sjp.poly: Plot polynomials for (generalized) linear regression

In sjPlot: Data Visualization for Statistics in Social Science

Plot polynomials for (generalized) linear regression

Description

This function plots a scatter plot of a term poly.term against a response variable x and adds - depending on the amount of numeric values in poly.degree - multiple polynomial curves. A loess-smoothed line can be added to see which of the polynomial curves fits best to the data.

Usage

sjp.poly(
  x,
  poly.term,
  poly.degree,
  poly.scale = FALSE,
  fun = NULL,
  axis.title = NULL,
  geom.colors = NULL,
  geom.size = 0.8,
  show.loess = TRUE,
  show.loess.ci = TRUE,
  show.p = TRUE,
  show.scatter = TRUE,
  point.alpha = 0.2,
  point.color = "#404040",
  loess.color = "#808080"
)

Arguments

x	A vector, representing the response variable of a linear (mixed) model; or a linear (mixed) model as returned by lm or lmer.
poly.term	If x is a vector, poly.term should also be a vector, representing the polynomial term (independent variabl) in the model; if x is a fitted model, poly.term should be the polynomial term's name as character string. See 'Examples'.
poly.degree	Numeric, or numeric vector, indicating the degree of the polynomial. If poly.degree is a numeric vector, multiple polynomial curves for each degree are plotted. See 'Examples'.
poly.scale	Logical, if TRUE, poly.term will be scaled before linear regression is computed. Default is FALSE. Scaling the polynomial term may have an impact on the resulting p-values.
fun	Linear function when modelling polynomial terms. Use fun = "lm" for linear models, or fun = "glm" for generalized linear models. When x is not a vector, but a fitted model object, the function is detected automatically. If x is a vector, fun defaults to "lm".
axis.title	Character vector of length one or two (depending on the plot function and type), used as title(s) for the x and y axis. If not specified, a default labelling is chosen. Note: Some plot types may not support this argument sufficiently. In such cases, use the returned ggplot-object and add axis titles manually with labs. Use axis.title = "" to remove axis titles.
geom.colors	user defined color for geoms. See 'Details' in plot_grpfrq.
geom.size	size resp. width of the geoms (bar width, line thickness or point size, depending on plot type and function). Note that bar and bin widths mostly need smaller values than dot sizes.
show.loess	Logical, if TRUE, an additional loess-smoothed line is plotted.
show.loess.ci	Logical, if TRUE, a confidence region for the loess-smoothed line will be plotted.
show.p	Logical, if TRUE (default), p-values for polynomial terms are printed to the console.
show.scatter	Logical, if TRUE (default), adds a scatter plot of data points to the plot.
point.alpha	Alpha value of point-geoms in the scatter plots. Only applies, if show.scatter = TRUE.
point.color	Color of of point-geoms in the scatter plots. Only applies, if show.scatter = TRUE.
loess.color	Color of the loess-smoothed line. Only applies, if show.loess = TRUE.

Details

For each polynomial degree, a simple linear regression on x (resp. the extracted response, if x is a fitted model) is performed, where only the polynomial term poly.term is included as independent variable. Thus, lm(y ~ x + I(x^2) + ... + I(x^i)) is repeatedly computed for all values in poly.degree, and the predicted values of the reponse are plotted against the raw values of poly.term. If x is a fitted model, other covariates are ignored when finding the best fitting polynomial.

This function evaluates raw polynomials, not orthogonal polynomials. Polynomials are computed using the poly function, with argument raw = TRUE.

To find out which polynomial degree fits best to the data, a loess-smoothed line (in dark grey) can be added (with show.loess = TRUE). The polynomial curves that comes closest to the loess-smoothed line should be the best fit to the data.

Value

A ggplot-object.

Examples

library(sjmisc)
data(efc)
# linear fit. loess-smoothed line indicates a more
# or less cubic curve
sjp.poly(efc$c160age, efc$quol_5, 1)

# quadratic fit
sjp.poly(efc$c160age, efc$quol_5, 2)

# linear to cubic fit
sjp.poly(efc$c160age, efc$quol_5, 1:4, show.scatter = FALSE)


# fit sample model
fit <- lm(tot_sc_e ~ c12hour + e17age + e42dep, data = efc)
# inspect relationship between predictors and response
plot_model(fit, type = "slope")
# "e17age" does not seem to be linear correlated to response
# try to find appropiate polynomial. Grey line (loess smoothed)
# indicates best fit. Looks like x^4 has the best fit,
# however, only x^3 has significant p-values.
sjp.poly(fit, "e17age", 2:4, show.scatter = FALSE)

## Not run: 
# fit new model
fit <- lm(tot_sc_e ~ c12hour + e42dep + e17age + I(e17age^2) + I(e17age^3),
          data = efc)
# plot marginal effects of polynomial term
plot_model(fit, type = "pred", terms = "e17age")
## End(Not run)

sjPlot documentation built on June 22, 2024, 12:19 p.m.