Trees 311 and NYC

Informing Natural Disaster Response with Census Data  

Jonathan Auerbach

Chris Eshleman

Constituent Generated Data Agency Generated Data

Linking Information

Agency Generated Data Constituent Generated Data

Parks Department Data

Linking Information

Constituent Generated Data

Picture from New York Times

Picture from AP

Picture from Reuters

Pictures  from  Reuters/CBS  Picture from Reuters

Real-Time Policy Decisions

•  Should a state of emergency be declared?

•  How large should relief funds be? Where should they be distributed?

•  Should the marathon be canceled?

Linking Information

Constituent Generated Data Parks Department Data

Linking Information

Service Request Data Parks Department Data

Service Request Data

Linking Information

Parks Department Data

Census Bureau Planning Data

Service Request Data Parks Department Data

Model Considerations

Parks Department Experts Say service requests consistent among resident groups

Model Considerations

Economists Say resident contacts city because benefit exceeds the cost

Parks Department Experts Say service requests consistent among resident groups

Model Considerations

Economists Say resident contacts city because benefit exceeds the cost

Sociologists Say benefit or cost due to heterogeneity in stake or social efficacy level of group

Parks Department Experts Say service requests consistent among resident groups

ith resident jth group

ith resident jth group

ith resident jth group

ith resident jth group

Few White, Owner Occupied, Single Unit

Moderate White, Owner Occupied, Single Unit

Many White, Owner Occupied, Single Unit

Many Hispanic, Renter Occupied, Multiple Unit

Moderate Hispanic, Renter Occupied, Multiple Unit

Few Hispanic, Renter Occupied, Multiple Unit

Census-Wide Association Study

Challenges:

1.  High pairwise correlation among census variables

2.  The number of variables are large

3.  The regression model uses a noncanonical link function

Census-Wide Association Study

Solution:

1. & 2. Add L1 and L2 penalties to impose group selection on the regression coefficients. 3. Use statistical software Stan to implement Bayesian Elastic Net.

Census-Wide Association Study