What I Did
For this experiment I ran a series of tests to determine how effectively I could tell apart two sets of stars and periods. There were two different examples, the A example had about 46 or less stars and the B example had 56 or more on average. I had to quickly assess which option was presented to me without counting the stars to periods ratio.
For the first round I did a series of three tests, one in which the two options would show up an equal amount, and two others where one option would be presented in a three to one ratio to the other. The fourth and final test I changed the visual distinctness between the two example types. I did this test myself, because it is a learning process. Ideally I would have recruited someone to take it and analyse their data, but I am still learning and wanted to focus on the analysis rather than recruitment. All of these tests were done in a typical working environment. In this case that means a quiet place with few distractions. This is done to simulate the participants average working environment. As I am the participant, I used my actual working environment.
The objective of this experiment is to determine what would make the two types of examples visually distinct enough to be easily identifiable. The design of the experiment is largely the same as the Mueller and Weidemann Dot Classification test. I wanted to use this test to learn more about how distinct visual items need to be for graphical interfaces. In the field of User Experience, it is extremely important that the user be able to tell what an interface does and that they can distinguish the functions of different parts of the interface.
What I Found
For test 3:1, there were three times as many type A examples. For test 1:3, there were three times as many type B examples.
Here are the data for the first three trials I ran for this experiment. We can learn valuable information from dissecting them. In this data I compare how many times I was presented with A or B, and then according to that, how many times I selected either A or B. These confusion matrices show that the general accuracy was close for all three of these trials. An interesting thing to note is that, during the 3:1 and 1:3 trial, I was more likely to choose the type of example which showed up more, even if it was wrong. We can infer from this that changing the base rate of the experiment biased the participant, should they have known this info, to select for the type with the higher rate. The sensitivity of the experiment remained largely similar through the initial three experiments.
For the fourth experiment, I returned to the 50:50 base rate for showing both type A and type B. However I diversified each to look less similar. Type A in the first three experiments had on average 46 stars, and type B had 56. I pushed the bounds of these to 40 stars for type A and 60 stars for type B. What happened is you can clearly see that it was much easier to distinguish between the two types presented. The sensitivity of the experiment was almost triple that from the previous three. This experiment could be tuned even more by reigning in one of the two variables I adjusted to see how close they could be while still being visually distinct.
Whats In Store?
I’m sure we have all been to a website which had a fake advertisement masquerading as the button you actually needed to click. Or have you ever wondered if that noise was Zoom telling you someone joined or left the call? These are all signals in use. I find the area of automotive acoustics of particular interest. Recently I have noticed a large amount of audio feedback coming from cars. When the door is open, the trunk unlocks, or if a car in front of you is suddenly stopping for example. This tiny orchestra of experience can easily get jumbled and give the user a false idea of what is happening. Currently composers can elicit different emotions with different short sounds from the vehicle to indicate if that sound is good or bad. I’m sure no small amount of signal detection theory was used to determine whether the drivers understood A, if those were good noises or bad noises, and B, what those noises meant for the driver.