|
| 1 | +#' @keywords internal |
| 2 | +#' @noRd |
| 3 | +.check_custom_contrasts_and_filter <- function( |
| 4 | + estimate, |
| 5 | + by, |
| 6 | + original_contrast, |
| 7 | + comparison |
| 8 | +) { |
| 9 | + # setup message to tell user that results must be cross-checked |
| 10 | + if (estimate == "average" && !all(.grep_cleaned_by_vars(by) == by)) { |
| 11 | + # first, extract contrast with filtering, which doesn't work |
| 12 | + wrong_contrast <- setdiff(original_contrast, .grep_cleaned_by_vars(original_contrast)) |
| 13 | + # clean contrast and by, used to show correct example |
| 14 | + original_contrast <- .grep_cleaned_by_vars(original_contrast) |
| 15 | + original_by <- setdiff(.grep_cleaned_by_vars(by), original_contrast) |
| 16 | + msg1 <- paste0( |
| 17 | + "Selecting specific levels or values in the `contrast` or `by` arguments ", |
| 18 | + if (length(wrong_contrast)) { |
| 19 | + paste0( |
| 20 | + "(e.g., ", |
| 21 | + paste0( |
| 22 | + "`contrast = c(", |
| 23 | + paste0("\"", wrong_contrast, "\"", collapse = ", "), |
| 24 | + ")`) " |
| 25 | + ) |
| 26 | + ) |
| 27 | + }, |
| 28 | + "is error-prone for custom contrasts like ", |
| 29 | + paste0("`comparison = \"", comparison, "\"`"), |
| 30 | + " in combination with `estimate = \"average\"`. This can yield incorrect results,", |
| 31 | + " even if the output suggests the correct comparisons. It is strongly recommended", |
| 32 | + " to use only bare variable names in `contrast` and `by`, e.g.\n\n" |
| 33 | + ) |
| 34 | + msg2 <- insight::color_text( |
| 35 | + paste0( |
| 36 | + " estimate_contrasts(\n", |
| 37 | + " contrast = c(", |
| 38 | + paste0("\"", original_contrast, "\"", collapse = ", "), |
| 39 | + "),\n", |
| 40 | + if (length(original_by)) { |
| 41 | + paste0(" by = c(", paste0("\"", original_by, "\"", collapse = ", "), "),\n") |
| 42 | + }, |
| 43 | + " estimate = \"average\",\n comparison = ...\n )" |
| 44 | + ), |
| 45 | + color = "green" |
| 46 | + ) |
| 47 | + msg3 <- "\n\n and update your `comparison` argument accordingly.\n Run\n\n" |
| 48 | + msg4 <- insight::color_text( |
| 49 | + paste0( |
| 50 | + " estimate_means(\n", |
| 51 | + " c(", |
| 52 | + paste0("\"", c(original_contrast, original_by), "\"", collapse = ", "), |
| 53 | + "),\n estimate = \"average\"\n )" |
| 54 | + ), |
| 55 | + color = "green" |
| 56 | + ) |
| 57 | + msg5 <- "\n\n first to find out the correct rows to specify the `b`-coefficients for the `comparison` argument.\n" |
| 58 | + warning( |
| 59 | + insight::format_message(msg1), |
| 60 | + msg2, |
| 61 | + msg3, |
| 62 | + msg4, |
| 63 | + insight::format_message(msg5), |
| 64 | + call. = FALSE |
| 65 | + ) |
| 66 | + } |
| 67 | +} |
| 68 | + |
| 69 | + |
| 70 | +#' @keywords internal |
| 71 | +#' @noRd |
| 72 | +.check_standard_errors <- function( |
| 73 | + out, |
| 74 | + by = NULL, |
| 75 | + contrast = NULL, |
| 76 | + model = NULL, |
| 77 | + model_name = "model", |
| 78 | + verbose = TRUE, |
| 79 | + ... |
| 80 | +) { |
| 81 | + if (!verbose || is.null(out$SE)) { |
| 82 | + return(NULL) |
| 83 | + } |
| 84 | + |
| 85 | + if (all(is.na(out$SE))) { |
| 86 | + # we show an example code how to resolve the problem. this example |
| 87 | + # code only works when we have at least `by` or `contrast`. If both |
| 88 | + # are NULL, we ignore the example code (see below) |
| 89 | + code_snippet <- paste0("\n\nestim <- estimate_relation(\n ", model_name) |
| 90 | + by_vars <- c(by, contrast) |
| 91 | + if (!is.null(by_vars)) { |
| 92 | + code_snippet <- paste0( |
| 93 | + code_snippet, |
| 94 | + ",\n by = ", |
| 95 | + ifelse(length(by_vars) > 1, "c(", ""), |
| 96 | + paste0("\"", by_vars, "\"", collapse = ", "), |
| 97 | + ifelse(length(by_vars) > 1, ")", "") |
| 98 | + ) |
| 99 | + } |
| 100 | + code_snippet <- paste0(code_snippet, "\n)\nestimate_contrasts(\n estim") |
| 101 | + if (!is.null(contrast)) { |
| 102 | + code_snippet <- paste0( |
| 103 | + code_snippet, |
| 104 | + ",\n contrast = ", |
| 105 | + ifelse(length(contrast) > 1, "c(", ""), |
| 106 | + paste0("\"", contrast, "\"", collapse = ", "), |
| 107 | + ifelse(length(contrast) > 1, ")", "") |
| 108 | + ) |
| 109 | + } |
| 110 | + code_snippet <- paste0(code_snippet, "\n)") |
| 111 | + # setup message |
| 112 | + msg <- insight::format_message( |
| 113 | + "Could not calculate standard errors for contrasts. This can happen when random effects are involved." |
| 114 | + ) |
| 115 | + # add example code, if valid |
| 116 | + if (!is.null(by_vars)) { |
| 117 | + msg <- c( |
| 118 | + paste(msg, "You may try following:"), |
| 119 | + insight::color_text(code_snippet, "green"), |
| 120 | + "\n" |
| 121 | + ) |
| 122 | + } |
| 123 | + message(msg) |
| 124 | + |
| 125 | + # disable message for now, see |
| 126 | + # https://github.com/easystats/modelbased/issues/526 |
| 127 | + # } else if (length(out$SE) > 1 && isTRUE(all(out$SE == out$SE[1])) && insight::is_mixed_model(model)) { |
| 128 | + # msg <- "Standard errors are probably not reliable. This can happen when random effects are involved. You may try `estimate_relation()` instead." |
| 129 | + # if (!inherits(model, "glmmTMB")) { |
| 130 | + # msg <- paste(msg, "You may also try package {.pkg glmmTMB} to produce valid standard errors.") |
| 131 | + # } |
| 132 | + # insight::format_alert(msg) |
| 133 | + } |
| 134 | +} |
| 135 | + |
| 136 | + |
| 137 | +#' @keywords internal |
| 138 | +#' @noRd |
| 139 | +.check_offset <- function( |
| 140 | + model, |
| 141 | + estimate, |
| 142 | + offset = NULL, |
| 143 | + my_args = NULL, |
| 144 | + verbose = TRUE |
| 145 | +) { |
| 146 | + model_offset <- insight::find_offset(model) |
| 147 | + # check if model has an offset at all |
| 148 | + if (!is.null(model_offset) && !any(startsWith(my_args$by, model_offset)) && verbose) { |
| 149 | + msg <- NULL |
| 150 | + if (is.null(offset)) { |
| 151 | + # if no offset argument was specified, tell user what this means |
| 152 | + msg <- switch( |
| 153 | + estimate, |
| 154 | + specific = , |
| 155 | + typical = paste( |
| 156 | + "Model contains an offset-term, which is set to its mean value.", |
| 157 | + "If you want to average predictions over the distribution of the offset", |
| 158 | + "(if appropriate), use `estimate = \"average\"` or `estimate = \"population\"`.", |
| 159 | + "If you want to fix the offset to a specific value, for instance `1`,", |
| 160 | + "use `offset = 1`." |
| 161 | + ), |
| 162 | + average = , |
| 163 | + population = paste( |
| 164 | + "Model contains an offset-term and you average predictions over the", |
| 165 | + "distribution of that offset. If you want to fix the offset to a", |
| 166 | + "specific value, for instance `1`, use `offset = 1`." |
| 167 | + ) |
| 168 | + ) |
| 169 | + # if offset term is log-transformed, tell user. offset should be fixed then |
| 170 | + log_offset <- insight::find_transformation(insight::find_offset( |
| 171 | + model, |
| 172 | + as_term = TRUE |
| 173 | + )) |
| 174 | + if (!is.null(log_offset) && startsWith(log_offset, "log")) { |
| 175 | + msg <- c( |
| 176 | + msg, |
| 177 | + paste( |
| 178 | + "We also found that the model has a log-transformed offset term.", |
| 179 | + "If you use the `offset` argument, the log-transformation will", |
| 180 | + "automatically be applied to the provided offset-value. I.e., consider", |
| 181 | + "using, for instance, `offset = 10` and not `offset = log(10)`." |
| 182 | + ) |
| 183 | + ) |
| 184 | + } |
| 185 | + } |
| 186 | + if (!is.null(msg)) { |
| 187 | + insight::format_alert(msg) |
| 188 | + } |
| 189 | + } |
| 190 | +} |
| 191 | + |
| 192 | + |
| 193 | +#' @keywords internal |
| 194 | +#' @noRd |
| 195 | +.check_dots_data <- function(dots, verbose) { |
| 196 | + if (!is.null(dots$data)) { |
| 197 | + if (!is.null(dots$newdata) && verbose) { |
| 198 | + insight::format_alert( |
| 199 | + "Both 'data' and 'newdata' were provided. Please specify only one. Ignoring 'newdata' and using 'data' instead." |
| 200 | + ) |
| 201 | + } |
| 202 | + dots$newdata <- dots$data |
| 203 | + dots$data <- NULL |
| 204 | + } |
| 205 | + dots |
| 206 | +} |
| 207 | + |
| 208 | + |
| 209 | +#' @keywords internal |
| 210 | +#' @noRd |
| 211 | +.check_for_inequality_comparison <- function(comparison) { |
| 212 | + # check whether we have a formula definition of inequality comparisons, |
| 213 | + # and convert it to a string |
| 214 | + # |
| 215 | + # the default formulas are converted to a string: |
| 216 | + # ~inequality -> "inequality" |
| 217 | + # inequality ~ pairwise -> "inequality_pairwise" |
| 218 | + # ratio ~ inequality -> "inequality_ratio" |
| 219 | + # ratio ~ inequality + pairwise` -> "inequality_ratio_pairwise" |
| 220 | + # |
| 221 | + # we may have other formulas that control grouping and averaging, like |
| 222 | + # `~ inequality | grp1 + grp2`. In this case, the formula is returned as is |
| 223 | + # and processed later in ".process_inequality_formula()" |
| 224 | + if (inherits(comparison, "formula")) { |
| 225 | + # parse variables into a string |
| 226 | + out <- paste(all.vars(comparison), collapse = "_") |
| 227 | + # handle special cases |
| 228 | + out <- switch( |
| 229 | + out, |
| 230 | + ratio_inequality = "inequality_ratio", |
| 231 | + ratio_inequality_pairwise = "inequality_ratio_pairwise", |
| 232 | + out |
| 233 | + ) |
| 234 | + if (.is_inequality_comparison(out)) { |
| 235 | + return(out) |
| 236 | + } |
| 237 | + } |
| 238 | + comparison |
| 239 | +} |
| 240 | + |
| 241 | + |
| 242 | +#' @keywords internal |
| 243 | +#' @noRd |
| 244 | +.check_format_backend <- function(...) { |
| 245 | + # we allow exporting HTML format based on "gt" or "tinytable" |
| 246 | + dots <- list(...) |
| 247 | + if (identical(dots$backend, "tt")) { |
| 248 | + "tt" |
| 249 | + } else { |
| 250 | + "html" |
| 251 | + } |
| 252 | +} |
| 253 | + |
| 254 | + |
| 255 | +#' @keywords internal |
| 256 | +#' @noRd |
| 257 | +.check_predict_arg <- function(predict, valid_types, error_arg) { |
| 258 | + if (isTRUE(is.na(predict))) { |
| 259 | + # add modelbased-options to valid types |
| 260 | + valid_types <- unique(c("response", "link", valid_types)) |
| 261 | + insight::format_error(paste0( |
| 262 | + "The option provided in the `", |
| 263 | + error_arg, |
| 264 | + "` argument is not recognized.", |
| 265 | + " Valid options are: ", |
| 266 | + datawizard::text_concatenate(valid_types, enclose = "`"), |
| 267 | + "." |
| 268 | + )) |
| 269 | + } |
| 270 | +} |
| 271 | + |
| 272 | + |
| 273 | +# handle errors from marginaleffects ----------------------------------------- |
| 274 | +# |
| 275 | +# This helper function processes errors that occur during calls to the |
| 276 | +# {marginaleffects} package. It creates more informative and user-friendly |
| 277 | +# error messages by inspecting the original error and suggesting potential |
| 278 | +# solutions for common problems, such as using a different `estimate` option |
| 279 | +# or switching to the `emmeans` backend. |
| 280 | +# |
| 281 | +# Arguments: |
| 282 | +# - out: The error object returned from the `tryCatch` block. |
| 283 | +# - fun_args: A list of arguments that were passed to the failing |
| 284 | +# {marginaleffects} function. |
| 285 | +# |
| 286 | +# returns: A character vector containing the formatted, user-friendly error |
| 287 | +# message, which is then passed to `insight::format_error()`. |
| 288 | +# |
| 289 | +#' @keywords internal |
| 290 | +#' @noRd |
| 291 | +.check_marginaleffects_errors <- function(out, fun_args) { |
| 292 | + # what was requested? |
| 293 | + if (is.null(fun_args$hypothesis)) { |
| 294 | + fun <- "marginal means" |
| 295 | + } else { |
| 296 | + fun <- "marginal contrasts" |
| 297 | + } |
| 298 | + # clean original error message |
| 299 | + out$message <- gsub("\\s+", " ", gsub("\n", "", out$message, fixed = TRUE)) |
| 300 | + # setup clear error message |
| 301 | + msg <- c( |
| 302 | + paste0("Sorry, calculating ", fun, " failed with following error:"), |
| 303 | + insight::color_text(gsub("\n", "", out$message, fixed = TRUE), "red") |
| 304 | + ) |
| 305 | + # handle exceptions ------------------------------------------------------ |
| 306 | + # we get this error when we should use counterfactuals |
| 307 | + if (grepl("not found in column names", out$message, fixed = TRUE)) { |
| 308 | + msg <- c( |
| 309 | + msg, |
| 310 | + "\nIt seems that not all required levels of the focal terms are available in the provided data. If you want predictions extrapolated to a hypothetical target population, try setting `estimate=\"population\"." |
| 311 | + ) |
| 312 | + } |
| 313 | + # we get this error for models with complex random effects structures in glmmTMB, |
| 314 | + # or when the data grid is too large |
| 315 | + if ( |
| 316 | + grepl("map factor length must equal", out$message, fixed = TRUE) || |
| 317 | + grepl("cannot allocate", out$message, fixed = TRUE) |
| 318 | + ) { |
| 319 | + msg <- c( |
| 320 | + msg, |
| 321 | + paste0( |
| 322 | + "\nYou may try using the `emmeans` backend, e.g. `estimate_means(model, by = c(", |
| 323 | + toString(paste0("\"", fun_args$by, "\"")), |
| 324 | + "), backend = \"emmeans\")`, or use `estimate_relation(model, by = c(", |
| 325 | + toString(paste0("\"", fun_args$by, "\"")), |
| 326 | + "))` instead. For contrasts or pairwise comparisons, save the output of `estimate_relation()` and pass it to `estimate_contrasts()`, e.g.\n" |
| 327 | + ), |
| 328 | + paste0( |
| 329 | + "out <- estimate_relation(model, by = c(", |
| 330 | + toString(paste0("\"", fun_args$by, "\"")), |
| 331 | + "))" |
| 332 | + ), |
| 333 | + paste0( |
| 334 | + "estimate_contrasts(out, contrast = c(", |
| 335 | + toString(paste0("\"", fun_args$by, "\"")), |
| 336 | + "))" |
| 337 | + ) |
| 338 | + ) |
| 339 | + } |
| 340 | + msg |
| 341 | +} |
| 342 | + |
| 343 | + |
| 344 | +#' @keywords internal |
| 345 | +#' @noRd |
| 346 | +.check_filter_args <- function(result, prefix = "") { |
| 347 | + if (nrow(result) == 0) { |
| 348 | + # small helper, because we have the same error message in several places |
| 349 | + insight::format_error( |
| 350 | + prefix, |
| 351 | + "Please check your `by` and `contrast` arguments, or try one of the following options:", |
| 352 | + "1. Use a different option for the `estimate` argument, e.g. `estimate = \"typical\"`.", |
| 353 | + "2. Use the `newdata` argument to provide a data grid of predictor values at which to evaluate predictions." |
| 354 | + ) |
| 355 | + } |
| 356 | +} |
| 357 | + |
| 358 | + |
| 359 | +#' @param trend The `trend` argument |
| 360 | +#' @return updated `trend` variable |
| 361 | +#' @keywords internal |
| 362 | +#' @noRd |
| 363 | +.check_trend_arg <- function(trend, verbose = TRUE) { |
| 364 | + if (length(trend) > 1) { |
| 365 | + if (verbose) { |
| 366 | + insight::format_alert(paste0( |
| 367 | + "More than one numeric variable was selected for slope estimation. Keeping only `", |
| 368 | + trend[1], |
| 369 | + "`. ", |
| 370 | + "If you want to estimate the slope of `", |
| 371 | + trend[1], |
| 372 | + "` at different values of `", |
| 373 | + trend[2], |
| 374 | + "`, use `by=\"", |
| 375 | + trend[2], |
| 376 | + "\"` instead." |
| 377 | + )) |
| 378 | + } |
| 379 | + trend <- trend[1] |
| 380 | + } |
| 381 | + trend |
| 382 | +} |
0 commit comments