umx!

Statistical helper functions

There are dozens of useful helper functions in the umx library. A great overview of these is sitting in the package help (?umx). It’s organized into families of functions to help you navigate.

umxAPA

Many functions in R present data in ways designed for programming pipelines, and to be highly modular. This is great for programming, but not so good in helping you quickly create tables, and text-summaries of the results of a hypothesis test.

umxAPA is a flexible function to summarize, process, and present data in ways that are suitable for viewing or inclusion in publications. If we walk through a writeup, it’s value can soon be seen.

First, you might need descriptive statistics. Given a dataframe, umxAPA will return a table of correlations, with the mean and SD of each variable as the last row. So, umxAPA(mtcars[,c(“cyl”, “wt”, “mpg”, )] yields a table of correlations, means and SDs thus:

Next, having run a linear model, you’ll want to tell readers what the result was, with an effect, CI, and p-value formatted correctly.

Given an lm model, umxAPA will return a formatted effect, including 95% CI in square brackets, for each effect. or all of the effects (specified by name in se). e.g.:

m1 = lm(mpg~wt, data = mtcars)
umxAPA(m1, "wt")

β = -5.34 [-6.49, -4.2], t = -9.56, p < 0.001

You can also just ask for one effect, by setting the name that appears in the anova table. You can also standardize the model. An example of this is:

m1 = lm(mpg ~ wt + disp, data = mtcars)
umxAPA(m1, "wt", std = TRUE)

β = -0.54 [-0.93, -0.16], t = -2.88, p = 0.007

Perhaps while writing the introduction, you hit data from another paper that gave only a beta and se, and you’d like the CI. umxAPA returns a CI based on 1.96 times the se. e.g. umxAPA(.3, se=.1) yields: β = 0.3 [0.1, 0.5].

While the digits= parameter of umx functions often gives you numbers to the precision you need, often you’ll need to format a p-value. Just give umxAPA a single value, it will be treated as a p-value as returned in APA format. e.g., umxAPA(.000034) gives “< 0.001”

You will often want to report a correlation. Pass in a cor.test object to umxAPA.

m1 = cor.test(~mpg + wt, data = mtcars)
umxAPA(m1)

yields: r = -0.87 [-0.93, -0.74], t(30) = -9.56, p = < 0.001

umxAPA works the same way for several other objects, including t.test:

m1 = t.test(extra ~ group, data = sleep)
umxAPA(m1)

Means were ‘0.75’ and ‘2.33’. CI[-3.37, 0.21]. t(17.78) = -1.86, p = 0.08

On my TODO list are tutorial blogs about:

• umx_aggregate
• umx_names
• umx_print
• umx_time
• FishersMethod
• umxWeightedAIC
• tmx_is.identified
• tmx_show