Agresti's first book: Categorical Data Analysis, Appendix B.
If you're interested in probability and random processes look at he notes
at the end of each chapter in "Real Ananlysis and Probability" by R. M
Dudley.
Antonio,
You basically need the cross-correlation function in R which is very easy
to use - just look up the examples in ?ccf.
So, if any of the books you mention deal with it - you'll be ok.
If you know about ARIMA models (or know someone who does in a maths/stats
dept) and stuff like that then
Here's some code to implement the cheese data table using the proportinal
odds model given in Generalised Linear Models by McCullagh & Nelder (Ch5).
You will have to adapt this to handle your case data - but i'll give you
something to go on.
cheese
library(MASS)
options(contrasts =
Hi,
didn't get your name.
For large tables (5 X 5) or bigger the dist of the log of the cross product
ratios tends to normality. there are (nC2)**2/2 of these (200 in a 5X5
table. The chi-sq test for independence fits a main effects loglinear model
to the table and this can be expressed in terms o
My thoughts exactly.
?FFT should do the job.
And define the dominant term - a_n**2 + b_n**2 - the Parseval Relation.
stephen sefick
James,
look up "errors in variables" models or "instrumental variable" models in
econometrics.
The statistics alternative is a "random effects" or "mixed effects" model
which plugs the variation in the x's into a randomly varying parameter -
these are available in R (?lmer or glmm - I think).
Som
Emmanuel,
Friedman's (Annals of Stats 1991) MARS program implements recursive
partitioning in a regression context - a version of it written by Trevor
Hastie was available in R but I don't know what package it's now in - I
only have base stuff available (long story).
MARS, like recursive partitio
Dear all,
The standard call to ARIMA in the base package such as
arima(y,c(5,0,0),include.mean=FALSE)
gives a full 5th order lag polynomial model with for example coeffs
Coefficients:
ar1ar2 ar3 ar4 ar5
0.4715 0.067 -0.1772 0.0
Fit an ARIMA(0,2,2) model - it's the same thing and you'll get the MLE of
the smoothing parameter for free.
Use logs if you want a multiplicative model.
Gerard
Stephan Kolassa
Keon,
why not fit a loglinear independence model which as far as I know is the
same.
Gerard
Here's an example from Agresti - Intro to Cat Data analysis
Example: Alcohol, cigarette, marijuana use
|--+--+|
| Alcohol |
Dear All,
1. Does anyone have experience of running R on a server inside a Citrix
shell - I'd like to get R onto the server and would be greatful for any
tips or direction on the matter.
2. This may seem like a silly question so forgive my ignornace.
Most of the data I currently work with is h
Dear All,
1. Does anyone have experience of running R on a server inside a Citrix
shell - I'd like to get R onto the server and would be greatful for any
tips or direction on the matter.
2. This may seem like a silly question so forgive my ignornace.
Most of the data I currently work with is hel
Ajay ohri
To
Greg Snow
03/03/2009 04:58
Greg Snow
To
Sent by: "Gerard M. Keogh"
r-help-boun...@r- , Frank E
project.org
Frank E Harrell
Jr
"Gerard M. Keogh"
27/02/
Frank,
I can't see the code you mention - Web marshall at work - but I don't think
you should be too quick to run down SAS - it's a powerful and flexible
language but unfortunately very expensive.
Your example mentions doing a vector product in the macro language - this
only suggest to me that th
Saji,
This may help.
Your model is
(1,0,1)X(0,1,1)S
giving difference polynomials
nonseasonal (1,0,1) = (1-ar1*B) = (1-ma1*B)
seasonal (0,1,1)S = (1-B**S)= (1-sma1*B**S)
giving: (1-ar1*B)X(1-B**S) x_t = (1-ma1*B)X(1-sma1*B**S) a_t
multiplying out:
x_t - x_(t-S) - ar1*x_(
Surely, this sounds like a bug in the optim function.
The rule of thumb with ts data is to scale so that data have mean 0 and
unit variance and then fit
a) for non-seasonal data the IMA model (0,1,1); and
b) for seasonal data so-called Airline Model (0,1,1)X(0,1,1)S
see for example A course
Quick response on the binomial:
If possible I would suggest you should model
pi = (number/freq of type A) / (total_freq of type A)
veg.glm = glm ( pi ~ x, weights = total_freq, family=binomial)
The glm method is supposed to work only on the natural numbers (inc 0!) but
also works for decimal da
Hi,
here's a possibility!
Your problem can be restated as "given 2 observers giving 2 measures what
is their level of agreement" - the classical measure here is the Kappa (see
sec 10.5 of Categorical Data Analysis by Alan Agresti (in Ed 1 - Ed 2
should also have it!) and you can also model the si
Dear all,
sorry to bother you all with this but I've been trying to use the loglm in
MASS package (v2.8.0) and cannot get any sensible output.
I'm wondering am I doing something very foolish or missing something
obvious.
For example, I tried the documentation help(loglm) example - here's the
cod
Tim,
Given you have so little data - I would try
a) prefilter and forecast
Fit a fairly simple ARIMA(0,1,1) model to A and treat 8987 as an outlier -
then predict the fitted series AF with the 3 missing points as forecasts.
Fit another ARIMA(0,1,1) with 7688 as an outlier - then back cast to fill
Dear all,
I've replicated the cheese tasting example on p175 of GLM's by McCullagh
and Nelder. This is a 4 treatment (rows) by 9 ordinal response (cols)
table.
Here's my simple code:
cheese
library(MASS)
options(contrasts = c("contr.treatment", "contr.poly"))
y = c
ma use lm on lagged values.
Gerard
"Stephen Oman"
To
Hi all,
I have a multiple queing situation I'd like to simulate to get some idea of
the distributions - waiting times and allocations etc.
Does R has a package available for this - many years ago there used to be a
language called "simscript" for discrete event simulation and I was
wondering if
Sorry,
but this gives me the shivers!
Are all your time series linear?
For each model you should check the residuals and their squares to see if
they are uncorrelated (Box-ljung Chi-sq).
Another useful check is to test for a trend in the coefficient of variation
of the residuals.
If the series is
Hi all,
Just on this question :
can I assume any R internal defined function can be used to describe the
link (e.g. = "arctan") so long as its increasing and monotone?
How might abs work for example - (except at 0)?
And/or finally, can I define any old function in R called "myfun" and use
link="
re pseudo inverse
On the point of generalised inverses - GINV is usually taken to mean the
moore-penrose pseudo inverse - this is the least squares projection.
There are others - e.g. the Drazin inverse which amounts to diagonalisation
- of course this inverse may not be available in R.
Gerard
Hi all,
apologies if this is obvious - but I can't see it and would appreciate some
quick help!
the matrix mhouse is 26x3 and I'm computing odds ratios. The simple code
below "should" compute the odds vector for every pair (325) i.e. 26C2 in
cols 1 and 2.
On the first i=1 outer loop the inner j
V interesting point Greg.
But are you not just suggesting left and right truncation?
It strikes me that if the data are Poission then a mixture is likely to be
better - something akin to zero-deflated.
Neg binomail works for greater variance == mix a gamma and poisson, but I'm
unsure what to mix t
Hi all,
simple Q:
how do I extract the upper and lower CI for predicted probabilities
directly for a glm - I'm sure there's a one line to do it but I can't find
it.
the predicted values I get with the predict (.. "response")
Thanks
Gerard
*
Q1: Quick answer
a) you need to remove the seasonality - there s/b a tool in the time series
package to do this - though I'm not familar enough with R to know this.
b) check the resulting series to see if it is stationary - acf decays
quickly i.e. within a couple of lags.
c) if two series are st
Sara,
look carefully at the acf again and increase the lag.
Lags outside the envelope indicate differencing may be necessary.
If the data are "seasonal" you could be seeing a cycle with period 8 -
you'll see alias peaks at 16 and 24; ideally plot the periodiogram which
will show spikes at these fr
Hi everyone,
newbee query!
I've installed R 2.8.0 and tried to run this simple glm -
x is no of cars in a given year, y is the number voted in an election
that year while n is the population 18+:
votes <- data.frame(x = c(0.62,0.77,0.71,0.74,0.77,0.86,1.13,1.44),
+
34 matches
Mail list logo
