A Study of Improvement on IPW Estimator  

Oleh:

Nagano, Rena ; Nagata, Yasushi

Jenis: Article from Proceeding
Dalam koleksi: 12th ANQ Congress in Singapore, 5-8 Agustus 2014, page 1-7.
Topik: Observational studies; Covariates; Propensity Score; Inverse Probability Weighting Estimator; Doubly Robust Estimator
Fulltext: QP2-2.5-P0257.pdf (753.45KB)

Isi artikel

Rosenbaum and Rubin (1983) introduced a new concept of ‘propensity score’. The propensity score is defined as the conditional probability assignment to a particular treatment given a vector of observed covariates. Rubin (1985) proposed the method for estimating the causal effect by using propensity score, which is the weighted mean calculated by the use of the inverse propensity score. Bang and Rubin (2005) defined doubly robust estimator using a regression function which explains dependent variables by covariates. The doubly robust estimator is also included in this research. In this paper, we propose three kinds of methods: (1) modified IPW estimator by adding a small value to the propensity scores, (2) modified IPW estimator by deleting the data of which the propensity scores are close to zero or one, and (3) modified IPW estimator by taking the logarithm of propensity term. We conduct Monte Carlo simulations to compare the performances of the methods with those of two common competitors: the original IPW estimator and the doubly robust estimator. It is shown that the proposed methods have comparable or lower biases than the competing estimators when both of the propensity score and outcome models are properly specified and when one of the models is misspecified, the proposed methods are superior to IPW estimator

Opini Anda

Klik untuk menuliskan opini Anda tentang koleksi ini!

Kembali