tobiaskauer
commented
7 years ago 
- I went though NY open data sale and tax repo as seen here https://data.cityofnewyork.us/City-Government/Revised-Notice-of-Property-Value-RNOPV-/8vgb-zm6e
- Then I mapped all the Trueskill rated beauty scores from NY into one of the 8000 odd tract boundaries , and did the same for sale prices.
- In the end I have Beauty-> tract mapping and Sales -> tract mapping
- Then took all the tracts where at-least 1 house has been sold and at-least one beauty rating exist (As you can see this thins down the heard by a lot and is not in my opinion a very good method , but I made what I could with the data I got, any good ideas are much appreciated)
- Finally I was left with 59 tracts with both beauty and sale scores, for which I then found correlation (median beauty vs median sale price ) :
One big issue for me still is the outcome variable. if i understand sagars analysis on the houseprices correctly then we have a) not enough data and the data we have b) suggests no correlation of houseprices and beauty. Beside the existing design (result visualization and explanatory scrollytelling), there is no narrative for the exploratory part. Would love to further discus what other data (anything not harvested from the image analysis), like houseprices or income or poverty or whatever we will base our narrative on.