validate: Common regression fit statistics in a vector.

In robertschnitman/schnitr: Diagnostic tools for regression modeling

Common regression fit statistics in a vector.

Description

Common regression fit statistics in a vector.

Usage

validate(model, dataframe = FALSE, ...)

Arguments

model	An lm, glm, or nls object.
dataframe	Logical. FALSE (default) outputs a matrix; TRUE outputs a dataframe.
...	Arguments passed to resid().

Details

The broom library's glance() had a vague label for the F statistic (simply "statistic") and lacked the pseudo R-squared, which is commonly based on McFadden's version (i.e. 1 - (residual deviance / null deviance)). While the same function is friendly for data frames, it's wide form is cumbersome for quickly ascertaining model validity. Thus, validate() produces similar output as a column vector. Those who wish to have the values in broom's format can always transpose the vector.

Value

Vector or dataframe. Includes F-statistic, R-squared, RMSE, and others.

Output definitions (alphabetical order)

adj.rsq = Adjusted R-Squared.

aer = Apparent Error Rate, calculated as the proportion of misclassifications (i.e. number of incorrect / total cases). Cutoff is the proportion of positive cases.

AIC = Akaike Information Criterion.

BIC = Bayesian Information Criterion.

convergence_tolerance = Tolerance of convergence, calculated from summary(model)$convInfo$finTol

df.den = degrees of freedom, denominator.

df.null = Degrees of freedom for the null deviance.

df.num = degrees of freedom, numerator.

df.sigma = degrees of freedom for sigma.

F.stat = F statistic

iterations = Number of iterations for NLS model to converge.

loglik = Log Likelihood.

mad = Median Absolute Deviation.

mae = Mean Absolute Error.

mpe = Mean Percentage Error.

medianpe = Median Percentage Error.

n = number of observations used in the model.

null.deviance = Null Deviance.

p.value = p-value for the F statistic.

pseudo.rsq.mcfad = McFadden's Pseudo R-Squared, calculated as 1 - (residual.deviance/null.deviance).

residual.deviance = Residual Deviance.

residual.mean = Mean of the residual.

residual.median = Median of the residual.

residual.sd = Standard deviation of the residual.

residual.se = Standard error of the residual.

rmse = Root Mean Square Error, calculated as sqrt(mean(resid(model)^2)).

rsq = R-squared.

sdpe = Standard Deviation of the Percent Error.

sepe = Standard Error of the Percent Error (sd(residuals

sigma = Standard deviation of the NLS model, calculated from summary(model)$sigma

See Also

https://github.com/robertschnitman/diagnoser

Examples

model.lm <- lm(data = mtcars, formula = mpg ~ wt + gear)
validate(model.lm, TRUE)

model.glm <- glm(data = mtcars, am ~ mpg + wt, family = binomial(link = 'logit'))
validate(model.glm)

model.nls <- nls(Ozone ~ theta0 + Temp^theta1, airquality)
validate(model.nls)

robertschnitman/schnitr documentation built on Aug. 18, 2022, 8:39 p.m.