Movie study

Data

The directory dat/tmdb_raw/tmdb-5000-movie-dataset/ contains the raw TMDB 5000 Dataset, downloaded from https://www.kaggle.com/tmdb/tmdb-movie-metadata.

Environment

python 2

tensorflow 1.5.0

edward 1.3.5

Scripts

Run src/tmdb_movie_preprocess.ipynb to preprocess data.

Run src/movie-actor-py2-subconf.ipynb to fit factor models.

Run src/movie-actor-py2-causalest.ipynb to estimate causal effect.

Output

The file res/factorfit/ contains prefit substitute confounders.

The file src/movie-actor-py2-causalest.ipynb uses these prefit substitute confounders to estimate causal effects.

Differences from Wang and Blei (2019)

The results in src/movie-actor-py2-causalest.ipynb differ from the results in Section 6.3 and Tables 16-18 in Wang and Blei (2019). However, the captions of those tables apply to these results as well. There are two differences in this implementation:

1. This implementation uses coordinate ascent variational inference (CAVI) for Poisson factorization. The original implementation fitted all factor models with the Edward implementation of black box variational inference.

2. This implementation uses OLS in scikit-learn to estimate causal effects. The original implementation performed a Bayesian linear regression with black box variational inference, as implemented in Edward.

A note on conditioning on observed covariates

Grimmer et al. (2020) points out that conditioning on observed covariates can be important for causal estimation. Using the movie data, they illustrate how the deconfounder without conditioning on observed covariates can produce unreasonable causal estimates.

The exploratory results on actors in Wang and Blei (2019) do not condition on covariates. The "intervened test set" analyses in the supplement of Wang and Blei (2019) includes such conditioning.

For users that are interested in exploring different analyses, the causal estimation section of movie_py/movie-actor-py2-causalest.ipynb includes a block of code that conditions on observed covariates in addition to the substitute confounder.