flatr is a package designed to make the analysis of
contingency tables easier.
Contingency tables are a popular means of presenting categorical data
in textbooks, as they take up very little space, while still allowing to
present all the data. However, this means makes it tough to run analysis
on them. flatr helps ease this pain by turning
i × j × k contingency tables into “tidy”
data.
flatten_ct() takes an
i × j × k contingency table, and turns it
into a tibble.
goodness_of_fit() takes a logistic or probit
regression model, and does a χ2 Goodness of Fit
Test. The test statistic is one of:
flatr is designed to work with the tidyverse series of packages. Tidy
data is data in a “long” format, where each variable has its own
column.
lung_cancer
#> , , City = Beijing
#>
#> Lung
#> Smoking Y N
#> Y 126 100
#> N 35 61
#>
#> , , City = Shanghai
#>
#> Lung
#> Smoking Y N
#> Y 908 688
#> N 497 807
#>
#> , , City = Shenyang
#>
#> Lung
#> Smoking Y N
#> Y 913 747
#> N 336 598
#>
#> , , City = Nanjing
#>
#> Lung
#> Smoking Y N
#> Y 235 172
#> N 58 121
#>
#> , , City = Harbin
#>
#> Lung
#> Smoking Y N
#> Y 402 308
#> N 121 215
#>
#> , , City = Zhengzhou
#>
#> Lung
#> Smoking Y N
#> Y 182 156
#> N 72 98
#>
#> , , City = Taiyuan
#>
#> Lung
#> Smoking Y N
#> Y 60 99
#> N 11 43
#>
#> , , City = Nanchang
#>
#> Lung
#> Smoking Y N
#> Y 104 89
#> N 21 36
lung_tidy <- flatten_ct(lung_cancer)
lung_tidy
#> # A tibble: 8,419 x 3
#> Smoking Lung City
#> <fctr> <fctr> <fctr>
#> 1 Y Y Beijing
#> 2 Y Y Beijing
#> 3 Y Y Beijing
#> 4 Y Y Beijing
#> 5 Y Y Beijing
#> 6 Y Y Beijing
#> 7 Y Y Beijing
#> 8 Y Y Beijing
#> 9 Y Y Beijing
#> 10 Y Y Beijing
#> # ... with 8,409 more rows
lung_logit <- glm(Lung ~ Smoking + City, family = binomial, data = lung_tidy)
goodness_of_fit(model = lung_logit, response = "Lung", type = "Chisq")
#>
#> Chi-squared Goodness of Fit Test
#>
#> model: lung_logit
#> Chi-squared = 5.19987, df = 7, p-value = 0.63559
lung_tidy %>%
glm(
Lung ~ Smoking + City
,family = binomial(link = "probit")
,data = .
) %>%
goodness_of_fit(response = "Lung", type = "Gsq")
#>
#> G-squared Goodness of Fit Test
#>
#> model: .
#> G-squared = 5.15871, df = 7, p-value = 0.6406