HI
I am trying to analyse data which is left-censored (i.e. has values below the
detection limit). I have been using the NADA package of R to derive summary
statistics and do some regression. I am now trying to carry out regression on
paired data where both my X and Y have left-censored data within them.
I have tried various commands in R:
rega = cenreg(Cen(conc, cens_ind) ~ Gp_ident))?
with all X and Y data stacked and using a group identifier to look at the
differences
this doesn't take account of the paired data though.
I have also tried splitting the data and regessing one on the other
rega = cenreg(Cen(conc1, censind1) ~ Cen(conc2,censind2))
which doesn't work.
Does anyone know of a command that will work - or perhaps suggest another
package that I could use?
I have also looked at multiple imputation packages but they all seem to impute
data depending on other columns - whereas I would want to impute data between
zero and the censored value.
Any guidance/advice would be very much appreciated.
Laura
?
Dr Laura MacCalman Msci MSc PhD Gradstat
Senior Statistician
Institute of Occupational Medicine
Research Avenue North
Riccarton
Edinburgh
EH14 4AP
Tel:?0131 449 8078
Fax: 0131 449 8084
Mob:?07595 054 881
Email: Laura.MacCalman at iom-world.org
Web: http://www.iom-world.org
?
--------------------------------------------------------------------------
?
The Institute of Occupational Medicine (IOM) is a company limited by guarantee,
registered in Scotland (No.SC123972) and a?Registered Scottish Charity
(No.SC000365). IOM Consulting Ltd is a wholly owned subsidiary of IOM and a
private limited company registered in Scotland (No. SC205670). Registered
Office: Research Avenue North, Riccarton, Edinburgh, EH14 4AP, Tel +44 (0)131
449 8000.
This email and any files transmitted with it are confide...{{dropped:18}}
I would probably start with maximum likelihood estimation. I suppose you could impute X and Y separately using ros() from the NADA package, and then run you ordinary regression on the imputed values. Obviously, this ignores any relationship between X and Y, since each is imputed independently of the other. I have no idea whether ordinary inferences on the parameter estimates would be valid. Probably not. Probably, MLE would be better. -Don -- Don MacQueen Lawrence Livermore National Laboratory 7000 East Ave., L-627 Livermore, CA 94550 925-423-1062 On 4/15/13 8:55 AM, "Laura MacCalman" <Laura.MacCalman at iom-world.org> wrote:> >HI > >I am trying to analyse data which is left-censored (i.e. has values below >the detection limit). I have been using the NADA package of R to derive >summary statistics and do some regression. I am now trying to carry out >regression on paired data where both my X and Y have left-censored data >within them. > >I have tried various commands in R: > >rega = cenreg(Cen(conc, cens_ind) ~ Gp_ident)) >with all X and Y data stacked and using a group identifier to look at the >differences > >this doesn't take account of the paired data though. > >I have also tried splitting the data and regessing one on the other > >rega = cenreg(Cen(conc1, censind1) ~ Cen(conc2,censind2)) > >which doesn't work. > >Does anyone know of a command that will work - or perhaps suggest another >package that I could use? > >I have also looked at multiple imputation packages but they all seem to >impute data depending on other columns - whereas I would want to impute >data between zero and the censored value. > >Any guidance/advice would be very much appreciated. > >Laura > > > >Dr Laura MacCalman Msci MSc PhD Gradstat >Senior Statistician > >Institute of Occupational Medicine >Research Avenue North >Riccarton >Edinburgh >EH14 4AP > >Tel: 0131 449 8078 >Fax: 0131 449 8084 >Mob: 07595 054 881 >Email: Laura.MacCalman at iom-world.org > >Web: http://www.iom-world.org > > >-------------------------------------------------------------------------- > >The Institute of Occupational Medicine (IOM) is a company limited by >guarantee, registered in Scotland (No.SC123972) and a Registered Scottish >Charity (No.SC000365). IOM Consulting Ltd is a wholly owned subsidiary of >IOM and a private limited company registered in Scotland (No. SC205670). >Registered Office: Research Avenue North, Riccarton, Edinburgh, EH14 4AP, >Tel +44 (0)131 449 8000. > >This email and any files transmitted with it are confide...{{dropped:18}} > >______________________________________________ >R-help at r-project.org mailing list >https://stat.ethz.ch/mailman/listinfo/r-help >PLEASE do read the posting guide >http://www.R-project.org/posting-guide.html >and provide commented, minimal, self-contained, reproducible code.
On Apr 15, 2013, at 8:55 AM, Laura MacCalman wrote:> > HI > > I am trying to analyse data which is left-censored (i.e. has values below the detection limit). I have been using the NADA package of R to derive summary statistics and do some regression. I am now trying to carry out regression on paired data where both my X and Y have left-censored data within them. > > I have tried various commands in R: > > rega = cenreg(Cen(conc, cens_ind) ~ Gp_ident)) > with all X and Y data stacked and using a group identifier to look at the differences > > this doesn't take account of the paired data though. > > I have also tried splitting the data and regessing one on the other > > rega = cenreg(Cen(conc1, censind1) ~ Cen(conc2,censind2)) > > which doesn't work. > > Does anyone know of a command that will work - or perhaps suggest another package that I could use? > > I have also looked at multiple imputation packages but they all seem to impute data depending on other columns - whereas I would want to impute data between zero and the censored value. > > Any guidance/advice would be very much appreciated.The `survival::Surv` function allows left censoring and the `coxph` function allows strata or clusters to be specified. My understanding is that for many years analysts used the Cox regression machinery to crank out conditional logistic regression by creating two-member (or 1+n) member strata/cluster. I'm wondering if that could be made to work here, since this seems even closer to a "real" survival analysis problem. This response from Terry Therneau (looked up with Markmail) to a question about a right-censored situation suggests that the use of cluster() rather than strata() might be be more powerful: http://markmail.org/message/c2oqqd34nujxvuvi?q=list:org%2Er-project%2Er-help+paired+survival+coxph+strata -- David.> > Laura > > > > Dr Laura MacCalman Msci MSc PhD Gradstat > Senior Statistician > > Institute of Occupational Medicine > Research Avenue North > Riccarton > Edinburgh > EH14 4AP > > Tel: 0131 449 8078 > Fax: 0131 449 8084 > Mob: 07595 054 881 > Email: Laura.MacCalman at iom-world.org > > Web: http://www.iom-world.org > > > -------------------------------------------------------------------------- > > The Institute of Occupational Medicine (IOM) is a company limited by guarantee, registered in Scotland (No.SC123972) and a Registered Scottish Charity (No.SC000365). IOM Consulting Ltd is a wholly owned subsidiary of IOM and a private limited company registered in Scotland (No. SC205670). Registered Office: Research Avenue North, Riccarton, Edinburgh, EH14 4AP, Tel +44 (0)131 449 8000. > > This email and any files transmitted with it are confide...{{dropped:18}} > > ______________________________________________ > R-help at r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code.David Winsemius Alameda, CA, USA