[6551] in s-news-athena
New and updated functions installed in statlib
daemon@ATHENA.MIT.EDU (Frank Harrell)
Thu Mar 2 11:28:56 1995
Date: Thu, 2 Mar 1995 10:23:28 -0500 (EST)
From: Frank Harrell <feh@biostat.mc.duke.edu>
Reply-To: feh@biostat.mc.duke.edu
To: s-news@utstat.toronto.edu
Mike Meyer has installed updates to several of my functions as well as
two new function:
----------------------------------------------------------------------
New functions: find.matches, cpower/ciapower (stored under name cpower)
find.matches Find matches between columns of matrices
This function is useful for drawing matched samples from a reference
sample using exact matching, matching within specified tolerances,
and Euclidean distance matching with specified scale constants.
Example:
> set.seed(17) # so can replicate results
> x <- matrix(runif(500), ncol=2)
> y <- matrix(runif(2000), ncol=2)
> w <- find.matches(x, y, maxmatch=5, tol=c(.02,.03))
> w$matches[1:5,]
Match #1 Match #2 Match #3 Match #4 Match #5
[1,] 383 0 0 0 0
[2,] 710 283 585 143 67
[3,] 935 343 682 47 659
[4,] 919 857 491 810 557
[5,] 558 8 675 0 0
> w$distance[1:5,]
Distance #1 Distance #2 Distance #3 Distance #4 Distance #5
[1,] 1.0160329 NA NA NA NA
[2,] 0.4180267 0.5695871 0.6080781 0.8480735 1.1498693
[3,] 0.1263243 0.1358317 0.4338589 0.4425930 0.7466168
[4,] 0.1120088 0.4246693 0.4630336 0.5731505 0.6278675
[5,] 0.1407727 0.6417053 1.1753261 NA NA
> x[1,]
[1] 0.9596826 0.6678012
> y[383,]
[1] 0.9467001 0.6446666
> summary(w)
Frequency table of number of matches found per observation
0 1 2 3 4 5
25 59 60 48 34 24
Median minimum distance by number of matches
1 2 3 4 5
0.650781 0.3803045 0.2224013 0.2467452 0.1326343
Observations selected first more than once (with frequencies)
5 17 32 185 243 315 383 394 501 535 621 657 708 745 770 817 848 899
2 2 2 2 2 2 2 3 2 2 2 2 4 2 2 2 2 2
# For many applications would do something like this:
> attach(df1)
> x <- cbind(age, sex)
# Just do as.matrix(df1) if df1 has no factor objects
> attach(df2)
> y <- cbind(age, sex)
> mat <- find.matches(x, y, tol=c(5,0)) # exact match on sex, 5y on age
cpower - Compute power of two-sample Cox or logrank survival test
ciapower- Compute power of interaction test for exponential survival
---------------------------------------------------------------------------
Bug corrections: cut2, impute
---------------------------------------------------------------------------
New features:
sas.get: Added uncompress option, sasprog option, where option,
is.special.miss function, [.special.miss function,
print.special.miss,
sas.codes: prints original SAS codes that were given value
labels with PROC FORMAT
Fixed bug in reading very long records when SAS dataset converted
to ASCII
Note: sas.get still only runs on UNIX. A couple of MS Windows users have
been experimenting with modifying sas.get to work there. Will keep you
posted.
bystats: Allowed the variable being described to be a matrix
Added LaTeX support
Added bystats2 function for printing 2-way tables
Examples:
> a <- rep(1:5,100)
> b <- c(rep(1,200),rep(2,300))
> set.seed(13)
> x <- runif(500)
> table(a)
1 2 3 4 5
100 100 100 100 100
> table(b)
1 2
200 300
> options(digits=2)
> bystats(x>.5, a, b)
Fraction of x > 0.5 by a, b
N Fraction
1 1 40 0.47
2 1 40 0.42
3 1 40 0.45
4 1 40 0.55
5 1 40 0.47
1 2 60 0.50
2 2 60 0.42
3 2 60 0.57
4 2 60 0.67
5 2 60 0.50
ALL 500 0.51
> bystats(x, a)
Mean of x by a
N Mean
1 100 0.49
2 100 0.46
3 100 0.53
4 100 0.57
5 100 0.49
ALL 500 0.51
> bystats(x, a, fun=quantile)
quantile of x by a
N 0% 25% 50% 75% 100%
1 100 0.0050 0.20 0.49 0.72 0.99
2 100 0.0234 0.21 0.38 0.75 0.98
3 100 0.0029 0.26 0.53 0.82 1.00
4 100 0.0103 0.26 0.63 0.82 1.00
5 100 0.0088 0.21 0.49 0.74 0.99
ALL 500 0.0029 0.22 0.51 0.76 1.00
> bystats(x, a, fun=function(x) c(Mean=mean(x), Median=median(x)))
c(Mean = mean(x), Median = median(x)) of x by a
N Mean Median
1 100 0.49 0.49
2 100 0.46 0.38
3 100 0.53 0.53
4 100 0.57 0.63
5 100 0.49 0.49
ALL 500 0.51 0.51
> x2 <- runif(500)
> bystats(cbind(x,x2), a, fun=function(y) apply(y, 2, quantile, c(.25, .75)))
quantile of cbind(x, x2) by a
N x 25% x 75% x2 25% x2 75%
1 100 0.20 0.72 0.23 0.71
2 100 0.21 0.75 0.31 0.81
3 100 0.26 0.82 0.24 0.72
4 100 0.26 0.82 0.25 0.81
5 100 0.21 0.74 0.29 0.72
ALL 500 0.22 0.76 0.25 0.75
> x[1] <- NA
> a <- cut2(a, c(2,4)) #cut2 in statlib
> bystats2(sqrt(x), a, b, fun=function(x) c(Mean=mean(x), quantile(x)))
c(Mean = mean(x), quantile(x)) of sqrt(x) by a, b
+-------+
|N |
|Missing|
|Mean |
| 0% |
| 25% |
| 50% |
| 75% |
|100% |
+-------+
a |b
|1 |2 |ALL |
-----+-----+-----+-----+
[1,2)| 39 | 60 | 99 |
|1 |0 |1 |
|0.64 |0.66 |0.66 |
|0.071|0.207|0.071|
|0.44 |0.48 |0.45 |
|0.66 |0.70 |0.70 |
|0.84 |0.85 |0.85 |
|0.99 |1.00 |1.00 |
-----+-----+-----+-----+
[2,4)| 80 |120 |200 |
|0 |0 |0 |
|0.67 |0.65 |0.66 |
|0.057|0.054|0.054|
|0.50 |0.42 |0.47 |
|0.69 |0.70 |0.69 |
|0.87 |0.89 |0.88 |
|1.00 |1.00 |1.00 |
-----+-----+-----+-----+
[4,5]| 80 |120 |200 |
|0 |0 |0 |
|0.68 |0.69 |0.69 |
|0.169|0.094|0.094|
|0.47 |0.47 |0.47 |
|0.73 |0.78 |0.74 |
|0.83 |0.89 |0.87 |
|1.00 |1.00 |1.00 |
-----+-----+-----+-----+
ALL |199 |300 |499 |
|1 |0 |1 |
|0.67 |0.67 |0.67 |
|0.057|0.054|0.054|
|0.48 |0.45 |0.47 |
|0.69 |0.73 |0.71 |
|0.85 |0.89 |0.87 |
|1.00 |1.00 |1.00 |
-----+-----+-----+-----+
varclus: Added support for formula language, data, subset, na.action
Added new function naclus, which depicts the similarities
between variables on which observations are NA
----------------------------------------------------------------------------
Frank E Harrell Jr feh@biostat.mc.duke.edu
Associate Professor of Biostatistics
Division of Biometry Duke University Medical Center
Box 3363 Durham NC 27710 USA
