waldt function

T-test for the difference in 2 regression parameters

This is the one the students call the "fancy t test". It is just the simplest, most easy to use version of the t test to decide if 2 coefficients are equal. It is not as general as other functions in other packages. This is simpler to use for beginners. The car package's function linearHypothesis is more general, but its documentation is much more difficult to understand. It gives statistically identical results, albeit phrased as an F test.

waldt(parm1, parm2, model, model.cov = NULL, digits = getOption("digits"))

Arguments

• parm1: A parameter name, in quotes!
• parm2: Another parameter name, in quotes!
• model: A fitted regression model
• model.cov: Optional, another covariance matrix to use while calculating the test. Primarily used for robust (or otherwise adjusted) standard errors
• digits: How many digits to print? This affects only the on-screen printout. The return object is numeric, full precision.

Returns

A vector with the difference, std. err., t-stat, and p value. Prints a formatted output statement.

Details

I did this because we have trouble understanding terminology in documentation for more abstract functions in other R packages.

It has an additional feature, it can import robust standard errors to conduct the test.

Examples

mdat <- data.frame(x1 = rnorm(100), x2 = rnorm(100))
stde <- 2
mdat$y <- 0.2 * mdat$x1 + 0.24 * mdat$x2 + stde * rnorm(100)
m1 <- lm(y ~ x1 + x2, data = mdat)
waldt("x1", "x2", m1)
waldt("x1", "x2", m1, digits = 2)
## Returned object is not "rounded characters". It is still numbers
stillnumeric <-  waldt("x1", "x2", m1, digits = 2)
stillnumeric
## Equivalent to car package linearHypothesis:
if(require(car)){
    linearHypothesis(m1, "x1 = x2")
}
## recall t = sqrt(F) for a 1 degree of freedom test.
## If we could understand instructions for car, we probably
## would not need this function, actually.

Author(s)

Paul Johnson pauljohn@ku.edu