Add method for Chi2 (#393)

* add full support for chi2

* update snapshots, styler

* lints

R/report.htest.R

@@ -65,13 +65,13 @@ report_effectsize.htest <- function(x, ...) {
     }
   }
 
-  # For correlations ---------------
+  ## For correlations ---------------
 
   if (model_info$is_correlation) {
     out <- .report_effectsize_correlation(x, table, dot_args)
   }
 
-  # For Chi2 ---------------
+  ## For Chi2 ---------------
 
   if (model_info$is_chi2test) {
     out <- .report_effectsize_chi2(x, table, dot_args)
@@ -131,6 +131,7 @@ report_table.htest <- function(x, ...) {
   } else if (model_info$is_chi2test) {
     # chi2 test
     out <- .report_table_chi2(table_full, effsize)
+    attr(out$table_full, "table_footer") <- attr(attr(effsize, "table"), "table_footer")
   } else if (model_info$is_correlation) {
     # correlation test
     out <- .report_table_correlation(table_full)
@@ -250,12 +251,15 @@ report_parameters.htest <- function(x, table = NULL, ...) {
     # t-tests
   } else if (model_info$is_ttest) {
     out <- .report_parameters_ttest(table, stats, effsize, ...)
+    # Friedman
   } else if (model_info$is_ranktest &&
     grepl("Friedman", attributes(x$statistic)$names, fixed = TRUE)) {
     out <- .report_parameters_friedman(table, stats, effsize, ...)
-
-    # TODO: default, same as t-test?
+    # chi2
+  } else if (model_info$is_chi2test) {
+    out <- .report_parameters_chi2(table, stats, effsize, ...)
   } else {
+    # TODO: default, same as t-test?
     out <- .report_parameters_htest_default(table, stats, effsize, ...)
   }
 

R/report_htest_chi2.R

@@ -9,18 +9,17 @@
 # report_effectsize ---------------------
 
 .report_effectsize_chi2 <- function(x, table, dot_args, rules = "funder2019") {
-  if (grepl("Pearson", x$method, fixed = TRUE)) {
+  if (chi2_type(x) %in% c("pearson", "probabilities")) {
     args <- c(list(x), dot_args)
     table <- do.call(effectsize::effectsize, args)
+    table_footer <- attributes(table)$table_footer
     ci <- attributes(table)$ci
     estimate <- names(table)[1]
     rules <- ifelse(is.null(dot_args$rules), rules, dot_args$rules)
 
     args <- list(table, rules = rules, dot_args)
     interpretation <- do.call(effectsize::interpret, args)$Interpretation
     rules <- .text_effectsize(attr(attr(interpretation, "rules"), "rule_name"))
-  } else if (grepl("given probabilities", x$method, fixed = TRUE)) {
-
   } else {
     stop(insight::format_message(
       "This test is not yet supported. Please open an issue at {.url https://github.com/easystats/report/issues}."
@@ -29,6 +28,8 @@
 
   if (estimate == "Cramers_v_adjusted") {
     main <- paste0("Adjusted Cramer's v = ", insight::format_value(table[[estimate]]))
+  } else if (estimate == "Fei") {
+    main <- paste0("Fei = ", insight::format_value(table[[estimate]]))
   } else if (estimate == "Tschuprows_t") {
     main <- paste0("Tschuprow's t = ", insight::format_value(table[[estimate]]))
   } else if (estimate == "Tschuprows_t_adjusted") {
@@ -62,6 +63,7 @@
   )
 
   table <- table[c(estimate, paste0(estimate, c("_CI_low", "_CI_high")))]
+  attributes(table)$table_footer <- table_footer
 
   list(
     table = table, statistics = statistics, interpretation = interpretation,
@@ -72,13 +74,62 @@
 # report_model ----------------------------
 
 .report_model_chi2 <- function(x, table) {
-  vars_full <- paste0(names(attributes(x$observed)$dimnames), collapse = " and ")
+  if (chi2_type(x) == "pearson") {
+    type <- " of independence between"
+    vars_full <- paste0(names(attributes(x$observed)$dimnames), collapse = " and ")
+  } else if (chi2_type(x) == "probabilities") {
+    type <- " / goodness of fit of "
+    dist <- ifelse(
+      grepl("non", attr(table, "table_footer"), fixed = TRUE), "a uniform distribution",
+      paste0("a distribution of [", paste0(
+        names(x$expected), ": n=", x$expected,
+        collapse = ", "
+      ), "]")
+    )
+
+    vars_full <- paste(x$data.name, "to", dist)
+  }
 
   text <- paste0(
     trimws(x$method),
-    " testing the association between ",
-    vars_full
+    type,
+    paste0(" ", vars_full)
   )
 
   text
 }
+
+chi2_type <- function(x) {
+  if (grepl("probabilities", x$method, fixed = TRUE)) {
+    out <- "probabilities"
+  } else if (grepl("Pearson", x$method, fixed = TRUE)) {
+    out <- "pearson"
+  }
+  out
+}
+
+# report_parameters ----------------------------
+
+.report_parameters_chi2 <- function(table, stats, effsize, ...) {
+  text_full <- paste0(
+    "statistically ",
+    effectsize::interpret_p(table$p, rules = "default"),
+    ", and ",
+    attributes(effsize)$interpretation,
+    " (",
+    stats,
+    ")"
+  )
+
+  text_short <- paste0(
+    "statistically ",
+    effectsize::interpret_p(table$p, rules = "default"),
+    ", and ",
+    attributes(effsize)$interpretation,
+    " (",
+    summary(stats),
+    ")"
+  )
+
+  list(text_short = text_short, text_full = text_full)
+}

tests/testthat/_snaps/windows/report.htest-chi2.md

@@ -140,9 +140,9 @@
     Output
       Effect sizes were labelled following Funder's (2019) recommendations.
       
-      The Pearson's Chi-squared test testing the association between gender and party
-      suggests that the effect is positive, statistically significant, and small
-      (chi2 = 30.07, p < .001; Adjusted Cramer's v = 0.10, 95% CI [0.07, 1.00])
+      The Pearson's Chi-squared test of independence between gender and party
+      suggests that the effect is statistically significant, and small (chi2 = 30.07,
+      p < .001; Adjusted Cramer's v = 0.10, 95% CI [0.07, 1.00])
 
 ---
 
@@ -151,9 +151,9 @@
     Output
       Effect sizes were labelled following Funder's (2019) recommendations.
       
-      The Pearson's Chi-squared test testing the association between gender and party
-      suggests that the effect is positive, statistically significant, and small
-      (chi2 = 30.07, p < .001; Adjusted Cramer's v = 0.10, 95% CI [0.07, 1.00])
+      The Pearson's Chi-squared test of independence between gender and party
+      suggests that the effect is statistically significant, and small (chi2 = 30.07,
+      p < .001; Adjusted Cramer's v = 0.10, 95% CI [0.07, 1.00])
 
 ---
 
@@ -162,9 +162,9 @@
     Output
       Effect sizes were labelled following Gignac's (2016) recommendations.
       
-      The Pearson's Chi-squared test testing the association between gender and party
-      suggests that the effect is positive, statistically significant, and small
-      (chi2 = 30.07, p < .001; Adjusted Cramer's v = 0.10, 95% CI [0.07, 1.00])
+      The Pearson's Chi-squared test of independence between gender and party
+      suggests that the effect is statistically significant, and small (chi2 = 30.07,
+      p < .001; Adjusted Cramer's v = 0.10, 95% CI [0.07, 1.00])
 
 ---
 
@@ -173,9 +173,9 @@
     Output
       Effect sizes were labelled following Cohen's (1988) recommendations.
       
-      The Pearson's Chi-squared test testing the association between gender and party
-      suggests that the effect is positive, statistically significant, and small
-      (chi2 = 30.07, p < .001; Adjusted Cramer's v = 0.10, 95% CI [0.07, 1.00])
+      The Pearson's Chi-squared test of independence between gender and party
+      suggests that the effect is statistically significant, and small (chi2 = 30.07,
+      p < .001; Adjusted Cramer's v = 0.10, 95% CI [0.07, 1.00])
 
 ---
 
@@ -184,9 +184,9 @@
     Output
       Effect sizes were labelled following Evans's (1996) recommendations.
       
-      The Pearson's Chi-squared test testing the association between gender and party
-      suggests that the effect is positive, statistically significant, and very weak
-      (chi2 = 30.07, p < .001; Adjusted Cramer's v = 0.10, 95% CI [0.07, 1.00])
+      The Pearson's Chi-squared test of independence between gender and party
+      suggests that the effect is statistically significant, and very weak (chi2 =
+      30.07, p < .001; Adjusted Cramer's v = 0.10, 95% CI [0.07, 1.00])
 
 ---
 
@@ -195,9 +195,9 @@
     Output
       Effect sizes were labelled following Lovakov's (2021) recommendations.
       
-      The Pearson's Chi-squared test testing the association between gender and party
-      suggests that the effect is positive, statistically significant, and very small
-      (chi2 = 30.07, p < .001; Adjusted Cramer's v = 0.10, 95% CI [0.07, 1.00])
+      The Pearson's Chi-squared test of independence between gender and party
+      suggests that the effect is statistically significant, and very small (chi2 =
+      30.07, p < .001; Adjusted Cramer's v = 0.10, 95% CI [0.07, 1.00])
 
 ---
 
@@ -206,9 +206,9 @@
     Output
       Effect sizes were labelled following Funder's (2019) recommendations.
       
-      The Pearson's Chi-squared test testing the association between gender and party
-      suggests that the effect is positive, statistically significant, and small
-      (chi2 = 30.07, p < .001; Adjusted Cramer's v = 0.10, 95% CI [0.07, 1.00])
+      The Pearson's Chi-squared test of independence between gender and party
+      suggests that the effect is statistically significant, and small (chi2 = 30.07,
+      p < .001; Adjusted Cramer's v = 0.10, 95% CI [0.07, 1.00])
 
 ---
 
@@ -217,9 +217,9 @@
     Output
       Effect sizes were labelled following Funder's (2019) recommendations.
       
-      The Pearson's Chi-squared test testing the association between gender and party
-      suggests that the effect is positive, statistically significant, and small
-      (chi2 = 30.07, p < .001; Pearson's c = 0.10, 95% CI [0.07, 1.00])
+      The Pearson's Chi-squared test of independence between gender and party
+      suggests that the effect is statistically significant, and small (chi2 = 30.07,
+      p < .001; Pearson's c = 0.10, 95% CI [0.07, 1.00])
 
 ---
 
@@ -228,9 +228,9 @@
     Output
       Effect sizes were labelled following Funder's (2019) recommendations.
       
-      The Pearson's Chi-squared test testing the association between gender and party
-      suggests that the effect is positive, statistically significant, and very small
-      (chi2 = 30.07, p < .001; Tschuprow's t = 0.09, 95% CI [0.06, 1.00])
+      The Pearson's Chi-squared test of independence between gender and party
+      suggests that the effect is statistically significant, and very small (chi2 =
+      30.07, p < .001; Tschuprow's t = 0.09, 95% CI [0.06, 1.00])
 
 ---
 
@@ -239,9 +239,9 @@
     Output
       Effect sizes were labelled following Funder's (2019) recommendations.
       
-      The Pearson's Chi-squared test testing the association between gender and party
-      suggests that the effect is positive, statistically significant, and very small
-      (chi2 = 30.07, p < .001; Adjusted Tschuprow's t = 0.08, 95% CI [0.06, 1.00])
+      The Pearson's Chi-squared test of independence between gender and party
+      suggests that the effect is statistically significant, and very small (chi2 =
+      30.07, p < .001; Adjusted Tschuprow's t = 0.08, 95% CI [0.06, 1.00])
 
 ---
 
@@ -250,9 +250,9 @@
     Output
       Effect sizes were labelled following Funder's (2019) recommendations.
       
-      The Pearson's Chi-squared test testing the association between gender and party
-      suggests that the effect is positive, statistically significant, and small
-      (chi2 = 30.07, p < .001; Cohens_w = 0.10, 95% CI [0.07, 1.00])
+      The Pearson's Chi-squared test of independence between gender and party
+      suggests that the effect is statistically significant, and small (chi2 = 30.07,
+      p < .001; Cohens_w = 0.10, 95% CI [0.07, 1.00])
 
 ---
 
@@ -261,8 +261,8 @@
     Output
       Effect sizes were labelled following Funder's (2019) recommendations.
       
-      The Pearson's Chi-squared test with Yates' continuity correction testing the
-      association between Diagnosis and Group suggests that the effect is positive,
+      The Pearson's Chi-squared test with Yates' continuity correction of
+      independence between Diagnosis and Group suggests that the effect is
       statistically significant, and large (chi2 = 31.57, p < .001; Adjusted Phi =
       0.36, 95% CI [0.25, 1.00])
 
@@ -273,8 +273,8 @@
     Output
       Effect sizes were labelled following Savilowsky's (2009) recommendations.
       
-      The Pearson's Chi-squared test with Yates' continuity correction testing the
-      association between Diagnosis and Group suggests that the effect is positive,
+      The Pearson's Chi-squared test with Yates' continuity correction of
+      independence between Diagnosis and Group suggests that the effect is
       statistically significant, and medium (chi2 = 31.57, p < .001; Cohen's h =
       0.74, 95% CI [0.50, 0.99])
 
@@ -285,8 +285,8 @@
     Output
       Effect sizes were labelled following Chen's (2010) recommendations.
       
-      The Pearson's Chi-squared test with Yates' continuity correction testing the
-      association between Diagnosis and Group suggests that the effect is positive,
+      The Pearson's Chi-squared test with Yates' continuity correction of
+      independence between Diagnosis and Group suggests that the effect is
       statistically significant, and medium (chi2 = 31.57, p < .001; Odds ratio =
       4.73, 95% CI [2.74, 8.17])
 
@@ -297,8 +297,31 @@
     Output
       
       
-      The Pearson's Chi-squared test with Yates' continuity correction testing the
-      association between Diagnosis and Group suggests that the effect is positive,
+      The Pearson's Chi-squared test with Yates' continuity correction of
+      independence between Diagnosis and Group suggests that the effect is
       statistically significant, and (chi2 = 31.57, p < .001; Risk_ratio = 2.54, 95%
       CI [1.80, 3.60])
 
+---
+
+    Code
+      report(x)
+    Output
+      Effect sizes were labelled following Funder's (2019) recommendations.
+      
+      The Pearson's Chi-squared test of independence between mtcars$cyl and mtcars$am
+      suggests that the effect is statistically significant, and very large (chi2 =
+      8.74, p = 0.013; Adjusted Cramer's v = 0.46, 95% CI [0.00, 1.00])
+
+# report.htest-chi2 for given probabilities
+
+    Code
+      report(x)
+    Output
+      
+      
+      The Chi-squared test for given probabilities / goodness of fit of
+      table(mtcars$cyl) to a distribution of [4: n=3.2, 6: n=9.6, 8: n=19.2] suggests
+      that the effect is statistically significant, and (chi2 = 21.12, p < .001; Fei
+      = 0.27, 95% CI [0.17, 1.00])
+

tests/testthat/test-report.htest-chi2.R

@@ -186,4 +186,22 @@ test_that("report.htest-chi2 report", {
     report(x, type = "riskratio")
   ) # riskratio has no interpretation in effectsize
   # Watch carefully in case effectsize adds support
+
+  # Mattan example comparison
+  x <- suppressWarnings(chisq.test(mtcars$cyl, mtcars$am))
+
+  expect_snapshot(
+    variant = "windows",
+    report(x)
+  )
+})
+
+test_that("report.htest-chi2 for given probabilities", {
+  # Mattan example comparison
+  x <- suppressWarnings(chisq.test(table(mtcars$cyl), p = c(0.1, 0.3, 0.6)))
+
+  expect_snapshot(
+    variant = "windows",
+    report(x)
+  )
 })