blog/chi-square-test-of-independence-in-r/

[utterances-bot]

Chi-square test of independence in R - Stats and R

Learn when and how to use the Chi-square test of independence in R. See also how it works in practice and how to interpret the results of the Chi-square test

https://statsandr.com/blog/chi-square-test-of-independence-in-r/

[AntoineSoetewey]

Comment written by Herivelto Cordeiro dos Santos on November 30, 2020 11:55:35:

Hi Antoine! I realy liked your blog, I think is clear and focus! Let me ask you about this one... I have tried the scripts method #1 and #2 and I got different p-values. Do you know what could be the reason for that as they should be equal? My table is matrix (c(63,78,94,65), ncol=2).

[AntoineSoetewey]

Comment written by Herivelto Cordeiro dos Santos on November 30, 2020 11:55:35:

Hi Antoine! I realy liked your blog, I think is clear and focus! Let me ask you about this one... I have tried the scripts method #1 and #2 and I got different p-values. Do you know what could be the reason for that as they should be equal? My table is matrix (c(63,78,94,65), ncol=2).

Comment written by Antoine Soetewey on November 30, 2020 12:20:12:

Thank you for your feedback!

I just tried on my side, and I have the same p-values with your data, see my code here.

One potential reason you have different p-values is due to the fact that first method use the Yate's continuity correction by default. Add the argument correct = FALSE in the chisq.test() function to prevent from applying this continuity correction.

(I've added a note at the end of this section following your comment.)

Hope this helps.

Regards,
Antoine

[Clarice3]

Hi Antoine! Thank you so much for your blog, it's very helpful! I have a question for you: I have a data frame with several (10) different categorical variables that I would like to test for possible correlations between each other. Is there a way that I can test them all at the same time, like you explained for the quantitative variables? Or is it really only possible for two variables at the same time? Not sure how I would do this for 10 variables...
Thanks in advance!
Regards,
Clarice

[AntoineSoetewey]

Hi Antoine! Thank you so much for your blog, it's very helpful! I have a question for you: I have a data frame with several (10) different categorical variables that I would like to test for possible correlations between each other. Is there a way that I can test them all at the same time, like you explained for the quantitative variables? Or is it really only possible for two variables at the same time? Not sure how I would do this for 10 variables...
Thanks in advance!
Regards,
Clarice

Dear Clarice,

Do you want to compute correlation coefficients or perform chi-square tests? You mentioned correlations but you posted the comment on the article about chi-square test, so I'm not sure.

For correlation, if your categorical variables are ordinal, you can simply use cor(dat, method = "spearman"), where dat is the name of your dataframe. See more details in this article about correlation coefficient in R.

The standard Chi-square test for independence (with the chisq.test() function and presented in this article) is only possible between two categorical variables at the same time, so you'd need to tweak your code a bit to do it for all possible pairs of variables. Or if your dataset contains a relatively small number of variables, you can copy paste your code for each pair of variables.

Hope this helps.

Regards,
Antoine

[Clarice3]

Hey Antoine! Thank you so much for your quick answer. My goal is to find out whether the different variables correlate with each other or not, so I can exclude them before computing a model. From your blog I learned that it’s not possible to compute correlation coefficients between two categorical variables (if I understood that correctly?) but only to do a contingency analysis. Some of my categorical variables are ordinal with 3-4 levels, most of them are nominal though. I tried to use the corr() function that you suggested, but unfortunately I just can’t make it work with my R version… not sure why. So I’ll have to find another way I guess! Best wishes, Hanna

[AntoineSoetewey]

Hey Antoine! Thank you so much for your quick answer. My goal is to find out whether the different variables correlate with each other or not, so I can exclude them before computing a model. From your blog I learned that it’s not possible to compute correlation coefficients between two categorical variables (if I understood that correctly?) but only to do a contingency analysis. Some of my categorical variables are ordinal with 3-4 levels, most of them are nominal though. I tried to use the corr() function that you suggested, but unfortunately I just can’t make it work with my R version… not sure why. So I’ll have to find another way I guess! Best wishes, Hanna

You understood correctly:

• You can compute the correlation between your ordinal variables (thanks to the cor() function, with only one r and not two as you wrote in your comment),
• But for your nominal variables, you cannot compute the correlation. You'll need to apply the Chi-square test of independence (with the chisq.test() function).

Regards,
Antoine

[Clarice3]

Alright, I get it. Sorry for the confusion and thanks for your help! Best wishes, Hanna

[venkatpgi]

Hello Mr Antoine,
It was a nice article about Chi squared test.
I am trying to replicate your earlier blog on using t.test and ANOVA on multiple columns at the same time in the case of Chisquared test. But unable to.
I tried using looping
comparison1 <- lapply(df[, 1:4], function(x)t.test(x~df$var)) - this worked for t.test but not for chi squared test
Any suggestions?

[AntoineSoetewey]

Hello Mr Antoine,
It was a nice article about Chi squared test.
I am trying to replicate your earlier blog on using t.test and ANOVA on multiple columns at the same time in the case of Chisquared test. But unable to.
I tried using looping
comparison1 <- lapply(df[, 1:4], function(x)t.test(x~df$var)) - this worked for t.test but not for chi squared test
Any suggestions?

Hello,

Here is a reproducible example using for loop:

df <- data.frame(sex = sample(c("male", "female"), size = 100, replace = TRUE),
                 smoke = sample(c("smoker", "non smoker"), size = 100, replace = TRUE),
                 sport = sample(c("athlete", "non athlete"), size = 100, replace = TRUE))

for (i in 2:ncol(df)) {
  print(names(df)[i])
  print(chisq.test(table(df[, 1], df[, i])))
}

Hope this helps.

Regards,
Antoine

[MSIB76]

Hi Antoine, really like your blog and the amazing work you do here.
If you don't mind I have a question about the same topic regarding looping. I know how to do the Chi square through multiple columns thanks to your explanation. However, I still need to do two more thing here. I want to include in this loop a way to bring out in addition to the chi value and p value the Phi and Cramer'V, and the expected values. Because I want to check these expected values to detect which test needs to be done by Fisher.
Another thing, how I can deal with the NAs with <chisq.test(table())> or other functions for Chi square?
Sorry for the troubles.
Thanks

Mudathir

[AntoineSoetewey]

I want to include in this loop a way to bring out in addition to the chi value and p value the Phi and Cramer'V, and the expected values. Because I want to check these expected values to detect which test needs to be done by Fisher.

You can do so thanks to the assocstats() function in the {vcd} package. It gives the test statistic and the p-value for the Chi-square test and the Phi and Cramer's V. Expected values are found with $expected

set.seed(42)
df <- data.frame(sex = sample(c("male", "female"), size = 100, replace = TRUE),
                 smoke = sample(c("smoker", "non smoker"), size = 100, replace = TRUE),
                 sport = sample(c("athlete", "non athlete"), size = 100, replace = TRUE))

library(vcd)

for (i in 2:ncol(df)) {
  print(names(df)[i])
  print(assocstats(table(df[, 1], df[, i])))
  print(chisq.test(table(df[, 1], df[, i]))$expected)
}

Another thing, how I can deal with the NAs with <chisq.test(table())> or other functions for Chi square?

Nothing to do here with the NA. Try by replacing a value by NA.

Hope this helps.

Regards,
Antoine

[open-source-rs]

Trouble installing summarytools on r studio server

I'm building a shiny dashboard on an r studio server
I'd like to try out the chi square example in the application i'm building
However, I cannot get the summarytools to install

here's the error

installation method:

This is the recommended method, as some minor fixes and improvements are regularly added.

install.packages("remotes") # Using devtools is also possible
library(remotes)
install_github("rapporter/pander") # Necessary for optimal results!
install_github("dcomtois/summarytools")

result from install in console:

> runApp('dashboard/dashboard')
Error in library(summarytools) : 
  there is no package called ‘summarytools’

✔ Package 'summarytools' successfully installed.
Warning messages:
1: In utils::install.packages("magick", repos = "https://cloud.r-project.org") :
  installation of package ‘magick’ had non-zero exit status
2: In utils::install.packages("summarytools", repos = "https://cloud.r-project.org") :
  installation of package ‘magick’ had non-zero exit status
3: In utils::install.packages("summarytools", repos = "https://cloud.r-project.org") :
  installation of package ‘summarytools’ had non-zero exit status

If i try to run the app after attempting to install summarytools it crashes with:

Error in library(summarytools) : 
  there is no package called ‘summarytools’

installing them in r has no affect

jimbo

[open-source-rs]

also - here's the libraries i currently have installed for the dashboard

#
# This is a Shiny web application. You can run the application by clicking
# the 'Run App' button above.
#
# Find out more about building applications with Shiny here:
#
#    http://shiny.rstudio.com/
#

# Libraries to load

library(shiny)
library(shinydashboard)
library(googlesheets4)
library(markdown)
library(DT)
library(ggplot2)
library(plotly)
library(rmarkdown)
library(knitr)
library(pander)

[open-source-rs]

so i was able to install magick in the background using command line
after that i tried to install summarytools
looks like it worked

libraries in the code now look like this:

# Libraries to load

library(shiny)
library(shinydashboard)
library(googlesheets4)
library(markdown)
library(DT)
library(ggplot2)
library(plotly)
library(rmarkdown)
library(knitr)
library(pander)
library(summarytools)

dashboard launches without error
will get to testing some chi square now

[AntoineSoetewey]

Dear Jimbo,

Glad you fixed it!

[open-source-rs]

with regard to this piece of code in the example:

library(summarytools)
library(dplyr)

# fourth method:
dat %$%
  ctable(Species, size,
    prop = "r", chisq = TRUE, headings = FALSE
  ) %>%
  print(
    method = "render",
    style = "rmarkdown",
    footnote = NA
  )

i don't know what this means

dat %$%

sorry for my newbness...i just use something like this in any of my calculations:

    output$plotRegression <- renderPlot({
        plotRegressionModel <- lm(dbis ~ ciatas, data = dbrp_data1)
        summary(plotRegressionModel)
        ggplot(dbrp_data1, aes(dbis, ciatas)) + geom_point() + stat_smooth(method = lm) +
            labs(title = "Regression Model", x = "some lable", y = "another lable")
    })

cheers ~!

[AntoineSoetewey]

Hello Jimbo,

%$% and %>% are just pipe operators from {magrittr} package.

So in some sense dat %$% ctable(), simply means "take the dataset called dat and execute the ctable() function on this data". It is equivalent than with(dat, ...).

If you are unfamiliar with pipes, you can simply run the following code and it will give the same results:

ctable(dat$Species, dat$size, prop = "r", chisq = TRUE, headings = FALSE)

Hope this helps.

Regards,
Antoine

[auva1146]

Hi Antoine, regarding this:
"Gather some levels (especially those with a small number of observations) to increase the number of observations in the subgroups." how do you do this?

[AntoineSoetewey]

Thanks for your question.

For example you have a categorical variable of 3 levels, with the following occurrences:

Level 1: 2 observations
Level 2: 4 observations
Level 3: 17 observations

You can gather levels 1 and 2 to make:

Level 1: 6 observations
Level 2: 17 observations

This gathering makes even more sense if the levels that you combine are "similar".

Hope this helps.

Regards,
Antoine