# Make Pretty Summary and Anova Tables Improve summary tables by replacing variable names with labels and separating variable names and value labels of factor variables. Additionally, confidence intervalls are added to summaries per default and p-values are formated for pretty printing. ```r ## generic function called by all prettify.summary.xxx functions ## S3 method for class 'data.frame' prettify(object, labels = NULL, sep = ": ", extra.column = FALSE, smallest.pval = 0.001, digits = NULL, scientific = FALSE, signif.stars = getOption("show.signif.stars"), ...) ## S3 method for class 'summary.lm' prettify(object, labels = NULL, sep = ": ", extra.column = FALSE, confint = TRUE, level = 0.95, smallest.pval = 0.001, digits = NULL, scientific = FALSE, signif.stars = getOption("show.signif.stars"), ...) ## S3 method for class 'summary.glm' prettify(object, labels = NULL, sep = ": ", extra.column = FALSE, confint = TRUE, level = 0.95, OR = TRUE, smallest.pval = 0.001, digits = NULL, scientific = FALSE, signif.stars = getOption("show.signif.stars"), ...) ## S3 method for class 'summary.coxph' prettify(object, labels = NULL, sep = ": ", extra.column = FALSE, confint = TRUE, level = 0.95, HR = TRUE, smallest.pval = 0.001, digits = NULL, scientific = FALSE, signif.stars = getOption("show.signif.stars"), env = parent.frame(), ...) ## S3 method for class 'summary.lme' prettify(object, labels = NULL, sep = ": ", extra.column = FALSE, confint = TRUE, level = 0.95, smallest.pval = 0.001, digits = NULL, scientific = FALSE, signif.stars = getOption("show.signif.stars"), ...) ## method for mixed models fitted with lme4 (vers. < 1.0) ## S3 method for class 'summary.mer' prettify(object, labels = NULL, sep = ": ", extra.column = FALSE, confint = TRUE, level = 0.95, smallest.pval = 0.001, digits = NULL, scientific = FALSE, signif.stars = getOption("show.signif.stars"), simulate = c("ifneeded", TRUE, FALSE), B = 1000, ...) ## method for mixed models fitted with lme4 (vers. >= 1.0) ## S3 method for class 'summary.merMod' prettify(object, labels = NULL, sep = ": ", extra.column = FALSE, confint = TRUE, level = 0.95, smallest.pval = 0.001, digits = NULL, scientific = FALSE, signif.stars = getOption("show.signif.stars"), method = c("profile", "Wald", "boot"), B = 1000, env = parent.frame(), ...) ## S3 method for class 'anova' prettify(object, labels, smallest.pval = 0.001, digits = NULL, scientific = FALSE, signif.stars = getOption("show.signif.stars"), ...) ## helper function for pretty p-values prettifyPValue(object, smallest.pval = 0.001, digits = NULL, scientific = FALSE, signif.stars = getOption("show.signif.stars"), ...) ``` ## Arguments - `object`: object of class `data.frame` resulting (most likely) from a call to `summary` or directly the output from `summary`, `anova` or `Anova` (the latter from package `car`). - `labels`: specify labels here. For the format see `labels`. - `sep`: separator between variable label and value label of a factor variable (default: `": "`). - `extra.column`: logical: provide an extra column for the value labels of factors (default: `FALSE`). - `confint`: logical value indicating if confidence intervals sould be added or the confidence intervals themself. Using `confint = TRUE` is experimental only and special care needs to be taken that the data set used for fitting is neither changed nor deleted. See Details and Examples . - `level`: confidence level; Per default 0.95% confidence intervals are returned - `OR`: logical. Should odds ratios be added? Only applicable if a logistic regression model was fitted (i.e., with `family = "binomial"`). - `HR`: logical. Should hazard ratios be added? - `smallest.pval`: determines the smallest p-value to be printed exactly. Smaller values are given as \< smallest.pval . This argument is passed to the `eps` argument of `format.pval`. See there for details. - `digits`: number of significant digits. The default, `NULL`, uses `getOption("digits")` for formating p-values and leaves all other columns unchanged. If `digits` are specified, all columns use this number of significant digits (columnwise). See also argument `digits` in `format`. - `scientific`: specifies if numbers should be printed in scientific format. For details and possible values see `format`. - `signif.stars`: logical (default = TRUE). Should significance stars be added? Per default system options are used. See `getOption("show.signif.stars")`. - `simulate`: should the asysmptotic or simulated confidence intervals be used? See `confint.mer` for details. - `B`: number of samples to take in `mcmcsamp`. See `confint.mer` for details. - `method`: Determines the method for computing confidence intervals; One of `"profile"` (default), `"Wald"`, `"boot"`. For details see `confint.merMod` in package `lme4`. - ``...``: further options. Currently not applicable. - `env`: specify environment in which the model was fitted. Needed to find the correct data for refitting the model in order to obtain confidence intervals. ## Details Specialized functions that prettify summary tables of various models exist. For the `data.frame` method, `extra.column` and `sep` can only be used if `labels` are specified as variable names need to be known in order to split variable name and factor level. For `summary` objects, variable names can be extracted from the objects. To compute confidence intervalls, the model is refitted internally extrating the call and environment from the model summary. All functions then use `confint` on the refitted model. For `mer` models special `confint` functions are defined in this package (for backward compatibility). For details see there. Note that is it highly important not to modify or delete the data in the fitting environment if one wants to obtain correct confidence intervals. See examples for what might happen. We try ourt best to find changes of the data and to warn the user (but without any warranty). Alternatively, one can directly specify the confidence intervals using e.g. `confint = confint(model)`, where `model` is the fitted model. This does not rely on refitting of the model and should always work as expected. In this case, arguments `level`, `simulate` and `B` are ignored. Note that in this case it is adviced to also specify the labels by hand! `prettifyPValue` is a helper function used within the prettify functions but can also be used directly on a `data.frame` object. The function tries to (cleverly) guess the column of p-values (based on the column names) and formats them nicely. Additionally, significance stars are added if requested. ## Returns `data.frame` with prettier variable labels. For summary functions additionally confidence intervalls (if requested), odds ratio (for logistic regression models, if requested), p-values formated for pretty printing and significance stars (if requested) are attached. ## Author(s) Benjamin Hofner ## See Also `summary`, `summary.lm`, `summary.glm`, `summary.lme`, `summary.merMod` (or `summary.mer-class` in lme4 < 1.0) and `summary.coxph` for summary functions. `anova` and `Anova` for ANOVA functions. `confint` and `ci` for confidence intervals. Special functions are implemented for mixed models: `confint.mer`. ## Examples ```r ## Example requires package nlme to be installed and loaded if (require("nlme")) { ## Load data set Orthodont data(Orthodont, package = "nlme") ###################################################################### # Linear model ###################################################################### ## Fit a linear model linmod <- lm(distance ~ age + Sex, data = Orthodont) ## Extract pretty summary prettify(summary(linmod)) ## Extract anova (sequential tests) anova(linmod) ## now prettify it prettify(anova(linmod)) ###################################################################### # Random effects model (nlme) ###################################################################### ### (fit a more suitable model with random effects) ## With package nlme: require("nlme") ## Fit a model for distance with random intercept for Subject mod <- lme(distance ~ age + Sex, data = Orthodont, random = ~ 1 | Subject) summary(mod) ## Extract fixed effects table, add confidence interval and make it pretty prettify(summary(mod)) ## Extract fixed effects table only and make it pretty prettify(summary(mod), confint = FALSE) ###################################################################### # Random effects model (lme4) ###################################################################### set.seed(130913) ## With package lme4: if (require("lme4") && require("car")) { ## Fit a model for distance with random intercept for Subject mod4 <- lmer(distance ~ age + Sex + (1|Subject), data = Orthodont) summary(mod4) ## Extract fixed effects table and make it pretty prettify(summary(mod4)) ## Extract and prettify anova (sequential tests) prettify(anova(mod4)) ## Better: extract Anova (partial instead of sequential tests) library("car") Anova(mod4) ## now prettify it prettify(Anova(mod4)) } ###################################################################### # Cox model ###################################################################### ## survival models if (require("survival")) { ## Load data set ovarian (now part of cancer) data(cancer, package = "survival") ## fit a Cox model mod5 <- coxph(Surv(futime, fustat) ~ age, data=ovarian) summary(mod5) ## Make pretty summary prettify(summary(mod5)) ## Make pretty summary prettify(Anova(mod5)) } ###################################################################### # ATTENTION when confint = TRUE: Do not modify or delete data ###################################################################### ## Fit a linear model (same as above) linmod <- lm(distance ~ age + Sex, data = Orthodont) ## Extract pretty summary prettify(summary(linmod)) ## Change the data (age in month instead of years) Orthodont$age <- Orthodont$age * 12 prettify(summary(linmod)) ## confidence intervals for age have changed ## but coefficients stayed the same; a ## warning is issued ## Remove data in fitting environment rm(Orthodont) prettify(summary(linmod)) ## confidence intervals are missing as no ## data set was available to refit the model ###################################################################### # Use confint to specify confidence interval without refitting ###################################################################### ## make labels without using the data set labels <- c("distance", "age", "Subject", "Sex") names(labels) <- labels ## usually easier via: labels(Orthodont) prettify(summary(linmod), confint = confint(linmod), labels = labels) } ```