search

blei-lab / causal-text-embeddings

Software and data for "Using Text Embeddings for Causal Inference"
MIT License
123 stars 17 forks source link

Ground truth #4

Closed srn284 closed 4 years ago

srn284 commented 4 years ago

Dear @vveitch and Team, Re the paper, I had a question for the ground truth values in table 2 for confounding low med high. I was wondering how ground truth is calculated and if so, what methods of verification/robustness checking are endorsed for measurements of ground truth?

Thank you very much. Warm wishes, Shishir

vveitch commented 4 years ago

Hi,

The ground truth here is known in table 2 because the data is semi-synthetic. We simply simulate both the factual and counterfactual outcomes for each unit and then compute their mean difference :)

srn284 commented 4 years ago

Dear Victor, Many apologies for the late reply - was buried with something unavoidable in the last week. Thank you very much for the response! I get what you’re saying now about how we know it’s ground truth and the value is ATE naive without any of the other stats estimators.

Much appreciated Victor! :) Thank you, Shishir

On Dec 6, 2019, at 9:03 PM, Victor Veitch notifications@github.com wrote:

Hi,

The ground truth here is known in table 2 because the data is semi-synthetic. We simply simulate both the factual and counterfactual outcomes for each unit and then compute their mean difference :)

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

vveitch commented 4 years ago

Glad that cleared it up :)