In robertgambrel/tabler: Make Publication Tables from Multiple Model Results

This vignette will show a few examples of how to use this package and alter what is output. It will also highlight some restrictions on formats that the user can request.

Default values

The main function used from this package is tablify. This function accepts a series of model output objects as its main arguments, converts the outputs to dataset columns (using the tidy function from the broom package), and then outputs two objects:

• A cleaned dataset where each column contains one model and each row contains regression results
• A string caption that explains the values of significance stars in the model

library(tabler)

lm1 <- lm(mpg ~ wt, data = mtcars)
lm2 <- lm(mpg ~ wt + cyl, data = mtcars)

tablify(lm1, lm2)

If the user wants to manually manipulate the dataset of regression output:

start_table <- tablify(lm1, lm2)[[1]]

In general, the internal R object output is meant to be intermediary, or used as a preview before exporting the table to share. Exporting is achieved by:

library(WriteXLS)
tablify(lm1, lm2, file = "myresults.csv")
tablify(lm1, lm2, file = "myresults.xls")
tablify(lm1, lm2, file = "myresults.xlsx")

Outputs to xls or xlsx formats requires the WriteXLS package.

Customizing Outputs

The default output can be altered in several ways:

• teststat: by default is "p.value", but can also be "std.error" or "statistic"
• "statistic" will be a t-value or z-value and is model-dependent
• digits: by default is 3, and specifies the number of decimal rounding places
• digits_coef and digits_teststat take on the digits value, unless they are explicitly specified. These allow more control over rounding coefficient vs. significance test outputs
• cutoffs: by default c(0.1, 0.05, 0.01), these specify the cutpoints at which significance markers are displayed. They must be presented in decreasing order, and any number are allowed.
• stars: by default c("*", "**", "***"), this list must be the same length as cutoffs and again in order of weakest significance to strongest significance
• intercept_last: by default TRUE, moves the intercept output to the bottom of the table (but before observation count and model fit statistics). If FALSE, reports it first.
• N: by default TRUE, adds a row to the table displaying observation count
• fit: by default NULL, can also be a string or list of fit statistics available in a regression output ("r.squared", "adj.r.squared")
• file: by deafult NULL, specifies the output file name and path if desired

lm3 <- lm(mpg ~ cyl + disp + hp + drat + wt + qsec + vs, data = mtcars)
tablify(lm3)
tablify(lm3, cutoffs = c(0.2, 0.1, 0.05))

# error:
# tablify(lm3, cutoffs = c(0.9, 0.7, 0.5, 0.3, 0.1))

# correct - must supply enugh significance denoters
tablify(lm3, cutoffs = c(0.9, 0.7, 0.5, 0.3, 0.1), 
        stars = c('?', '+', '*', '**', '***'))

tablify(lm1, lm2, lm3, fit = c("r.squared", "adj.r.squared"))

Final note on model types

Since this package relies on broom to tidy model outputs into the columns for tables, it is currently limited to model types handled by broom. For the latest list, see the developer's site. Most frequently-used models should be ready to go, though.

Models of different types (i.e. OLS and ordinal / binary logit) can be combined into one table, so long as the requested outputs are common between all of them. For example, requesting coefficients and their p-values will work, but requesting $R^2$ will fail, as ordinal models do not include this in their outputs.

robertgambrel/tabler documentation built on May 27, 2019, 10:32 a.m.