#

Winsteps Tutorial 1
Mike Linacre, instructor - June 2012

1.

Save this PDF to your desktop to avoid re-downloading it from the Internet
Tutorial 1. Software operation and basic concepts
Welcome!
Winsteps software installation and operation
Basic measurement and Rasch concepts
Simple dichotomous analysis
Constructing data files
This lesson includes a quick run-through of the operation of the computer program, Winsteps. If you run
into difficulties, please ask questions and get guidance on the Discussion Forum. No question is too
dumb! http://www.winsteps.com/forum
I am also including important aspects of theory and practice, so please go through the material
carefully. Think of yourself as the star of this movie. You are reading the script, and I am the director. I
can point you in the right direction, but it is up to you to win the Oscar!
Unfamiliar words? Rasch measurement has its own technical words. Please see the explanations at
http://www.winsteps.com/winman/glossary.htm
You can click on the blue links to access the Internet.

2.

A. Introductory Video

3.

Please view a few minutes of a movie introduction to

Rasch and Winsteps by yours truly ...
it is 50MB so may take a few minutes to download ...
http://www.winsteps.com/a/intro.wmv
This is a big file. Add it to your Media Player playlist.
Or you may need to use a free download manager such as
FDM Lite
This movie was originally requested by the energetic
Rasch folks in Hong Kong - a great idea!
The movie contains slides, separately viewable at
http://www.winsteps.com/a/slides.htm

4.
5.

B. Winsteps Software Installation

If you have the current version of Winsteps, then you
only need the Course data-files. Please download them
from:
http://www.winsteps.com/a/course-data.zip

6.

Please download and run WinstepsPasswordInstall.exe

7.

Save the installation file, WinstepsPasswordInstall.exe to

your desktop. Double-click on it if you need to re-install
this software. You can delete it when you have installed
Winsteps.

8.

To install Winsteps,
Run the downloaded file,
or double-click on the WinstepsPasswordInstall.exe on
your desktop.

9.

The installation screen for Winsteps displays. The

installation password is (copy-and-paste) in your purchase
receipt:
Then click on Next

10. Please install Winsteps in

C:\Winsteps
or the folder of your choice
Then click on Next

11. Click on Next until the Installation is finished. Click on

Finish to launch Winsteps for the first time.
If you run into difficulties during the installation, please
refer to http://www.winsteps.com/problems.htm

12. You can also launch Winsteps at any time by doubleclicking the short-cut on your desktop.
Or by dragging your control file onto the icon.
2

13.

C. Running the Winsteps Program

14. The main Winsteps window displays.

At the top of the screen is the Winsteps menu bar.
In a dialog box, Winsteps asks if we want to enter the
Setup procedure. Yes, we want to this time! So
Click on Control + Data Setup Procedure
15. The Winsteps Control File Set-Up screen displays. This
is one of the many ways of constructing your Winsteps
control and data files.
We will use a different way, but this is the easiest one to get
started for simple analyses.
Well look at one of the example data files. Please
click on Files

16. Click on Read control (and data) file

17. The directory listing of c:\Winsteps-time-limited\examples

displays. This contains the standard Winsteps examples.
They begin exam....
The first data file is Bond4.xls. It is an example from the
book by Bond & Fox Applying the Rasch Model. They
begin Bond&Fox. Ill call this book B&F.
Click on exam1.txt
If you only see exam1, but want to see exam1.txt
please tell Windows to display suffixes. See Error!
Reference source not found.. Appendix 1 toward the end of
this document.
Click on Open

18. The Winsteps Control File Set-Up screen displays, now

populated with the information from exam1.txt
The top part of the screen (inside the blue box) contains the
control instructions - you can see there are 11 control
variables displayed. We tell the Set-Up screen how many
rows and columns there are so that the input spreadsheet is
the correct size.
The lower part of the screen (inside the red box) contains
the person and item information and the responses.
At the bottom of the screen (light blue box) are additional
control variables. Winsteps has about 150 control
variables. So you will often have variables to enter here.
19. Exam1.txt is the Knox Cube Test discussed in Wright &
Stones book Best Test Design (1979), available from
http://www.rasch.org/btd.htm - It is a screening test
designed by Dr. Howard Knox on Ellis Island in New York
harbor. It assesses attention span and short-term memory.
Dr. Knox tapped out patterns on 4 blocks and the newly
arrived immigrants were signaled to repeat the tapping
patterns. There is more information at:
Hold down shift key and please click on
www.rasch.org/rmt/rmt133j.htm - to open the webpage in a
new window. www hyperlinks may not always work from
this PDF file. If necessary, copy-and-paste the link into
your Internet browser..
20. Lets see what Winsteps needs to know about your data.
First off is TITLE=. The Winsteps control variables are
identified as a word followed by =. Here the TITLE= is
KNOX CUBE TEST. This will appear on the top line of
each report output Table.
21. Lets personalize this a tiny bit.
Type your name into the Title box
22. Look down at the data spreadsheet.
The person identifying information is in columns 1 to 9.
These are highlighted with the blue stripe. In column 9 is
M or F. This is the person-gender indicator.
The item responses are in columns 11 to 28. There are 18
items they are indicated by the green stripe.
The observations are 1 for success and 0 for failure in
repeating Dr. Knoxs tapping pattern.
Observation = a response by a person to an item
4

23. To see the item label,

Position the mouse pointer on the green stripe line between
columns 11 and 12
Hold down the left mouse-button
Drag to the right
The label for the first item is 1-4. This means Tap on
the top of cube 1 and then tap cube 4.
24. To restore the look of the spreadsheet
Click on Refresh Data Display

25. The members of our person sample are called KID,

meaning child. The control variable name is PERSON=.
The location of the person identifying label in the data
spreadsheet is controlled by NAME1= and
NAMELEN(GTH)=.
Variable names can be abbreviated or lengthened and can
be upper or lower or mixed case: PERSON=, PERS=,
person=, Person=, etc.
26. Lets experiment,
change NAME1= from 1 to 10 and
change NAMELEN= from 9 to 15
click on Refresh Data Display
27. The blue person identifying stripe now overlaps the green
item response stripe. This is valid. It can be helpful to
include the responses in the person label, so that the
responses always appear when the persons are reported.
28. Reset to the original values:
change NAME1= from 10 to 1 and
change NAMELEN= from 15 to 9
click on Refresh Data Display
29. The control variables for the items are
ITEM= for the name give to the items
ITEM1= for the first item response column in the
spreadsheet
NI= for the number of items in the test or instrument.
30. To find out more about the control variables,
on the menu bar, click on Help
when the Help screen displays,
click on Contents

31. Winsteps Help for the Control and data file setup window
displays.
Lets find out about the control variable about which we
need more information.
In the left-hand panel, click on Control Variables

32. Scroll down the left-hand panel

Click on ITEM1=
The Help information for ITEM1= displays. The Help
information for each control variable includes a definition
and examples of its use.
Notice the Search tab in the left-hand panel. Questions and
suggestions from Winsteps users are constantly improving
the Winsteps Help. You will find that Help contains a
wealth of information, but you may need so search for it.
The information in the Help file is more current than in the
Winsteps User Manual which you will find in
c:\winsteps-time-limited\winsteps.pdf

33. Close Help:

Click on

34. Back on the Winsteps setup screen,

XWIDE= defines how many columns there are per
response. Use 1 column whenever possible as we do in
exam1.txt
CODES= contains the valid observations 0 and 1 for
this dataset. Any other values in the data, such as X
would be treated as missing data and ignored.
If you are not sure what codes are in your data file, then
click on Scan data for codes
Experiment by blanking out the CODES= value and
clicking on the green button.
35. Click on the green button Item Labels Enter/Edit

36. This Window lets you enter the item label information. 1-4
means tap onto cube 1 and then onto cube 4. As you look
down the list you can see that the tapping patterns become
longer and more complicated.
Click on Item Labels OK to close the Window.
37. Lets now do a Rasch analysis.
On the Setup menu bar, click on Winsteps
We changed the Title=, so
Click on Save control with data file and exit to Winsteps
Analysis
38. On the Save screen,
Type in a new file name. It can be whatever you like, Im
calling mine mikes1.txt
Click on Save
Notice that the suffix is .txt. All Winsteps control and
data files are standard text files with line-feeds. They
can be edited and created with Word, NotePad, WordPad
and many other programs.
39. The Winsteps Set-up procedure confirms that it saved your
control and data file, and asks if you want to analyze these:
Click on Yes

40. The Winsteps analysis screen displays again.

The analysis is now identified (arrowed) as being of
mikes1.txt (or your choice of file name).
And the Name of control file (red box) has been set to
c:\Winsteps-time-limited\examples\mikes1.txt.
41. Lets see what our Control and Data file looks like
internally ....
Click on the Winsteps Edit menu,
Click on Edit Control File= ...

42. The Control and Data file displays in a NotePad window. It

is a text file. You can change it with WordPad, Word, etc.
We need to display in Courier New, with a reasonably
small font. If NotePad does not do this, see 184. Appendix
5.
The top section (blue box) contains the control variables.
They are variable = value. You will recognize many of
them, for instance TITLE=. The control variables end with
&END - but do notice that there is no identification of the
number of persons or cases. In the Control file, you do
not need to tell Winsteps how many persons there are.
Winsteps discovers that for itself!
Then come the item labels (green box) with one item label
per line in the same order as the columns of item responses.
You will recognize that the first item is 1-4.
The item labels end with END LABELS (or END
NAMES)
Then comes the data (red box). Each line has a person
identification label, e.g., Anne, and a string of responses,
1100...
43. Oops! If you see that the item responses dont line up
neatly in columns, then please set the font to Courier New,
with a small font-size. For assistance, see 184. Appendix 5.

44. Thats enough for the moment, so close this NotePad

window.
45. Back to the Winsteps analysis again.
Winsteps is asking for the Report output file name.
Press Enter (blue arrow) for a temporary file name.
You can always Save As temporary files to make them
permanent, if you want to keep them.

46. Winsteps asks for Extra specifications. These are extra

control instructions. We dont have any, so
Press Enter

47. If you dont want to use the Enter key, then you can
click on File
click on Enter

48. The Winsteps analysis starts.

Green box: the control and data file is processed (green
box). The first data record is displayed. This is useful for
verifying that the data is correctly laid out.
^P = start of person label
^I = start of item responses
^N = end of item responses
The number of data records is reported. Here it is 35 KID
Records input. Wright & Stone called the children in their
sample, kids, and there were 35 of them. All is well.
Red box: Winsteps starts making its initial estimates (red
box).
49. The estimation process continues (red box). Each
iteration is one pass through the data file. After each
pass, the current set of Rasch measure estimates is
improved. This continues until the improvement has
become negligible. The estimation process has
converged.
When estimation ceases, the fit statistics are computed
(green boxes). The fit statistics report how well the data
correspond to the measure estimates.
The results of the analysis are summarized and reported in
the blue box. Usually we want to see much more
information than is reported here.
50. Of course, Winsteps outputs lots of numbers. Lets take a
quick look at a couple of Tables.
On the Winsteps menu bar,
click on Output Tables
click on 13. TAP: measure

51. Table 13 shows the tapping items in measure order with

the most difficult at the top. They are displayed in a
NotePad window. If the Table text is wrapped or ragged,
please see 184. Appendix 5.
If you prefer to use a different text editor, see 158.
Appendix 2 of this document.
The Rasch measures are in the red box, headed
MEASURE. The raw scores are in the blue box, headed
TOTAL SCORE.
Well talk more about measures in a few minutes.
52. On the Windows taskbar (usually on the bottom of your
screen), click on 2 Winsteps
Then click on Winsteps mikes1.txt on the Windows task
bar.
53. On the Winsteps menu bar,
click on Output Tables
click on 17. KID: measure

54. Winsteps Table 17, the person measure Table displays.

You can see it has the same format as the item measure
Table, Table 13. We know the difference between items
and persons, but Rasch analysis doesnt. In Rasch analysis,
persons and items are usually treated the same way.
Numbers are very useful, but difficult to visualize, so lets
look at a picture of these measures.
55. Click on Winsteps on the Windows task bar
56. If you see a Winsteps icon, click on it, and then on
mikes1.txt

57. Lets take a look at a picture, a map, of the measures we

have estimated.
On the Winsteps menu bar,
click on Output Tables
click on 1. Variable maps

10

58. Table 1.0 is a text file displayed in a NotePad window.

Winsteps has 34 Tables and most of these have sub-tables.
If the Table is ragged: in NotePad, Format,
Font, Courier New
The Table number is in the first line of the Table (red box).
A count of persons and items is shown on the second line.
If these numbers look wrong in your own analysis, then
investigate your control and data files carefully.
59. Ctrl+End takes you to the bottom of this Table.
Ctrl+Home back to the top
Here is sub-table 1.0
In the blue box on the left are the linear measures - like a
meter rule. Well talk about what these are very soon.
In the green box are the children located by their ability.
In the red box are the items, located by their difficulty.
When a child aligns with an item, then the child is
predicted to have a 50% chance of succeeding (or failing)
on the item. For instance, at +2, we predict child Dav to
have a 50% chance of succeeding on item 1-4-3-. The
labels Dav and 1-4-3- have been abbreviated so fit this
sub-Table. We could look at the data file to find out what
actually happened, but the purpose of measurement is to
make inferences (i.e., be useful) beyond the data. So, based
on these data, we predict that children like Dav- have a
50% chance of success on items like 1-4-3-.
We can already see improvements to make to this version
of the Knox Cube Test. Children in the yellow box near 0
have no items targeted at their ability level (mauve ???).
And, at +5 at the top, there are 4 very difficult items, but no
children. So this version of the Knox Cube Test needs
more middle difficulty items and fewer very hard items.
60.

11

61.

D. More Readable Output Maps

62. Lets look at the variable map again ... to find out
about the child-to-item targeting.
Click on Winsteps menu bar
Click on Output Tables
Click on 1. Variable maps
63. Here is Table 1.0
Oops! It is too big to conveniently look at on my
screen.
Lets squash it (reduce its size vertically) ....
64. Winsteps menu bar
Click on Specification
In the Dialog Box, click on Help

65. I want to find out how to squash Table 1, so

In the Help window
In the Contents panel
Click on Output Tables
Click on Table 1.0
The Help entry for Table 1.0 lists a number of
relevant control variables. You can click on them
to find out what they are about. Click on:
MAXPAGE=
66. MAXPAGE= (same thing as MAXPAG=, because
only the first few letters are needed) sets the
number of lines per page.
So lets set MAXPAGE=40
The default is MAXPAGE=0 (which means
Winsteps chooses the number), so we will need to
remember to reset MAXPAGE=0 back to that
value.
Close the Help window
67. In the Specification box,
MAXPAGE=40
Click OK
68. In your analysis window, the control variable is
reported back to you

12

69. Click on Winsteps menu bar

Click on Output Tables
Click on 1. Variable maps
70. The variable map (Wright map) is now squashed
...
Scroll down to Table 1.1
We can now see it clearly on our screen. The Xs
show the children (left-hand-side) and item (righthand-side) distributions.
Do you notice a problem with this test? There are
many kids at XXXX...., one for each X, but there
are few items close to them. We need items at the
???. It is an item-targeting problem.
For dichotomies: No nearby items Less
precision of measurement Larger person
Standard Errors
But for polytomies, this is more complicated,
because rating scales have a wider operational
range, equivalent to two or more dichotomies.
71. Lets reset MAXPAGE=0 - so that it has its
standard function.
Windows Menu bar
Click on Specification
MAXPAGE=0
Click on OK
You can see
MAXPAGE=0
on your Winsteps analysis window
72. Lets make the Winsteps Table more publication
quality.
Winsteps menu bar
Click on Specification
In the Dialog Box, click on Help
73. In Help, find ASCII=
There are three options:
ASCII = Yes - this produces the Tables we have
seen
ASCII = No - this produces the Tables in the old
MS-DOS format. It requires the Letter Gothic
Line font (installed with Winsteps)
ASCII = Webpage - this produces webpage
output.
Lets try the Webpage!
13

74. Winsteps Specification box, type

ASCII=Webpage
Click on OK
75. Again ...
Click on Table menu
Click on 12. TAP (item): map

76. Table 12.2 displays. It is a webpage displayed by

Internet Explorer (or your web browser).
Table 12 is like Table 1, but the item (TAP) labels
are shown in full on the right-hand-side. The Xs
on the left are the person abilities.
You can copy and paste this into Microsoft Word.
Then highlight the text there and change the font
size.
77. You can use the ASCII options whenever you
want to, but lets return to the standard one
Winsteps Specification box, type
ASCII=Yes
Click on OK

14

78.

E. Pathways & Bubble Charts

79. Bond & Fox have popularized pathway charts ...

Theres one on the front cover of their book. Here it is.
The circles are items. The squares are people.
The vertical axis is the measure, the location on the
latent variable.
The size of the circles and squares indicates the
precision (standard errors) of the measures along the
vertical axis, the latent variable.
The horizontal axis is the fit. Overfit on the left (the
responses are too predictable), underfit on the right
(the responses are too unpredictable). The fit indicates
the accuracy of the measures. We will discuss this in
Lesson 2.
This Figure was drawn by an artist for B&F. Well
have Excel do approximately the same thing.
80. To keep the pathway from being too crowded, lets
select a few people and items to display.
Winsteps Menu Bar:
Click on Specification
In the Specification box, type (or copy-and-paste from
here)
idelete=+4 +5 +10 +15
Click on OK and again
81. Your Winsteps Analysis window reports that
4 items are selected 4 items for reporting: 4, 5, 10, 15
82. In the Specification box, type (or copy-and-paste)
pdelete=+1-5
Click on OK

83. This has selected the first 5 children: 1, 2, 3, 4, 5

84. Winsteps Menu Bar
Click on Plots
Click on Bubble chart (Pathway)

15

85. Click on Persons and Items to check (or tick)

them.
Click on OK

86. Click on the Label button - we want to see the

person and item labels displayed

87. On the Winsteps Analysis screen it tells us that Bubble

Chart processing is underway
88. After a few moments, the Bubble Chart (pathway) plot
is displayed by Excel. There are many different
versions and options in Excel, so your plot may not
look exactly like this. You have full Excel editing
capabilities for this plot.
Most importantly, Excel has a mind of its own, and
has displayed the bubbles with the incorrect sizes. See
177. Appendix 4. Changing the size of the bubbles.

16

89. Practical Challenge: Heres a version of the pathway

chart after some adjustments using Excels own
capabilities. See if you can do bette.r
To adjust the labels in the bubbles, please install the
free Excel Add-in XY Chart Labeler from
http://www.appspro.com/Utilities/ChartLabeler.htm
A huge amount of meaning is squeezed into one
picture. The bubble chart shows the locations
(measures) of the persons and items on the latent
variable vertically. These indicate the predictive
validity and the construct validity of the instrument.
We talk about these in this Lesson.
It shows the behavior (fit) of the persons and items
horizontally. In Lesson 2, we talk about fit.
It shows the certainty (precision, standard error of
measurement) of the persons and items by the size of
the bubble. In Lesson 3 we talk about precision.

How about trying differently shaped bubbles?

In Excel 2007 and later,
www.winsteps.com/winman/bubblecharts.htm
section Autoshapes - also in Winsteps Help
90. Close the Excel window

17

91.

F. Practical Additive Measurement: Norman Campbell and the Social Scientists

92. There was a big argument back in the 1930s. The question being debated was Is psychological
measurement possible?
The hard scientists, the physicists, led by Norman Campbell said No, measurement requires a
deliberate action, a concatenation (such as putting sticks end-to-end to measure length, or piling bricks
one on top of another to measure weight), and you cant concatenate peoples heads!.
The soft scientists, the social scientists led by Stanley Stevens, could not answer that, so they devised
their own definition for a new kind of measurement: Measurement is the assignment of numbers to
objects or events according to rule (Stevens, S. S. 1946. On the theory of scales of measurement.
Science, 103, 677-680). The social scientists call measures whatever numbers they happen to have
acquired.
This has caused confusion ever since. For social scientists, any number is a measurement, provided you
can imagine some rule that it follows. For everyone else, measures have to conform to strict objective
criteria. But Georg Rasch solved the problem. He showed how the strict criteria of the physical
scientists can be applied to social science by means of Rasch models, which Georg Rasch called
Models for Measurement.
93. Lets go back to the debate with Norman Campbell. He laid out the philosophy underlying physics in
his book Physics: The Elements. Measurement is central to physics, and Norman Campbell, a
practical man, took an operational view of additive measurement. To paraphrase, Measurement
means that adding one more unit adds the same amount extra, no matter how much there is
already. Adding one more unit means the simple addition of 1 to the measurement number. But
adds the same amount is more complicated. It means defining an operation that increases the amount
we have by a certain, constant amount. With weight the constant amount is one gram or one pound, and
the operation is piling up. With length, the constant amount is one meter or one foot, and the
operation is putting end to end (technically concatenation). With temperature the constant amount
is one degree, and the operation is heating. Norman Campbell could not imagine how we could define
a unit of attitude and then construct a process that would increase an attitude by a known amount. Nor
could the social scientists of the 1930s, so they gave up .....
94. Campbell: "measurement requires a deliberate action, a concatenation". Measurement requires
deliberate construction.
Stevens:"Measurement is the assignment of numbers according to rule ..." The social scientists call
measures whatever numbers they happen to have acquired.
Example: 3 minutes on the clock is a Campbell measurement. The clock is deliberately constructed to
concatenate equal-units of time.
A score of 3 on a Math test is a Stevens "measurement". The Math test is an accidental collection of
items. There is no deliberate concatenation of equal-units of item difficulty.

18

95.

Georg Rasch (1900-1980)

96. Georg Rasch was a poorly-paid mathematics instructor in Denmark who consulted in statistics in order
to support himself. He didnt know that social-science measurement is impossible. Instead, he was faced
with the practical problems of his clients, problems that needed practical solutions. The Danish
Department of Defense had a particular problem with their educational tests. Rasch came up with a
practical solution based on log-odds transformations. It worked well, but the social scientists found it
too complicated (as many still do) and the mathematical statisticians found it too simplistic (as many
still do). So Georg Rasch made little progress with his measurement ideas until he was invited to
present a course of lectures at the University of Chicago in 1960. Only one person attended all the
lectures and that was Benjamin D. Wright, and even he had more pressing things to do for the next
several years. Then, in 1964, Ben had some data analysis problems of his own, and he thought that
perhaps Georg Rasch would have some ideas about how to solve them, and, besides, a trip to Denmark
would be fun. This got Ben interested again and ... but you can read the story at
www.rasch.org/rmt/rmt0.htm - and it was Ben Wright who spread the Rasch model around the world ...
97. So, what had Georg Rasch discovered, invented, constructed, that was able to overcome Norman
Campbells objections? Georg Rasch didnt put it this way himself, but we can now say that Georg
Rasch devised a way of concatenating heads in a psychological way that parallels Campbells way of
concatenating rods in a physical way. Rasch implements additive measurement: Adding one more
unit means the same amount extra, no matter how much there is already. If you want to read about
how this is done, and also obtain some other insights into what measurement really means, please
read my paper Measurement, Meaning and Morality www.rasch.org/memo71.pdf The linearity (= additivity) that the Rasch model constructs differs from the assumed linearity of
Classical Test Theory and much of Item Response Theory (IRT). For more about IRT, see
http://en.wikipedia.org/wiki/Item_response_theory
Read more about the asserted (rather than constructed) linearity of the widely-adopted interval
measurement classification of S.S. Stevens in www.rasch.org/rmt/rmt111n.htm "Stevens Revisited".
98. You may be wondering: I signed up to learn a statistical method, why are we wasting time with history
and philosophy? Let Lord Acton (1832-1902) answer: The knowledge of the past, the record of truths
revealed by experience, is eminently practical, is an instrument of action, and a power that goes to the
making of the future. Especially in the area of social-science measurement, the misconceptions of days
gone by continually dog our steps, and we find ourselves in disputes that should have been resolved
long ago. Indeed, you will undoubtedly encounter sceptics who will tell you the Rasch model is
wrong! or the Rasch model doesnt work! Make a note of www.rasch.org/rmt/rmt103e.htm The
Rasch Model Cannot Be Disproved as a resource for such occasions.
99.

19

100.

G. The Rasch Dichotomous Model

101. Georg Rasch presents several mathematical

models in his book Probabilistic Models for
Some Intelligence and Attainment Tests
(Chicago, 1980). His model for dichotomous
(right/wrong, yes/no, present/absent) responses by
persons to items has become known as the Rasch
model. This is our starting point, and is the model
that we used to analyze exam1.txt, the Knox Cube
Test. Georg Rasch wrote his model in
multiplicative form using Greek letters. Later
authors wrote it in exponential form also using
Greek letters. But both of these obscure the
essential additive process, so I prefer to write the
model in log-odds form using the more familiar
Latin (Roman) letters.

Georg Rasch writes his model:

Wikipedia writes his model:

102. The Rasch dichotomous model specifies the probability, P, that

person n of ability Bn succeeds on item i of difficulty Di.
Success means exhibits more of our intended latent
variable. Failure means exhibits less of our intended
variable. So we must score the observations in accordance with
this intention, no matter what values are assigned to the
observations during data collection.

P
loge ni
1 Pni

Bn Di

Log-odds of person n succeeding on

item i = Ability of person n Difficulty of item i

103. Loge is the natural logarithm: if you would like a quick refresher about logarithms, please study 161.
Appendix 3 of this document.
104. Probability: I think of this in frequentist terms. The probability of an event is the proportion of times
the event would happen if we could repeat the operation a great many times. A probability is always
between 0 (never happen) and 1 (always happen). P is the probability of success, and 1-P is the
probability of failure. Since either success or failure must always happen, when we add their
probabilities they must sum to 1:
(Probability it happens) + (Probability it doesnt happen) = (P) + (1-P) = 1
105.

Probabilities, Logarithms and the Rasch Model

106. To help with our understanding of Rasch models, lets be

explicit that success is a score of 1, and failure is a score of
0 on an item. Then the Rasch dichotomous model specifies
the probability, Pni1, that person n of ability Bn scores 1 on item i
of difficulty Di and Pni0 is the probability of scoring 0.
Ability is what we are looking for in the people. In your
situation, you can use other words Proclivity, Motivation,
Health status, Disability.
Difficulty is what we identify in the items. In your situation,
you can use other words, Challenge, Impediment

20

loge(Pni1 / Pni0 ) = Bn - Di
Pni1 + Pni0 = 1

107. An essential concept: qualitatively-ordered data:

"qualitative ordering" is "good, better, best"
Success means more of what we are looking for
Failure means less of what we are looking for
The difference between Success and Failure is qualitative.
The qualitative ordering of these different qualities is shown by
scoring them 1 and 0.
1 means indicating more of the latent variable.
0 means indicating less of the latent variable.

Standard Rasch item-scoring:

Success = 1 = more of what we seek
Failure = 0 = less of what we seek
what we seek is the latent
variable

Rasch models analyze qualitatively-ordered data. These data

can be found everywhere.
108. Bn and Di are distances in logits along the latent variable relative to the local origin
A latent variable is something which we can have more or less of, but which we cannot measure
directly. It is a variable such as mathematics ability or patient quality of life. We conceptualize it to
be a straight line marked out in equal-interval units. This line is infinitely long. We can always imagine
something (or someone) with more of the attribute than anything (or anyone) we have encountered so
far, and also something (or someone) with less of the attribute. We conceptualize each observation in
the data to indicate less or more of this latent variable.
109. Logits are log-odds units. Look at the Rasch equations: loge(Pni/(1-Pni)) = Bn - Di
On the left is loge, this means the natural logarithm. If you dont know about logarithms, please
look at 161. Appendix 3 to this document.
The logarithm is of Pni1 divided by Pni0 which is the ratio of two probabilities. A ratio of probabilities is
called the odds. So, on the left-side of the equation we have log(odds). These provide the units for the
right-side of the equation, so Bn and Di are measured in log-odds units, logits - pronounced low-jits.
In Table 1.0 above, the logit values have been linearly rescaled (multiplied by a constant and then had
another constant added) to produce numbers that are easier for most people to think with. This is like
going from Celsius to Fahrenheit on a temperature scale. The meaning hasnt changed only the
numbering.
The local origin is the place we are measuring from. In physical measurement: for length, it is one
end of a tape measure. For mountains, 0 is at sea level. For temperature, it is 0 on a thermometer. In
Rasch measurement, 0 is usually in the center of the range of item difficulties, but it can be wherever we
prefer it to be - as long as we make it clear to ourselves and everyone else where that 0 point is.
110. What if the probability of success is the same as the
probability of failure? Then both probabilities are 0.5.
The odds of success are 0.5/0.5 = 1, and the logarithm
of the odds is log(1) = zero. So the ability and the
difficulty are the same. Exactly what we expect! When
I encounter an item of exactly the same difficulty as
my ability, I cant predict whether Im going to
succeed of fail. My prediction would be like tossing a
coin ... Heads or Tails?

21

Pni1 + Pni0 = 1
Pni1 = Pni0 = 0.5
loge(Pni1 / Pni0 ) = loge( 0.5 / 0.5 ) =
loge(1) = 0 = Bn - Di
Bn = Di

111. If Im sure to succeed, then my probability of success

is 1 and my probability of failure is zero. My logit
ability will be plus-infinity relative to any item of
finite difficulty.

Pni1 + Pni0 = 1
Pni1 = 1, Pni0 = 0
loge(Pni1 / Pni0 ) = loge( 1 / 0 ) =
loge() = = Bn - Di
Bn =

112. If Im sure to fail, then my probability of success is 0

and my probability of failure is one. My logit ability
will be minus-infinity relative to any item of finite
difficulty.

Pni1 + Pni0 = 1
Pni1 = 0, Pni0 = 1
loge(Pni1 / Pni0 ) = loge( 0 / 1 ) =
loge(0) = - = Bn - Di
Bn = -

113. Suppose that I generally succeed 3 times out of 4 in

hitting a target with an arrow. Then my probability of
success is P = 3/4 = 0.75. And my probability of
failure is 1-P = 1/4 = 0.25. Then my odds of success is
P / (1-P) = 0.75 / 0.25 = 3.
And my ability is loge(3) = 1.1 logits more than the
target is difficult.
114. Extending this idea, we can draw a picture of the
relationship between ability (relative to an item) and
probability-of-success. the relationship is called a
logistic ogive.
You can see on the plot that a zero logit difference
(dotted line) corresponds to a 0.5 probability of
success.
An ability advantage of 1.1 logits (arrowed) is
equivalent to a probability of success of 0.75.
This curve is monotonic ascending - the probability
of success always increases with increasing measure
difference, monotonously.
115. Table of logits to probability
Notice that the range 0-100% approximates 10 logits.

116. Lets see this curve for exam1.txt.

On the Winsteps menu, click on Graphs
Click on Category Probability Curves

22

Pni1 = 0.75
Pni0 = 0.25
Pni1 / Pni0 = 3
loge(Pni1 / Pni0 ) = loge(3) = 1.1
1.1 = Bn - Di = 1.1

117. The probability curves for categories 0 and 1 display.

These are for item 4. 1-3-4 but they look the same
for all dichotomous (right-wrong) items.
The x-axis is in logits relative to the difficulty of item
4. It has an infinite range from - to +
The y-axis is the probability. This always has the range
0 to 1.
The red arrows show that 1.1 logits above the item
difficulty corresponds to 75% chance of success on the
item.
118. Theres much, much more to see ... but lets make it
easier for you to set up your own data for Winsteps.
So close all windows
119. You may see the WINSTEPS closing dialog. This is
because you are closing the Winsteps analysis window
while Tables are displaying.
Click Yes because we want to close everything.
120.

23

121.

H. Winsteps Control and Data Setup with Excel, R, SAS, SPSS and STATA

122. It really simplifies Winsteps control+data-file set-up if

you already have the response-level data on your disk
in a convenient form.
Lets start with an Excel spreadsheet.
Here is the layout:
One row for each person.
One column for each person identifying variable.
One column per item.
The first row has the identifying labels
Heres an Excel spreadsheet for B&F Chapter 4.
It has one person identifying column A, and 35 items
columns, B - AJ
123. Double-click on the Winsteps icon on your desktop

124. We dont need to see the Welcome dialog every

time,
Click on Dont ask again
If we need the Data Setup screen, we can launch it
from Data Setup (green ellipse)
Now click No
125. On the Winsteps menu,
Click on EXCEL/RSSST
Click on EXCEL
R statistics, SAS, SPSS, STATA: Your procedure is the
same as Excel. Your variable names are not in row 1,
but the usual ones.
126. When the Excel dialog box displays,
Click on Select Excel file
In the file dialog box,
Click on Bond4.xls
Click on Open
There are also datasets in the other formats for you to
experiment with.

24

127. After a few moments, the labels in the first row of the
Excel spreadsheet display. Scroll up and down to see
them. They are shown here in the red box.
Sorry, if you dont have Excel on your computer, this
wont happen!
Look down the list of column labels. Some are person
variables. For instance, column A. Some are item
labels. We need to tell Winsteps which are which.
We do this by copying-and-pasting the column
information into the correct place in this control
dialog. The person variable columns go under Person
Label Variables (blue arrow) and the item columns go
under Item Response Variables (green arrow)
Do not change anything else
128. OK - so go ahead and copy-and-paste.
This is what happens when I do it ....
Item 1 is in column B of the Excel spreadsheet. It is
Negation. I pasted in Excel B - K as the Item
Response Variables.
Column A of the Excel spreadsheet in the person
number. I pasted it in as the Person Label Variable.
You dont have to use all the Excel columns. You can
change their order and you can also use the same
column in both the person and item variable list.
129. To construct the Winsteps control and data file:
Click on Construct Winsteps File
Enter the name of a new output file to be the
Winsteps control and data file. My name is
bond4control.txt
Click on Save
The Excel interface starts processing. This can be
slow, because Winsteps is communicating with Excel
through Windows.

25

130. Then the Winsteps control and data file displays. It is a

text file in the familiar format. You can edit it if you
want.

131. Close the bond4control.txt window

132. Lets analyze the file we have just created ....

On the Excel Processing for Winsteps
Click on Launch Winsteps

133. The usual Winsteps analysis screen displays with the

control file already active.
We dont want to perform this analysis right now so ...

134. Close all windows

135. You can experiment doing the same procedure for

SPSS .sav files, SAS .sas7bdat files and STATA
.dta files and R .rda or .rdata files using datasets
in the Examples folder. SPSS, STATA and R should
work for everyone.

136. SAS: If this fails, you probably need the SAS Local Data Provider free (with free registration) from
http://www.sas.com/apps/demosdownloads/92_SDL_sysdep.jsp?packageID=000608
137.

26

138.

I. More to think about ....

139. It is now time to hear from that long-time advocate of

Rasch measurement, Ben Wright. Unfortunately he
was incapacitated by a stroke in October 2001 and is
struggling to do any productive work.
Read the document listed here that matches your
interests, but everyone glance at the first one - it has a
typical picture of Ben in action proclaiming useful
measuring.
140. In the literature you will find that the Rasch
Dichotomous model is sometimes called the 1-PL
IRT (One-Parameter Item Response Theory)
model. This is because these models have a
superficial mathematical similarity, but they also have
a fundamental conceptual difference: the Rasch model
deliberately implements the properties of additive
measurement. The data must conform to the Rasch
model if they are to support additivity. The 1-PL IRT
model of Frederic Lord is a hoped-for description of
a set of data. If the data dont fit it, then another model
is tried. http://www.rasch.org/rmt/rmt193h.htm
See also http://www.rasch.org/rmt/rmt61a.htm
141. Thats the end of the Lesson. Well done!

27

Fundamental Measurement in Social

Science and Education
http://www.rasch.org/memo33a.htm
Fundamental Measurement for
Psychology
http://www.rasch.org/memo64.htm
Fundamental Measurement for Outcome
Evaluation
http://www.rasch.org/memo66.htm

142.

J. Supplemental Readings on Rasch Measurement

143. In B&F, 2nd Edn, Chapters 1 and 2 correspond to what

we have looked at so far.
All questions are based on the Lessons or free
downloaded documents. The books are for those who
wish to do additional study on their own.
144. In B&F, 1st Edn, Chapters 1 and 2 and sections of the
Preface correspond to what we have looked at so far.

145. In Best Test Design (Wright & Stone), chapter 1 is the

place, but it is much more mathematical than our
discussion here.

146. Free PDF:

Wu, M. & Adams, R. (2007). Applying the Rasch
model to psycho-social measurement: A practical
approach. Educational Measurement Solutions,
Melbourne.
www.edmeasurement.com.au/Learning.html
147. Free PDF:
Wright, B.D. & Stone M.H. (1999) Measurement
Essentials.
www.rasch.org/memos.htm - measess

28

148. Appendix 1. Displaying file name suffixes

149. On your desktop, click on My Computer or any
folder
150. XP: Click on Tools

151. XP: Click on Folder Options

152. Vista:
1. Click the Layout button (to the left of the
Views button)
2. Click Folder Options

153. Click on View

154. Un-check Hide extensions for known file

types

155. Click on OK

156. Close unwanted windows

29

157. Suffixes will now display in folder file lists

30

158. Appendix 2. Changing the text-editor used by Winsteps

159. In the Winsteps main window,
Click on Edit
Then
Click on Edit initial settings

160. In the top box, type in the path to your preferred text
editor, or use the Browse button to locate it.
Ive entered the path to Wordpad.
Then click OK to save this setting.
For NotePad please blank out the Editor box:

31

161. Appendix 3. What are Logarithms?

PPT: http://teachers.henrico.k12.va.us/math/HCPSAlgebra2/Documents/10-2/2006_10_2.ppt
Video: http://www.youtube.com/watch?v=mQTWzLpCcW0
100 * 1000 = 102 * 103 = 105 = 100000
this is the same as:
log(100) + log(1000) = log(100000)

Much of what we do during this Course is based

on logarithms. One reason is because logarithms
transform multiplication into addition:
162. Lets start with squares, 2 x 2 = 4
and cubes, 2 x 2 x 2 = 8
We can see that it takes two 2s to make 4 so lets
write them 22 and it takes three 2s to make 8 so
lets write them 23. Then what about 22+3 = 25?

22 = 2 x 2 = 4
23 = 2 x 2 x 2 = 8
2+3
5
2 = 2 = 2 x 2 x 2 x 2 x 2 = 32 = 4 x 8 = 22 x 23

163. The superscripts 2 and 3 are called powers, and

weve discovered that multiplying numbers is the
same as adding powers. We can do this with nonintegers:

1.4142 x 1.4142 = 2 = 21 = 21/2 * 21/2

1.4142 = 21/2 = square-root(2)

164. And powers work for negative numbers

21 = 2 = 4 / 2 = 22 / 21 = 22 * 2-1 = 22-1
2x+y = 2x * 2y

165. So, we have the general rule of powers of 2:

166. This works exactly the same way if we change the
base value from 2 to 10.

X = 10x
Y = 10y
X * Y = 10x+y
10x+y = 10x * 10y

167. Now x and y are called exponents and 10 the

base. So this is an exponential form.

X * Y = 10x * 10y = 10x+y

log10(X) = log10(10x) = x
log10(Y) = log10(10y) = y
log10(X*Y) = log10(10x+y) = x+y

168. We can rewrite this exponential form into

logarithmic form. Log is short for
Logarithm.
169. This is very useful. We can transform a
multiplication X*Y into and addition log(X) +
log(Y). This saves a huge amount of effort when
the multiplication is done by hand, and was why
logarithms were invented around 1617.

log10(10x) + log10(10y) = x+y = log10(10x+y)

log10(X) + log10(Y) = x+y = log10(X*Y)

170. In Rasch work, we use a special base, called e.

This has the value 2.718.... And the logarithms are
then called natural or Napierian logarithms.

2.3026 * log10(X) = loge(X)

log10(X) = 0.4343 * loge(X)

171. Once we alert the reader what base we are using,

e from here on, we can omit it.

log(X) + log(Y) = log(X*Y)

172. Here are some important facts about logarithms:

the logarithm of 1 is zero

log(1) + log(1) = log(1*1) = log(1)

so log(1) = 0
this is the same as saying x0 = 1
32

173. the logarithm of plus infinity is infinity

log() =

174. the logarithm of zero is minus infinity

log(0) = -

175. the logarithm to the base e of e is 1

loge(e) = 1

176. the logarithms of negative numbers dont exist

log (-1) = Oops!

33

177. Appendix 4. Changing the size of Bubbles in Excel Bubble Charts (Pathway maps)
178. This section is tricky ...
The Excel plot displays. The bubbles are located
vertically by measure and horizontally by fit.
Their diameters are supposed to be twice their
Standard Errors of measurement. But Excel has
chosen its own bubble-sizing.
179. Click on the Worksheet Tab below the plot. The
standard errors are in Column D for the persons
and Column I for the items.
Look at the biggest bubble, item 1-3-2-4-1-3. Its
Model S.E. is 1.07, so that the bubble diameter
= 2*S.E. = should be 2.14.

180. Click on the Bubble Chart Tab.

Look at the biggest bubble. Its diameter is
currently about 3.1 on the vertical axis on my
screen. That may be different on yours.
The bubbles is too large, 3.1 on the vertical axis
instead of 2.1. This is because Excel plots relative,
not absolute, sizes.
Let's use the Excel functions to correct the plot.
If you are familiar with Excel, turning on
Gridlines helps with this.
181. Right-click on the bubble.
Click on "Format Data Series" (not "Format Data
Labels" or "Format Data Points").
If "Format Data Series" does not display, move
the mouse-pointer a little lower down in the
bubble and right-click again.

34

182. Click on "Options" tab

"Scale bubble size to:" Type in a value to reduce
the bubble size, such as "65"
Click on "OK"
All the bubbles change size.
Excel rescales all bubbles by the same amount
both when it originally displays the bubbles and
when you rescale them.
183. And we see a much better looking pathway. The
diameter of the biggest bubble is about 2.2 logits
(vertically).
You may need to go back-and-forth several times
adjusting the bubble size to get it to look right.
This is now your plot. You can beautify the plot
using other Excel functions for coloration, layout,
etc.
See Bond & Fox Chapter 3 for a further
explanation of this pathway.

35

184. Appendix 5. Changing the text appearance in NotePad

185. Oops! A Table may display too big or ragged or
wrapped. We need to display this text in a fixedspace font, such as Courier New, and also a
smaller font so everything fits in the window.
186. On the NotePad menu bar,
Alt+O or click on Format pull-down menu
Alt+F to change the Font
Font: Courier New
Font style: Regular
Font size: 7 (You can type in values not listed)
Click on OK
187. The Table now displays neatly
Alter the Font size if the Table is too big or too
small.
188. To make permanent (default) changes in NotePad font face and/or size:
189. Windows "Start"

Click on "Run"
Type in "regedit"
Click on OK

190. Registry Editor:

Click on the + in front of "HKEY_CURRENT_USER

Click on the + in front of Software
Click on the + in front of Microsoft
Click on Notepad"
For the type face:
Double-click on "IfFaceName"
Type in Courier New (or Letter Gothic Line)
Click on OK

191. For the font size:

Double-click on "iPointSize"
Click on "Decimal"
Type in 80 (for point-size 8 multiplied by 10)
Click on OK

Close registry
Click on top right

36

192. Appendix 6. Print-Screen, then email - in case of problems

193. If you have a problem, and need to send me a
picture of what is on your screen.
Press the PrintScreen key on your keyboard
This copies a picture of your screen on to the
Windows clipboard
194. Open a new Word document

195. Then Paste (Ctrl+v) the clipboard into the

document

196. Save the document

Attach it to an email to me
Be sure to tell me the exact section # in the
Tutorial where the problem happens.

37