optmatch

http://optmat.ch/
Admin
optmatch is an R package that implements optimal full matching algorithms for bipartite matching...
Tokens:2,204
Snippets:39
Trust Score:2.4
Update:1 month ago
Show doc for...
Tokens:
Raw
### Install and load devtools

Source: http://optmat.ch

Prepares the environment for installing development packages.

```R
> install.packages("devtools")
> library("devtools")
```

--------------------------------

### Install and load optmatch

Source: http://optmat.ch

Install the package from CRAN and load it into the R session.

```R
> install.packages("optmatch")
> library("optmatch")
```

--------------------------------

### Install development version

Source: http://optmat.ch

Installs the master branch of Optmatch from GitHub while in dev_mode.

```R
d> install_github("markmfredrickson/optmatch")
```

--------------------------------

### Install local binary package

Source: http://optmat.ch

Installs a locally built binary version of the optmatch package.

```r
install.packages("../optmatch_VERSION.tgz", repo=NULL)
```

--------------------------------

### Install package dependencies

Source: http://optmat.ch

Ensures all required dependencies for the optmatch package are installed.

```r
devtools::install_deps(dependencies = TRUE)
```

--------------------------------

### Development commands via Make

Source: http://optmat.ch

Common command-line tasks for managing the package build and test lifecycle.

```bash
make test
```

```bash
make document
```

```bash
make interactive
```

```bash
make interactive-emacs
```

```bash
make check
```

```bash
make build
```

```bash
make vignette
```

```bash
make clean
```

```bash
make release
```

--------------------------------

### Load Optmatch in dev_mode

Source: http://optmat.ch

Loads the development version of the package.

```R
 d> library(optmatch)
```

--------------------------------

### Generate simulated data for matching

Source: http://optmat.ch

Create a dataset with treated and control groups based on covariates to demonstrate matching workflows.

```R
set.seed(20120111) # set this to get the exact same answers as I do
n <- 26 # chosen so we can divide the alphabet in half
W <- data.frame(w1 = rbeta(n, 4, 2), w2 = rbinom(n, 1, p = .33))

# nature assigns to treatment
tmp <- numeric(n)
tmp[sample(1:n, prob = W$w1^(1 + W$w2), size = n/2)] <- 1
W$z <- tmp

# for convenience, let's give the treated units capital letter names
tmp <- character(n)
tmp[W$z == 1] <- LETTERS[1:(n/2)]
tmp[W$z == 0] <- letters[(26 - n/2 + 1):26]
rownames(W) <- tmp
```

--------------------------------

### Print match results

Source: http://optmat.ch

Displays the match object with units grouped by factor level.

```R
print(mahal.match, grouped = T)
```

--------------------------------

### Toggle dev_mode

Source: http://optmat.ch

Activates or deactivates the development mode for package management.

```R
> dev_mode()
d>
```

```R
d> dev_mode()
>
```

--------------------------------

### Revert to CRAN version

Source: http://optmat.ch

Removes the local development version and reinstalls the package from CRAN.

```r
remove.packages("optmatch") install.packages("optmatch")
```

--------------------------------

### Generate pair matches

Source: http://optmat.ch

Creates a one-to-one match using the Mahalanobis distance matrix.

```R
mahal.match <- pairmatch(distances$mahal, data = W)
```

--------------------------------

### Evaluate distance specifications

Source: http://optmat.ch

Applies fullmatch to a list of distance specifications.

```R
matches <- lapply(distances, function(x) { fullmatch(x, data = W) })
```

--------------------------------

### Inspect group balance

Source: http://optmat.ch

Examine the distribution of covariates across treatment groups using tables and density plots.

```R
table(W$w2, W$z)
library(lattice) ; densityplot(W$w1, groups = W$z)
```

--------------------------------

### Append match results to data

Source: http://optmat.ch

Joins the match factor back to the original data frame.

```R
W.matched <- cbind(W, matches = mahal.match)
```

--------------------------------

### Combine distance metrics

Source: http://optmat.ch

Aggregates multiple distance matrices into a single metric using standard arithmetic operations.

```R
distances$all <- with(distances, euclid + mahal + propensity)
```

--------------------------------

### Calculate Euclidean distances

Source: http://optmat.ch

Generate a distance matrix for matching based on Euclidean distance between covariates.

```R
distances <- list()
distances$euclid <- match_on(z ~ w1 + w2, data = W, method = "euclidean")
```

--------------------------------

### Compute Mahalanobis distance

Source: http://optmat.ch

Calculates Mahalanobis distance using the match_on function with a formula interface.

```R
distances$mahal <- match_on(z ~ w1 + w2, data = W)
```

--------------------------------

### Stitch distance matrices

Source: http://optmat.ch

Combines different distance specifications using cbind to merge results from subsets of data.

```R
W.n.to.s <- W[c(LETTERS[1:13], letters[14:19]),]
W.t.to.z <- W[c(LETTERS[1:13], letters[20:26]),]
mahal.n.to.s <- match_on(z ~ w1 + w2, data = W.n.to.s)
ps.t.to.z <- match_on(glm(z ~ w1 + w2, data = W.t.to.z, family = binomial()))
distances$combined <- cbind(mahal.n.to.s, ps.t.to.z)
```

--------------------------------

### Compute Mahalanobis distance with caliper

Source: http://optmat.ch

Calculates Mahalanobis distance restricted to units differing by less than one on the propensity score.

```R
distances$mahal.trimmed <- match_on(z ~ w1 + w2, data = W,
  within = match_on(propensity.model, caliper = 1))
```

--------------------------------

### Apply a caliper to distances

Source: http://optmat.ch

Trims a distance matrix by setting values exceeding a specified threshold to infinity.

```R
distances$median.caliper <- caliper(distances$all, median(distances$all))
distances$all.trimmed <- with(distances, all + median.caliper)
```

--------------------------------

### Compute propensity score distance

Source: http://optmat.ch

Uses a GLM model to estimate propensity scores and compute distances based on the model output.

```R
propensity.model <- glm(z ~ w1 + w2, data = W, family =
  binomial())
distances$propensity <- match_on(propensity.model)
```

--------------------------------

### Filter distance computation with exact matching

Source: http://optmat.ch

Uses the within argument to restrict distance calculations to valid pairs defined by an exact match structure.

```R
tmp <- exactMatch(z ~ w2, data = W)
distances$exact <- match_on(z ~ w1, data = W, within = tmp)
```
Tokens:
Raw