outer(fm1, fm1, FUN = function(x, y) {
  as.numeric(x == y)
})
blah2 <- outer(fm2, fm2, FUN = function(x, y) {
  as.numeric(x == y)
})
par(mfrow = c(2, 2), mar = c(3, 3, 3, 1))
plot(graph_from_adjacency_matrix(blah0, mode = "undirected", diag = FALSE),
  vertex.color = c("white", "green")[meddat$nhTrt + 1], main = "Min Ctrls=0, Max Ctrls=Inf"
)
plot(graph_from_adjacency_matrix(blah1, mode = "undirected", diag = FALSE),
  vertex.color = c("white", "green")[meddat$nhTrt + 1], main = "Min Ctrls=1, Max Ctrls=Inf"
)
plot(graph_from_adjacency_matrix(blah2, mode = "undirected", diag = FALSE),
  vertex.color = c("white", "green")[meddat$nhTrt + 1], main = "Penalties,Min Ctrls=1, Max Ctrls=Inf"
)

[josherrickson]

Another idea I came up with.

Distance is created with mean, but offers other options. X-axis gives a good visual of how far from 1:1 a match is; y-axis can help identify poor matches.

Messy code below. One downside is that the match is needed for distance calculations; perhaps revisit our choice to store only a hash of the distance matrix in an optmatch object?

library(ggplot2)


plot.optmatch <- function(optm, match, distance_function = mean) {

  # Calculate matched distances and apply `distance_function` to them
  mtchdists <- matched.distances(optm, match)
  mtchdists <- as.data.frame(do.call(rbind,
                                     lapply(mtchdists, distance_function)))
  names(mtchdists) <- "dist"
  mtchdists$names <- row.names(mtchdists)

  # Calculate table of # treatment and # control
  txtctl <- tapply(names(optm), optm, function(x) x)
  txtctl <- lapply(txtctl, function(x) {
    as.numeric(x %in% row.names(match))
  })
  txtctl <- data.frame(txt = vapply(txtctl, sum, numeric(1)),
                       ctl = vapply(txtctl, function(x) sum(1-x), numeric(1)))
  txtctl$names <- row.names(txtctl)

  # Merge to single data set
  alldata <- merge(txtctl, mtchdists, by.x = "names")

  # Generate X position. 1:1 is at 0, 1:k is at 1, 2, etc, and j:1 is at -1, -2,
  # etc
  alldata$x <- alldata$ctl - 1
  revdir <- alldata$ctl < alldata$txt
  alldata$x[revdir] <- -1*(alldata$txt[revdir] - 1)

  # Breaks and labels for x axis
  brks <- c(-1*rev(seq_len(max(alldata$txt) - 1)),
            0,
            seq_len(max(alldata$ctl) - 1))
  lbls <- c(paste0(rev(seq_len(max(alldata$txt))), ":1"),
            paste0("1:", seq_len(max(alldata$ctl)))[-1])

  ggplot(alldata, aes(x = x, y = dist)) +
    geom_vline(aes(xintercept = 0), size = 2, alpha = .2) +
    geom_point() + 
    scale_x_continuous(breaks = brks,
                       labels = lbls, name = "Match Sizes") +
    scale_y_continuous(name = "Distance in Match")

}

match <- match_on(ct ~ cost, data = nuclearplants)
fm <- fullmatch(match, data = nuclearplants)

plot(fm, match, mean)
plot(fm, match, max)

[jwbowers]

Hi All,

I've implemented drafts of two visualization ideas using ggplot2/ggraph to the jake_visualization branch. I'm calling this a feature since this is not urgent. I just didn't want to forget these ideas --- which I used in our ICPSR class this summer.

In covariate_matched_dist_plot we have a violin plot of within-set absolute differences:

fm1 <- fullmatch(pr ~ t1 + t2, data = nuclearplants)

nuclearplants$fm1 <- fm1

## The below description shows the following about differences within set on t2:
## half of the sets differ by less than 7
## 90% of the sets differ by less than 10.8
## The set with the largest difference had a difference of 13.
quantile(abs(t2_diffs$cov_diffs), c(0, .1, .25, .5, .9, 1))

## This next is a plot visualizing that distribution of within-stratum
## absolute differences

g0 <- covariate_matched_dist_plot(
  strata_indicator = fm1, trt = nuclearplants$pr,
  covariate = nuclearplants$t2
)
print(g0)

And in stratification_graph we have a visualization of the sets using nodes and edges:

g0 <- stratification_graph(fm1, node_color = nuclearplants$pr) + ggtitle("Full MH Stratification")
g0

[benthestatistician]

It looks like covariate_matched_distances_plot() is showing signed within set differences, as opposed to absolute distances. Unless I'm missing something.

I suppose signed distances would be better here, but I'm not sure.

How about including on the plot an indication of the diff of means w/o matching, as an anchor point?