ivBounds: Instrumental Variables causal effects bounds Get...
In rbas2015/CausalFX: Methods for Estimating Causal Effects from Observational Data

Instrumental Variables causal effects bounds

Get generalization of Balke and Pearl's IV bounds for model with W -> X -> Y.

1	ivBounds(p, epsilons = c(0, 1, 1, 1, 1, 1))

`p`	an array or vector of probabilities P(X=x, Y=x, W=w) or P(X=x, Y=x \| W=w), with X changing fastest, W slowest
`epsilons`	vector of relaxation parameters (see details); defaults to Balke-Pearl bounds

Joint probabilities should be specified, but conditional probabilities are sufficient (and permitted) if epsilon[5] = epsilon[6] = 1. If either doesn't hold and conditional probabilities are supplied, then an error is returned.

epsilons is a vector of six positions corresponding to the relaxation parameters. In order:

the maximum difference in the conditional probability of the outcome given everything else, as the instrument changes levels;
the maximum difference in the conditional probability of the outcome given everything else, and the conditional distribution excluding latent variables for the instrument set at 0;
the maximum difference in the conditional probability of the outcome given everything else, and the conditional distribution excluding latent variables for the instrument set at 1;
the maximum difference in the conditional probability of the treatment given its causes, and the conditional distribution excluding latent variables
the maximum ratio between the conditional distribution of the latent variable given the instrument and the marginal distribution of the latent variable. This has to be greater than or equal to 1;
the minimum ratio between the conditional distribution of the latent variable given the instrument and the marginal distribution of the latent variable. This has to be in the interval (0, 1].

Setting this to c(0,1,1,1,1,1) corresponds to the ordinary IV model.

rbas2015/CausalFX documentation built on May 27, 2019, 2:06 a.m.