[Feature Request]: Scatter/Jitter Plots could show approximate accuracy of measurement by dot thickness

[tomtomme]

still valid for 0.19 beta

Description

Being able to adjust how thick a dot is within a scatter or jitter plot

Purpose

Inform the reader about the expected accuracy of the measurement (thicker dots for low accuracy, smaller dots for high accuracy)

Use-case

This maybe useful foremost if you have manifest measurements like time, weight, height, where you know from the calibration of the measuring tool how accurate it is.

Is your feature request related to a problem?

At the moment the thickness of the dots do not mean anything. But it varies across tools, e.g. the flexplot module produces smaller dots, the descriptive module thicker ones.

Describe the solution you would like

Add a slider or box to enter a value, that changes the dot thikness.

Describe alternatives that you have considered

This slider could be a) a global slider hidden in jasps main menu => preferences => results => plot options - but that would mean that your complete dataset would need to contain variables of the exact same accuracy. b) reserved for advanced plotting modules like Flexplot c) could be added somewhere to the data view, one for every variable. This would bring the best flexibility. Dot thickness could even differ on x and y axiss, resulting in oval dots!

[juliuspfadt]

well, if this is associated with the visual modeling module, I think we should assign dustinfife, although I am unable to do that somehow. Dont know what's wrong with GitHub. I dont think changing the overall preferences is feasible. anyways. I dont really understand the issue though. Could you give an example?

[tomtomme]

There is no software I know of that does exactly this right now. Or pictures that I could think of. But it is related to this: https://flowingdata.com/2018/01/08/visualizing-the-uncertainty-in-data/

I will describe in more detail below. I think @dustinfife would like it. It fits into the aims and concepts of flexplot. It might be a nice feature for bio-psychology, physics, chemistry etc.

Detailed description:

Imagine 2 different research groups measuring almost the same thing, but with two different measurement instruments, which have (roughly) known, but different degrees of accuracy.

Example:

• lab 1 uses a very precise while expensive timer to measure reaction times. Known measurement error of +-1 milliseconds.
• lab 2 uses a not that expensive timer to measure reaction times. Known measurement error of +-5 ms

Now both labs produce a scatterplot: Reaction time on x-axis, with a lenght of 100 millimeters showing values of 0 to 100 ms. So a measurement that would be exact to the millisecond would be shown by a dot that is 1 mm wide (resulting in an ellipse instead of a round dot, depending on whats on the y-axis and how precise that was measured). But the measurement error for lab one is +-1ms, so the lab would want to make the dot/ellipse 3 mm wide, maybe with increasing transparency to the border. Lab 2 then would want their dots/ellipses 11 mm wide. When those scatterplots are compared, the different measurement accuracy would be obvious. Same could be done with any physical measurement, be it temperature, weight etc. and probably also with psychometric measurements where the measurement error is approximatly known.

We do a similar thing all the time showing confidence intervals and bands that show unsystematic errors. But systematic errors like the measurement error might have a much bigger impact, so I think it would be important to visualize it.

[tomtomme]

I just realized I could ask the KI, if something like this is available as an R-Package....

"Hello, this is Bing. I searched the web for R packages or other software that can do what you described. Here are some possible options:

• The errors package1 provides a class for associating uncertainty metadata, automated propagation and reporting of measurement errors in R. It also has a function called plot_errors that can create scatterplots with error bars or ellipses to represent the measurement uncertainty2.
• The mecor package3 implements measurement error correction methods for linear models with continuous outcomes. It also has a function called plot_mecor that can create scatterplots with error ellipses to show the measurement error in the covariate or the outcome4.
• The ggplot2 package is a popular and versatile package for creating graphics in R. It has a function called geom_errorbar that can add error bars to a scatterplot to show the measurement error in the x or y direction. It also has a function called geom_ellipse that can add ellipses to a scatterplot to show the measurement error in both directions. I hope this helps you find the right tool for your task. 😊"

So yeah. Nothing novel about my idea... who would have guessed that ;)

[dustinfife]

This could be a cool feature. I'll keep this in mind.