diff --git a/NAMESPACE b/NAMESPACE
index e8b95d7..5d57361 100644
--- a/NAMESPACE
+++ b/NAMESPACE
@@ -1,14 +1,22 @@
 # Generated by roxygen2: do not edit by hand
 
 S3method("[",causal_tbl)
+S3method("[",indices)
 S3method("names<-",causal_tbl)
 S3method(ctl_new_pillar,causal_tbl)
+S3method(format,indices)
 S3method(tbl_format_header,causal_tbl)
 S3method(tbl_format_setup,causal_tbl)
 S3method(tbl_sum,causal_tbl)
+S3method(vec_cast,indices.list)
+S3method(vec_cast,list.indices)
+S3method(vec_ptype2,indices.list)
+S3method(vec_ptype2,list.indices)
+S3method(vec_ptype_abbr,indices)
 export("causal_cols<-")
 export(add_causal_col)
 export(as_causal_tbl)
+export(as_indices)
 export(causal_cols)
 export(causal_tbl)
 export(get_causal_col)
@@ -18,7 +26,11 @@ export(get_treatment)
 export(has_outcome)
 export(has_panel)
 export(has_treatment)
+export(indices)
+export(is_causal_tbl)
+export(is_indices)
 export(new_causal_tbl)
+export(new_indices)
 export(set_causal_col)
 export(set_outcome)
 export(set_panel)
@@ -32,3 +44,6 @@ importFrom(pillar,tbl_format_setup)
 importFrom(pillar,tbl_sum)
 importFrom(rlang,enquo)
 importFrom(rlang,enquos)
+importFrom(vctrs,vec_cast)
+importFrom(vctrs,vec_ptype2)
+importFrom(vctrs,vec_ptype_abbr)
diff --git a/R/causal_cols.R b/R/causal_cols.R
index 5656cdc..bccc2f3 100644
--- a/R/causal_cols.R
+++ b/R/causal_cols.R
@@ -231,6 +231,10 @@ multi_col_name <- function(expr, data, arg) {
         cli_abort("Must select a column for {.arg {arg}}", call=parent.frame())
     }
     out <- names(data)[idx]
-    names(out) = names(idx)
+    nms <- names(idx)
+    if (!all(nms == out)) {
+        nms[nms == out] = ""
+        names(out) <- nms
+    }
     out
 }
diff --git a/R/causal_tbl.R b/R/causal_tbl.R
index 00e4950..414a97b 100644
--- a/R/causal_tbl.R
+++ b/R/causal_tbl.R
@@ -122,10 +122,12 @@ reconstruct.causal_tbl <- function(data, old) {
 #' @return A `causal_tbl` object
 #'
 #' @examples
-#' causal_tbl(
+#' data <- causal_tbl(
 #'   milk_first = c(0, 1, 0, 1, 1, 0, 0, 1),
 #'   guess = c(0, 1, 0, 1, 1, 0, 0, 1)
 #' )
+#' is_causal_tbl(data)
+#' print(data)
 #'
 #' @export
 causal_tbl <- function(..., .outcome=NULL, .treatment=NULL) {
@@ -142,10 +144,10 @@ causal_tbl <- function(..., .outcome=NULL, .treatment=NULL) {
 
 
 #' @describeIn causal_tbl Coerce a data frame to a `causal_tbl`
-#' @param x A data frame to be coerced
+#' @param x A data frame to be checked or coerced
 #' @export
 as_causal_tbl <- function(x) {
-    if (inherits(x, "causal_tbl")) {
+    if (is_causal_tbl(x)) {
         x
     } else if (is.data.frame(x)) {
         reconstruct.causal_tbl(x)
@@ -154,8 +156,14 @@ as_causal_tbl <- function(x) {
     }
 }
 
+#' @describeIn causal_tbl Return `TRUE` if a data frame is a `causal_tbl`
+#' @export
+is_causal_tbl <- function(x) {
+    inherits(x, "causal_tbl")
+}
+
 assert_causal_tbl <- function(data, arg) {
-    if (!inherits(data, "causal_tbl")) {
+    if (!is_causal_tbl(data)) {
         cli_abort("{.arg {deparse(substitute(data))}} must be a {.cls causal_tbl}.",
                   call=parent.frame())
     }
diff --git a/R/indices.R b/R/indices.R
new file mode 100644
index 0000000..b199c98
--- /dev/null
+++ b/R/indices.R
@@ -0,0 +1,90 @@
+#' @describeIn indices Construct an `indices` list with minimal checks
+#' @export
+new_indices <- function(x = list()) {
+    if (!is.list(x)) {
+        cli_abort("{.arg x} must be a list.")
+    }
+
+    vctrs::new_vctr(x, class="indices")
+}
+
+#' Construct a list of indices
+#'
+#' An `indices` list is a list with integer vector entries which represent
+#' indices in some other object (like a data frame). Generically it can
+#' be used to represent a graph structure, such as an interference network
+#' or a collection of matched objects.
+#' The main feature of `indices` is that the index references are preserved
+#' through slicing and reordering. Indices that no longer refer to elements
+#' (because of subsetting) are set to NA.
+#'
+#' @param x
+#' * For `indices()` and `new_indices()`: A list of indices
+#' * For `is_indices()`: An object to test
+#' * For `as_indices()`: An object to coerce
+#'
+#' @returns An `indices` object.
+#'
+#' @examples
+#' idx <- indices(list(2, c(1, NA, 3), 2))
+#' print(idx)
+#' idx[1:2] # subsetting
+#' idx[c(2, 1, 3)] # reordering
+#' @export
+indices <- function(x = list()) {
+    # convert each element to an integer
+    x <- relist(vctrs::vec_cast(unlist(x), integer(), x_arg="x"), x)
+    new_indices(x)
+}
+
+#' @describeIn indices Return `TRUE` if an object is an `indices` list
+#' @export
+is_indices <- function(x) {
+    inherits(x, "indices")
+}
+
+#' @describeIn indices Coerce an object to an `indices` list
+#' @export
+as_indices <- function(x) {
+    vctrs::vec_cast(x, new_indices())
+}
+
+
+# printing
+#' @export
+format.indices <- function(x, ...) {
+    vapply(vctrs::vec_data(x), format_index_line, "")
+}
+format_index_line <- function(y) {
+    paste0("(", paste0(formatC(y[!is.na(y)]), collapse=","), ")")
+}
+
+# vctrs -------------------------------------------------------------------
+
+#' @export
+`[.indices` <- function(x, i) {
+    lookup <- match(seq_along(x), i)
+    out <- NextMethod()
+    for (j in seq_along(out)) {
+        out[[j]] = lookup[out[[j]]]
+    }
+    out
+}
+
+#' @importFrom vctrs vec_ptype_abbr
+#' @method vec_ptype_abbr indices
+#' @export
+vec_ptype_abbr.indices <- function(x, ...) {
+    "idx" # nocov
+}
+
+#' @importFrom vctrs vec_ptype2
+#' @export
+vec_ptype2.list.indices <- function(x, y, ...) list()
+#' @export
+vec_ptype2.indices.list <- function(x, y, ...) list()
+#' @importFrom vctrs vec_cast
+#' @export
+vec_cast.list.indices <- function(x, y, ...) as.list(x)
+#' @export
+vec_cast.indices.list <- function(x, y, ...) indices(x)
diff --git a/man/causal_tbl.Rd b/man/causal_tbl.Rd
index f4c3ec4..2f69c00 100644
--- a/man/causal_tbl.Rd
+++ b/man/causal_tbl.Rd
@@ -4,6 +4,7 @@
 \alias{new_causal_tbl}
 \alias{causal_tbl}
 \alias{as_causal_tbl}
+\alias{is_causal_tbl}
 \title{Build a causal data frame}
 \usage{
 new_causal_tbl(..., .outcome = NULL, .treatment = NULL)
@@ -11,6 +12,8 @@ new_causal_tbl(..., .outcome = NULL, .treatment = NULL)
 causal_tbl(..., .outcome = NULL, .treatment = NULL)
 
 as_causal_tbl(x)
+
+is_causal_tbl(x)
 }
 \arguments{
 \item{...}{passed on to \code{\link[=tibble]{tibble()}}}
@@ -21,7 +24,7 @@ Can be set later with \code{\link[=set_outcome]{set_outcome()}}.}
 \item{.treatment}{the column containing the treatment variable (tidy-selected).
 Can be set later with \code{\link[=set_treatment]{set_treatment()}}.}
 
-\item{x}{A data frame to be coerced}
+\item{x}{A data frame to be checked or coerced}
 }
 \value{
 A \code{causal_tbl} object
@@ -45,6 +48,8 @@ in this package, like \code{\link[=get_outcome]{get_outcome()}} and \code{\link[
 
 \item \code{as_causal_tbl()}: Coerce a data frame to a \code{causal_tbl}
 
+\item \code{is_causal_tbl()}: Return \code{TRUE} if a data frame is a \code{causal_tbl}
+
 }}
 \section{Internal structure}{
 The \link{causal_cols} attribute is considered mostly internal, and end users
@@ -78,9 +83,11 @@ that a \code{causal_tbl} has the necessary columns set via helpers like
 }
 
 \examples{
-causal_tbl(
+data <- causal_tbl(
   milk_first = c(0, 1, 0, 1, 1, 0, 0, 1),
   guess = c(0, 1, 0, 1, 1, 0, 0, 1)
 )
+is_causal_tbl(data)
+print(data)
 
 }
diff --git a/man/indices.Rd b/man/indices.Rd
new file mode 100644
index 0000000..149c5fe
--- /dev/null
+++ b/man/indices.Rd
@@ -0,0 +1,51 @@
+% Generated by roxygen2: do not edit by hand
+% Please edit documentation in R/indices.R
+\name{new_indices}
+\alias{new_indices}
+\alias{indices}
+\alias{is_indices}
+\alias{as_indices}
+\title{Construct a list of indices}
+\usage{
+new_indices(x = list())
+
+indices(x = list())
+
+is_indices(x)
+
+as_indices(x)
+}
+\arguments{
+\item{x}{\itemize{
+\item For \code{indices()} and \code{new_indices()}: A list of indices
+\item For \code{is_indices()}: An object to test
+\item For \code{as_indices()}: An object to coerce
+}}
+}
+\value{
+An \code{indices} object.
+}
+\description{
+An \code{indices} list is a list with integer vector entries which represent
+indices in some other object (like a data frame). Generically it can
+be used to represent a graph structure, such as an interference network
+or a collection of matched objects.
+The main feature of \code{indices} is that the index references are preserved
+through slicing and reordering. Indices that no longer refer to elements
+(because of subsetting) are set to NA.
+}
+\section{Functions}{
+\itemize{
+\item \code{new_indices()}: Construct an \code{indices} list with minimal checks
+
+\item \code{is_indices()}: Return \code{TRUE} if an object is an \code{indices} list
+
+\item \code{as_indices()}: Coerce an object to an \code{indices} list
+
+}}
+\examples{
+idx <- indices(list(2, c(1, NA, 3), 2))
+print(idx)
+idx[1:2] # subsetting
+idx[c(2, 1, 3)] # reordering
+}
diff --git a/tests/testthat/_snaps/indices.md b/tests/testthat/_snaps/indices.md
new file mode 100644
index 0000000..923b96e
--- /dev/null
+++ b/tests/testthat/_snaps/indices.md
@@ -0,0 +1,18 @@
+# indices printing
+
+    Code
+      print(idx)
+    Output
+      <indices[3]>
+      [1] (2)   (1,3) (2)  
+
+---
+
+    Code
+      str(idx)
+    Output
+      idx [1:3] 
+      $ : int 2
+      $ : int [1:3] 1 NA 3
+      $ : int 2
+
diff --git a/tests/testthat/test-indices.R b/tests/testthat/test-indices.R
new file mode 100644
index 0000000..83375bb
--- /dev/null
+++ b/tests/testthat/test-indices.R
@@ -0,0 +1,36 @@
+test_that("indices constructor", {
+    idx <- indices(list(2, c(1, NA, 3), 2))
+    expect_s3_class(idx, "indices")
+    expect_type(idx, "list")
+    expect_true(is_indices(idx))
+
+    expect_error(indices(list("a")), "character")
+    expect_error(indices(5), "must be a list")
+})
+
+test_that("indices conversion", {
+    idx <- indices(list(2, c(1, NA, 3), 2))
+
+    expect_s3_class(as_indices(as.list(idx)), "indices")
+    expect_type(as.list(idx), "list")
+    expect_type(c(idx, list()), "list")
+    expect_type(c(list(), idx), "list")
+})
+
+test_that("indices slicing", {
+    idx <- indices(list(2, c(1, NA, 3), 2))
+
+    expect_equal(idx[1:3], idx)
+    expect_equal(idx[1:2],
+                 indices(list(2, c(1, NA, NA))))
+    expect_equal(idx[2:1],
+                 indices(list(c(2, NA, NA), 1)))
+})
+
+
+test_that("indices printing", {
+    idx <- indices(list(2, c(1, NA, 3), 2))
+
+    expect_snapshot(print(idx))
+    expect_snapshot(str(idx))
+})