Fitt s Law Study Report Amia Oberai Overview of the study The aim of this study was to investigate the effect of different music genres and tempos on people s pointing interactions. 5 participants took part in the study, 3 female. All participants were aged 19-20 years old and were sophomores at Brown University. The participants were recruited through opportunity sampling, by the experimenter simply asking them to take part in a study for her HCI class. The study was conducted in the MS lab in Brown University s Computer Science building. Each participant was given a consent form, and then sat in front of a desktop computer (the same computer was used for all participants to avoid experimental error.) They clicked circular targets on the screen, formatted in a circular ring, provided by Wobbrock s FittsStudy software. The target amplitudes and widths of the targets were combinations of 255, 384, 512 (A) and 9, 24, 64 (W). The participants were given 10 conditions in each condition 1 of 9 songs played in the background while the participant wore headphones, and in 1 condition no song was played. Participants did 23 trials for each condition (the first 3 were practice.) The order of the songs were randomized for each participant. The 9 songs were of a range of genres and tempos: Classical o Claire de Lune Debussy (68 BMP) o Mozart March ala Turk (115 BMP) o Brahms Hungarian Dance No. 5 in G Minor (142 BMP) Pop o Katy Perry Dark Horse (66 BMP) o Shakira ft. Rihanna - Can t Remember to Forget You (138 BMP) o Pharell Williams Happy (160 BPM) Electro o Chili Banks London (Gash, Bay B Kane VIP Remix) (86 BPM) o David Guetta & Showtek ft. Vassy Bad (128 BPM) o Ellie Goulding Burn (174 BPM) Expectations I expected that participants pointing behavior would match the song being played. In other words, they would click targets faster (while still being relatively accurate) while listening to faster songs. I also thought that if they liked the song, this would also increase speed, as the task would seem like less of a chore and therefore they would be able to concentrate better. I was expecting movement time to be highest for the No Music condition, as the participants would not have background noise to keep them engrossed in the task.
Results Results were collected through the software used, which analyzed logs that could be uploaded into Excel (see https://docs.google.com/spreadsheets/d/1oyzm8tgzww8bzeyczcvcumzwigr DwBsrBRdEXLzhi24/edit?usp=sharing for the 5 Excel sheets.) The results were analyzed as followed: First, I plotted a scatter plot graph for each participant, showing the IDe and MTe for each trial. Each graph contains 10 lines, one for each condition (9 songs, and no song.) It was important to exclude the data for the 3 practice trials. Another precaution I took was to plot each of the individual A and W pairs, as opposed to averaging the IDe values, to avoid the pitfall mentioned in Shoemaker et al. This pitfall is the concept that averaging the ID (or IDe) values can hide the data and ultimately leads to a higher R^2 value, that does not actually take into account all the data. Although this meant I had to plot 9*20*10 data points, it shows how the data was spread rather than collapsing them. The graphs can be viewed below (*Note that the white line represents the No Song condition in all 5 graphs, but the colors/line style used for the other conditions lines are not consistent; instead the legend order represents the order in which the conditions were presented to the participant. Also note that the background is black to allow clearer viewing of the lines and data points): Figure 1: Participant 1
Figure 2: Participant 2 Figure 3: Participant 3
Figure 4: Participant 4 Figure 5: Participant 5
The regression plots were not what I expected, as they do not represent movement time as a function of the speed of the song (BMP). Also, the No Music line is not displayed with the highest MTe values, expect for Participants 2 and 3 for higher IDe values, but in these cases the lines have lower MTe values than other conditions for higher IDe values. I did, however, somewhat expect this spread in data points, due to some confounding variables that I will discuss later. Though it is worthwhile to mention that the R^2 values for all the regression lines are fairly strong approximately 0.5 or 0.6. Furthermore, the error rates % were very low all below 0.1% (according to the software, but this is under some speculation.) I also calculated the throughput (TP) for each condition, to see which regression represents fits Fitt s Law the best. I used the following equation for TP: (Soukoreff & MacKenzie.) This was achieved by first calculating the average IDe and average MTe values for each A and W pair (but the IDe values for matching A and W pairs were consistent), and then dividing average IDe by average MTe, and then averaging this result for the 9 A and W pairs. Finally I averaged this result for across all 5 participants for the same song. This can be seen below: Figure 6: Figure 7: Figure 8:
Figure 6 shows the first few rows of IDe and corresponding MTe values for Participant 1, and the third column shows the average MTe (of the 20 MTe values, representing the 20 trials per IDe, excluding the 3 practice trials, which were filtered out.) Figure 7 shows all the average MTe values for Participant 1 s first condition. Figure 8 shows the first 4 (and a half) conditions for Participant 1 and the corresponding TP value for each condition is highlighted in lilac... If we use Shannon s formulation of Fitt s Law, we can see whether or not the values we get for a and b from the regression in the graphs are valid. So let s take the first regression (Shakira ft. Rihanna) for Participant 1 (See Figure 1). The equation of the line is y=176.73x + 113.5. Because the graph displays MTe in milliseconds, we need to divide the values in the equation by 1000, to display the appropriate values for MTe in seconds. So a is 0.1135 and b is 0.17673. Now we look at the first data point for this regression as seen in the first row in Figure 6 (3.8031, 824). We must divide the MTe value (y value) by 1000 again for seconds, so 0.824. According to Shannon s formulation, and using the values above, we have: 0.824 = 0.1135 + 0.17673 * 3.8031 BUT 0.1135 + 0.17673 * 3.8031 is actually 0.7856, so the regression, which has given us the values for a and b, is not perfect, but is pretty close. This is reflected by the R^2 value of 0.58563 (as can be seen in Figure 1.). Finally, I calculated the total TP for each of the 10 conditions, by averaging the TP values for each condition for each participant, like so: Figure 9:
Eventually I came up with a table that shows the conditions in descending total TP order: Figure 10: If we look at the BMP of these songs (in descending TP order): 115 BMP 138 BMP 66 BMP 68 BMP 142 BMP 160 BMP 128 BMP 174 BMP 0 BMP 86 BMP It seems that beats per minute does not seem to affect performance. If we look at the genres of these songs (in descending TP order): Classical Pop Pop Classical Classical Pop Electro Electro No Music Electro
It seems that participants performed better with Classical or Pop songs. It is possible that this is because Electro music generally has fewer lyrics, or less emotion in the song, as the vocals are performed in a deadpan manner with electronic distortion, so participants were less likely to feel connected to the song and more likely to go at the pace they normally would without music (which is supported by the fact that the No Music condition falls in between Electro songs.) Of course, this is just speculation. Qualitative Data After participants had completed the experiment, I asked them some questions. I did not want to ask if participants knew or liked the song straight after each condition because this might have led to demand characteristics, and could have altered performance in the next condition. When I did ask at the end of the experiment, they were able to remember if they liked the song or not but generally could not remember how the song affected their mood or performance. Figure 11: Here is an image of whether participants knew or liked a song (Y means they knew the song, N means they did not know the song. + means they liked the song, - means they found the song neutral, and a downward arrow means the disliked the song.)
If you compare these ratings (especially the positive, negative and neutral ratings of each song) to the participants pointing performance in Figures 1-5, we can see that whether or not a participant knew or liked a song did not seem to have an affect on performance. Some more useful qualitative data: Participant 3 made an interesting suggestion that there are more errors horizontally because you must use both parts of the brain to view the target (when moving from one side of the screen to the other.) This participant also claimed that he was not paying attention to the songs. This could be true, as his graph shows higher R^2 values compared to graphs for the other participants, which suggests the data is less spread out and thus that he was not getting distracted by the music. Participant 4 claimed that she sometimes zoned out of the songs, particularly those without lyrics, and started thinking about her life while doing the task. This gives us some insight into why, perhaps, Electro music (that has fewer unique lyrics) shows the worst TP. Also, she said that in the no music condition she got distracted by the people talking around her (as there were other people in the lab.) In general, though, she seemed to be getting really into the songs she knew, bobbing her head or mouthing the lyrics if she knew them. Participant 2 occasionally stopped to move the mouse in a more comfortable position, so her movements were not always very smooth. Also, she wore contacts, and said her eyes were hurting so she had to take a short break between some conditions. Participant 5 seemed like he didn t want to be there, he said he wanted to be outdoors because it was a nice day. So he may have been less invested in the experiment. Participants may have placed more importance on either speed or accuracy, even though they were all given the same instructions at the beginning ( we are testing speed but you must still be accurate. ) For example, Participant 3 seemed like he was really focusing on being accurate, rather than fast, especially for small targets. All participants seemed to be getting bored or tired as the experiment went on.
Limitations Most of the qualitative feedback and things I noticed about the participants that I mentioned above are limitations to the study. Another related limitation is the fact that the experiments were conducted at different times of the day for each participant. At nighttime, there were usually more people in the MS lab, so this could have been more distracting. Results might have been different if participants were alone in a small room. Conclusion In general, it does not seem that music seems to have much of an effect on pointing performance. However, it might be worthwhile further investigate why TP was lowest for Electro songs. More songs need to be tested in order to make any fair claims. In general though, perhaps we, as students (who often work in all sorts of noisy, distracting environments), are used to background noise and have therefore learnt to ignore it for the most part when we have to concentrate on a task.