Change regression units

[griffinsharps]

What

The units of the multinomial regression in our analysis need to be switched from households to daily observations.

Why

I was originally assigning individual households to clusters based on a dominance score based their modal membership to a cluster based on daily profiles over a month. This throws away a lot of information about usage patterns, so we are eliminating this step and running the multinomial regression on the clusters of usage profiles (households x days). This allows for a more detailed analysis and better results for the client.

How

1. Load cluster assignments without aggregation, treating each row as a household-day assignment.
2. Attach block groups by joining ZIP+4→BG crosswalk.
3. Aggregate to BG × cluster.
4. Compute cluster share = number of assignments / total assignments
5. Fit MNLogit with:
   • outcome constructed from cluster share
   • Frequency weights

Deliverables

• PR with: Functioning update to the code that passes the clustering results through a multinomial regression and delivers initial results.

[linear]

SMA-31