ANALYSIS: assessing geographic diversity of reading for TI

[kanarinka]

The question: Did users show increased geographic diversity of reading after installing TI?

Method:

1. Compute HHI for world countries using population (which was how we determined # cities/country in TI) -- I've done this already and it's around 750. This is the diversity level we would "expect" if people read perfectly about countries in relation to their % of world population.
2. Compute preinstallation HHI for users with >= 5 days preinstallation and >= 5 days postinstallation history in the system.
3. Compute postinstallation HHI for users with >= 5 days preinstallation and >= 5 days postinstallation history in the system.
4. Did postinstallation HHI go up? For what % of users and by how much on average?

NOTE: exclude Catherine & Matt's user ids

[kanarinka]

For this we need to make:

• server side - /exportgeo/ function to download spreadsheet of preinstallation and postinstallation country counts normalized by day, e.g. for each user we have
  • totals of urls visited associated with a country preinstallation / Number of preinstallation days for which we have data
  • totals of urls visited associated with a country postinstallation / Number of postinstallation days for which we have data
• CSV looks like user.id, countrycode, preinstallation.per.day, postinstallation.per.day
• R script calculates HHI preinstallation and HHI postinstallation, and summary stats for the difference across users and difference from the "expected" population HHI

[kanarinka]

Preliminary Results for Changes in Geographic Diversity of Reading (6/11/14)

48.6% of users showed an increase in geographic diversity of newsreading after installing Terra Incognita

For users who increased diversity, 1610.4 is the average number of increased diversity points on the Herfindahl–Hirschman Index scale of 1-10000

For users who increased diversity, 864.9 is the median number of increased diversity points on the Herfindahl–Hirschman Index scale of 1-10000