From: Roland R Regoes <roland.regoes_at_env.ethz.ch>

Date: Mon 16 Jan 2006 - 00:04:19 EST

Error: subscript out of bounds

In addition: Warning message:

step size truncated: out of bounds

*>
*

There were 33 warnings (use warnings() to see them)

*> warnings()
*

1: step size truncated: out of bounds

...

*>
*

*>
*

Date: Mon 16 Jan 2006 - 00:04:19 EST

Dear R users,

I am having some problems with glm. The first is an error message "subscript out of bounds". The second is the fact that reasonable starting values are not accepted by the function.

To be more specific, here is an example:

*> success <- c(13,12,11,14,14,11,13,11,12)
**> failure <- c(0,0,0,0,0,0,0,2,2)
**> predictor <- c(0,80*5^(0:7))
**> glm(cbind(success,failure) ~ predictor + 0,
*

+ family=binomial(link="log"), + control=glm.control(epsilon=1e-8,trace=TRUE,maxit=50))Deviance = 3.348039 Iterations - 1

Error: subscript out of bounds

In addition: Warning message:

step size truncated: out of bounds

The model with intercept yields:

*> glm(cbind(success,failure) ~ predictor ,
*

+ family=binomial(link="log"), + control=glm.control(epsilon=1e-8,trace=FALSE,maxit=50)) Call: glm(formula = cbind(success, failure) ~ predictor, family = binomial(link = "log"), control = glm.control(epsilon = 1e-08, trace = FALSE, maxit = 50))

Coefficients:

(Intercept) predictor

-5.830e-17 -4.000e-08

Degrees of Freedom: 8 Total (i.e. Null); 7 Residual

Null Deviance: 12.08 Residual Deviance: 2.889 AIC: 11.8

There were 33 warnings (use warnings() to see them)

1: step size truncated: out of bounds

...

31: step size truncated: out of bounds 32: algorithm stopped at boundary value in: glm.fit(x = X, y = Y, weights = weights, start = start, etastart = etastart, ... 33: fitted probabilities numerically 0 or 1 occurred in: glm.fit(x = X, y = Y, weights = weights, start = start, etastart = etastart, ...

Since the intercept in the above fit is fairly small I thought I could use -4e-8 as a reasonable starting value in a model without intercept. But to no avail:

*> glm(cbind(success,failure) ~ predictor + 0, start=-4e-8,
*

+ family=binomial(link="log"), + control=glm.control(epsilon=1e-8,trace=FALSE,maxit=50)) Error in glm.fit(x = X, y = Y, weights = weights, start = start, etastart = etastart, : cannot find valid starting values: please specify some

I am stuck here. Am I doing something wrong when specifying the starting value? I would appreciate any help. (I could not find anything relevant in the documentation of glm and the mailing list archives, but I did not read the source code of glm yet.)

Roland

PS: I am using R Version 2.2.0 (R Cocoa GUI 1.13 (1915)) on MacOSX 10.4.4

-- ----------------------------------------------------------------------- Roland Regoes Theoretical Biology Universitaetsstr. 16 ETH Zentrum, CHN H76.1 CH-8092 Zurich, Switzerland tel: +41-44-632-6935 fax: +41-44-632-1271 ______________________________________________ R-help@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.htmlReceived on Mon Jan 16 00:10:57 2006

*
This archive was generated by hypermail 2.1.8
: Fri 03 Mar 2006 - 03:42:04 EST
*