
Frequently Asked Questions
 How can I read my own data into Arc?
 See Section A.5 of the text book. In addition, you can read the data
directly from Microsoft Excel.
 How can I change the font or font size in Arc?
 Changing the font requires modifying a file. With Windows, start
Arc, and then type the command
(load "config")
In the resulting dialog, you can specify the name for a font and the point
size, and then select "Configure and Quit." This will automatically change
the file wxls32.ini or wxls.ini in your Windows
directory. When you restart, the font will
be changed. However, you should always select a text font like Courier New so
all letters have the same width. You can use any font for graphics.
On Macintosh, you need to edit the file XLS Preferences , again
changing the name of the font if you wish and the point size. When you
restart, the fonts will change.
 What does the Print item on the Windows 95/98/NT File menu do?
 The print item allows you to print the graph in the top window on your
printer. It does not allow printing of text. The method described in the
book of copying a graph to the clipboard and pasting into a word processing
document will be more useful for most users.
 On the Macintosh, the program tells me that I have insufficient node
space and to close graphs. What should I do?
 Quit from Arc, and find the icon for
XlispStat .
Click once on the icon, and then select Get Info from the File menu. In the
resulting dialog, you may need to select Memory from the Show menu. Increase
the values of the minimum and preferred size from 6000K and 8000K to larger
numbers, depending on the amount of memory you have available.
 How can I display the ordered values of a variable?
 If you have a variable called
X , then typing
(qsort X) will display the ordered values. If X
consists of words rather than numbers, they will be alphabetized.
 In Inverse regression, how can I get access to all the eigenvalues
and eigenvectors?
 If the name of the regression menu is
I2.SIR , then drop
everything after the period, and type (send I2 :eigenvectors) or
(send I2 :eigenvalues) .
 When I try to download the updates.lsp file, the text appears in the
browser. How do I save it on my disk?
 From the browser's file menu, select "Save as", and then save the file
with the name updates.lsp in your Arc/Extras directory. Warning: Some browsers, notably Internet Explorer, may automatically add the suffix ".txt" to the downloaded file, so you might end up with a file called "updates.lsp.txt". If this happens, you need to find the file and change its name by deleting the ".txt" or Arc will not be able to read the file.
 Help! The program can't read my datafile correctly!
 Be sure that you have NOT USED a period "." as a missing value character.
This is a frequent problem because a period is the missing value character
for SAS. For more hints, see section A.5.5. of the book.
 The typed command
(standarddeviation x)
returns the standard deviation of the values in x. How do I get the
variance?
(^ (standarddeviation x) 2) returns the variance.
 How can I add leverage contours, like those in Figure 7.7 on page 162
of the book, to a 2D graph?
 There is a typed command to do this. If the plot is named
plot1 (the name is given in square brackets on the title of the
plot), type the message
(send plot1 :leveragecontours)
The contours are at h = 0.2, .4, .8 and 0.1. If you want other contours,
type, for example,
(send plot1 :leveragecontours :h '(.1 .3 .5 .7 .9))
 Explain the text on page 479, which states that the fitted
probabilities in logistic regression "are called
:fitprob ."
 If the name of the binomial model is B1, then the item
B1:fitprob will appear in the list of quantities to plot. Also,
if you type (send B1 :fitprob) , the fitted probabilities will
be displayed in the text window. If you know how to use lisp, you can use
these values in further calculations of your choice.
 Please explain how the transformation slidebars work.

On the slide bars, the transformations are (y^{lambda}1)/lambda. This is
done so that as
lambda is varied from positive to negative, one does not get a jump from
positive correlation to negative (or vice versa).
However, when the transformations are saved, the quantity saved is
y^{lambda} because this is what users expect. Consequently, the sign of the
relationship between the two variables may be the opposite of what it was on
the plot of the transformed variables.
 Is there a way to select a random or fixed subset of cases?

If you have a dataset called d (the name of the dataset's menu), then
(send d :subsample 50)
will select a sample of 50 of the cases for fitting and
graphing.
If you have a fixed subsample (say cases for
which the variable z is between 3 and 12), you can type the following:
(send d :included (ifelse (<= 3 z 12) t nil))
The ifelse function returns t for TRUE if 3 <= z <=12 and nil for FALSE otherwise for all values of
z. The :included method takes the vector of t's and nil's and does all analyses
only on the cases with a corresponding t.

How can I generate a "random response" corresponding to one of the
models I have fit?

If L1 is the name of a model, then the typed command
(send l1 :addrandomresponse) will create a new variable that is
the random response
computed as if the fitted model L1 were true.
If the name of the original response were, for
example, "m", the name of the created variable will be "msim".
For example, if L1 is a linear regression
model, the random response is equal to the fitted values plus sigmahat times
a vector of standard normal random numbers; if the model is binomial, then
the random response is a binomial random number with number of trials as in
the model, and probability of success equal to the fitted probability of
success.

In problem 3.1.3, we "extract mean" from the popup menu on the smoother
slidebar, and get a new variable. I can't figure out what this variable is.

In problem 3.1.3, you have added a smooth line (the lowess curve with
smoothing parameter 0.6) to a plot. This is an estimate of E(yx), or the
"mean function." What "Extract mean" does is:
(1) For each value of x in the data, compute the corresponding point on the
smooth line on the plot. Call this yfit = estimate of E(yx) accoring to the
smoother.
(2) Save the values of yfit as a new variable.
If you draw a graph of yfit vs x, you will see exactly what I mean by
this.

I have a dataset named gender and in it a variable named
gender. Will this cause any problems?

Yes! Because both the data set and the variable are made into local variables,
one will overwrite the other, and you may get unexpected (and possibly
unpredictable) results. Never use the same name for a data set and a
variable!

With Windows, can I use the right mouse button?

The right mouse button can be used to get items in the Edit menu, but for the
text window ONLY. To copy a graph to the clipboard, you must use the Edit
menu on the menu bar (or the keyboard shortcuts either AltEC, or CtrlC).
 The graphs on page 19 of the book illustate the Central Limit
Theorem. Can I draw graphs like these for different data?

You can use the function cltdemo that is included with Arc to get these graphs. If you have a
variable called x, the command (cltdemo x 16 1000) will draw a
histogram of the means of 1000 samples of size 16 (without replacement) from
x. To sample with replacement, add :replace t to the command.
 I cannot reproduce the values for the multipliers given in equation
(20.2), page 451, for the mulitpliers obtained using GREG for the mussels
data.
 The process of fitting using GREG is described on pages 448451. The
very last step is omitted. To get the multipliers for the final direction
selected by GREG, select the item "Display active predictors" from the
graphical regression menu. The values you get for the multipliers in
(20.2) can be quite different from the values given in the book, due to
variation in making graphical judgements. Small variation in the
graphical judgement results in large variation in the multipliers because
the predictors in the mussels data are highly correlated. In any case,
if you draw a
scatterplot of the linear combination given by (20.2) versus the linear
combination you find using GREG, you will discover that they are very
highly correlated.
 The arcsine of the squareroot of a sample proportion can be used as a
variance stabilizing transformation (see page 318). How is this computed in Arc?
 If a variable p contains sample proportions, in the "Add a
variate" dialog, type ps = asin(sqrt(p)). The lisp function
asin computes the arc sine. There are also functions
acos and atan, for the arc cosine and tangent,
respectively.
 I have a data file with 5000 observations. Can I use Arc for big
data sets like this one?
 Arc has no set limit on the size of a data set, either number of
cases or number of variables, but working with large data sets can be
slow, and graphics may be somewhat less informative because of
overplotting of points. For exploratory purposes, it may be helpful to
work with a subset of the cases rather than all of them. Suppose for
example, that the name of your data set is BigData. You can work with a
random subset of, say, 1000 of the cases by typing the command
(send BigData :subsample 1000) . This is like selecting all
but 1000 of the points in a graph and then selecting the "Delete selection
from data set" item on the Case deletions menu. You can get a different
random sample of 1000 by repeating the command, and you can restore all
cases to the data set either using the "Restore all" item in the Case
deletions menu on a graph, or by typing
(send BigData :togglecases) .
 I use both Arc and either Splus or R. Is there an easy way to
transfer data between these programs?
 To transfer a data set from Arc to R or Splus, select the item
"Display data" from the dataset's menu. Select the variables you want to
save, and then push the button for "Save in R/Splus format." You will be
prompted give a file name. If you select, for example
mydata , then two files mydata.txt and
mydata.Rd will
be created. mydata.txt can be read into either R and Splus using
the read.table command. mydata.Rd saves the documentation
about the data set in a standard format for R.
To transfer from R/Splus to Arc, use the write.table function.
If you have a matrix X , use the command
write.table(data.frame(X),"mydata.txt",quote=F,row.names=F,na="?")
If X is a data frame, you can use
write.table(X,"mydata.txt",quote=F,row.names=F,na="?")
 Arc reports that a variable is a text variable when I know that it
is numeric. What is the problem?
 First, check to be sure you have entered the numbers for the variable
correctly, making sure all zeroes are not the letter "Oh", and ones not the
letter "ell." Next, be sure that everything in the variable is
recognizable as a number; even one stray character will turn a variate into
a text variable. A stray character could be a nonprinting character that is put in the file by the program that created it. If none of these is the problem, and your
variable has missing values, be sure that you are using the correct missing
value indicator. The default value is a question mark, but this may be
changed in two ways, as described in Section A.5.9, page 559, of the
textbook. If you are using a public computer, for example, it is possible
that someone changed the default missing value indicator to something else.
 I use the getpic addon to produce graphs that can be compiled and printed using LaTeX. When I use the "Choose response transform..." item in a regression model menu, the LaTeX menu items don't appear in the resulting graph's menu. What can I do?
 You can add the menu items to the graph's menu by typing a command. First draw the graph. At the top of the graph, there will be a name, such as "[Plot2]AIS(L1)Confidence curve", meaning that this is a plot named "Plot2" based on model named "L1" using data set "AIS", and it is a confidence curve. You need the name of the graph, which in this example is "plot2". In the text window, type the following:
> (send plot2 :latexmenuitem)
and the menu items will be added to the menu.
Contact us!
arc@stat.umn.edu
Last modified September 8, 2004
