regressionTable {Publish}

Regression table

Description

Tabulate the results of a regression analysis.

Usage

regressionTable(
  object,
  param.method = "coef",
  confint.method = c("default", "profile", "robust", "simultaneous"),
  pvalue.method = c("default", "robust", "simultaneous"),
  factor.reference = "extraline",
  intercept = 0L,
  units = NULL,
  noterms = NULL,
  probindex = 0L,
  ...
)

Arguments

object

Fitted regression model obtained with lm, glm or coxph.

param.method

Method to obtain model coefficients.

confint.method

Method to obtain confidence intervals. Default is 'default' which leads to Wald type intervals using the model based estimate of standard error. 'profile' yields profile likelihood confidence intervals, available from library MASS for lm and glm objects. 'robust' uses the sandwich form standard error to construct Wald type intervals (see lava::estimate.default). 'simultaneous' calls multcomp::glht to obtain simultaneous confidence intervals.

pvalue.method

Method to obtain p-values. If 'default' show raw p-values. If 'robust' use p-value corresponding to robust standard error as provided by lava::estimate.default. If 'simultaneous' call multcomp::glht to obtain p-values.

factor.reference

Style for showing results for categorical variables. If 'extraline' show an additional line for the reference category. If 'inline' display as level vs. reference.

intercept

Logical. If FALSE suppress intercept.

units

List of units for continuous variables. See examples.

noterms

Position of terms that should be ignored. E.g., for a Cox model with a cluster(id) term, there will be no hazard ratio for variable id.

probindex

Logical. If TRUE show coefficients on probabilistic index scale instead of hazard ratio scale.

...

Not yet used

Details

The basic use of this function is to generate a near publication worthy table from a regression object. As with summary(object) reference levels of factor variables are not included. Expansion of the table with such values can be performed using the "fixRegressionTable" function. Forest plot can be added to the output with "plotRegressionTable".

regressionTable produces an object (list) with the parameters deriveds. The summary function creates a data frame which can be used as a (near) publication ready table.

The table shows changes in mean for linear regression, odds ratios for logistic regression (family = binomial) and hazard ratios for Cox regression.

Value

List of regression blocks

Author(s)

Thomas A. Gerds <[e-mail]>

Examples

# linear regression
data(Diabetes)
f1 <- glm(bp.1s~age+gender+frame+chol,data=Diabetes)
summary(regressionTable(f1))
summary(regressionTable(f1,units=list("chol"="mmol/L","age"="years")))
## with interaction
f2 <- glm(bp.1s~age*gender+frame+chol,data=Diabetes)
summary(regressionTable(f2))
#Add reference values
summary(regressionTable(f2))
f3 <- glm(bp.1s~age+gender*frame+chol,data=Diabetes)
publish(f3)
regressionTable(f3)

# logistic regression
Diabetes$hyp1 <- factor(1*(Diabetes$bp.1s>140))
l1 <- glm(hyp1~age+gender+frame+chol,data=Diabetes,family="binomial")
regressionTable(l1)
publish(l1)
plot(regressionTable(l1))

## with interaction
l2 <- glm(hyp1~age+gender+frame*chol,data=Diabetes,family="binomial")
regressionTable(l2)
l3 <- glm(hyp1~age*gender+frame*chol,data=Diabetes,family="binomial")
regressionTable(l3)

# Cox regression
library(survival)
data(pbc)
pbc$edema <- factor(pbc$edema,levels=c("0","0.5","1"),labels=c("0","0.5","1"))
c1 <- coxph(Surv(time,status!=0)~log(bili)+age+protime+sex+edema,data=pbc)
regressionTable(c1)
# with interaction
c2 <- coxph(Surv(time,status!=0)~log(bili)+age+protime*sex+edema,data=pbc)
regressionTable(c2)
c3 <- coxph(Surv(time,status!=0)~edema*log(bili)+age+protime+sex+edema+edema:sex,data=pbc)
regressionTable(c3)


if (requireNamespace("nlme",quietly=TRUE)){ 
## gls regression
library(lava)
library(nlme)
m <- lvm(Y ~ X1 + gender + group + Interaction)
distribution(m, ~gender) <- binomial.lvm()
distribution(m, ~group) <- binomial.lvm(size = 2)
constrain(m, Interaction ~ gender + group) <- function(x){x[,1]*x[,2]}
d <- sim(m, 1e2)
d$gender <- factor(d$gender, labels = letters[1:2])
d$group <- factor(d$group)

e.gls <- gls(Y ~ X1 + gender*group, data = d,
             weights = varIdent(form = ~1|group))
regressionTable(e.gls)
summary(regressionTable(e.gls))
}