This site is an archived version of Voteview.com archived from University of Georgia on May 23, 2017. This point-in-time capture includes all files publicly linked on Voteview.com at that time. We provide access to this content as a service to ensure that past users of Voteview.com have access to historical files. This content will remain online until at least January 1st, 2018. UCLA provides no warranty or guarantee of access to these files.

POLS 6386 MEASUREMENT THEORY
Eleventh Assignment
Due 15 April 2003

1. In this problem we are going to use the Multidimensional Scaling routine that is part of the pcurve package in R (a bow to Noah Kaplan for pointing this out to me!). Download the example R program and data files below that show how it works with the Color Circle data: R Program to do Kruskal MDS (mdscal_color_1.r)
A listing of mdscal_color_1.r is below:
```#
#
# mdscal_color_1.r -- Tests R Version of Kruskal MDS
#
#           This needs MASS and pcurve packages
#
# The Data Must Be Transformed to Squared Distances Below
#
library(MASS)
library(pcurve)
#
T <- matrix(scan("D:/R_Files/colors.txt",0),ncol=14,byrow=TRUE)
attach(colornames)
TT <- T
nrow <- length(T[,1])
ncol <- length(T[1,])
xrow <- NULL
xcol <- NULL
matrixmean <- 0
#
# Transform the Matrix into Squared Distances The R MDS program assumes that
#                                             the data are distances
i <- 0
while (i < nrow) {
i <- i + 1
j <- 0
while (j < ncol) {
j <- j + 1
TT[i,j] <- ((100 - T[i,j])/50)**2
}
}
T <- TT
#
#  Kruskal MDS Routine
#
# T -- Input
# dim=2 -- number of dimensions
# maxit = number of iterations
# The program returns the configuration in points and the Stress in stress:
#     for example, colormds\$points and colormds\$stress
#
colorsmds <- isomds(T,dim=2, maxit=50)
#
#
plot(colorsmds\$points[,1],colorsmds\$points[,2],type="n",asp=1,
main="The Color Circle\nFrom MDS Program in R",
xlab="First Dimension",ylab="Second Dimension",
xlim=c(-3.0,3.0),ylim=c(-3.0,3.0))
text(-2.0,2.5,paste("Stress = ",
0.01*round(colorsmds\$stress, 2)),col="blue")
#Note this neat trick -- Stress is returned as a percentage so that
#multiplying it by 0.01 converts it to the KYST style.  The "round(..)" command
#gets us 4 digits after the decimal point
```
Below is the graph that the program generates: 1. Repeat problem 1a of Homework 4 using mdscal_color_1.r. Namely, run Tables 1.1 (crime data), 1.3 (nations data), 1.4 (one spoked wheels), 4.1 (color data), and 4.4 (Faces data -- note that these are distances!) in one, two, and three dimensions. Produce two-dimensional plots for each dataset and make a table showing the Stress in one, two, and three dimensions for isomds(...) and KYST for all the matrices. Notes:

• You have already run KYST on all the above matrices except 4.4.

• You will have to use Epsilon to change Tables 1.1, 1.3, 1.4, and 4.4 into square symmetric matrices!

• To run mdscal_color_1.r in one dimension, simply comment out the plotting commands!

• If the input data are distances/dissimilarites simply comment out the transformation line.

2. Make a Shepard Gradient of Generalization graph for each two-dimensional configuration from part (a). Use the R program below to do the plots: R Program to do Shepard Gradient of Generalization graph (mdscal_color_3.r)

Below is a listing of the program:
```
#
# mdscal_color_3.r -- Tests R Version of Kruskal MDS -- This version
#                      Produces Shepard's Gradient of Generalization
#
#           This needs MASS and pcurve packages
#
# The Data Must Be Transformed to Squared Distances Below
#
library(MASS)                   Note that this program is the same as the one
library(pcurve)                 above through the call to isomds(...)
#
T <- matrix(scan("F:/R_Files_Office/colors.txt",0),ncol=14,byrow=TRUE)
attach(colornames)
TT <- T
#
#  Save Copy of Original Data
#
TTT <-T
nrow <- length(T[,1])
ncol <- length(T[1,])
#
y <- rep(0,((nrow*(nrow+1))/2)*5)   This simply creates a matrix that will hold
dim(y) <- c(((nrow*(nrow+1))/2),5)  the "DATA" and "DIST" as in KYST
#
#
# Transform the Matrix to Squared Distances
#
i <- 0
while (i < nrow) {
i <- i + 1
j <- 0
while (j < ncol) {
j <- j + 1
#
#  Adjust This Transformation According to the MAXIMUM similarity
#    Value!  The Diagonal of the Matrix Should be all zeroes!
#    If the data are correlations this would be:
#     TT[i,j] <- (1.0 - T[i,j])**2
#    If the maximum value for the Similarity was 10 this would be:
#     TT[i,j] <- (10.0 - T[i,j])**2 or
#     TT[i,j] <- ((10.0 - T[i,j])/5.0)**2
#
TT[i,j] <- ((100 - T[i,j])/50)**2
}
}
T <- TT
#
#
# T -- Input
# dim=2 -- number of dimensions
#
dim <- 2
colorsmds <- isomds(T,dim, maxit=50)
#
#
# Create Data for Gradient of Generalization Plots
#
i <- 1                           This code creates the Euclidean distances
kk <- 0                          between the points estimated by isomds(...).
while (i <= nrow) {              These are stored in the matrix y(,) along with
j <- i                         the original Similarities and the Dissimilarities
while (j <= nrow) {            computed above.
k <- 0
dist <- 0.0
while (k < dim) {
k <- k+1
dist <- dist + (colorsmds\$points[i,k]-colorsmds\$points[j,k])^2
}
kk <- kk +1
y[kk,1] <- dist
y[kk,2] <- T[i,j]
y[kk,3] <- TTT[i,j]
y[kk,4] <- i
y[kk,5] <- j
j <- j + 1
}
i <- i + 1
}
#             This produces a plot of the Distances against the Dissimilarities
plot(y[,1],y[,2],
xlab="Psychological Distance",ylab="Observed Distance/Dissimilarity",col="blue")
mtext(side=3,line=1.5,"Shepard's Theory of Generalization\nColor Data: Dissimilarities",font=2)
lines(lowess(y[,1],y[,2],f=.2),lwd=3)
text(10,2,"Line estimated \nUsing Lowess")
#
windows()               This Command allows us to create the second plot
#                       The Last One Drawn will be on Top
#
#             This produces a plot of the Distances against the Similarities
plot(y[,1],y[,3],ylim=c(0,100),
xlab="Psychological Distance",ylab="Observed Similarity",col="red")
mtext(side=3,line=1.5,"Shepard's Theory of Generalization\nColor Data: Similarities",font=2)
lines(lowess(y[,1],y[,3],f=.2),lwd=3)
text(10,80,"Line estimated \nUsing Lowess")```
For the color data, you should get the following two graphs:  Turn in both graphs for each matrix except for the Faces data (Table 4.4). Just turn in the Dissimilarities against the Distances plot for the Faces data (you can comment out the code for the other plot for this data).

2. In this problem we are going to revisit some of the analyses we did in Problem 2.c and 2.d of Homework 9. Use Epsilon to change the number of dimensions from "2" to "1" in PERFSTRT.2000 and change the last number in the second line from "1" to "2" (this number tells the program that respondent number 2 is on the left or liberal side of the dimension -- it ensures that the rank ordering goes from liberal to conservative) then rename it PERFSTRT.DAT (do not overwrite any other PERFSTRT.DAT you may have). Your PERFSTRT.DAT file should look this:
```VOTE_THERM_2000.ORD
NON-PARAMETRIC MULTIDIMENSIONAL UNFOLDING OF THERMOMETER DATA
1   91   20   40    2
(40A1,3900I1)
(I5,1X,40A1,2I5,50F8.3)```
We are going to use a modified version of Optimal Classification more suited to analyzing roll calls from rank order data. The program is PERFLRANK. Download the program and place it in the same directory as your PERFSTRT.DAT and the candidate "roll call" file VOTE_THERM_2000.ORD. Maximum Classification Scaling Program (PERFLRANK)

1. Run PERFLRANK using the above PERFSTRT.DAT and VOTE_THERM_2000.ORD. Turn in the PERF21.DAT file.

2. The PERF25.DAT shows the rank ordering of the 1,443 respondents included in the scaling along with the rank positions of the cutpoints for the candidate "roll calls." Note that because tied ranks are allowed, the respondents are ordered from 17.5 to 1,419.5. The respondents are first shown in their rank ordering from 17.5 to 1,419.5 and then in the order that they appear in VOTE_THERM_2000.ORD. The 91 cutpoints corresponding to the candidate "roll calls" are at the bottom of PERF25.DAT and the rank position of the cutpoints are in the last column. This part of the file looks like this:
```
1    1 436 568 291  0.333 1 6   392.500
2    2 691 609  80  0.869 1 6   809.000
3    3 781 369  99  0.732 1 6  1114.000
4    4 515 453 191  0.578 1 6  1058.750
5    5 503 539 166  0.670 1 6   688.000
6    6 494 462 178  0.615 1 6   922.500
7    7 428 463 212  0.505 1 6   797.750
8    8 558 488  84  0.828 1 6   809.000
9    9 588 427 328  0.232 6 1   171.000
10   10 422 604 340  0.194 1 6   779.250
11   11 720 567  65  0.885 1 6   788.250
12   12 662 313  95  0.696 1 6  1047.250
13   13 693 504 151  0.700 1 6   797.750
etc etc etc
85   85 633 573 169  0.705 1 6   715.750
86   86 678 488  60  0.877 1 6   896.750
87   87 707 235  91  0.613 1 6  1199.000
88   88 701 387 110  0.716 1 6   980.750
89   89 644 265 117  0.558 6 1   359.000
90   90 565 468 147  0.686 6 1   602.000
91   91 249 600 208  0.165 6 1  1349.500```
Use Epsilon to merge in the 91 pairs of candidate names that you created in question 1.a of Homework number 10 into this file. The first 13 lines of your file should look exactly like this!!!
``` CLINTON      GORE              1 436 568 291  0.333 1 6   392.500
CLINTON      BUSH              2 691 609  80  0.869 1 6   809.000
CLINTON      BUCHANAN          3 781 369  99  0.732 1 6  1114.000
CLINTON      NADER             4 515 453 191  0.578 1 6  1058.750
CLINTON      MCCAIN            5 503 539 166  0.670 1 6   688.000
CLINTON      BRADLEY           6 494 462 178  0.615 1 6   922.500
CLINTON      LIEBERMAN         7 428 463 212  0.505 1 6   797.750
CLINTON      CHENEY            8 558 488  84  0.828 1 6   809.000
CLINTON      HILLARY           9 588 427 328  0.232 6 1   171.000
CLINTON      DEMPARTY         10 422 604 340  0.194 1 6   779.250
CLINTON      REPUBPARTY       11 720 567  65  0.885 1 6   788.250
CLINTON      REFORMPTY        12 662 313  95  0.696 1 6  1047.250
CLINTON      PARTIES          13 693 504 151  0.700 1 6   797.750
etc etc etc```
Turn in the Epsilon macro you used to create your file and a complete listing of the file.

3. There are 13 cutpoints associated with former President Clinton. The cutpoint between former President Clinton and former Vice-President Gore is 392.5. The "1 6" to the immediate left of the rank tells you that Yea (vote for Clinton) was below 392.5 and Nay (vote for Gore) was above 392.5. This implies that former President Clinton's rank position is below 392.5 and former Vice-President Gore's rank position is above 392.5. Although the data has some noise in it, you should be able to pin down a range within which former President Clinton's rank should lie. For example, note that the cutpoint between former President Clinton and Hillary Clinton is 171.0 and the "6 1" means that Hillary Clinton is below 171.0 and former President Clinton is is above 171.0.

Using the above reasoning, locate a range of ranks for former President Clinton, former Vice-President Gore, President Bush, and Hillary Clinton. Defend you reasoning for each person!

4. Use Epsilon to merge the "vote for" variable in VOTE_THERM_2000.ORD into the respondent rank file (cut the respondents out of PERF25.DAT first). Turn in a listing of the Epsilon macro you use to create the file.

5. Make smoothed histograms of the Bush and Gore voters from the above file. These histograms should look similar to those you made in question 1.c of Homework 8.