Rasch Model to Measure Change

Applying the Rasch Model to Measure Change in Student Performance over Time

Jessica D. Cunningham; Western Carolina University

Kelly D. Bradley1; University of Kentucky

1
Please use Kelly D. Bradley as author of contact. 131 Taylor Education Building, Lexington, KY 40506-
0001, [email protected] or (859)257-4923

1
 Rasch Model to Measure Change

Applying the Rasch Model to Measure Change in Student Performance over Time

Abstract

Newton’s Universe (NU), an innovative teacher training program, strives to obtain

measures from rural, middle school science teachers and their students to determine the impact

of its distance learning course on understanding of temperature. Although limited research exists

examining the usefulness of the Rasch model in measuring change over time, this study applied

the method of racking and stacking the NU data to examine change in student performance over

time. Locally, researchers involved in the NU course development can use these findings to

further improve the course and the student assessment instrument. Researchers across various

fields can use the methods shown in this study to improve assessment instruments, course

development, and measures of change over time.

DESCRIPTORS: Rasch Model, Racking and Stacking, Measuring Change in Student

Performance, Newton’s Universe teacher training

2
 Rasch Model to Measure Change

Applying the Rasch Model to Measure Change in Student Performance over Time

Challenges regarding student achievement in math and science education remain an issue

for rural educators in general and particularly in the rural regions of Appalachia. Newton’s

Universe (NU)2 is an inquiry-based teacher training program striving to obtain measures from

rural, middle school science teachers and their students to determine the impact of its distance

learning course on understanding of temperature. The goals were to enhance the conceptual base

of science educators in state and national content standards in physical science, introducing and

developing inquiry strategies, integrating technology into teaching and learning, and ultimately

improving student learning. Teacher knowledge is instrumental in impacting student

achievement (Darling-Hammond, 1999; NCLB, 2002; Rowan, Correnti, & Miller, 2002), but

rarely discussed as impacts of teacher training, are measures of change in student understanding.

This is likely due to the many potential issues associated with measuring change over time in

psychological constructs such as isolating intervention effects. Although typical for a

professional development program to administer an assessment to measure its impact on

teachers, essential to the evaluation of professional development in education is evidence of

student learning associated with an intervention (Guskey, 2000), such as student baseline and

post instruction measures.

NU provided participating teachers a distance learning course during the summers of

2006 through 2008 with an emphasis on inquiry-based learning of temperature topics. In addition

to obtaining teacher baseline and post instruction measures from control and experimental

groups, teachers administered baseline and post instruction assessments to their middle school

science students. Measures for teacher (Bradley & Sampson, 2006) and student understanding of
2
Newton’s Universe funded by the National Science Foundation Grant No. 0437768 (J. Osborn, P.I.). Further
information can be found at the project website: http://www.uky.edu/NewtonsUniverse.

3
 Rasch Model to Measure Change

topics covered in the course were taken using a quality assessment developed by the NU research

team. This study illustrates the process of applying the Rasch model to measure change in

student understanding of temperature over the time and ways these results can be used to inform

course developers as well as teachers participating in the project.

Purpose of Study

Limited research exists featuring the Rasch model and its usefulness in depicting change

in measures over time. The purpose of this study is to apply the method of racking and stacking

the NU student data within the Rasch model to address the following research questions:

1. Are items on the baseline and post instruction science teacher assessments functioning as

expected within the racked analysis? In other words, are post instruction item estimates less

than corresponding baseline item estimates?

2. How well are the items distributed along the continuum of the “understanding of

temperature” variable in the racked analysis? How does this distribution change from

baseline to post instruction assessment?

3. Are persons functioning as expected from baseline to post instruction assessment? In other

words, are student estimates increasing from baseline to post instruction assessment?

4. What are the plausible implications of these results for NU and others?

The NU student data were examined under the assumptions of the Rasch model, specifically

assessed for unidimensionality. In the racked analysis, the purpose was to examine the change in

items over time while the focus of the stacked analysis was to examine change in student

performance over time (Wright, 2003). Items and persons were placed along a continuum where

examinees are measured based on their performance in comparison to where the items are

measured based on their difficulty. A wider spread of items allows for the measurement of a

4
 Rasch Model to Measure Change

larger range of person abilities. Investigating item difficulties on the same ruler within the racked

analysis illustrates change in items over time. The expectation would be that item estimates

decreased over time. In other words, item difficulties should be higher for the baseline

assessment compared to the post assessment. Placing baseline and post assessment student

performance estimates along the same continuum allows the examination of whether each

student measure increased from baseline to post assessment. Transforming measures onto the

same ruler enables meaningful implications of these results for NU and others.

Literature Review

The launching of Sputnik I by the Soviet Union in 1957 spawned the appropriation of

federal funds to place an emphasis on science education, deemed imperative for our national

security. Building on the foundation of inquiry-based learning (Hunkins, 1972), NU is a distance

learning program focused on the inquiry-based learning of physical science for middle school

teachers with an interest in discovering its impact on middle school student learning. While the

impacts on teachers are typically measured, it is a rare opportunity to also measure transferability

of teacher knowledge to students as a result of an intervention. Equally important to developing

assessments to measure growth in understanding is responsibly reporting the results to accurately

reflect the measures in a meaningful way for not only the course developers but also its

participants.

Although more than one method of analysis exists to determine growth in understanding

over time, item response theory (IRT) has noteworthy additions to the traditional classical test

theory (CTT) approach (Embretson & Reise, 2000). This study features the application of the

Rasch model equivalent to the one parameter, specifically item difficulty, IRT model. The

following provides a review of literature to build a foundation for applying the Rasch model to

5
 Rasch Model to Measure Change

measure change in understanding over time. The review begins with outlining important

additions of the Rasch model compared to CTT followed by ways to determine fit of the data to

the Rasch model. After theoretical and empirical fit of the data to the model have been

established, a conceptual foundation for racking and stacking the data is presented.

From a theoretical standpoint, the Rasch model provides important additions to the

measurement of change to CTT. The Rasch model consists of "model-based measurement in

which trait level estimates depend on both the persons' responses and on the properties of the

items that were administered" (Embretson & Reise, 2000, p. 13). The Rasch model theorizes an

additive linear structure underlies the observed data, that both participants and items can be

arrayed on a continuum, and that the items are equivalent in item discrimination (Bond & Fox,

2001). Rasch measurement produces a standard error (SE) for each person and item, specifying

the range within which each person’s ‘true’ performance and each item’s ‘true’ difficulty fall

when a person or item fits the model and the corresponding SE is small. Individual errors can

then be used to produce a more accurate average error variance for the sample (Wright &

Masters, 1982). The property of invariance – the expectation that the measures remain the same

regardless of which items are used as well as independent of the sample used – is an essential

component of Rasch measurement allowing interpretations over useful range of time and place

(Wright & Masters, 1982).

Prefacing any analysis, the data must uphold the required assumptions of the model. In

this case, the underlying assumption of the Rasch model is unidimensionality, referring to

measuring only one attribute at a time (Bond & Fox, 2001). Theoretically, the NU student

assessment was designed to measure a unidimensional construct – specifically understanding of

temperature. Although content experts argue that unidimensionality holds from a theoretical

6
 Rasch Model to Measure Change

standpoint, fit statistics along with a principal component analysis can determine if the

assumption indeed holds empirically (Bond & Fox, 2001). Wright and Stone (1994b)

recommend the guideline of 0.7-1.3 as an acceptable range for infit and outfit mean square

statistics for a multiple choice assessment. In other words, fit statistics outside a literature-based

established cutoff indicate a violation of expected response patterns central to Rasch

measurement (Smith, 2004): that a person with a greater ability has a higher probability of

correctly answering any item and that an item of lesser difficulty has a greater probability of

being answered correctly (Wright and Masters, 1982). Assuming the data fit the model,

interpretations of results may proceed.

The assessment was examined using the Rasch model similar to Bradley and Sampson

(2006), and appropriate changes were made to improve the quality of the instrument. Therefore,

this study used the restructured assessment and focused on applying the Rasch model to examine

change in measures over time. Wright (2003) posed a method of racking and stacking the data

within the Rasch model to determine change in student performance (stacking) as well as change

in item difficulties (racking) over time. Although the usefulness of racking and stacking the data

has been posited by Wright (2003), limited research exists applying the Rasch model for this

purpose.

In order to employ the Rasch model to depict change in measures over time, the data

must be restructured according to Wright (2003) by racking and stacking the data. Stacking the

data is done by appending the person measures for the post instruction onto the baseline

measures resulting in twice as many persons being measured (see Figure 1).

7
 Rasch Model to Measure Change

Figure 1. Conceptual visualization of stacking and racking the data

STACK DATA RACK DATA

Items Baseline Post-

Persons Items Instruction
Baseline Persons Items

Post Instruction Persons

Conceptually, this transforms both the baseline and post instruction assessment measures

into measures on the same ruler to make interpretations about the effect of the instruction on

student understanding. The assumption is that the students have changed over time as a result of

instruction prior to the post assessment. Similarly, racking the data can be done by attaching the

post instruction item responses onto the baseline item responses resulting in each person taking

twice as many items on the assessment (see Figure 1). This conceptual approach illustrates item

difficulty estimates from baseline and post instruction assessment on the same ruler to make

conclusions about the effect of the intervention on the change in items. Racking the data allows

the inspection of change in the items rather than students by the end of instruction. In summary,

"stacking the data, we see who has changed" while "racking the data, we see what has changed"

(Wright, 2003, p. 906). This study applies the Rasch model to the restructured NU data to

illustrate its appropriateness in examining change in measures over time. Followed by assessing

the data for unidimensionality, results are presented and interpretations are discussed to provide

information to NU and its participants.

8
 Rasch Model to Measure Change

Methodology

Although analysis of variance is often the method of choice in educational research to

depict change over time, others exist with accurate results and potentially more meaningful

interpretations for interested parties. The importance of choosing accurate and meaningful

analyses requires attention in order to evaluate impacts of interventions such as a teacher training

program. The purpose of this study was to illustrate the usefulness of the Rasch model in

depicting change in measures over time and implications of these results.

Sample

The selection of teachers is first detailed because the student sample is dependent on the

participation of their teachers in the NU program. The target population for the NU project was

middle school science teachers from the sampling frame of all public high schools in rural

regions of Appalachia, excluding teachers who had participated in any Appalachian Mathematics

and Science Partnership (AMSP) training workshops of physical science. As a participant,

teachers were asked to complete surveys and assessments on a voluntary and confidential basis

in exchange for stipends and classroom materials. In an effort to isolate the NU course impacts,

control teachers were selected under the same criteria as participants and offered stipends with

the opportunity to enter the project as an experimental teacher the following summer. Descriptive

statistics of demographic data for the teachers have been produced (see Table 20).

9
 Rasch Model to Measure Change

Table 20

Descriptive statistics for teacher demographics

Number of Teachers
Demographic Variable (n = 47)

Grade Level Currently Teaching

6th only 12
7th only 8
8th only 11
6th and 7th 10
7th and 8th 1
6th, 7th, and 8th 1
Other 3

Subjects Teaching
Science only 31
Science and other 15
Other 1

Highest Level of Formal Education Completed

B.A. 4
B.S. 10
Master’s 21
Ed.D./Ph.D. 2
Other 9

Science Certifications Held

Physics 2
Chemistry 5
Physical science 3
Biology/life science 16
Earth/space science 4
General science 17
No certification 11
Other

Years Teaching Science

0-2 11
3-5 10
6-10 10
11-15 6
16-20 8
Over 20 2

10
 Rasch Model to Measure Change

The student sample was a relatively homogeneous group with similar backgrounds from

rural Appalachia. These students are in middle school science classes (6th, 7th, and 8th grades) that

are taught temperature topics by control and experimental NU teachers. Each student was

administered a subset of an assessment based on the temperature topics taught by their teacher.

The number of students administered either the entire assessment or a subset of the assessment

has been summarized (see Table 1).

Table 1

Number of teachers by cohort across years

2006 2007 2008

Cohort 1:

Teachers 36 23 17

Students ABCDE/All ABCDE/All

Base 1072/1642 573/1325
Post 1091/1495 531/919

Cohort 2:

Teachers 5 8 8

Students ABCDE/All ABCDE/All

Base 295/762 311/667
Post 267/634 398/567

Cohort 3:

Teachers 0 17 13

Students ABCDE/All ABCDE/All

Base 0 499/1248
Post 0 310/912

However, only students assessed both baseline and post instruction were used in this study.

11
 Rasch Model to Measure Change

Data Source

The research team piloted the 39 multiple-choice items and employed a Rasch analysis,

as Bradley and Sampson (2006), to revise and improve the quality of the student assessment

prior to collection. The temperature construct can be partitioned into five subtopics: foundations,

properties of matter, energy transfer, phase change and thermal energy. Students were

administered only subtopics of temperature concepts for which their teachers reported having

taught. Administration of the student assessments occurred at the beginning of the academic

semester prior to any instruction. The teachers were asked to administer the same assessment

within at least two weeks after the temperature unit was covered to serve as a post instruction

measure.

Data Collection Procedure

The experimental design for this project influences the timing of the data because the

teachers were partitioned into three cohorts. The first cohort entered the project in the summer of

2006 and immediately had the opportunity to participate in the first NU course covering topics of

temperature. The second cohort volunteered to serve as controls first and participate in the course

the following year in the summer of 2007. The final cohort of teachers entered as controls during

the summer of 2007 followed by the first NU course in the summer of 2008 (see Figure 2).

12
 Rasch Model to Measure Change

Figure 2. Experimental Design for Newton’s Universe

Teachers were given detailed instructions to assign student identification numbers –

comprised of the unique teacher identification number, class period number, and unique student

number within the class. Teachers kept record of these identification numbers so that all students

remain anonymous throughout the research project. Detailed instructions were also given to

teachers regarding to administration of the assessment to students (see Appendices).

Administration of the student assessments occurred at the beginning of the academic year prior

to any instruction in August or January depending on traditional or block scheduling. Teachers

were asked to administer the same assessment using the same student identification numbers

within at least two weeks after the temperature unit was covered.

Prior to scanning the data for electronic storage in an Excel file, the data were checked to

ensure its quality. Answer sheet were checked for stray marks and to make sure the answers were

bubbled in correctly for scanning. Since teachers had the option of giving a subset of the

assessment based on what topics were covered during instruction, two independent checks were

in place to guarantee the appropriate key sheet was included for grading the assessment. Students

13
 Rasch Model to Measure Change

without meaningful identification numbers, students not completing more than 20% of the

assessment, and students with answers bubbled in past the number on the assessment were not

deemed as quality data and are not kept in the data set for analysis.

Data Analysis

The data were collected from the NU baseline and post instruction multiple-choice

assessment designed to measure middle school science student understanding of temperature.

The dichotomous Rasch (1960) model was applied to racked and stacked data matrix, which is

 Pni1 
represented with Loge   Bn  Di . Here, Pni1 and Pni 0 are the probability that person n
 Pni 0 

encountering item i is observed in category 1 or 0, Bn is the ability measure of person n, and Di is

the difficulty measure of item i. WINSTEPS software, version 3.65 (Linacre, 2006) was used for

the analysis.

Fit statistics were examined for the racked and stacked analyses to determine fit of the

data to the model. Wright and Linacre (1994) recommend an acceptable range of 0.7-1.3 for infit

and outfit mean square statistics for multiple-choice assessments. Average person measures and

spread of items along the continuum were examined within the racked analysis to determine how

the assessment targeted the ability of the examinees. Variable maps for the racked data were

examined to determine changes in item difficulties from baseline to post assessment. Student

measures for the stacked analysis were examined to determine changes in performance from

baseline to post assessment.

Results and Discussion

Experimental and control teachers administered a student assessment to all classes being

taught temperature concepts. This analysis includes only students that were assessed both

14
 Rasch Model to Measure Change

baseline and post instruction. These data include all of the following categories of students:

students having teachers with no NU course (Year 0), students having teachers with one NU

course (Year 1), and students having teachers with two NU courses (Year 2) (see Table 6).

Table 6

Number of students with teachers by cohort and number of NU courses

Year 0 Year 1 Year 2 All

(No NU (1 NU course) (2 NU courses)
course)
Cohort 1 0 1212 779 1991
Cohort 2 404 479 0 883
Cohort 3 792 0 0 792
All 1196 1691 779 3666

According to Wright (2003), there are two ways to structure the data matrix and extend

the Rasch model to measure change in understanding over time provided the data fit the model.

By racking the data, the Rasch model can be used to examine changes in item difficulty

estimates over time. On the other hand, stacking the data within the Rasch model allows the

investigation of changes in person estimates over time. These two methods of analysis answer

very different research questions, one focusing on changes in student performance while the

other focuses on changes in content understanding over time. By applying the Rasch model to

the racked and stacked data, interpretations can be made regarding change in student

performance and in item functioning over time, even at the individual level.

Prior to any interpretations, inspecting fit of the data to the model is essential. The

separation and reliability for the persons and items were examined laying the foundation for the

validity of interpretations. Wright and Stone (2004) provided an accepted separation cutoff of 2.

15
 Rasch Model to Measure Change

The person and item separation for the racked and stacked data are summarized (see Table 13).

The item separation for the racked data was acceptable (greater than 2) for the purpose of this

analysis (Wright & Stone, 2004). The person separation for the stacked data was less than 2 for

both analyses.

Table 13

Person and item summary statistics for racked and stacked data

Person Item
Separation Reliability Separation Reliability
Racked 1.63 0.73 23.11 1.00
Stacked 1.27 0.62 32.15 1.00

Such a low person separation was not a surprise considering there are only 39 items to separate a

very large number of students. In addition, the person separation and reliability was not expected

to be high considering all students are in similar regions of rural Appalachia forming a

homogeneous group of middle school science students.

The underlying assumption of the Rasch model is unidimensionality, referring to

measuring only one attribute at a time which in this study is temperature understanding (Bond &

Fox, 2001). The construct can be partitioned into five subtopics encompassed by understanding

of temperature: foundations (1-14), properties of matter (15-17), energy transfer (18-26),

property changes (27-32), and thermal energy (33-39). A principal component analysis

confirmed that the five concepts making up the construct of temperature understanding did not

violate the assumption of unidimensionality. Empirical fit of the data to the model was inspected

for the stacked and racked data to ensure valid interpretations of the NU data using these

analyses. In examining infit and outfit mean square statistics exceeding 1.3 (Wright & Linacre,

16
 Rasch Model to Measure Change

1994b), no items were highlighted within the racked analysis as misfitting. The only item

resulting in a negative point measure correlation for the racked analysis is baseline item 6, which

indicates students estimated with higher levels of understanding of temperature are unexpectedly

missing this item. Fit of the data to the model must also be examined to make interpretations

using person estimates. For the stacked data, 506 students out of 7332 resulted in high fit

statistics. The response patterns for these misfitting students will be further investigated

according to the misfit diagnosis guidelines of Wright and Linacre (1994a).

Stacked Results and Discussion

A primary interest of researchers is typically to determine whether the NU teacher

training course impacted the change in student understanding. To address this question, the data

were stacked to explore changes in student performance from baseline to post assessment. Here

the Rasch model was employed to calibrate the raw scores onto the same ruler under

consideration for fit of data to the model, followed by analysis of variance to compare results

across control and experimental students. The interpretations are then focused to results unique

to the application of the Rasch model.

The stacked analysis allows the comparison of baseline and post student measures on the

same performance continuum. The person estimates were rescaled to resemble the raw score

scale of 0-39 questions to ease the presentation of the Rasch results while maintaining the

properties of the newly calibrated logit measures. The expectation was for person estimates to

increase from baseline to post assessment which indicates students performed better on the

assessment after instruction. Descriptively, 2394 out of 3666 students resulted in a higher person

estimate from baseline to post assessment. Of these, 1704 students were experimental students;

furthermore, 552 students of 779 had teachers that participated in two NU course while 1152

17
 Rasch Model to Measure Change

students of 1691 had teachers participating in one NU course. An analysis of variance on these

calibrated estimates revealed a significant increase in the average change in student performance

estimates for experimental students compared to control students (see Table 14).

Table 14

Descriptive statistics separated by number of NU courses

Std.
N Mean Dev. Min Q1 Median Q3 Max
Baseline Exp Y1 1691 17.189 2.074 10.39 15.69 17.27 18.69 23.90
Y2 779 17.072 2.068 9.27 15.69 17.04 18.29 26.90
Control Y0 1196 17.094 2.161 9.93 15.81 17.18 18.46 24.65
Post Exp Y1 1691 18.635 2.782 9.10 16.77 18.46 20.50 31.80
Y2 779 18.583 2.820 10.40 16.78 18.29 20.10 38.10
Control Y0 1196 17.785 2.467 7.80 16.24 17.77 19.53 27.89
Post- Exp Y1 1691 1.447 2.556 -6.30 -0.20 1.41 3.14 11.80
Baseline Y2 779 1.511 2.525 -7.20 0.00 1.42 3.12 11.30
Control Y0 1196 0.691 2.442 -9.80 -0.70 0.60 2.30 9.90

Assuming control and experimental teachers are similar with the exception of

participating in the NU teacher training course, a much larger percentage of students increased

their performance after instruction implying the professional development ultimately impacted

student learning of temperature. While these results and interpretations can be made using raw

scores and an analysis of variance, the Rasch model followed by analysis of variance provides

accurate results for these data in that students with the same raw score can be distinguished based

on individual item responses and that missing data is not treated as incorrect since not all

students were given the entire assessment.

18
 Rasch Model to Measure Change

Although the use of the Rasch model coupled with analysis of variance provides

appropriate results for these data, this method of analysis extends the conversation beyond

traditional interpretations. With individual person and item standard error estimates,

interpretations can go beyond the aggregate data and include qualitative, individual student

interpretations. Individual student performance estimates with its own standard error enables the

discussion of changes in individual student performance in relation to other students, items, or

proficiency levels along the continuum. Another powerful attribute of the Rasch model is the

estimation of items and persons invariant of the other but calibrated on the same ruler enabling

the conversation of person estimates compared to items on the assessment. From the person and

item map, interpretations can be made about how the assessment is targeting student

understanding of temperature over time. These interpretations can be used by teachers to inform

their instruction, at the individual student or classroom level, as well as curriculum developers of

the NU teacher training course.

Recall that the items assessing five concepts are encompassed by the construct of

understanding of temperature: foundations (F), properties of matter (PM), energy transfer (ET),

property changes (PC), and thermal energy (TE). Interpretations regarding the relation between

the students and the items on the assessment can be made using the item and person map within

the stacked analysis. The item mean is higher than the person mean which indicates the

assessment in its entirety was too difficult for the students. Six of the fourteen foundations items

are estimated at or below the student mean. Gaps in the items reveal where the assessment is not

spanning the student performance continuum adequately. Notice gaps occur in between the item

and person means and one standard deviation below the person mean (see Figure 6).

19
 Rasch Model to Measure Change

Figure 6. Item Map of Stacked Analysis

<more>|<rare>
32 . +
. |
31 +
|
30 . +
|
29 . +
|
28 . +
|
27 . +T
. |
26 . +
. |
25 . + PC27
. | TE37 F4 F6
24 . + ET25
. |S TE36 F8
23 .# + ET24
.# T| F12 TE34
22 .# + ET21 ET23 PC29 TE39
.## | PM17
21 .## + PM15 ET19 ET20 F9
.### | F11 PC30 PC31
20 #### S+ PC28
.##### |M ET22 F7
19 .###### +
.######## | F10 TE38
18 .############ +
.######## M| F13
17 .######### + F3 PC32 TE35
.#### | ET18
16 ####### +S F2
.###### | PM16
15 .##### S+ ET26
.### |
14 .# +
.# | F1 F14 F5
13 .# T+
.# |T
12 . +
. |
11 . + TE33
. |
10 . +
. |
9 . +
|
8 . +
<less>|<frequ>
EACH '#' IS 71.

20
 Rasch Model to Measure Change

Four foundation items (4, 6, 8, and 12) are estimated to be greater than two standard

deviations above the person mean which is troublesome if mastery of these concepts is expected

to ultimately grasp a proficient understanding of temperature. A way to explore these gaps along

the continuum is to consider the possibility that these gaps illustrate proficiency levels in

understanding of temperature. Perhaps each gap illustrates a theoretical concept hierarchy and

levels of knowledge required to master the understanding of temperature.

Within the stacked analysis, there is an expectation that once students have been

instructed on the material they respond significantly different on the baseline assessment

compared to the post assessment. However, 17 of the 39 items did not result in differential

responding from baseline to post assessment. Specifically, these items were 2, 6, 8, 13, 14, 15,

16, 17, 20, 23, 25, 29, 31, 32, 33, 37, and 38. There are two possible interpretations for this

finding. Either the students for the most part answered these items correctly on baseline and post

assessment or vice versa. Conversely, the items resulting in differential responding from baseline

to post assessment can be interpreted as the topics that overall students learned by the post

assessment.

Racked Results and Discussion

Limited attention is usually given to investigating specific topics becoming less difficult

by the end of the course. Within the racked analysis, item difficulty estimates are inspected and

utilized to make interpretations about changes in content rather than student performance. The

conversation under these analyses focuses more on the content of the items rather than the

general discussion of whether experimental students perform better than control students. For

curriculum developers, this information is meaningful in determining potential content areas

where the professional development is impacting teachers more effectively and potentially their

21
 Rasch Model to Measure Change

students. Unlike person estimates, the goal is for item difficulty estimates to decrease by the post

assessment; in other words, the expectation is that once students are instructed on the material

the items become easier.

Similar to the findings from the principal contrast analysis, there was no significant

difference in the average change in item difficulty estimates from baseline to post assessment

across the five topics related to the understanding of temperature. Therefore, students did not

perform better overall on any topic from baseline to post assessment. The item map produced by

the racked data shows the items from baseline and post instruction assessments overlaid on the

same item map allowing examination of changes in distribution of the items along the student

performance continuum (see Figure 7).

22
 Rasch Model to Measure Change

Figure 7. Item Map of Racked Analysis

<more>|<rare>
|
|
2 +
|T
| B4
. |
. | B6
. | B37
. | B27
. | B24 P27
. | B12 B25
. | B21 B8 P6
1 . +S B36 P36 P37
. | P25
. | B17 B23 B29
. | B34 P34 P4 P8
.# T| B19 B39 B9
.# | B15 B20 P39
.## | B30 P24 P29
.### | B31 P23
.##### | B22 P11 P12 P17 P21
.#### | B11 B28 P15 P20 P28 P30 P31 P7
0 .###### S+M P19 P9
.##### |
.######### | B38 P10
.########## | B7
.########## | B13 B35 P22
.############ M| P38
.######## | B10 B18 P3
.############ | B32
.######### | P13
.######### | B16 B2 B3 P32
-1 .########## +S B26 P35
.###### S|
.###### | P16 P18 P2
.#### |
.## | B5
.# | B1 B14
.# | P26
.# T|
. |
. |T P14 P5
-2 . + P1
. | B33
. |
. |
| P33
|
|
+
<less>|<frequ>
EACH '#' IS 24.

23
 Rasch Model to Measure Change

The items within the racked data are labeled B1 for baseline assessment item 1 corresponding to

P1 for post assessment item 1. Another way to visualize the change in item difficulty estimates is

to plot the baseline post-instruction estimates against each other (see Figure 8).

Figure 8. Scatterplot of item difficulty estimates from racked analysis

Scatterplot of Baseline versus Post Item Difficulty Estimates

1.5

1.0
Post Item Difficulty Estimates

0.5

0.0

-0.5

-1.0

-1.5

-2.0

-2.5
-2 -1 0 1 2
Baseline Item Difficulty Estimates

The line (y = x) indicating no change in item difficulty estimates has been drawn to

illustrate increases or decreases in item difficulty estimates over time. Items above the line

represent items estimated as more difficult by the post-instruction assessment while items below

the line represent items estimated as less difficult by the post-instruction assessment. If the items

only assessed topics covered in the course, the expectation would be that the majority of items

would be less difficult by the post instruction assessment (below the line).

24
 Rasch Model to Measure Change

Overall the assessment did not target the student ability as well as was hoped –

specifically ten baseline items and nine post items were estimated below the person mean. The

expectation would be for more items from the post assessment to be estimated as easier after

instruction. The distribution of post instruction assessment items shifted downward indicating the

assessment was in general less difficult for examinees by the post assessment. Item difficulty

estimates for foundational items 4 and 12 decreased by greater than one standard deviation from

baseline to post assessment (see Figure 7). Within the racked analysis, six items out of thirty-

nine resulted in higher difficulty estimates from baseline to post assessment – namely 3, 7, 10,

11, 34, and 36 (see Figure 7). Considering the first fourteen items on the assessment are

considered to be foundations items, it is troublesome that four of these items are more difficult

for students on the post assessment. The concern, here, is that students are developing

fundamental misconceptions throughout instruction. While caution must be taken with

interpretations based on a few items, it is also important to remember the stability of these item

estimates.

Linking content of the student assessment with topics from the teacher assessment and

the NU teacher training course can inform curriculum developers of content areas in need of

more attention. A differential item functioning (DIF) analysis across teachers with student

assessment items linked to the teacher assessment or teacher training may lead to significant

findings in terms of specific impacts of the professional development. These interpretations can

also be appropriate and useful to the participants at the individual classroom level to inform

classroom instruction and for teachers to develop instructional strategies based on results.

25
 Rasch Model to Measure Change

Limitations

The restructured assessment was not administered again and analyzed with the Rasch

model to determine quality of the instrument. The low variability among the response frame also

affects the internal validity of this study due to introducing dependency with the stacked data

into the model (Wright, 2003). Even though evidence in this analysis suggests an increase in

student understanding as a result of the NU distance learning course, neither change in student

performance or item difficulty can be absolutely attributed to NU. The lack of random sampling

and low variability among the response frame is a threat to the external validity of this study.

Conclusion

In summary, the results from the stacked data indicated students of teachers participating

at some level in an NU teacher training course performed significantly better after instruction

than students of teachers having no NU course experience. Not only can this analysis provide

meaningful interpretations to evaluate the impact of NU on student learning overall, qualitative

attributes for student performance at the classroom levels can be useful to teachers in informing

instruction. The results from the racked data indicated the majority of items demonstrated a

decrease in item difficulty estimates by the post instruction assessment. Characteristics of the

items unexpectedly increasing in difficulty by the post assessment should be investigated by the

research team to develop potential reasons for this occurrence. Results will be presented to the

NU course developers to determine how these results link to the NU teacher training course as

well as the teacher assessment.

Educational Importance

The importance of this study is to extend the conversation beyond the overarching

question of whether students whose teachers participated in NU performed significantly better on

26
 Rasch Model to Measure Change

the post assessment after instruction compared to the students of teachers without NU

experience. Instead this study provides a method of analysis encompassing interpretations of this

nature as well as others enabling conversations at the individual student or item level. This study

illustrates a way to apply the Rasch model to examine change among examinees and items over

time in conjunction with traditional statistical analyses in CTT. The racking and stacking of data

illustrates an effective and meaningful way to examine assessment data collected at multiple

points of time. Locally, researchers involved in the NU course development can use these

findings to further improve the course and the student assessment instrument. The interpretations

can also be focused at the classroom level where teachers could use these results to inform their

instruction of temperature. In a broader sense, researchers across various fields can use the

methods shown in this study to improve assessment instruments, course development, and

measures of change over time.

References
Andrich, D. (2004). Controversy and the Rasch model: A characteristic of incompatible
paradigms? Medical Care, 41, I7-I16.
Beck, W. (2001). Testing a non-competency inservice education model based on humanistic or
third force psychology. Education. 98(3), 337-343. Retrieved November 21, 2006, from
EBSCO database.
Bereiter, C. (1963). Some persisting dilemmas in the measurement of change. In C.W. Harris
(Ed.), Problems in measuring change (pp.3-20). Madision: The University of Wisconsin
Press.
Bond, T., & Fox, C. (2001). Applying the Rasch model: Fundamental measurement in the human
sciences. Mahwah, NJ: Lawrence Erlbaum Associates.
Bradley, K. D., & Sampson, S. O. (2006). Utilizing the Rasch model in the construction of
science assessments: The process of measurement, feedback, reflection and change. In X.

27
 Rasch Model to Measure Change

Liu & W. Boone (Eds.), Applications of Rasch measurement in science education (pp.
23-44). Maple Grove, MN: JAM Press.
Duffy, J.R., & Dale, B.J. (1977). The PICA scoring scale: Do its statistical shortcomings cause
clinical problems? In R.H. Brookshire (Ed.), Collected proceedings from clinical
aphasiology (pp.290-296). Minneapolis, MN: BRK.
Embretson, S. E., & Reise, S. P. (2000). Item response theory for psychologists. New Jersey:
Lawrence Erlbaum Associates, Publishers.
Fischer, G.H. (1976). Some probabilistic models for measuring change. In D.N>M De Gruijter &
L.J.T. Van der Kemp (Eds.). Advances in psychological and educational measurement
(pp.97-110). New York: Wiley.
Fischer, G.H. (1983). Logistic latent trait theory models with linear constraints. Psychometrika,
48, 3-26.
Fischer, G.H. (1987). Applying the principles of specific objectivity and generalizability to the
measurement of change. Psychometrika, 52, 565-587.
Fischer, G.H. (1989). An IRT-based model for dichotomous longitudinal data. Psychometrika,
54, 599-624.
Fischer, G.H. (1995a). The Linear Logistic Test Model. In G.H. Fischer & I. Molenaar (Eds.),
Rasch models – Foundations, recent developments, and applications (pp. 131-180). New
York: Springer.
Fischer, G.H. (1995b). Linear logistic models for change. In G.H. Fischer & I. Molenaar (Eds.),
Rasch models – Foundations, recent developments, and applications (pp. 157-180). New
York: Springer.
Fischer, G.H., & Formann, A.K. (1982). Some applications of logistic latent trait models with
linear constraints on the parameters. Applied Psychological Measurement, 4, 397-416.
Green, K.E., & Smith, R.M. (1987). A comparisons of two methods of decomposing item
difficulties. Journal of Educational Statistics, 12, 369-381.
Guskey, T.R. (2000). Evaluating professional development. Thousand Oaks: Corwin Press, Inc.
Hill, L. (2004). Changing minds: Developmental education for conceptual change. Journal of
Adult Development. 11(1), 29-40. Retrieved November 21, 2006, from EBSCO database.
Hunkins, (1972). Questioning strategies and techniques. Boston: Allyn and Bacon, Inc.

28
 Rasch Model to Measure Change

Linacre, J.M. (2006). WINSTEPS Rasch measurement computer program. Chicago:

Winsteps.com.
Linacre, J.M., & Wright, B.D. (1994a). Chi-square fit statistics. Rasch Measurement
Transactions, 8(2), 360. Retrieved from http://www.rasch.org/rmt/rmt82.htm.
Linacre, J.M., & Wright, B.D. (1994b). Reasonable mean-square fit values. Rasch Measurement
Transactions, 8 (3), 370. Retrieved from http://www.rasch.org/rmt/rmt83.htm.
Luce, R.D., & Tukey, J.W. (1964). Simultaneous conjoint measurement: A new type of
fundamental measurement. Journal of Mathematical Psychology, 1, 1-27.
McNeil, M.R., Dionigi, C.M., Langlois, A., & Prescott, T.E. (1989). A measure of Revised
Token Test ordinality and intervality. Aphasiology, 3, 31-40.
Rasch, G. (1960) Probalistic models for some intelligence and achievement tests. Copenhagen:
Danish Institute for Educational Research (Expanded edition, 1980). Chicago: University
of Chicago Press.
Rost, J. (1989). Rasch Models in latent class models for measuring change with ordinal
variables. In R. Coppi & S. Bolasco (Eds.), Multiway data analysis (pp. 473-483).
Amsterdam: Elsevier.
Rost, J. (1990). Rasch Models in latent classes. An integration of two approaches to item
analysis. Analysis Psychological Measurement, 14, 271-282.
Rost, J., & Davier, M. (1995). Mixture distributed Rasch Models. In G. Fischer & I. Molenaar
(Eds.), Rasch models – Foundations, recent developments, and applications (pp.257-
268). New York: Springer.
Smith, E. (2004). Evidence for the reliability of measures and validity of measure interpretation:
A Rasch measurement perspective. In E. Smith & R. Smith (Eds.), Introduction to Rasch
measurement (pp. 93-122). Maple Grove, MN: JAM Press.
Stevens, S.S. (1946, June 7). On the theory of scales of measurement. Science, 103, 677-680.
U. S. Department of Education (2002). No Child Left Behind. Retrieved September 2, 2005,
from http://www.ed.gov/nclb/landing.jhtml?src=pb.
Wainer, H., & Mislevy, R.J. (2000). Item response theory, item calibration, and proficiency
estimation. In H. Wainer, N.J. Dorans, D. Eignor, R. Flaugher, B.F. Green, & R. J.
Mislevy, et al. (Eds.), Computerized adaptive testing: A primer (2nd ed., pp. 61-100).
Mahwah, NJ: Erlbaum.

29
 Rasch Model to Measure Change

Wright, B.D. (1984). Despair and hope for educational measurement. Contemporary Education
Review, 3(1), 281-288.
Wright, B.D. (1999) Fundamental measurement for psychology. In S.E. Embretson & S.L.
Hershberger (Eds.), The new rules of measurement: What every psychologist and
educator should know (pp. 65-104). Mahwah, NJ: Erlbaum.
Wright, B. D. (2003). Rack and Stack: Time 1 vs. Time 2. Rasch measurement transactions.
17(1), 905-906. Retrieved October 15, 2006, from
http://www.rasch.org/rmt/rmt171a.htm.
Wright, B.D., & Linacre, J.M. (1989). Observations are always ordinal; measurements, however
must be interval. Archives of Physical Medicine and Rehabilitation, 70, 857-860.
Wright, B.D., & Masters, G.N. (1981). The measurement of knowledge and attitude (Research
Memorandum No. 30). Chicago: University of Chicago, MESA Psychometric
Laboratory.
Wright, B. D., & Masters, G. N. (1982). Rating scale analysis: Rasch measurement. Chicago, IL:
MESA Press.
Wright, B.D., & Stone, M.H. (1979). Best test design. Chicago: MESA Press.
Wright, B. D., & Stone, M. H. (2004). Making Measures. Chicago, IL: The Phaneron Press.

30
 Rasch Model to Measure Change

Appendix A: Cover Letter for Student Assessment Instructions

Dear Newton’s Universe Teachers:

As part of the Newton’s Universe project, we are collecting student achievement data on selected science topics
addressed in the Newton’s Universe course, Temperature, Heat, and Energy. To accomplish this, we will be
administering pre and posttests three times this year. Enclosed in this packet are the Temperature, Heat, and Energy
Assessment (THESA) materials for the pretest. It is important that you follow the enclosed instructions to maintain
quality and anonymity of the student results.

In this box, you will find the following.

 Instructions for assigning student IDs (blue sheet)

 Instructions for administering & returning the test materials (yellow sheets)
 1 envelope for each class to be tested. Each envelope contains:
o Scantron answer sheets
o Tests
 Classroom set of #2 pencils
 UPS return label and plastic sleeve for sending back all test materials including the answer sheets and all
tests. You may re-use the same box.

Please note that the number of envelopes and tests in each box are based upon your responses on the post-course
survey. The envelopes are labeled with your own teacher ID number, and the quantity of tests enclosed. Please
contact Sally Shafer immediately if the quantities of tests supplied are insufficient.

Please carefully read over the instructions for assigning 5-digit student IDs (blue sheet) and for administering the
test (yellow sheet). Before the test day, follow the instructions to assign student IDs to all students enrolled in
classes that you teach heat, temperature, and energy concepts. Please be certain to record each 5-digit student
ID in a secure place since these IDs will be needed for each test and survey in the project.

On the day of testing, it is important that students use the unique ID numbers you have assigned them. Follow the
yellow sheet instructions for test administration, explaining each step that students should complete. Students should
not write on the test itself, but should mark their answers on the separate answer sheets.

Please administer the test to students as soon as possible. Testing should be completed no later than
September 29, 2006. The deadline for returning all testing materials to Newton’s Universe is October 4, 2006.

If you have additional questions, please feel free to contact the project manager, Sally Shafer. Her contact
information is below.

Sally Shafer
177 Chemistry-Physics Building
Lexington, KY 40506-0055
Phone: 859-257-5131
Email: [email protected]

We appreciate your participation in this project and your assistance in this important partnership with UK and the
National Science Foundation.

Sincerely,

Newton’s Universe Staff

31
 Rasch Model to Measure Change

Appendix B
DIRECTIONS FOR ADMINISTERING NEWTON’S UNIVERSE TEMPERATURE, HEAT, AND
ENERGY STUDENT ASSESSMENT

The Newton’s Universe project is partnering with teachers to improve student understanding of concepts
related to temperature and heat. As a participating teacher, your students will be completing the Newton’s
Universe Temperature, Heat, and Energy Student Assessment (THESA) three times during the school
year.

Complete the blue sheet instructions for ASSIGNING STUDENT IDs before testing day

WHEN YOU ADMINISTER THE TEST, DO SO AS OUTLINED BELOW:

 Please administer the exam on a day that most of your class is present (at least 85%). If there is a
school function or illness leading to many absences, please postpone the test until you can have most
of the class present.

 If a student has an IEP for standardized tests, please follow the procedures outlined in their IEP for
state standardized assessments.

BEGINNING THE TESTING:

The tests and bubble answer sheets are in unsealed envelopes, one envelope for each science class in
which you will be teaching heat and temperature concepts. Do not open testing materials until the time of
the test.

 Provide each student a scantron (answer sheet) and a # 2 pencil (not pen).*
*Please note: Even though answer sheets explicitly state pen or pencil can be used, the machine used
to read the answer sheets will only recognize marks made by a #2 pencil.

 Ask each student to fill in their five-digit STUDENT ID in the space provided at the top right corner
of the form. They should write their number in the first five boxes starting on the left, filling in the
appropriate matching bubbles. Leave the remaining ID boxes blank. It is very important that
student IDs are correct, so please double check that students have accurately completed this
task. Students should not complete the name or test name spaces on the answer sheet. The section
marked For Teacher Use Only should remain blank.

 Tell students to not write on the test. To mark an answer they should carefully darken the
bubble for that choice on the answer sheet. Please remind students to double check that their
answer sheet number matches the test question. Mark only one answer for each question.

 Tell students that the last page of the test is a four-part short answer question. They should
write their answers, and label them A, B, C & D, in the block provided on the answer sheet.
 Hand out the tests. Instruct students to answer all questions and to do their best work.

32
 Rasch Model to Measure Change

Appendix B continued

UPON COMPLETION OF TESTING:

When a class completes the test, immediately place all the scantron (answer) sheets and tests back
in the envelope, including all unused tests and answer sheets. Immediately seal the envelope and
write the class number on the label. It is important that you do not review the students’ responses.

Do not keep any copies of the test or test questions; return them all to Newton’s Universe.

RETURNING THE TEST MATERIALS TO NEWTON’S UNIVERSE:

After all of your classes have finished testing, put all of the sealed envelopes, pencils, and any other
testing materials in the shipping package in which they arrived.

Use the pre-printed UPS shipping label to return the test materials to the Newton’s Universe project
office. There is an adhesive backed plastic sleeve enclosed for the shipping label. After sealing your box
with sturdy tape, attach the label in its plastic sleeve to the outside of the box. Be sure to remove any
address or barcode information that might remain on the box from its initial journey to you.

For your convenience, the boxes can be sent directly from your school’s office by calling UPS to arrange
for a pickup, and giving them the information that they request from the return label. If there are other
teachers in your school completing testing, they should return their materials separately, in their own
shipping package.

33