QWK incorrect for non-integer values

[burakaydin]

ScoreQuadraticWeightedKappa produces results for non-integer values but results are incorrect. I could not figure out how it handles non-integer values but it does not round them for sure.

[mfrasco]

You are correct. ScoreQuadraticWeightedKappa does not work if the inputs are non-integer values. However, the documentation does specify that the inputs should be two integer vectors. In cases like this, I feel like there are three broad options

1. Assume that the user provided the desired inputs and execute the code
2. Check if the user provided the desired inputs. If the inputs do not meet expectation, throw an error.
3. Check if the user provided the desired inputs. If the inputs do not meet expectation, try to change the inputs to meet the expectation. For example, round floats to integers.

The code in this package takes the first approach across all of its functions. In my experience, that's how a lot of functions in R work. Instead of raising errors, they might do things that you don't expect.

I can definitely see the argument for taking approach 2.

I'm struggling to see why approach 3 is the best. There are lots of ways to round floats to integers. Should we round up, down, away from 0.5, towards even numbers, or towards odd numbers? Of course, we could add an argument to the function that lets the user control how the rounding is done. However, that doesn't seem like the core responsibility of this function. To me, implementing rounding behavior is something that should be done elsewhere in the code based on the specific problem you are trying to solve.

What do you think? I'm curious to hear your perspective. Thanks for raising the issue. I appreciate you using the package.

[burakaydin]

Thanks for the reply.
In my experience the first approach almost never happens with credible packages. It should have at least thrown a warning if not an error.
That being said, frankly, I do not see why this function only works with integers. I compared its results using python and manuel computations. That is how I realized this issue. My point is maybe it is a quick fix?
I appriciate your effort to support open science.