Using chaid_table

In CGPfunctions: Powell Miscellaneous Functions for Teaching and Learning Statistics

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

Chi-square automatic interaction detection (CHAID) is a decision tree technique, based on adjusted significance testing (Bonferroni testing). The technique was developed in South Africa and was published in 1980 by Gordon V. Kass, who had completed a PhD thesis on this topic. Wikipedia

I've written a few blog posts about using this venerable technique. The default output using print() and plot() is sparse, elegant and useful as is, and can be adjusted in many ways that I have documented here.

The code chunk below is taken straight from the example in the help page for CHAID::chaid

The dataset is based on data from a post-election survey on persons who voted for either Bush or Gore in the 2000 U.S. election. The specific variables are related to the publication of Magidson and Vermunt (2005).

Further information (and datasets) about the 2000 U.S. election and other National Election Studies is available on the American National Election Studies Web site (https://electionstudies.org/).

library(CGPfunctions)
# library(CHAID)
library(dplyr)
library(knitr)

### fit tree to subsample see ?chaid
##  set.seed(290875)
##  USvoteS <- USvote[sample(1:nrow(USvote), 1000),]

##  ctrl <- chaid_control(minsplit = 200, minprob = 0.1)
##  chaidUS <- chaid(vote3 ~ ., data = USvoteS, control = ctrl)
print(chaidUS) 
plot(chaidUS)

Getting more detailed

But what happens if you want to investigate your results more closely? I actually had at least one reader of my blog posts point out that some other statistical packages produce more detailed information in a tabular format.

Some additional things you may want to know that are actually contained in chaidUS but aren't especially easy to see from the default plot and print and aren't necessarily easy to ferret out of the chaidUS object:

1. It's clear from our call that we fed chaid 1,000 random rows of the dataset USvote. It must be safe to assume that's how many valid cases there were right? The documentation is mute on how it handles missing cases.

2. There's information about counts and frequencies in the terminal nodes (3, 4, 6, 8, 9) and we could manually calculate the answers to other questions like how many people voted for Bush in node #9 (115 * 40.9% = 47)? How many total voters are in node #2 (311 + 249 = 560)? But getting more information gets increasingly tedious. At node #2 what was the breakdown of votes for Bush -v- Gore (hint 245 -v- 315 respectively).

3. It would also be nice to have easy access to the results of the $\chi^2$ tests that are the inherent workhorse of CHAID. We know that marstat was selected as the first split by virtue of having the smallest p value after a Bonferroni adjustment, but what were the results?

chaid_table attempts to provide much more granular information in a tibble and also make it possible for you to derive even more nuanced questions through piping operations. The simplest call is just to feed it the name of the object after it has been processed by CHAID. That object will be of class "constparty" "party".

# simplest use --- chaidUS is included in the package
class(chaidUS)

chaid_table(chaidUS)

mychaidtable <- chaid_table(chaidUS)

I debated the wisdom of providing tables as output in the manner I found most useful and aesthetically pleasing but in the end decided to simply provide the tibble and let the user decide what and how to format.

Example uses

Some easy examples using kable and dplyr.

mychaidtable %>% 
  select(nodeID:ruletext) %>% 
  kable()

# Just node #2 show percentage
mychaidtable %>% 
  select(nodeID:ruletext) %>% 
  filter(nodeID == 2) %>% 
  mutate(pctBush = Bush/NodeN * 100) %>%
  kable(digits = 1)

# Just the children of node #5
mychaidtable %>% 
  select(nodeID:ruletext) %>% 
  filter(parent == 5) %>% 
  kable()

# stats for all splits including raw (unadjusted) p value
mychaidtable %>% 
  select(nodeID, NodeN, split.variable:rawpvalue) %>% 
  filter(!is.na(split.variable)) %>% 
  kable()

Hopefully those are enough examples to get your creative juices going.

Leaving Feedback

If you like CGPfunctions, please consider Filing a GitHub issue by leaving feedback here, or by contacting me at ibecav at gmail.com by email.

I hope you've found this useful. I am always open to comments, corrections and suggestions.

Chuck

License

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.

CGPfunctions documentation built on Nov. 12, 2020, 5:08 p.m.