tidy_add_n {broom.helpers}R Documentation

Add the (weighted) number of observations

Description

Add the number of observations in a new column n_obs, taking into account any weights if they have been defined.

Usage

tidy_add_n(x, model = tidy_get_model(x))

Arguments

x

a tidy tibble

model

the corresponding model, if not attached to x

Details

For continuous variables, it corresponds to all valid observations contributing to the model.

For categorical variables coded with treatment or sum contrasts, each model term could be associated to only one level of the original categorical variable. Therefore, n_obs will correspond to the number of observations associated with that level. n_obs will also be computed for reference rows. For polynomial contrasts (defined with stats::contr.poly()), all levels will contribute to the computation of each model term. Therefore, n_obs will be equal to the total number of observations. For Helmert and custom contrasts, only rows contributing positively (i.e. with a positive contrast) to the computation of a term will be considered for estimating n_obs. The result could therefore be difficult to interpret. For a better understanding of which observations are taken into account to compute n_obs values, you could look at model_compute_terms_contributions().

For interaction terms, only rows contributing to all the terms of the interaction will be considered to compute n_obs.

For binomial logistic models, tidy_add_n() will also return the corresponding number of events (n_event) for each term, taking into account any defined weights. Observed proportions could be obtained as n_obs / n_event.

Similarly, a number of events will be computed for multinomial logistic models (nnet::multinom()) for each level of the outcome (y.level), corresponding to the number of observations equal to that outcome level.

For Poisson models, n_event will be equal to the number of counts per term. In addition, a third column exposure will be computed. If no offset is defined, exposure is assumed to be equal to 1 (eventually multiplied by weights) per observation. If an offset is defined, exposure will be equal to the (weighted) sum of the exponential of the offset (as a reminder, to model the effect of x on the ratio y / z, a Poisson model will be defined as glm(y ~ x + offset(log(z)), family = poisson)). Observed rates could be obtained with n_event / exposure.

For Cox models (survival::coxph()), an individual could be coded with several observations (several rows). n_obs will correspond to the weighted number of observations which could be different from the number of individuals. tidy_add_n() will also compute a (weighted) number of events (n_events) according to the definition of the survival::Surv() object. Exposure time is also returned in exposure column. It is equal to the (weighted) sum of the time variable if only one variable time is passed to survival::Surv(), and to the (weighted) sum of time2 - time if two time variables are defined in survival::Surv().

The (weighted) total number of observations (N_obs), of events (N_event) and of exposure time (Exposure) are stored as attributes of the returned tibble.

See Also

Other tidy_helpers: tidy_add_coefficients_type(), tidy_add_contrasts(), tidy_add_estimate_to_reference_rows(), tidy_add_header_rows(), tidy_add_reference_rows(), tidy_add_term_labels(), tidy_add_variable_labels(), tidy_attach_model(), tidy_disambiguate_terms(), tidy_identify_variables(), tidy_plus_plus(), tidy_remove_intercept(), tidy_select_variables()

Examples

lm(Petal.Length ~ ., data = iris) %>%
  tidy_and_attach() %>%
  tidy_add_n()

lm(Petal.Length ~ ., data = iris, contrasts = list(Species = contr.sum)) %>%
  tidy_and_attach() %>%
  tidy_add_n()

lm(Petal.Length ~ ., data = iris, contrasts = list(Species = contr.poly)) %>%
  tidy_and_attach() %>%
  tidy_add_n()

lm(Petal.Length ~ poly(Sepal.Length, 2), data = iris) %>%
  tidy_and_attach() %>%
  tidy_add_n()

df <- Titanic %>%
  dplyr::as_tibble() %>%
  dplyr::mutate(Survived = factor(Survived, c("No", "Yes")))

df %>%
  glm(
    Survived ~ Class + Age + Sex,
    data = ., weights = .$n, family = binomial,
    contrasts = list(Age = contr.sum, Class = "contr.helmert")
  ) %>%
  tidy_and_attach() %>%
  tidy_add_n()

df %>%
  glm(
    Survived ~ Class * (Age : Sex),
    data = ., weights = .$n, family = binomial,
    contrasts = list(Age = contr.sum, Class = "contr.helmert")
  ) %>%
  tidy_and_attach() %>%
  tidy_add_n()

glm(response ~ age + grade * trt, gtsummary::trial, family = poisson) %>%
  tidy_and_attach() %>%
  tidy_add_n()

glm(
  response ~ trt * grade + offset(log(ttdeath)),
  gtsummary::trial,
  family = poisson
) %>%
  tidy_and_attach() %>%
  tidy_add_n()

[Package broom.helpers version 1.3.0 Index]