getUnivariate: Run "univariate" regressions for each variable in a model,...

In Dectech/DectechR: Collection of useful functions for Dectech Analysts

Run "univariate" regressions for each variable in a model, and copy to clipboard

Description

Based on the input model (or input formula), runs a series of models predicting the dependant variable with one independant variables at a time (univariate).

Usage

    getUnivariate(mod = NULL,
                  full_formula = NULL,
                  df = NULL,
                  model_class = NULL,
                  model_family = NULL,
                  returnIntercept = FALSE)

    # If starting with a fitted model:
    getUnivariate(mod,
                  returnIntercept = FALSE)

    # If starting with a formula and data:
    getUnivariate(full_formula,
                  df,
                  model_class,
                  model_family, # only required for glm models
                  returnIntercept = FALSE)

Arguments

mod	A model object. Support types include lm, glm, polr.
full_formula	A formula object, for use when starting without a fitted model
df	A data frame, for use when starting without a fitted model
model_class	A string describing the model class e.g. "lm", "glm", or "polr". For use when starting without a fitted model
model_family	A string describing the model familily required in the glm model e.g. "binomial". For use when starting without a fitted model
returnIntercept	(optional) if set to TRUE will also include a column for the intercept value (or for an ordinal logit regression, it will return columns for each boundary)

Details

Loops over each independent variable in the input model, predicting the dependent variable using that IV alone i.e. "univariate". Hence each row represents a seperate model*.

Input should be a fitted model with all the variables you want to test OR a formula object with all the variables you want to test. If a formula supplied then you also need to supply the data and the class of model e.g. "lm" (plus model family for glm models).

This is useful for comparing against a multivariate model containing the same IVs. The univariate results show the individual impact of each IV, while the full regression shows how they might interact. For example if two IV's are correlated, they might both come out as significant predictors in a univariate regression, but in a full regression their betas may impact each other in unpredictable ways (they may be suppressed or exagerated depending on the nature of their interaction). For lm and glm regressions you should also refer to the "tolerance" outputted by getOutput to identify collinearity.

* NB: For factor and categorical variables, each level is outputted on its own line (except the baseline), but it's important to remember only one model is actually run. For example, consider predicting age using an "Income" variable with levels "Low", "Med" and "High". The univariate model run would be: lm(age ~ Income), and there would be two betas added to the result table, one for "IncomeMed" and one for "IncomeHigh". In these cases, if returnIntercept is set to TRUE, there is only one model intecept to output, so this value would be repeated for each level of this factor.

See Also

See also runUnivariate which returns the result as a dataframe rather than copying to clipboard

Examples

    # Running a linear regression using the built in mtcars data set...
    m1 = lm(mpg ~ gear + factor(carb) + hp, mtcars)
    # to run unvariate results for each IV and copy result to clipboard:
    getUnivariate(m1)

    # to also return the intercept for each univariate model, run:
    getUnivariate(m1, returnIntercept = T)


    # Running a version with a formula input...
    getUnivariate(full_formula = mpg ~ gear + factor(carb) + hp,
                  df = mtcars,
                   model_class = "lm")

Dectech/DectechR documentation built on Feb. 15, 2024, 9:17 a.m.