A3-JonBarron

From CS 294-10 Visualization Sp10

Jump to: navigation, search

It being near Valentine's day when I started work on this assignment, I decided to use Google Trends to analyze the number of queries for related to Valentine's day in 2009 (under the assumption that the number of searches correlates with people's interest in / preparation for Valentine's day).

Image:JonBarron-A2-0.png

Unsuprisingly, we see a sharp peak at the week of Valentine's day, with a ramp up and a ramp back down. This is consistent with my experience, as I remembered Googling for ideas on Valentine's day presents several weeks before the day. I wondered: "do other American holidays have similar trends?". I scraped Google Trends for other American holidays:

Image:JonBarron-A2-1.png

This data was obtained by searching Google Trends for these five holidays (where five is unfortunately the maximum number allowed by Google trends), exporting the data into CSVs, and then parsing and transforming this data in Matlab (as the data was in a rather complicated format, that Tableau completely failed to parse), exporting that transformed data back to a CSV, and then importing it into Tableau. In Tableau, I did nothing but import and display the data.

Clearly, other holidays have similar trends, though they vary heavily in terms of overall interest in the holiday (the magnitude of the peak), and their location (the week in which the holiday falls). I began to wonder: "how do the increases in interest immediately before these holidays differ between holidays?" For easier comparison, I decided to normalize by the peak magnitude:

Image:JonBarron-A2-2.png

Tableau was not at all helpful for doing this, so I had to rescale the data in Matlab. Normalization helped a little, but it also seemed necessary to shift all holidays to fall on the same week, so I shifted everything such that all holidays appear to fall in the 52nd week:

Image:JonBarron-A2-4.png

Tableau was also incapable of doing this, so I again had to shift the data in Matlab. I did not find this visualization clear, as the vertical rescaling made small holidays appear to have very drastic and rapid ramp-ups, and large holidays appear to have very gradual ramp-ups. To make the data easier to compare, I decided to rescale the data along the X-axis by as much as I had rescaled it along the Y-axis, as well as shift it:

Image:JonBarron-A2-5.png

Again, Tableau was not helpful for doing this, so I did it manually in Matlab. All data has been rescaled such that each holiday is "as popular" as the most popular holiday, Christmas. As such, the X-axis has been relabeled in "Christmas weeks". This scaling seems extremely effective for comparing the profiles of different holidays. We can see that Halloween and New Years have, relatively speaking, the fastest ramp-up period (which is consistent with my experience, as I rarely think of these holidays until they are immediately upon us), and we see that Valentine's day and Thanksgiving have the heaviest tale, meaning that people plan for these holidays far more in advance than most others (which is also consistent with my experience, excluding Valentine's day). Another trend that suddenly becomes prominent is the large bump way before Thanksgiving. I believe this bump is either due to extremely well-prepared Thanksgiving hosts, or to Canadian Thanksgiving (which falls several weeks before American thanksgiving). I began to wonder: "Are there any similar trends in the immediate decrease in interest in holidays?" Because this plot cuts off that data, I made a new plot:

Image:JonBarron-A2-6.png

Interestingly, we see that interest in Christmas and Halloween drop off very rapidly, while interest in New Years, Valentine's day, and Thanksgiving are much more heavy tailed. Oddly enough, we see another bump in the Thanksgiving plot, which I can only assume is extremely ill-prepared Thanksgiving hosts, or some other nation's Thanksgiving celebration.



[add comment]
Personal tools