Univariate and Multivariate Regressions and Their Average Marginal Effects

Computes linear or binomial regressions in two steps : univariate regressions and a multivariate regressions. All the results are nicely displayed side by side with average marginal effects.

regtab(x, y, weights = NULL, continuous = "slopes", 
 show.ci = TRUE, conf.level = 0.95)

Arguments

x

data frame. The explanatory (i.e. independent) variables used in regressions. They can be numerical or factors.

y

vector. The outcome (i.e. dependent) variable. It can be numerical (linear regression) or a factor with 2 levels (binomial regression).

weights

numeric vector of weights. If NULL (default), uniform weights (i.e. all equal to 1) are used.

continuous

character. The kind of average marginal effects computed for continuous explanatory variables. If "slopes" (defaults), these are average marginal slopes. If "predictions", these are average marginal predictions for a set of values.

show.ci

logical. Whether to display the confidence intervals

conf.level

numerical value. Defaults to 0.95, which corresponds to a 95 percent confidence interval. Must be strictly greater than 0 and less than 1.

Details

This function is basically a wrapper for regression functions in the gtsummary function. It computes a series of univariate regressions (one for each explanatory variable), then a multivariate regression (with all explanatory variables) and displays the results side by side. These results are presented in the form of average marginal effects : average marginal predictions for categorical variables and average marginal slopes (or predictions) for continuous variables.

Besides, the function is compatible with the attribute labels assigned with labelled package : these labels are displayed automatically.

Value

an object of class tbl_merge from gtsummary package

Author

Nicolas Robette

References

Arel-Bundock V, Greifer N, Heiss A (Forthcoming). “How to Interpret Statistical Models Using marginaleffects in R and Python.” Journal of Statistical Software.

Larmarange J., 2024, “Prédictions marginales, contrastes marginaux & effets marginaux”, in Guide-R, Guide pour l’analyse de données d’enquêtes avec R, https://larmarange.github.io/guide-R/analyses/estimations-marginales.html

See also

cattab, catdesc, condesc, assoc.yx, darma, assoc.twocat, assoc.twocat.by

Examples

data(Movies)
regtab(x = Movies[, c("Genre", "Budget", "Festival", "Critics")],
       y = Movies$BoxOffice)
#> The variance-covariance matrix is not positive definite. Returning the
#>   nearest positive definite matrix now.
#>   This ensures that eigenvalues are all positive real numbers, and
#>   thereby, for instance, it is possible to calculate standard errors for
#>   all relevant parameters.
#> The variance-covariance matrix is not positive definite. Returning the
#>   nearest positive definite matrix now.
#>   This ensures that eigenvalues are all positive real numbers, and
#>   thereby, for instance, it is possible to calculate standard errors for
#>   all relevant parameters.
#> The variance-covariance matrix is not positive definite. Returning the
#>   nearest positive definite matrix now.
#>   This ensures that eigenvalues are all positive real numbers, and
#>   thereby, for instance, it is possible to calculate standard errors for
#>   all relevant parameters.
#> The variance-covariance matrix is not positive definite. Returning the
#>   nearest positive definite matrix now.
#>   This ensures that eigenvalues are all positive real numbers, and
#>   thereby, for instance, it is possible to calculate standard errors for
#>   all relevant parameters.