Week 8a: Interpreting Our Data

Pandora Survey

My bar graph illustrates total enjoyment of the songs Pandora chose next to the correspondance of the music's "actual" qualities as compared to the original qualities participants claimed to like. As the graph makes clear, enjoyment of the music is always higher than quality correspondance of the music. This relationship lets us draw several possible conclusions: 1. the five people participating in the survey aren't very aware of the qualities they "actually" like about music, 2. Pandora's choices did not (could not?) account for more qualitative, emotional or personal qualities that participants stated they liked, such as "thought-provoking lyrics", "witty lyrics," "brings back memories," "catchy hooks," and "soulful." The participants in the survey who stated more musically-savvy qualities like "varied rhythms," "piano-centric background," "nice harmonies" and "folk" (person 2, 4 and 5) were much more satisfied with the music choices and the qualities Pandora chose corresponded very closely to their own stated qualities. There is also an interesting connection, I think, between this relationship of how satisfied listeners were and how well the qualities matched, and how "negative" or "positive" participants' respective reactions were to the fact that Pandora might be able to make an algorithm for their music tastes. Again, those who were able to state the qualities they liked in more official and less personal terms felt much more positively (scoring 5 and 4) about this fact whereas the other, evidently less-satisfied group scored 2.5 and 3. Perhaps some people like to know what they like and have it be clear, while others prefer more mystery in their musical choices. The answers to the positive/negative question also could have been affected by the initial inability or refusal of some users (those who said 2.5 and 3) to use official music lingo that Pandora uses in their initial qualities/reasons for liking said artist.

Enjoyment of Music Next to Quality Correlation of Music; Person 1-5

Average ranking of songs provided by Pandora

Generally, my respondents were very pleased with the songs that Pandora suggested based on their input. However, as Jessica noted before, there was not a strong correlation between how much respondents enjoyed a song and how closely Pandora's criteria matched the criteria specified by the user before the experiment. This suggests several possibilities. Perhaps, as we noticed in class, users do not know how to articulate their preferences with the same level of detail as Pandora provides. It is also possible that the music Pandora provides has other characteristics in common with the music specified by the user besides those that are isolated by the Music Genome Project, such as the mood of the music or of the lyrics, or that these ideas are articulated with very different words by Pandora and by users.

In general, the selections made by Pandora seem to be accurate. According to my survey result, there were only 2 out of 15 songs selected by Pandora that scored 2 or lower on the scale of 1-5. Considering that there were 12 songs that were rated 5, which is the highest one can rate, it seems like Pandora did an excellent job of matching people with songs. However, it was interesting to me that answers to the Question 5 did not necessarily match the answers to the Question 6. It shows that Pandora's explanations to the selected songs did not always describe why the users liked or disliked the songs. It suggests that there are more sophisticated reasons that are not easy to be explained behind all the technicalities, which could be sufficiently described in a few words. All participants were not bothered by the fact that Pandora could give suggestions based on the users' favorite artists/songs. It shows that the participants perceived selections made by Pandora as if they were simple suggestions made by friends. Overall, the survey reflects that Pandora is an effective tool of identifying people's music tastes, though not a perfect one.

After completing this survey, I found mostly positive reviews of Pandora. There was some speculation that the algorithm could be more complex, allowing for more accuracy, but in general the program was fairly accurate. No one gave any response as less than 2/5, which indicated to me that Pandora was generally correct with its predictions, user-friendly, and inviting enough to not intimidate users with its seemingly "know-all" musical algorithm. There was a good variety of bands selected by the subjects, which also indicates that Pandora has a wide breath of musical genres and acts in its database. In conclusion, Pandora is a fun, easy, and dependable program.  

When running the Pandora survey, many of the girls that I surveyed were shocked by the music choices that Pandora made for them. For all of the stations, they started with the named artist and went off from there. Many of the girls either didn’t have an idea of what was playing at the moment, or knew exactly what it was. There were no reactions that centered towards the middle. As unknown songs kept playing, girls made comments like “I like this song” or “If I knew who they were before, I probably would already be listening to them.” The majority of the surveys had a similar experience with choice of songs, with an average of 4.45. I took out the one negative experience with Pandora. With this negative rating, the average would have been 4.12. Though there isn’t a large difference in the average, her personal answers and reactions were very different, comparing her to the other girls.

Each series corresponds with the songs that were played ie. series 1 = song 1.

As seen from the graph, four out of five people thought Pandora's predictions of their taste in music were accurate above or at average level, only one person felt that Pandora's choice did not fit her taste well. Indeed, two of these four people were very excited to discover Pandora and what it could do. They even recommended it to their friends instantly. I asked the respondent No.4 who rated a "2" about the cause of her dislikes for the music Pandora chose. She replied that  the songs Pandora chose for her was similar to the songs she loved in an element that was just not the one she preferred. She elaborated that her favorite artist was Eminem and the element of personal feelings in his songs. However, Pandora provided similar rap songs for her, focusing on the "gangsta" element in most of Eminem's songs, and not the personal part that she was looking for. From their responses, I have formed two ideas. First, overall, I think Pandora did an amazing job in finding songs for listeners, and I am quite impressed. Second, I think it is realistic to say that nothing can satisfy everyone's tastes and expectations. Especially in this case, where only the artist's name was provided for Pandora. We all expected it to give us exactly what we want, but now to think about it, it is just not completely doable with so little information given. I think that this is a reasonable explaination of why not everyone thought it was accurate, and it is a totally okay senario in real life.

In this graph, the majority of the subjects thought that Pandora did not find songs that had the same elements as their initial song had. No one voted a five (most accurate) although two people voted ones (completely inaccurate). My subject pool was some relatives (parents' age), an old friend (18 years) and my cello teacher (mid-late twenties). Interestingly, I found that people weren't as impressed with Pandora as much as I had anticipated. Most of the subjects, mainly my mom and my grandmother, didn't approve of Pandora. In their opinion Pandora missed the point of their songs/composer and choose newer and irrelevant. I was curious to see what my cello teacher would think of the program. Being in his mid-late twenties, he followed a similar trend as the younger subjects. He was critical of the choices Pandora made (the second song he deemed inaccurate 1) but he would invest heavily in the company. Here's my vast generalization: The younger the subjects were, the more positively they viewed Pandora overall.

This graph shows my subjects’ reactions to each of the five songs that Pandora selected. Overall, all five of my subjects enjoyed the song selections made by the website. I purposely chose to include one of my sisters in this survey because she is one of the most picky control freaks I have ever encountered. She was Subject 2. She was the only one that didn’t like the fact that someone else was choosing her music for her because of the lack of control. My sister completely disagreed with her second song choice, yet that didn't affect the rest of her Pandora experience. As seen in the red bars, she enjoyed the rest of the songs selected. The subjects found Pandora as a great site. It was something they found amazing yet slightly weird; that a computer selects music they think they would enjoy based on the selection of one artist. I found that since the subjects enjoyed the song selections, they were more likely to have a more positive feeling towards Pandora. The outlier, my sister, didn’t enjoy one of the five song choices, yet still agreed that Pandora was relatively accurate. Pandora was a success and they have all mentioned more use of it in the future.

      This was an interesting experience  because a few of the people I asked had a difficult time trying to list five reasons why they liked their specified artist. It led me to believe that they don't usually think about why they like the things that they do, which I can also relate to. Most of the people I asked were surprised by how well Pandora could predict their likes and dislikes. None of them were at all bothered by the idea that a website could guess these things about them from just knowing one favorite artist. In general, the songs all recieved very high ratings, and people agreed with the reasons. Although people liked the website, they were ambivalent about investing in the company. 

*Note: the scale of “Liking” a song was 1=complete dislike, 5=complete like

The results of my survey showed that in general, Pandora did a great job in finding songs that the person liked. For the graph above, I averaged the ‘score’ given to each song to find, on average, how well Pandora picked out a song.  Pandora scored in the range of 4 and above for all the participants, showing that in general, it picked out songs that the participants liked.  These findings can also support the idea that people are more or less ‘happy’ or feel ‘positively’ about a company that is able to write an algorithm to predict musical taste. All the responses to this question were “4” on the scale of Positive (5) to Negative (1). This makes sense because everyone who took this survey was overall satisfied with the songs picked out, and thus why would they not like a music site that helps them find other songs that they like (but they just don’t know about!).  The average response for the satisfaction in the reasons Pandora chose each song was 3.6, meaning that in general, people were slightly more satisfied than not. Pandora simply provides a name to an attribute that the participants liked but simply did not know what the technical term was. 

The responses to my data collections were mixed, and it was hard to see any real trend. The five participants in general had positive reactions to Pandora, whether or not the songs matched the reasons they had listed for question two. In the graph, I wanted to see whether there was a correlation between the responses to questions six and seven. Question six asked how accurately the reasons Pandora gave for playing the songs matched the reasons you had listed and question seven asked whether you had a positive or negative reaction to the fact that Pandora can predict your music taste. I hypothesized that these questions would be proportional, with positive reactions for number seven accompanied by high scores for question six. However, from the five data points there was no obvious correlation. In order to properly test this possible correlation it would be necessary to survey a much larger data set.

                         Attributes Emphasized!

  This is a comparison between the qualities pandora uses to discover music it thinks its users will like, versus how the users themselves would go about searching for new music.  On average, the numbers show how pandora and its users rate the relevance of different qualities for determining their musical taste.