linear.tools (version 1.3.0)

effect: evaluate the marginal effects of the selected raw variable on the dependent.

Description

evaluate the marginal effects of the selected raw variable on the dependent.

Usage

effect(model, data = NULL, focus_var_raw, focus_var_coeff = NULL, focus_var_model = NULL, focus_value = NULL, nonfocus_value = NULL, transform_y = NULL, PRINT = TRUE, PLOT = TRUE, Reverse = FALSE, bar_plot = NULL, intolerance_on_wrong_names = FALSE)

Arguments

model

an output of lm or glm

data

NULL (default) or a data.frame, a new dataset to evaluate the categorical variables. If NULL, then use the data used in model itself.

focus_var_raw

NULL or a character vector with maximum length of 2, in which you can choose raw vars you want to focus. See get_x for the meaning of raw var.

• If there is only one raw var in the vector focus_var_raw, then we will check the marginal impact of that raw var.
• If there is only two raw vars in the vector focus_var_raw, then we will check the marginal impact of the FIRST raw var (focus_var_raw[1]) under different values of SECOND raw var (focus_var_raw[2]).

See the example code for details.

focus_var_coeff

NULL or a character vector. Must be coeff vars containing focus_var_raw[1]. See get_x for the meaning of coeff var. After you set up the focus_var_raw, you can also choose to focus on effects of focus_var_raw[1] through only certain coeff vars, then all other unspecified coeff vars related focus_var_raw[1] will have coeff 0 by default, focus_var_coeff is null, which means we will check effect of focus_var_raw[1] on all coeff vars.

See the example code for details.

focus_var_model

NULL or a character vector. Must be model vars containing focus_var_raw[1]. See get_x for the meaning of model var. Similar use as argument focus_var_coeff, except here you can specify which model vars you want to focus.

See the example code for details.

focus_value

NULL or a list; each element of the list must have names in focus_var_raw. By default, we will check marginal effects of focus_var_raw[1] through seq(0.05,0.95,by = 0.05) quantiles of its values in the modelling data. But you can also specify the values you want to check here. See the sample code.

nonfocus_value

NULL or a list; each element of the list must have names in non-focused raw vars (not show up in focus_var_raw) The meaning of non-focus var is: When we check the marginal effect of focus var on dependent, we let the focus var vary and fix the non-focus vars. By default, for non-focused raw vars, we assume their values are fixed at mean (if numeric) or mode (if factor or character) in the modelling data. But you can also specify the fixed values you want. See the sample code.

transform_y

NULL or a function, used only for plot. Used as a function to recalculate y (a function on y (ex. log(y) )).

PRINT

a boolean, whether to print messages AND to plot.

PLOT

a bookean, whether to plot

Reverse

a boolean, whether to use reverse order in x-axis when plot. Default is FALSE.

bar_plot

NULL or a boolean, choose bar plot or line plot. If NULL, we will choose automatically.

intolerance_on_wrong_names

a boolean. If a name is wrong, either in focus_var_raw, focus_var_model, focus_var_coeff, focus_value or nonfocus_value, whether we delete the wrong names and go on (default), or report an error.

Value

a list:

• Focus_values: show the values of focus_var_raw we used to evaluate the marginal effects.
• data_and_predict: full dataset used to evaluate the marginal effects.
• summmary_glm: a summary of lm or glm model.
• Monoton_Increase: whether the marginal impact is Monotonic Increase.
• Monoton_Decrease: whether the marginal impact is Monotonic Decrease.

Details

This function will evaluate marginal impacts and show the monotonicity of marginal impacts of a selected variable on the dependent.

Note that the marginal impacts is not simply the sign of coeff: In a model like y~ x + x^2 + p + q, marginal impacts of x on y requires an evaluation of both x and x^2 at the same time.

Here the focus_var_raw is x, focus_var_coeff are x and x^2 nonfocus_value is p and q

Also the monotonicity of marginal impacts of x will be different for different range of x's values.

Another interesting case is when x is interacting with other variables, then its marginal impacts will also be dependent on the values of those interacted variables.

Level of marginal impacts: To make the level of marginal impacts of x realistic, by default we fixed all other right-hand-side variables fixed at their mean (numeric) or mode (character or factor). You can also provide fixed values for them. Also by default we let the interested variable (focused raw var) x to vary between its seq(0.05,0.95,by = 0.05) quantiles.

This function will take care those cases above and make evaluating marginal impacts easier.

Examples

##___ unit test ____

# __________________  One Dimension: the most basic case ____________________



set.seed(413)
traing_data = ggplot2::diamonds[runif(nrow(ggplot2::diamonds))<0.05,]
nrow(traing_data)

diamond_lm3 = lm(price~ cut + carat + I(carat^2) +
                   I(carat^3) + I(carat  * depth) + cut:depth, traing_data) # a GLM

# more carats, higher price.
effect(model = diamond_lm3,
       data = traing_data,
       focus_var_raw = c('carat'),
       Reverse = TRUE) # value in x-axis is reverse

# focus on only 'I(carat^3)', which means we will make all other coeff,
# including 'carat' and 'I(carat^2)' into 0
effect(model = diamond_lm3,
       data =traing_data,
       focus_var_raw =c('carat'),
       focus_var_coeff = 'I(carat^3)')
# __________________  One Dimension: Categorical ____________________

# selected model-var to focus: here not focus on cut:depth, only focus on cut
suppressWarnings(
  effect(model = diamond_lm3,
         data = traing_data,
         focus_var_raw = c('cut'),
         focus_var_model = 'cut'
         )
  )

# __________________  Double Dimensions ____________________

# here focus_var_raw has two values: "carat" and "cut"
# that means we will evaluate impact of "carat" on "price" through different value of "cut"

effect(model = diamond_lm3,data = traing_data, focus_var_raw=c('carat',"cut"))

# __________________  Provide Values to Focused vars  ____________________

# when evaluating impacts,
# we can provide the range of values for key variables

effect(model = diamond_lm3,data = traing_data,
       focus_var_raw = c('carat',"cut"),
       focus_value = list(carat=seq(0.5,6,0.1)))