Tableau Analytics: Dynamic Two-Sample t -Test with R

Inspiration

Need a dashboard that clients can perform any combination of two-sample for a t-test of group differences. Thus, Tableau developers can be relieved from writing too many R scripts for all possible pairs of two-group combinations that clients may be interested in.

What it does

Dynamically conducts a two-sample t-test for whichever two groups are selected.
Displays a customized & friendly, instead of the default, error message when the user fails to pick exactly two groups for the t-test.
Changeable alpha level for the hypothesis testing.

How we built it

Tableau, R, and Rserve

Challenges we ran into

A. Displaying the friendly error message took me a while to figure it out.

The hack is to use the if-else express in the R script to check how many distinct groups have been received. If the group count is not equal to 2, return a -9 error code back to Tableau; otherwise, conduct the t-test and return the p-value.
    if ( length(table(.arg2)) != 2){
          -9
    } else {
          ttest     <- t.test(.arg1 ~ .arg2, paired = FALSE, alternative="two.sided")
          ttest[["p.value"]]
    }

B. Get exactly whichever two groups of the selection to be sent into the Rserve for calculation.

When a user does select exactly two groups for the t-test. Use

        {FIXED : MIN([Group Name])}
        {FIXED : MAX([Group Name])}

to identify the two groups. This method works fine when users selected more than two groups too. So, it can prevent some errors when doing the R analytics.

C. The Rserve response speed is not satisfactory.

I have tried to write as few as possible R code.

Accomplishments that we're proud of

Using only a few lines of R script (including error prevention), for all possible pairs of two-group combinations.
Being able to display a friendly message to inform the user to select two groups only.

What we learned

More sophisticated Tableau Analytics can be make ready for and be utilized easily by our clients.
Since Tableau doesn't have a try and catch capability, developers need to implement the script to prevent the dashboard from displaying unfriendly messages when analytics running into errors.