Comparing two variables and measuring variance

gospel
gospel ⚪️

So, I am working with a dataset that contains questions from a form filled by two different users. Now I want to compare the results of these two fills based on the 'answer value' field - meaning what did user 1 select and what did user 2 select. My goal is to find the bias or variance between these two users in these questions and how they scored.

Here's an example - The interaction ID is my anchor point which tells me that it's the same evaluation these two evaluators looked at.

The output (or the Precision score) is what I am trying to get from my dataset.

Note: Orange is my expert evaluator and is evaluating if Apple is filling the forms correctly.

Using pivot tables I can construct a result like a table I want, but how to get the matched field?


My other thought is in beast mode do a calculation for Precision score like

Case when

Interaction id is the same

question id is the same

if the answer value is different - count distinct question id

then divide by

total number of distinct question id

Now if I select an evaluator can I see the precision score? - Like evaluator Apple's precision score is 60%. Note: Orange is my expert evaluator and is evaluating if Apple is filling the forms correctly.

Best Answers

  • GrantSmith
    GrantSmith Indiana 🔴
    Accepted Answer

    I'd recommend using a magic ETL to pivot your data into the table format you have above and then use a formula tile (Magic 2.0) to calculate if the answers are the same. You could do this calculation as a beast mode on the card as well but since you're already in the ETL it's better as the card doesn't have to process your static calculation every time it's loaded. That's likely the easiest option.

  • GrantSmith
    GrantSmith Indiana 🔴
    Accepted Answer

    Exactly.

Answers

  • GrantSmith
    GrantSmith Indiana 🔴

    Hi @gospel

    Are the two evaluators always the same two or will the number of evaluators be dynamic?

  • gospel
    gospel ⚪️

    @GrantSmith it will always be two. I have a primary evaluator and then someone who would randomly audit their work but it’s always two.

    one evaluator and one who audits the evaluator randomly

  • gospel
    gospel ⚪️

    Ah ok - Let me look into that. So it will look something like this? @GrantSmith