[R] question regarding logit regression using glm

Spencer Graves spencer.graves at pdf.com
Mon Aug 8 00:54:20 CEST 2005


	  The "problem" is that with 40 parameters, you are able to get a 
perfect fit for at least some of the observations.  To achieve this, it 
sends selected parameters to +/-Inf.  Of course, it quits before it gets 
to Inf, but most of your parameter estimates exceeded 1e13 in absolute 
value.

	  What do you want?  Do you really need MSA to be a factor, requiring 
you to estimate 39 parameters for MSA?  Does it make sense to 
parameterize it some other way, like latitude and longitude?  You could 
fit a polynomial in lat + lon and gain substantial insight, I suspect, 
that you can't get from the factor coefficients.

	  spencer graves

Haibo Huang wrote:

> I got the following warning messages when I did a
> binomial logit regression using glm():
> 
> Warning messages: 
> 1: Algorithm did not converge in: glm.fit(x = X, y =
> Y, weights = weights, start = start, etastart =
> etastart,  
> 2: fitted probabilities numerically 0 or 1 occurred
> in: glm.fit(x = X, y = Y, weights = weights, start =
> start, etastart = etastart,  
> 
> Can some one share your thoughts on how to solve this
> problem? Please read the following for details. Thank
> you very much!
> 
> Best,
> Ed
> 
> 
> 
>>Lease=read.csv("lease.csv", header=TRUE)
>>Lease$ET = factor(Lease$EarlyTermination)
>>SICCode=factor(Lease$SIC.Code)
>>TO=factor(Lease$TenantHasOption)
>>LO=factor(Lease$LandlordHasOption)
>>TEO=factor(Lease$TenantExercisedOption)
>>
>>RegA=glm(ET~1+MSA, 
> 
> + family=binomial(link=logit), data=Lease,
> weights=Origil.SQFT)
> Warning messages: 
> 1: Algorithm did not converge in: glm.fit(x = X, y =
> Y, weights = weights, start = start, etastart =
> etastart,  
> 2: fitted probabilities numerically 0 or 1 occurred
> in: glm.fit(x = X, y = Y, weights = weights, start =
> start, etastart = etastart,  
> 
>>summary(RegA)
> 
> 
> Call:
> glm(formula = ET ~ 1 + MSA, family = binomial(link =
> logit), 
>     data = Lease, weights = Origil.SQFT)
> 
> Deviance Residuals: 
>        Min          1Q      Median          3Q        
> Max  
> -6.038e+03  -2.066e-06   0.000e+00   0.000e+00  
> 6.720e+03  
> 
> Coefficients:
>                       Estimate Std. Error    z value
> Pr(>|z|)    
> (Intercept)          5.711e+00  8.466e-02  6.745e+01  
> <2e-16 ***
> MSAAnchorage        -6.493e+00  8.541e-02 -7.602e+01  
> <2e-16 ***
> MSAAtlanta           6.894e+14  2.310e+04  2.985e+10  
> <2e-16 ***
> MSAAustin           -9.362e+14  4.954e+04 -1.890e+10  
> <2e-16 ***
> MSABoston           -2.474e+15  2.151e+04 -1.150e+11  
> <2e-16 ***
> MSACharlotte        -2.150e+15  7.265e+04 -2.960e+10  
> <2e-16 ***
> MSAChicago          -1.174e+15  2.057e+04 -5.707e+10  
> <2e-16 ***
> MSACleveland        -7.607e+14  7.046e+04 -1.080e+10  
> <2e-16 ***
> MSAColumbus         -2.768e+15  1.685e+05 -1.642e+10  
> <2e-16 ***
> MSADallas            2.061e+14  3.261e+04  6.321e+09  
> <2e-16 ***
> MSADenver            5.470e+14  3.366e+04  1.625e+10  
> <2e-16 ***
> MSAEast Bay         -6.191e+01  1.344e+05  -4.61e-04  
>      1    
> MSAFt. Worth        -6.565e+00  8.483e-02 -7.739e+01  
> <2e-16 ***
> MSAHouston          -2.735e+15  3.576e+04 -7.648e+10  
> <2e-16 ***
> MSAIndianapolis     -7.483e+14  6.588e+04 -1.136e+10  
> <2e-16 ***
> MSALos Angeles      -1.388e+15  2.887e+04 -4.809e+10  
> <2e-16 ***
> MSAMinneapolis      -1.011e+15  2.731e+04 -3.702e+10  
> <2e-16 ***
> MSANashville         2.143e+01  9.395e+04   2.28e-04  
>      1    
> MSANew Orleans      -3.370e+15  5.038e+04 -6.689e+10  
> <2e-16 ***
> MSANew York         -2.526e+15  2.969e+04 -8.507e+10  
> <2e-16 ***
> MSANorfolk          -5.614e+01  2.020e+06  -2.78e-05  
>      1    
> MSAOakland-East Bay -2.272e+15  3.642e+04 -6.239e+10  
> <2e-16 ***
> MSAOrange County    -5.165e+14  2.428e+04 -2.128e+10  
> <2e-16 ***
> MSAOrlando          -3.215e+15  1.096e+05 -2.933e+10  
> <2e-16 ***
> MSAPhiladelphia     -8.871e+14  4.948e+04 -1.793e+10  
> <2e-16 ***
> MSAPhoenix          -1.156e+01  8.807e-02 -1.313e+02  
> <2e-16 ***
> MSAPortland          7.604e+14  3.841e+04  1.980e+10  
> <2e-16 ***
> MSARaleigh-Durham   -4.312e+01  1.294e+05  -3.33e-04  
>      1    
> MSARiverside         1.626e+15  4.645e+05  3.500e+09  
> <2e-16 ***
> MSASacramento       -9.873e+14  5.345e+04 -1.847e+10  
> <2e-16 ***
> MSASalt Lake City    1.793e+15  2.029e+05  8.839e+09  
> <2e-16 ***
> MSASan Antonio       9.451e+14  9.473e+04  9.977e+09  
> <2e-16 ***
> MSASan Diego        -3.740e+15  6.651e+04 -5.623e+10  
> <2e-16 ***
> MSASan Francisco     3.109e+14  2.394e+04  1.299e+10  
> <2e-16 ***
> MSASan Jose          7.392e+14  2.961e+04  2.497e+10  
> <2e-16 ***
> MSASeattle          -2.250e+15  1.581e+04 -1.423e+11  
> <2e-16 ***
> MSASt. Louis        -2.606e+15  1.801e+05 -1.447e+10  
> <2e-16 ***
> MSAStamford         -6.592e+00  8.469e-02 -7.784e+01  
> <2e-16 ***
> MSAWashington DC     8.460e+13  3.319e+04  2.549e+09  
> <2e-16 ***
> MSAWest Palm Beach  -3.924e+01  2.308e+05  -1.70e-04  
>      1    
> ---
> Signif. codes:  0 `***' 0.001 `**' 0.01 `*' 0.05 `.'
> 0.1 ` ' 1 
> 
> (Dispersion parameter for binomial family taken to be
> 1)
> 
>     Null deviance:  123111026  on 9302  degrees of
> freedom
> Residual deviance: 3028559052  on 9263  degrees of
> freedom
> AIC: 3028559132
> 
> Number of Fisher Scoring iterations: 25
> 
> 
>>anova(RegA)
> 
> Analysis of Deviance Table
> 
> Model: binomial, link: logit
> 
> Response: ET
> 
> Terms added sequentially (first to last)
> 
> 
>        Df   Deviance Resid. Df Resid. Dev
> NULL                      9302  123111026
> MSA    39          0      9263 3028559052
> 
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html

-- 
Spencer Graves, PhD
Senior Development Engineer
PDF Solutions, Inc.
333 West San Carlos Street Suite 700
San Jose, CA 95110, USA

spencer.graves at pdf.com
www.pdf.com <http://www.pdf.com>
Tel:  408-938-4420
Fax: 408-280-7915




More information about the R-help mailing list