| Type: | Package |
| Title: | Cox Regression (Proportional Hazards Model) with Multiple Causes and Mixed Effects |
| Version: | 1.1.1 |
| Date: | 2015-10-24 |
| Author: | Jing Peng |
| Maintainer: | Jing Peng <pengjing@live.com> |
| Description: | A high performance package estimating Cox Model when an even has more than one causes. It also supports random and fixed effects, tied events, and time-varying variables. |
| License: | GPL (≥ 3) |
| LazyData: | TRUE |
| Depends: | R (≥ 3.1.0), Rcpp (≥ 0.12.0) |
| Imports: | methods |
| LinkingTo: | Rcpp, RcppArmadillo |
| NeedsCompilation: | yes |
| Packaged: | 2015-10-24 07:11:57 UTC; Peng |
| Repository: | CRAN |
| Date/Publication: | 2015-10-24 09:32:41 |
Cox Regression (Proportional Hazards Model) with Multiple Causes and Mixed Effects
Description
A high performance package estimating Proportional Hazards Model when an even can have more than one causes, including support for random and fixed effects, tied events, and time-varying variables.
Usage
fastCox(head, formula, par = list(), data = NULL)
Arguments
head |
A data frame with 4~5 columns: start, stop, event, weight, strata (optional). |
formula |
A formula specifying the independent variables |
par |
A optional list of parameters controlling the estimation process |
data |
The dataset, a data frame containing observations on the independent variables |
Value
A list containing the estimated parameters
References
1. Jing Peng, Ashish Agarwal, Kartik Hosanagar, and Raghuram Iyengar. Towards Effective Information Diffusion on Social Media Platforms: A Dyadic Analysis of Network Embeddedness. Working Paper.
2. Jing Peng, Ashish Agarwal, Kartik Hosanagar, and Raghuram Iyengar. Toward Effective Social Contagion: A Micro Level Analysis of the Impact of Dyadic Network Relationship. In Proceedings of the 2014 International Conference on Information Systems.
Examples
# Simulate a dataset. lam=exp(x), suvtime depends on lam
x = rnorm(5000)
suvtime = -log(runif(length(x)))/exp(x)
# Censor 80% of events
thd = quantile(suvtime, 0.2)
event = as.numeric(suvtime <= thd)
suvtime[suvtime>thd] = thd
# The estimates of beta should be very close to 1, the true value
head = cbind(start=0,stop=suvtime,event=event,weight=1)
est = fastCox(head,~x)
print(est$result)