#################################################################################################

Aggregate-Latent-Segment-Logit: Demand Model (i.e., Aggregate Latent Class Logit)

Python and R code for the estimation of latent class logit model with aggregate POS data

By Minha Hwang (minha@alum.mit.edu)

#################################################################################################

There are 3 Python / R codes and 1 supporting synthetic data set.

1. lc-agglogit-main-availablity.R: Main R code - Note that this code accomodates varying choice sets (i.e. # of products are allowed to change over time/market.)

2. ll_adclc_ava.R: Log-Likelihood function for maximization

3. fake-data-gen-with-availability.R: Synthetic data generation code (for code validation)

4. lcl_agg_fake_data-with-availability.csv: Synthetic data generated from the code above (aggregated data). The code can generate individual-level data as well.

Reference: 1. A Segment-Level Model of Category Volume and Brand Choice

William R. Dillon and Sunil Gupta

Marketing Science, Vol. 15, No. 1 (1996), pp 38-59

2. The Recoverability of Segmentation Structure from Store-level Aggregate Data

Anand V. Bodapati, and Sachin Gupta

Journal of Marketing Research, August (2004)

Related package)

pyBLP support the estimation of random coefficient logit with aggregate data (under BLP demand model structure). However, pyBLP does not support latent segment

(i.e. latent class) with aggregate data yet.

In case of quesitons, please reach out to minha@alum.mit.edu