s-news
[Top] [All Lists]

Re: hurdle or two-step models

To: s-news@lists.biostat.wustl.edu
Subject: Re: hurdle or two-step models
From: Chris Barker <barker.chris@gene.com>
Date: Sat, 27 Mar 2004 07:39:53 -0800
Organization: Genentech, Inc.
References: <20040327112113.B24EC402724@smtp.biostat.wustl.edu>
Reply-to: barker.chris@gene.com
 An analysis that may complement an analysis of the "hurdle" model is a
TOBIT regression, (this can be set up in SPLUS)

or a "Two Part Model" (
Duan N, Manning WG Jr, Morris CN, Newhouse JP. Choosing between the sample selection model and the multi-part model. Journal
of Business and Economic Statistics. 1984 July; 2(3):283-289.
Duan N. Smearing estimate: A nonparametric retransformation method. J. of the American Statistical Assc. 1983 Sept; 78:605-610.
Duan N, Manning WG Jr, Morris CN, Newhouse JP. A comparison of alternative models for the demand for medical care. Journal of
Business and Economic Statistics. 1983 Apr; 1(2):115-126.)

I recall the HMISC() library has a smearing function.
 
Subject: Re: hurdle or two-step models
Date: Fri, 26 Mar 2004 15:06:55 -0400
From: "Smith, Stephen" <SmithSJ@mar.dfo-mpo.gc.ca>
To: "'Andrew Robinson'" <andrewr@uidaho.edu>, jadhavpr@vcu.edu,
     s-news@wubios.wustl.edu, Bill Shipley <bill.shipley@usherbrooke.ca>
You could also check out the following papers:

@ARTICLE{Stefansson96,
  author =       "Gunnar Stef\'{a}nsson",
  title =        "Analysis of groundfish survey abundance data: combining
the GLM and delta approaches",
  journal =      "ICES Journal of Marine Science",
  year =         "1996",
  volume =       "53",
  pages =        "577--588"
    }

@ARTICLE{Waiwoodetal91,
    author = "K. G. Waiwood and S. J. Smith and M. R. Petersen",
    title = "{Feeding of cod ({\it Gadus morhua}) at low temperatures}",
    journal = "Canadian Journal of Fisheries and Aquatic Science",
    year = "1991",
    volume = "48",
    pages = "824--831"

Stephen Smith



-----Original Message-----
From: Andrew Robinson [mailto:andrewr@uidaho.edu
Sent: March 26, 2004 1:44 PM
To: jadhavpr@vcu.edu; s-news@wubios.wustl.edu; Bill Shipley
Subject: [S] hurdle or two-step models


Off-list requests have suggested that more details might be useful.

The hurdle or two-step models are based on the following approach: 

Step 1 (the hurdle): divide the response variable into two classes: 0 and 
non-zero.  Predict the probability of 0 using a glm or glmm.

Step 2: analyze only the non-zero data using whatever model is appropriate
opr 
necessary.

See e.g. Ferguson, D. E., A. R. Stage, and R. J. Boyd.  1986.  Predicting 
regeneration in the grand fir-cedar-hemlock ecosystem of the northern Rocky 
Mountains.  Forest Science Monograph 1986, No 26.  

Other tools, such as ZIP (zero-inflated Poisson) etc. might also be 
appropriate.


<Prev in Thread] Current Thread [Next in Thread>