Institutions of higher learning send thousands of marketing emails to high school students as soon as those students' emails appear in various databases listing prospective university attendees.

I collected over 3000 such marketing emails from January 2011 to January 2013, and this a project to analyze the strategies used by these marketers in selling their institutions to prospective students.

What sort of language do these marketers use when advertising their school to prospective students? What words appear frequently? What words are certain schools more likely to use than others? Can this detail be used to develop a statistical model of how to write like a university marketer, or to predict which university a specific email body came from?

Current state: Incomplete. In progress.

What is the difference between a university that sends many emails and a university that sends few emails? Are there correlations between the number of emails a university sends a student and

Current state: Incomplete. In progress.

What time of day are these emails sent? Are they sent during a time of day when the marketers expect them to be most likely to be read?

What time of the year are these emails sent? Anecdotally, I saw a trickle of these emails in January 2011, when I was a tenth grader, and this trickle gradually intensified as I reached the twelfth grade, and then trickled off again as application deadlines had passed and high school graduation approached.

Current state: Not started.

The Python script download_gmail.py uses the Gmail API (accessed through the Python Google API library) to download all of the emails in a user-selected Gmail label.

The emails are pulled from Google in raw MIME format, and then heavily processed to turn rich multipart HTML emails into flat text messages, at which point they're written out to a CSV file.

To visualize and do more advanced processing on this data, we use Wolfram Mathematica. In addition to Mathematica's powerful general-purpose computing language, it's strong visualization features make it easy to create publication-worthy graphs from the raw CSV data.

Although many universities are represented in the sample, all 3000+ emails are only targeting one specific student.

It is entirely possible that a student of a different gender, race, academic standing, and/or geographical location would receive a different pool of emails from the universities interested in them. Because of this, it may be difficult to draw very specific conclusions about the marketing techniques of universities.

It is entirely possible that between January 2011 and January 2013 (when I was collecting the emails for this study) that not all of the college marketing emails I received were appropriately categorized. Some of them may be elsewhere in my inbox, marked as spam, or otherwise lost.

Additionally, it is very possible that emails that are not from college marketers (or are tainted by having specific user interaction with the college--such as applying to them) made it into the dataset. Although an effort has been made to sanitize the dataset, nobody can ever be perfectly certain...

Where can I download the marketing emails used in this analysis?

You can't... yet. With a little bit more post-processing, and some checking over to make sure that I am not accidentally publishing one or two important emails in the dataset of 3000+ emails, it will be available for download shortly.

If you want a copy of these emails for private analysis, then send me an email at [email protected]