Open biometrician opened 2 years ago
As a suggestion: At the beginning the response could be set as the linear effect.
So one can also see from the beginning the difference between linear effect and loess smoother and optimal fit.
Which color for the response would you suggest? It is not easily possible to make the coefficient range dependent on the dataset, this would make it too complex (hence too slow)
This is a good point. The response function is not easily recognizeable at 0. We chose black as color because it is neutral and cannot be confused with the colorful spline basis functions. Do you have another suggestion for the color? We can easily change that. The response function cannot really start with a linear fit because this cannot be shown in the splines input setting - there is no coefficient for a linear effect. I would keep the response consistent with the input settings. Generally we wanted to start at 0 because data could also be removed and then people can explore what happens to the response when using different parameter setting without focusing on fitting data - so a more theoretical view.
Thanks Christine for clarifying. I get your intention. Still I think the current implementation is not easy to grasp for our intended audience. There are many layers one has to understand at the same time. (If it is too difficult people will not use it.) However, maybe it is enough to explain how to use the app in the explanation tab. A commented walk through how to use it, where one explains the meaning of the different parameters, etc. at the same time.
For the color of the response function; why not red? This is the main line the user is interested.
Number 3b from #36:
As a first-time user it is frustrating, that if one tries to match the response with the optimal fit that if one changes the knots, powers and coefficients nothing happens....because first one has to adjust the intercept.
It would make more sense if the intercept is already adjusted. What do you think?
I would make the response line in not-black. It took me some time to realize that the black line at 0 is the response.
And the coefficient range should be so that changes in the coefficients lead to changes in the response from the beginning.