Merge pull request #182 from DeclareDesign/ls/list_names

Rename all list output to match tidy; remove any mention of margins

.Rbuildignore

@@ -18,7 +18,7 @@ update_repo.R
 ^_pkgdown\.yml$
 ^tests/testthat/mtcars\.csv$
 ^tests/testthat/run-stata-models\.do$
-^tests/testthat/test-broom\.R$
+^tests/testthat/test-zzzzzzbroom\.R$
 ^estimatr_.*\.tar\.gz$
 ^tests/testthat/test-texreg\.R$
 ^cran-comments\.md$

DESCRIPTION

@@ -28,8 +28,8 @@ Suggests:
     RcppEigen,
     fabricatr (>= 0.3),
     randomizr (>= 0.9),
-    margins,
-    prediction,
     AER
-Enhances: broom, texreg
+Enhances:
+    broom,
+    texreg
 Additional_Repositories: http://r.declaredesign.org

NAMESPACE

@@ -1,5 +1,10 @@
 # Generated by roxygen2: do not edit by hand
 
+S3method(coef,difference_in_means)
+S3method(coef,horvitz_thompson)
+S3method(coef,iv_robust)
+S3method(coef,lm_lin)
+S3method(coef,lm_robust)
 S3method(confint,difference_in_means)
 S3method(confint,horvitz_thompson)
 S3method(confint,iv_robust)

R/S3_coef.R

@@ -0,0 +1,18 @@
+coef_estimatr <- function(object, ...) {
+  object$estimate
+}
+
+#' @export
+coef.lm_robust <- coef_estimatr
+
+#' @export
+coef.iv_robust <- coef_estimatr
+
+#' @export
+coef.lm_lin <- coef_estimatr
+
+#' @export
+coef.horvitz_thompson <- coef_estimatr
+
+#' @export
+coef.difference_in_means <- coef_estimatr

R/S3_confint.R

@@ -53,20 +53,20 @@ get_ci_mat <- function(object, level, ttest = TRUE) {
   }
 
   cis <- cbind(
-    as.vector(object$ci_lower),
-    as.vector(object$ci_upper)
+    as.vector(object$ci.lower),
+    as.vector(object$ci.upper)
   )
 
-  if (is.matrix(object$ci_lower)) {
-    ny <- ncol(object$ci_lower)
-    p <- nrow(object$ci_lower)
+  if (is.matrix(object$ci.lower)) {
+    ny <- ncol(object$ci.lower)
+    p <- nrow(object$ci.lower)
     rownames(cis) <- paste0(
       rep(object$outcome, each = p),
       ":",
-      rep(object$coefficient_name, times = ny)
+      rep(object$term, times = ny)
     )
   } else {
-    rownames(cis) <- object$coefficient_name
+    rownames(cis) <- object$term
   }
 
   colnames(cis) <- paste((1 - level) / 2 * c(100, -100) + c(0, 100), "%")

R/S3_summary.R

@@ -2,21 +2,21 @@
 summary.lm_robust <- function(object,
                               ...) {
 
-  if (is.matrix(object$coefficients)) {
-    ny <- ncol(object$coefficients)
+  if (is.matrix(object$estimate)) {
+    ny <- ncol(object$estimate)
 
     ret <- setNames(
       vector("list", ny),
       paste("Response", object$outcome)
     )
 
     mat_objs <- c(
-      "coefficients",
-      "se",
+      "estimate",
+      "std.error",
       "df",
-      "ci_lower",
-      "ci_upper",
-      "p"
+      "ci.lower",
+      "ci.upper",
+      "p.value"
     )
 
     vec_objs <- c(
@@ -119,15 +119,15 @@ summarize_tidy <- function(object, test = "t", ...) {
   tidy_mat <- as.matrix(tidy_out[, !(names(tidy_out) %in% remove_cols)])
 
   ny <- length(object$outcome)
-  p <- length(object$coefficient_name)
+  p <- length(object$term)
   if (length(object$outcome) > 1) {
     rownames(tidy_mat) <- paste0(
       rep(object$outcome, each = p),
       ":",
-      rep(object$coefficient_name, times = ny)
+      rep(object$term, times = ny)
     )
   } else {
-    rownames(tidy_mat) <- object$coefficient_name
+    rownames(tidy_mat) <- object$term
   }
 
   return(tidy_mat)

R/S3_tidy.R

@@ -85,16 +85,6 @@ tidy.horvitz_thompson <- function(object, ...) {
 tidy_data_frame <- function(object, digits = NULL) {
 
   vec_cols <-
-    c(
-      "coefficients",
-      "se",
-      "p",
-      "ci_lower",
-      "ci_upper",
-      "df"
-    )
-
-  tidy_names <-
     c(
       "estimate",
       "std.error",
@@ -105,14 +95,14 @@ tidy_data_frame <- function(object, digits = NULL) {
     )
 
   tidy_mat <- do.call("cbind", lapply(vec_cols, function(x) {as.vector(object[[x]])}))
-  colnames(tidy_mat) <- tidy_names
+  colnames(tidy_mat) <- vec_cols
   return_frame <- data.frame(
-    term = object[["coefficient_name"]],
+    term = object[["term"]],
     tidy_mat,
     stringsAsFactors = FALSE
   )
 
-  return_frame$outcome <- rep(object[["outcome"]], each = length(object[["coefficient_name"]]))
+  return_frame$outcome <- rep(object[["outcome"]], each = length(object[["term"]]))
 
   rownames(return_frame) <- NULL
   return(return_frame)

R/estimatr_difference_in_means.R

@@ -64,13 +64,13 @@
 #'
 #' An object of class \code{"difference_in_means"} is a list containing at
 #' least the following components:
-#'   \item{coefficients}{the estimated coefficients}
-#'   \item{se}{the estimated standard errors}
+#'   \item{estimate}{the estimated difference in means}
+#'   \item{std.error}{the estimated standard error}
 #'   \item{df}{the estimated degrees of freedom}
-#'   \item{p}{the p-values from a two-sided t-test using \code{coefficients}, \code{se}, and \code{df}}
-#'   \item{ci_lower}{the lower bound of the \code{1 - alpha} percent confidence interval}
-#'   \item{ci_upper}{the upper bound of the \code{1 - alpha} percent confidence interval}
-#'   \item{coefficient_name}{a character vector of coefficient names}
+#'   \item{p.value}{the p-value from a two-sided t-test using \code{estimate}, \code{std.error}, and \code{df}}
+#'   \item{ci.lower}{the lower bound of the \code{1 - alpha} percent confidence interval}
+#'   \item{ci.upper}{the upper bound of the \code{1 - alpha} percent confidence interval}
+#'   \item{term}{a character vector of coefficient names}
 #'   \item{alpha}{the significance level specified by the user}
 #'   \item{N}{the number of observations used}
 #'   \item{outcome}{the name of the outcome variable}
@@ -320,7 +320,7 @@ difference_in_means <-
       N_overall <- with(block_estimates, sum(N))
 
       # Blocked design, (Gerber Green 2012, p73, eq3.10)
-      diff <- with(block_estimates, sum(coefficients * N / N_overall))
+      diff <- with(block_estimates, sum(estimate * N / N_overall))
 
       df <- NA
       n_blocks <- nrow(block_estimates)
@@ -329,20 +329,20 @@ difference_in_means <-
         if (is.null(data$cluster)) {
           design <- "Matched-pair"
           # Pair matched, unit randomized (Gerber Green 2012, p77, eq3.16)
-          se <-
+          std.error <-
             with(
               block_estimates,
-              sqrt((1 / (n_blocks * (n_blocks - 1))) * sum((coefficients - diff) ^ 2))
+              sqrt((1 / (n_blocks * (n_blocks - 1))) * sum((estimate - diff) ^ 2))
             )
         } else {
           design <- "Matched-pair clustered"
           # Pair matched, cluster randomized (Imai, King, Nall 2009, p36, eq6)
-          se <-
+          std.error <-
             with(
               block_estimates,
               sqrt(
                 (n_blocks / ((n_blocks - 1) * N_overall ^ 2)) *
-                  sum((N * coefficients - (N_overall * diff) / n_blocks) ^ 2)
+                  sum((N * estimate - (N_overall * diff) / n_blocks) ^ 2)
               )
             )
         }
@@ -351,7 +351,7 @@ difference_in_means <-
         df <- n_blocks - 1
       } else {
         # Block randomized (Gerber and Green 2012, p. 74, footnote 17)
-        se <- with(block_estimates, sqrt(sum(se ^ 2 * (N / N_overall) ^ 2)))
+        std.error <- with(block_estimates, sqrt(sum(std.error ^ 2 * (N / N_overall) ^ 2)))
 
 
         ## we don't know if this is correct!
@@ -367,8 +367,8 @@ difference_in_means <-
       }
 
       return_frame <- data.frame(
-        coefficients = diff,
-        se = se,
+        estimate = diff,
+        std.error = std.error,
         df = df,
         N = N_overall,
         stringsAsFactors = FALSE
@@ -465,24 +465,24 @@ difference_in_means_internal <-
         has_int = TRUE
       )
 
-      diff <- cr2_out$coefficients[2]
-      se <- cr2_out$se[2]
+      diff <- cr2_out$estimate[2]
+      std.error <- cr2_out$std.error[2]
       df <- cr2_out$df[2]
     } else {
       if (is.null(data$weights)) {
         diff <- mean(Y2) - mean(Y1)
 
         if (pair_matched) {
           # Pair matched designs
-          se <- NA
+          std.error <- NA
         } else {
           # Non-pair matched designs, unit level randomization
           var_Y2 <- var(Y2)
           var_Y1 <- var(Y1)
 
-          se <- sqrt(var_Y2 / N2 + var_Y1 / N1)
+          std.error <- sqrt(var_Y2 / N2 + var_Y1 / N1)
 
-          df <- se ^ 4 /
+          df <- std.error ^ 4 /
             (
               (var_Y2 / N2) ^ 2 / (N2 - 1) +
                 (var_Y1 / N1) ^ 2 / (N1 - 1)
@@ -513,16 +513,16 @@ difference_in_means_internal <-
           has_int = TRUE
         )
 
-        diff <- w_hc2_out$coefficients[2]
-        se <- w_hc2_out$se[2]
+        diff <- w_hc2_out$estimate[2]
+        std.error <- w_hc2_out$std.error[2]
         df <- w_hc2_out$df[2]
       }
     }
 
     return_frame <-
       data.frame(
-        coefficients = diff,
-        se = se,
+        estimate = diff,
+        std.error = std.error,
         df = df,
         stringsAsFactors = FALSE
       )

R/estimatr_horvitz_thompson.R

@@ -121,13 +121,13 @@
 #' An object of class \code{"horvitz_thompson"} is a list containing at
 #' least the following components:
 #'
-#'   \item{coefficients}{the estimated coefficients}
-#'   \item{se}{the estimated standard errors}
+#'   \item{estimate}{the estimated difference in totals}
+#'   \item{std.error}{the estimated standard error}
 #'   \item{df}{the estimated degrees of freedom}
-#'   \item{p}{the p-values from from a two-sided z-test using \code{coefficients}, \code{se}, and \code{df}}
-#'   \item{ci_lower}{the lower bound of the \code{1 - alpha} percent confidence interval}
-#'   \item{ci_upper}{the upper bound of the \code{1 - alpha} percent confidence interval}
-#'   \item{coefficient_name}{a character vector of coefficient names}
+#'   \item{p.value}{the p-value from a two-sided z-test using \code{estimate} and \code{std.error}}
+#'   \item{ci.lower}{the lower bound of the \code{1 - alpha} percent confidence interval}
+#'   \item{ci.upper}{the upper bound of the \code{1 - alpha} percent confidence interval}
+#'   \item{term}{a character vector of coefficient names}
 #'   \item{alpha}{the significance level specified by the user}
 #'   \item{N}{the number of observations used}
 #'   \item{outcome}{the name of the outcome variable}
@@ -545,13 +545,13 @@ horvitz_thompson <-
 
       n_blocks <- nrow(block_estimates)
 
-      diff <- with(block_estimates, sum(coefficients * N / N_overall))
+      diff <- with(block_estimates, sum(estimate * N / N_overall))
 
-      se <- with(block_estimates, sqrt(sum(se ^ 2 * (N / N_overall) ^ 2)))
+      std.error <- with(block_estimates, sqrt(sum(std.error ^ 2 * (N / N_overall) ^ 2)))
 
       return_frame <- data.frame(
-        coefficients = diff,
-        se = se,
+        estimate = diff,
+        std.error = std.error,
         N = N_overall
       )
     }
@@ -601,7 +601,7 @@ horvitz_thompson_internal <-
 
     N <- length(t2) + length(t1)
 
-    se <- NA
+    std.error <- NA
 
     collapsed <- !is.null(data$clusters)
     if (collapsed) {
@@ -639,7 +639,7 @@ horvitz_thompson_internal <-
 
       diff <- (sum(Y2) - sum(Y1)) / N
 
-      se <-
+      std.error <-
         sqrt(
           sum(Y2 ^ 2) +
             sum(Y1 ^ 2) +
@@ -687,7 +687,7 @@ horvitz_thompson_internal <-
           )
 
 
-          se <-
+          std.error <-
             sqrt(
               var_ht_total_no_cov(y1, data$condition_probabilities) +
                 var_ht_total_no_cov(y0, 1 - data$condition_probabilities) +
@@ -696,7 +696,7 @@ horvitz_thompson_internal <-
             ) / N
         } else {
           # Young's inequality
-          se <-
+          std.error <-
             sqrt(sum(Y2 ^ 2) + sum(Y1 ^ 2)) / N
         }
       } else {
@@ -733,24 +733,24 @@ horvitz_thompson_internal <-
           if (!is.nan(varN2)) {
             if (varN2 < 0) {
               warning("Variance below 0")
-              se <- NA
+              std.error <- NA
             } else {
-              se <- sqrt(varN2) / N
+              std.error <- sqrt(varN2) / N
             }
           } else {
             warning(
               "Variance is NaN. This is likely the result of a complex condition probability matrix"
             )
-            se <- NA
+            std.error <- NA
           }
         }
       }
     }
 
     return_frame <-
       data.frame(
-        coefficients = diff,
-        se = se,
+        estimate = diff,
+        std.error = std.error,
         N = N,
         stringsAsFactors = FALSE
       )