This idea of a theoretical background in dataviz is very inspiring.
I mean, it remembers me what Claude E. Shannon did with the information theory.
So, I was thinking that maybe there is a pdf of the visualization, a pdf of the reader and we can increase the mutual information or decrease some divergence in order to approximate if the visualization would be good at communicating or not.
If this is possible, you can define the set of reader and medium parameters, then start to build your chart and you're going to see how these information-theoretic descriptors increase or decrease as long as you modify the visualization 🤔
Interesting. I developed a theory of data visualization and a model that allows to make decisions about the type of data visualization easily and efficiently. I've been teaching her for the last year and she's really working! These days I am finishing the information video in English. I would love to hear your opinion on the subject.
I was trying to follow your logic concerning the power of prediction of theories in visualization and I’m wondering how that can be achieved without a model that would allow to quantify and evaluate thus theory’s performance. Moreover, even if a theory’s aim is to predict, it doesn’t necessarily mean that the prediction is highly accurate or even relevant.
The purpose of visualizations seems too general to me. I think it needs to be more restrictive, e.g. by adding effectively and efficiently, and the corollary which define what these restrictions mean would make more sense. Effectiveness means accomplishing something with the least amount of effort. This would mean that the effectiveness of a visualization can be defined in term of how we can convey the needed amount of information with the minimum visual elements by keeping the expected level of quality. Representing too much information can overload the receiver, obscuring maybe the important information. Then effectiveness would translate as the degree to which the visualization can transmit successfully the important information.
As soon we treat things in aggregation or apply other types of transformations on the data, this would lead to loss in quality. The question is how much quality we are allowed to lose.
When we consider the sender, medium and receiver in data visualizations, it is worth to know the maximum of information existing in a dataset, the information the sender choses to encode, the information the channel is able to transmit, respectively the volume of information the receiver is able to decode, integrate and understand.
This idea of a theoretical background in dataviz is very inspiring.
I mean, it remembers me what Claude E. Shannon did with the information theory.
So, I was thinking that maybe there is a pdf of the visualization, a pdf of the reader and we can increase the mutual information or decrease some divergence in order to approximate if the visualization would be good at communicating or not.
If this is possible, you can define the set of reader and medium parameters, then start to build your chart and you're going to see how these information-theoretic descriptors increase or decrease as long as you modify the visualization 🤔
Interesting. I developed a theory of data visualization and a model that allows to make decisions about the type of data visualization easily and efficiently. I've been teaching her for the last year and she's really working! These days I am finishing the information video in English. I would love to hear your opinion on the subject.
I was trying to follow your logic concerning the power of prediction of theories in visualization and I’m wondering how that can be achieved without a model that would allow to quantify and evaluate thus theory’s performance. Moreover, even if a theory’s aim is to predict, it doesn’t necessarily mean that the prediction is highly accurate or even relevant.
The purpose of visualizations seems too general to me. I think it needs to be more restrictive, e.g. by adding effectively and efficiently, and the corollary which define what these restrictions mean would make more sense. Effectiveness means accomplishing something with the least amount of effort. This would mean that the effectiveness of a visualization can be defined in term of how we can convey the needed amount of information with the minimum visual elements by keeping the expected level of quality. Representing too much information can overload the receiver, obscuring maybe the important information. Then effectiveness would translate as the degree to which the visualization can transmit successfully the important information.
As soon we treat things in aggregation or apply other types of transformations on the data, this would lead to loss in quality. The question is how much quality we are allowed to lose.
When we consider the sender, medium and receiver in data visualizations, it is worth to know the maximum of information existing in a dataset, the information the sender choses to encode, the information the channel is able to transmit, respectively the volume of information the receiver is able to decode, integrate and understand.