You are on page 1of 28

Graduate Statistics Education

Anxiety and Attitudes in Graduate Statistics Education:


Measuring and Monitoring Change During the Semester

Freda S. Watson
Melinda R. Hess
Anthony J. Onwuegbuzie
John M. Ferron
Jeffrey D. Kromrey
Kristine Y. Hogarty
Douglas Lunsford
Robert F. Dedrick
University of South Florida

Paper presented at the annual meeting of the Eastern Educational Research Association, February 11 –
13, 2004, Clearwater, Florida.
Graduate Statistics Education
2

Anxiety and Attitudes in Graduate Statistics Education:


Measuring and Monitoring Change During the Semester

The research literature leaves little doubt that affect and attitude are important in
statistics education. Onwuegbuzie and Wilson (2003) estimate that “. . . between 66 and 80
percent of graduate students are uncomfortably anxious about statistics” (p. 195), with many
students deeming statistics to be one of the most anxiety-inducing courses in their programs of
study (Blalock, 1987; Caine, Centa, Doroff, Horowitz, & Wisenbaker, 1978; Gaydosh, 1990;
Lundgren & Fawcett, 1980; Schacht & Stewart, 1990, 1991; Schau, 2003; Schau, Stevens,
Dauphinee, & Del Vecchio, 1995; Zeidner, 1991). Further, many students report negative
attitudes toward statistics (Dauphinee, Schau, & Stevens, 1997). Other researchers (for
example, Gal, Ginsberg, & Schau, 1997) emphasize that attitudes and beliefs can help or hinder
statistics learning and may determine, in part, whether students become proficient in statistics
and use that proficiency after their statistics education is complete. Tobias (1994) reported that
mathematics anxiety and negative attitudes often develop when students are unable to
comprehend, and then fail to find the help they need. Gal and Ginsburg (1994, p. 4) suggest
that a similar process likely occurs in statistics education and that “. . . it is important that
statistics educators have access to assessment instruments that enable an initial diagnosis of
their students’ attitudes and beliefs and also enable monitoring of the status of such attitudes
and beliefs during a course.”
The prevalence of anxiety associated with statistics has led most major publishing
houses to carry at least one title addressing statistics anxiety (SA) or a work designed to
present the concepts of statistics absent formulas. Sage Publications has Neil J. Salkind’s
Statistics for People Who Think They Hate Statistics and a third edition of Richard Jaeger’s
Statistics: A Spectator Sport. Prentice Hall has a third edition of Gerald Kranzler’s Statistics for
the Terrified—and the list goes on. The number and emotional intensity of the titles, the fact that
many are in their second or third edition and several other relatively new volumes on the same
subject have been authored in the last few years reflect the seriousness of this problem.
The prevalence and seriousness of anxiety associated with statistics has also led to the
development and implementation of programs designed to address the problem. One of these
programs is EncStat, a multimedia program designed to identify students with statistics anxiety
or negative attitudes toward statistics, ameliorate that anxiety using cognitive behavior therapy
techniques, and assist students to achieve more positive attitudes about statistics. Two of the
scales most frequently used in measuring statistics anxiety and attitudes, the SATS and the
Graduate Statistics Education
3

STARS, were deemed appropriate for EncStat. Because student attitudes (Elmore, Lewis, &
Bay, 1993; Zimmer & Fuller, 1996) and anxiety (Elmore et al., 1993; Fitzgerald, Jurs, & Hudson,
1996; Lalonde & Gardner, 1993; Onwuegbuzie, 2003; Onwuegbuzie & Seaman, 1995; Zanakis
& Valenza, 1997; Zeidner, 1991) play a crucial role in statistics achievement and frequency and
quality of statistics use by alumni, information provided by both scales was considered
essential. In the computerized pilot version of EncStat (the preliminary version was
paper/pencil), this information was used to tailor the intervention for each student, an
accommodation that would be impossible without information provided by the SATS and STARS
subscales. For example, if the STARS and SATs results show a student highly values statistics
but has poor study skills and a previous history of negative experiences with math, the EncStat
program for that student would include only brief statements on the worth of statistics but
significant detail about improving study habits and overcoming the effects of a negative past
history with math. Results of our work on the EncStat project have been presented several
times during the past years (see, for example, Watson, Lang, Kromrey, Ferron, Hess, &
Hogarty, 2003; Watson, Kromrey, Ferron, Dedrick, Hogarty, Lang, & Hess, 2003; and Watson,
Kromrey, & Hess, 2003) in our efforts to develop a comprehensive system for the study and
treatment of statistics anxiety.
As McLeod (1992, p. 578) observes, “Beliefs, attitudes, and emotions are used to
describe a wide range of affective responses to mathematics...beliefs and attitudes are
generally stable, but emotions change rapidly.” Further, “affective issues play a central role in
mathematics learning and instruction. . . . If research on learning and instruction is to maximize
its impact on students and teachers, affective issues need to occupy a more central position in
the minds of researchers” (McLeod, 1992, p. 575). However, Gal and Ginsburg (1994) observe
that much of the research on attitudes toward statistics (which, in their conceptualization,
includes statistics anxiety) has used Likert-type response scales that, although convenient for
generating broad descriptions and for reporting, produce little diagnostic information about
individual students, information that is critical to guide intervention and inform changes in
pedagogy.
The complexity of a phenomenon and its resistance to easy analysis are often related to
the proportion of emotional content. Few things are more emotional than fear (anxiety sans the
euphemistic cloaking) and few more resistant to change than emotionally charged attitudes.
Statistics anxiety is a complex problem, and to design effective instructional practices to
ameliorate even a small portion of the discomfort so many students feel, we must take our
Graduate Statistics Education
4

search to new depths, and we must do so with finer, more delicate methods that can detect the
qualitative differences with which we are grappling.
In an effort to develop a measure of affect that could be easily administered at multiple
points during a semester, Watson, Kromrey, Lang, Hess, Hogarty and Dedrick (2003)
developed the Statistics Anxiety Affect Checklist. This instrument consists of 21 emotion words
(which cohere around seven basic emotions, five of which were identified by Eckman, 1992),
each of which is rated on a 7-point scale (ranging from 1 = “Do Not Feel This Way at All” to 7 =
“Feel This Way Intensely”). In addition to the ratings of emotion words, space was provided for
open-ended commentary. Results of the analysis presented by Watson et al. (2003) suggest
that
 Students sitting in the same class, listening to the same instructor experience a broad
range of emotions and emotional intensity.
 In general, the most extreme levels of the negative emotions are being experienced by
the most anxious student.
 The students indicated the generally expected absolute and relative positions on each of
the emotion scales, i.e. following the pattern of their STARS and SATS total scores.
 Graphs for Anger, Fear, Sad, Confused, and Dumb generally vary together over the
course of the semester. The literature speaks of the SA student as feeling anxious and
not too bright, but none or rare mention is made of anger and sadness, emotions that
also have a significant impact on learning, concentration, and motivation.
 There is a great deal of variation, week to week, for each student.
 The student with the most intense feelings also evidenced the most variability week to
week.
 What is perhaps the most telling of all is how intensely students felt angry, afraid, and
dumb. Sadly, this affirms what most teachers of statistics had suspected for years and
which the work thus far with Likert scale type surveys has demonstrated: that some
students experience high levels of negative emotions.
 Open-ended comments provided by the participants, in general, supported their
responses to the 21 emotion words.

The preliminary work reported by Watson et al. (2003) supports the utility of a quick
affect check instrument such as the Statistics Anxiety Affect Checklist. In an effort to reduce the
response burden necessary with the 21-item instrument, a revised instrument was developed
and pilot tested with new samples of graduate students in statistics courses. As with the 2003
research, our hypothesis is that a pretest measure of attitudes and weekly measure of emotion
will help identify critical turning points in attitude formation when intervention would be most
beneficial in preventing the creation or exacerbation of statistics anxiety (SA) and negative
attitudes toward statistics (NAS). Thus, the purpose of this study was to identify trends and
relationships between changes in students’ self-reported emotion and increasing levels of SA
and NAS during a semester of instruction in graduate statistics.
Graduate Statistics Education
5

Method

Sample
Data were collected from 77 graduate students in three statistics classes in the College
of Education at a large metropolitan research university. Two of the participating classes were
sections of the first course in graduate statistics in the College (n = 56) and the other was a
single section of the subsequent, Statistics II, class (n = 21). Both of the Statistics I classes
were taught by the same instructor and the Statistics II class was taught by another member of
the faculty. Both faculty members have taught these courses at least 10 years at the same
institution. To encourage student participation in the project, course credit was provided.
Demographic information on the participants suggests that the average student was just
over 35 years of age (M = 35.3, SD = 8.4, range = 21 to 54 years). The number of graduate
credit hours earned at the time of the statistics course ranged from 0 to 85 with a mean of 19.9
and a SD of 21.1. Grade Point Average (GPA) was high, as expected with doctoral courses,
with a mean of 3.74 and a SD of 0.53. GPA values ranged from 1 to 4.0. Years of high school
mathematics ranged 1 to 5 years (grades 7 –12th) with a mean of 3.4 and a SD of 0.89, and the
number of college math or statistics courses ranged from 1 to 4 with a mean of 3.1 and a SD of
0.83.

Instrumentation
Affect Check.
A revised affect instrument was developed based on a semantic differential item format.
In this type of instrumentation, each item does not address a singular affect measure, rather
participants are provided with two bipolar adjectives intended to measure opposite feelings,
emotions, or states. The response scale consisted of seven points between the two polar
adjectives. The participant then marks a spot closest to what he or she is feeling at that
moment. For this study, items were developed intended to measure three domains: (1)
Cognitive Comfort vs. Distress, (2) Emotional Comfort vs. Distress, and (3) Interest vs.
Boredom. Examples of items used for each of the three domains are in Table 1.
Graduate Statistics Education
6

Table 1
Examples of Affect Check Items.

Negative Positive
Domain
Adjective Adjective

Cognitive Anxious ___ ___ ___ ___ ___ ___ ___ Calm

Emotional Frustrated ___ ___ ___ ___ ___ ___ ___ Satisfied

Boredom Indifferent ___ ___ ___ ___ ___ ___ ___ Fascinated

Once items had been constructed, reviewed, and finalized, they were randomly ordered
on the instrument and were not identified as belonging to a specific domain. In order to avoid
the bias of response set, the presentation of the individual adjective pairs regarding positive and
negative poles were mixed. That is, for some items the more negative adjective was on the left
side of the page and the positive adjective was on the right, whereas for others the more
positive measure of affect was on the left and the more negative on the right.
In addition to the affect check, two of the scales most frequently used in measuring
statistics anxiety and attitudes, the Statistics Anxiety Rating Scale (STARS) and the Survey of
Attitudes Towards Statistics (SATS) were also used in this study.
STARS
The STARS was developed by Cruise, Cash, and Bolton (1985) to measure statistics
anxiety, which the authors define as “feelings of anxiety encountered when taking a statistics
course or doing statistical analysis” (p. 92). The original STARS instrument created was
modified and pilot tested in a previous study (Watson, Lang, Kromrey, Ferron, Hess, & Hogarty,
2003). The revised instrument is comprised of six subscales, with a set of 36 items employing
a 5-point response scale. Depending on the nature of the question, values range from 1 = No
Anxiety to 5 = Very Much Anxiety; or 1 = Strongly Disagree to 5 = Strongly Agree. A high score
on the STARS indicates either a high level of anxiety and/or a negative attitude towards
statistics. The internal consistency of scores pertaining to this subscale was estimated to be
.93.
Worth of Statistics (Worth). This subscale is comprised of five items that address the
student’s perceived importance of statistics. Items in this subscale include: ‘Statistics takes
more time than it’s worth’, ‘Statistics is a waste of time’, and ‘Statistics is for people who have a
Graduate Statistics Education
7

natural talent in math’. Higher scores reflect a lower level of personal importance placed on
statistics by the individual. Cronbach’s alpha for the revised subscale was estimated to be .84.
Interpretation Anxiety (Interp). This set of 11 items was designed to measure anxiety as
a result of interpreting or making a decision utilizing statistical data. Among the items on this
scale, respondents are queried regarding their perceived anxiety when: ‘Making an objective
decision based on empirical data, ‘Interpreting the meaning of an obtained probability value’,
and ‘Deciding whether or not to reject the null hypothesis’. A high score on this subscale
indicates an inability to effectively use statistical procedures and strategies in the course of daily
life. The Cronbach’s alpha was estimated to be .90.
Test and Class Anxiety (Tstclss). The eight items comprising this subscale focus on
anxiety exhibited as a result of class or test participation. On this subscale, respondents
express their level of anxiety with respect to situations such as ‘Studying for a statistics exam’,
‘Doing statistics homework’, and ‘Taking a statistics exam. As with the two previous subscales,
a high score is synonymous with increased levels of anxiety. The Cronbach’s alpha was
estimated to be .92.
Computational Self-Concept (Compute). This subscale contains four items focusing on
anxiety related to one’s perceived knowledge and ability to use statistics and to complete
computations of mathematical procedures. Here respondents are asked if they enjoy math and
if they find statistics too mathematical. A high score reflects a greater level of anxiety related to
performing mathematical computations within statistical procedures as opposed to statistics
itself. The Cronbach’s alpha was estimated to be .72.
Fear of Asking for Help (Fearhlp). The four items comprising this subscale address
anxiety related to asking for help. Respondents are asked about their fear of asking for help
from statistics teachers, assistants in the computer lab, and fellow classmates. High scores
reflect greater amounts of anxiety perceived by the individual as a result of asking either
teachers or fellow students for help in understanding materials of a statistical nature.
Cronbach’s alpha was estimated to be .88.
Fear of Statistical Teachers (Feartch). This set of thee items was developed to address
an individual’s perception of the statistics teacher. Unlike the preceding subscale, these items
seek responses to questions regarding statistics teachers’ traits (i.e., are they too abstract, do
they talk a different language, and do they think differently than the rest of us). High scores
reflect an individual’s perception that the instructor lacks sufficient understanding to relate to the
individual’s predicament. The Cronbach’s alpha was estimated to be .81.
Graduate Statistics Education
8

SATS
The second instrument we utilized was an adaptation of the SATS, developed by Schau,
Stevens, Dauphinee and Del Vecchio (1995). The revised scale consists of 4 sub-sections
containing a total of 28 items. The items on this set of subscales employed a 5-point Likert scale
with 1 = Strongly Disagree and 5 = Strongly Agree. While a high score on the SATS originally
reflected a positive attitude or view towards statistics, for this study all items were reflected in
scoring so that a higher score reflects a more negative view or attitude towards statistics (to be
more consistent with scores on the STARS). Cronbach’s alpha was estimated to be .91.
Affect (Affect). This subscale contains five items and its focus is on feelings about
statistics. Respondents are asked about their comfort with statistics, if they find statistics
stressful, and if they enjoy taking statistics. A low score reflects a positive view of statistics.
Cronbach’s alpha was estimated to be .84.
Cognitive Competence (CogComp). The eight items in this subscale evaluate an
individual’s attitudes and intellectual knowledge relating to statistics. The items on this subscale
address the difficulty of statistical concepts, the tendency to make computational errors and
understanding statistical equations. A low score reflects a positive attitude and knowledge level.
Cronbach’s alpha was estimated to be .86.
Value (Value). This subscale reflects attitudes regarding the relevance and usefulness of
statistics. This subscale contains seven items. Here respondents are asked if they feel statistics
will make them more employable, the relevance and use of statistics in everyday life, and the
necessity of statistics in professional training. Cronbach’s alpha was estimated to be .66.
Difficulty (Diff). The eight items in this subscale address perceived difficulty of statistics
materials. Included in this subscale are items that explore areas such as statistical notation,
formulas and equations and statistical software. Lower scores reflect a more positive attitude
toward the field of statistics. Cronbach’s alpha was estimated to be .74.

Additional Scales
The authors developed two additional scales in an attempt to learn more about
participants’ study skills and past math history. Both of these instruments use a 5-point Likert
response scale.
Study Skills (Study). This subscale consists of eight items that focus on the individual’s
study habits. Reviewing statistical notes and texts, asking questions and preparing in advance
for class, are among the items on this subscale. A high score reflects better study habits as
reported by the individual. Cronbach’s alpha was estimated to be .70.
Graduate Statistics Education
9

Past Math History (MathPH). The four items making up this subscale evaluate the
individual’s perceptions of their past experiences with mathematics. A high score indicates a
positive history with the field of mathematics. Cronbach’s alpha was estimated to be .82.

Procedure
The revised Statistics Anxiety Affect Check (Appendix A) was administered over a 16-
week time period in three statistics classes taught in the graduate program of educational
measurement and research. The only weeks in which the instrument was not administered
were those of the mid-term and final exams. As such, data were available for most students at
14 points over the semester.
During the first class session of the semester, a member of the research team attended
each of the three classes to explain the study, its purpose, and the instrumentation. This
orientation was conducted by the same individual for all classes to ensure consistency across
classes. In addition to informing the students about the study in general and the affect check
instrumentation and purpose specifically, she also obtained written consent for voluntary
participation by the students and informed them of the anonymous nature of the process. This
anonymity was ensured by assigning each student a control number during orientation to be
used to track responses over the semester. Only one research assistant had the information
that could link an individual student with their number and that information was only used in the
event a student forgot his or her control number. Each week, the affect checks were gathered
in an envelope and provided to the research assistant. At no time did the instructors view the
instruments completed by their students, either at the data collection stage or later in the
process. All data entry for a specific class was performed by someone other than the instructor
of the class and only control numbers were entered into this data base and used in subsequent
analysis.
Over the ensuing 16 weeks, students in the classes completed one affect check
instrument each week (with the exception of the exam weeks). During the 20-minute mid-class
break each week, copies of the instruments were made available to the students to complete
regarding how they were feeling during class that evening. Upon completion, students
individually placed their affect check instrument in the envelope provided. At the end of each
class session, these envelopes were collected by the research assistant. The research
assistant then organized the affect checks by instructor, course, and point in time and reviewed
all forms to ensure that control numbers were entered and identifying information removed, if
necessary. Data entry was then conducted by a small group of the research team using an
Graduate Statistics Education
10

Excel template and coding protocol designed to capture course level, point in time of affect
check, instructor, and class. Once the data entry was completed, the data was analyzed using
the SAS statistical package, version 8.2.

Results
The results of this research are presented in four sections: exploratory factor analysis of
Student Affect Check, thematic analysis of open-ended responses, trends in affect during the
semester, and selected cases.

Exploratory Factor Analysis of Affect Check Data


An exploratory factor analysis was conducted via PROC FACTOR in SAS, using
principal factors extraction with squared multiple correlations as the initial communality
estimates. This analysis was based on 861 completed affect checks, which were collected
weekly during one semester in two sections of the introductory statistics course and one section
of the second statistics course. The initial communality estimates ranged from a low of .55 for
the adjective pair Fascinated -- Indifferent, to a high of .79 for the adjective pair Certain -- Lost.
Fifteen of the 18 items had initial communality estimates of .70 or higher. Prior to the factor
analysis, items were reflected so that higher scores represented responses closer to the
negative adjective in each pair.
Based upon the multiple criteria of a scree plot, proportion of variance accounted for and
interpretability, three factors were retained, accounting for 73% of the total variance in the items.
After extraction, the factors were rotated to improve interpretability. This was accomplished
using the promax procedure, which allows the factors to correlate. Both the factor pattern
matrix and the structure matrix for the rotated factors were examined (Table 2). The first factor
(Cognitive Distress), accounted for 54% of the total item variance and 12% of the unique
variance. This factor loaded nine adjective pairs such as Clear – Confused, Certain – Lost, and
Confident – Afraid. In contrast, the second factor (Emotional Distress), accounting for 53% of
the total item variance and 8% of the unique variance, loaded six adjective pairs that reflected
affect such as Sad – Happy, Mad – Pleased, and Ashamed – Proud. Finally, the third factor
(Boredom), accounting for 33% of the variance and 7% of the unique variance, loaded three
adjective pairs: Fascinated – Indifferent, Interested – Bored, and Attracted – Disgusted. Of the
item adjective pairs that comprise the Affect Check, only one pair evidenced complexity in
loading. Dumb – Smart, loaded on both the Cognitive Distress ( β = .54, r = .81) and the

Emotional Distress ( β = .53, r = .78) factors.


Graduate Statistics Education
11

Table 2
Pattern and Structure Coefficients for Three Factors Extracted from Student Affect Check.

Cognitive Distress Emotional Distress Boredom

Item Pattern Structure Pattern Structure Pattern Structure

Clear-Confused 92 * 87 * -9 59 4 42
Certain-Lost 89 * 90 * 2 65 -1 42
Confident-Afraid 74 * 88 * 22 73 * -4 44
Encouraged-Discouraged 70 * 80 * 5 63 13 49
Calm-Anxious 67 * 83 * 18 70 * 5 48
Tranquil-Worried 66 * 84 * 21 72 * 7 50
Dumb-Smart 54 * 81 * 53 * 78 * -24 33
Clever-Foolish 51 * 79 * 36 75 * 4 49
Peaceful-Irritated 44 * 77 * 29 76 * 26 64
Sad-Happy 10 67 75 * 86 * 7 55
Mad-Pleased 10 68 74 * 87 * 10 58
Depressed-Happy 3 64 73 * 86 * 17 62
Ashamed-Proud 28 71 * 68 * 81 * -13 41
Miserable-Delighted 5 64 66 * 84 * 25 66
Frustrated-Satisfied 31 76 * 55 * 84 * 11 58
Fascinated-Indifferent -1 35 -3 43 79 * 77 *
Interested-Bored -4 35 4 46 77 * 77 *
Attracted-Disgusted 18 61 15 67 66 * 84 *

Total Unique Total Unique Total Unique

Variance Accounted For 54% 12% 53% 8% 33% 7%

Note. All coefficients have been multiplied by 100 and rounded. Bolded coefficients represent items included on each of the
summative subscales.
Graduate Statistics Education
12

The correlations among the factors are provided in Table 3. The relationships among
factors were positive, with the largest correlation evident between Cognitive Distress and
Emotional Distress (r = .72). The correlations between each of these factors and the Boredom
factor were lower, but still of substantial magnitude (r = .47 with Cognitive Distress and r = .59
with Emotional Distress).

Table 3
Interfactor correlations and reliability estimates for Student Affect Check.

Cognitive Distress Emotional Distress Boredom

Cognitive Distress 96

Emotional Distress 72 95

Boredom 47 59 85

Note. Coefficient alpha reliability estimates are on the main diagonal. All coefficients have been
multiplied by 100 and rounded.

Internal Consistency Estimates. Following the factor analysis, scale scores were
constructed for each of the three factors by averaging the ratings on the items that loaded on
each. The bolded pattern coefficients in Table 2 indicate which items were used to represent
each factor. The internal consistency of the ratings on the items within each scale was then
estimated using Cronbach’s alpha. These estimates are provided in the diagonal of Table 3. All
of the scales evidenced satisfactory levels of internal consistency, with alphas ranging from .85
(Boredom) to .96 (Cognitive Distress).

Thematic Analysis
Themes identified in the write-in comments can be categorized as those focusing on
students’ feelings, and those focusing on other issues related to external conditions and
influences, with a subset of the former being those comments that seem indicative of potentially
serious difficulties developing or about to develop. The most common themes that emerged
related to students’ feelings were feelings of anxiety, confidence, understanding (or lack
thereof), effort exerted in studying statistics, stress, generally positive feelings, generally
negative feelings, and neutral comments.
Students electing to comment on their anxiety made such notations as, “Last night I
couldn’t sleep because I was worried about getting behind in this class”; “I need to get at least a
Graduate Statistics Education
13

B in this class to graduate in December and I am concerned that I will not get it”; “Very
concerned about research project; overwhelmed by information”; “Anxiety continues to mount”
and “Scared senseless about the final.” Responses from students who chose to elaborate on
their degree of confidence were in general positive, such as “Amazingly, I feel more confident
about the class than I did when anticipating taking the course. The professor has a lot to do
with the comfort zone.” and “Feeling confident because I understand what I’m doing so far - - I
feel exited and encouraged, for now.” Write-in comments addressing understanding were,
however, both positive and negative. For example, “So far I feel I’ve been able to understand
and make sense of the content I’ve studied.” and “I am becoming increasingly confused with the
abundance of material.”
It seemed likely that the students might have labeled some of their generally positive
feelings and negative feelings as they did because just having completed the semantic
differential scale brought these words to mind. For example, some students labeled their
feelings as encouraged, interested, proud, bored, or confused. Regarding the comments that
seemed neutral, e. g., “Nothing stressful or new”, it is possible that some students felt obliged to
at least write some commentary, just for the sake of commenting and continuing their
participation in the research effort. That is, of course, commendable if it is indeed what
occurred. Adding weight to the supposition that some students felt obliged to maintain their
weekly contribution to the research effort is the fact that several students made write-in
comments throughout the semester even though they consistently talked about being bored and
having had the material before. Because student populations in colleges of education have
extremely diverse educational backgrounds upon entry into master’s and doctoral programs,
some students are already familiar with many of the basic statistical concepts presented in other
classes. This seems to be a situation that most universities could not avoid, absent an
unexpected and highly unlikely funding surplus.
Since it was a subject not tapped by the Likert responses, it was notable that students
made a substantial number of comments about the effort they did or did not exert in their
studying. Their comments showed a good deal of metacognition that, appropriately guided,
could be useful. For example, “I have put some extra time in this past week learning the
material and I feel like things are clicking” and the other extreme “I feel like I need to do more
reviewing” and “I need to do my part!”
One of the most interesting themes to emerge was stress. It was interesting because of
its frequency of occurrence as well as the intensity and variety of responses given. Based on
this sample of student comments, it seems graduate students frequently arrive at their statistics
Graduate Statistics Education
14

classes with a high level of stress. It would be informative to compare degree of student stress
on arrival at statistics class with that experienced with other classes, such as those in the
student’s cognate, that would presumably be low stress.
Comments relating to external conditions and influences can be categorized as those
pertaining to content (difficulty understanding symbols, formulae, concepts, and math), grades,
classroom conditions, statistical software, and the instructor. Consistent with the literature
Wilson (1999) and results of previous work, students found instructor attitude helpful and
reassuring (“The professor is very good and therefore a lot of anxiety or stress has been
removed.“, . . . The professor has a lot to do with [my] comfort zone” and “. . . The instructor is
clear and presents information in a relevant and understandable manner.”
Some students seemed to be using the comments as a form of journaling, which has
been shown by Sgoutas-Emch and Johnson (1998) to reduce statistics anxiety. Instructions on
the Affect Check request students to explain why they are feeling as they have indicated and to
make note of anything experienced during the week or the day of class that was especially
stressful. This is useful information to have when interpreting scores as well as write-in
comments. During the semester in which data were collected for this study, some students
experienced, for example, serious family illness, personal health issues, and even the death of a
classmate. Such events, it seems, heavily influence student affect.

Trends in Affect During the Semester


To get a sense of how students’ affect changed from week to week, the scale scores for
Cognitive Distress, Emotional Distress, and Boredom were plotted for each student over the
semester. The plots for the Statistics I course are presented in Figures 1-3 and the plots for the
Statistics II course are presented in Figures 4-6. The most salient result is the variability in how
individual students experience a statistics course. Some students consistently experience very
high levels of negative affect, others experience high levels of negative affect on some weeks,
and others never experience high levels of negative affect.
Graduate Statistics Education
15

Figure 1
Cognitive Distress as a Function of Time for Statistics I Students

Figure 2
Emotional Distress as a Function of Time for Statistics I Students
Graduate Statistics Education
16

Figure 3
Boredom as a Function of Time for Statistics I Students

Figure 4
Cognitive Distress as a Function of Time for Statistics II Students
Graduate Statistics Education
17

Figure 5
Emotional Distress as a Function of Time for Statistics II Students

Figure 6
Boredom as a Function of Time for Statistics II Students
Graduate Statistics Education
18

HLM Results. To further examine the changes in affect over the semester hierarchical
linear models were examined using REML estimation via HLM, version 5. In the first level of
each of the two-level models, affect (cognitive distress, emotional distress, or boredom) was
modeled as a function of time. The simplest growth curve model assumes linear growth in
affect for the students. More formally,

affectit = π 0i + π 1i * time + eit ,

where affectit is the affect rating of the ith student at time t, π0i is the intercept, or affect rating
from week 1, π1i is the linear component which can be thought of as the slope of the growth
trajectory, and eit is the error term, which captures the degree to which affectit deviates from the
value predicted by the linear model. The errors were assumed to be normally distributed with a
covariance structure Σ = σ 2 I .
The second-level of the model, then focused on the degree to which the parameters of
the first-level model (π1i , and π2i) varied across students. More formally,

π 0i = β 00 + r0i
π 1i = β10 + r1i

where β 00 is the average intercept, β10 is the average slope, and r0i, and r1i, are the error terms
that allow the intercepts and slopes to vary from student to student, respectively. These errors
are assumed to be normally distributed with a covariance matrix, T, where T is assumed to be
unstructured. Modeling T to be unstructured lets the variance in each parameter be different,
and allows the components to covary. For example, one may find that students with lower
intercepts tend to show more pronounced changes.
Table 4 provides the results of modeling affect with a linear model for the Statistics I
students. We were particularly interested in whether there was evidence that the growth
trajectories varied across students, thus we conducted null hypothesis tests for intercept and
slope variability. The variance components were statistically significant for each outcome
(cognitive distress, emotional distress, boredom). This suggests that individual students differed
from each other in their feelings of cognitive distress, emotional distress, and boredom. We also
examined quadratic growth curve models and again would conclude there is variability in
individual trajectories.
Graduate Statistics Education
19

Table 4
Hierarchical Linear Model Results for Affect in the Statistics I Students.

Parameters Cognitive Distress Emotional Distress Boredom

Intercept Variance, Var(r0i) 1.005* 0.848* 0.970*

Slope Variance, Var(r1i) 0.061* 0.003* 0.003*

Covariance Cov(r0i, r1i) -0.002 -0.011 -0.033

Error Variance, Var(eit) 0.811 0.545 0.553

Average Intercept, β00 3.329 3.350 3.433

Average Slope, β10 0.0371 0.0177 0.007

* p<.05 using HLM’s chi-square test for variance components.

Selected Cases.
Given the variability from individual to individual, we decided to select five individual
cases for further consideration. Prior to selection we divided the sample into 9 groups based on
the level and variability in their weekly affect ratings. More specifically, we averaged across the
three affect scales (cognitive distress, emotional distress, and boredom) to created a total score,
and then for each participant we calculated the average and standard deviation of these total
scores across the weeks. We then sorted the participants into three groups based on average
level of affect: high (above Q3), medium (between Q1 and Q3), and low (below Q1). We then
classified them into three groups based on the variability of their affect ratings: high (above Q3),
medium (between Q1 and Q3), and low (below Q1). By crossing these two factors we created
nine subgroups representing different types of trajectories (e.g., low level of anxiety with low
variability from week to week, or medium level of anxiety with high levels of variability from week
to week). To consider the utility of using the Affect Check as an aid in identifying students
needing assistance as well as to identify particular points at which students need assistance,
comments that seemed to be turning points were identified. We then purposefully sampled 5
participants based on the verbal comments made on the weekly affect checks, under the
constraint that the sample would also include participants with different types of trajectories.
Graduate Statistics Education
20

The graphs showing the affect ratings for these five participants are displayed in Figures
6-9, where the different graphs correspond to the different affect scales (cognitive distress,
emotional distress, and boredom).

Figure 7
Cognitive Distress as a Function of Time for Selected Statistics I Students

On review of the comments, the students with medium or high affect values and those
with medium affect values but high variability (Students 1, 2, and 3) all made comments
relatively early in the semester that would have indicated a need for help that, if acted upon,
likely could have reduced anxiety, improved attitude, and increased achievement. This
suggests that a high degree of variability in emotionality might predispose a student to
experience intense levels of anxiety and related negative motions.
Graduate Statistics Education
21

Figure 8
Emotional Distress as a Function of Time for Selected Statistics I Students

Figure 9
Boredom as a Function of Time for Selected Statistics I Students
Graduate Statistics Education
22

Student 1 used some form of the word ‘overwhelmed’ four separate times (the first in
Week 3) and toward the end of the semester stated “I am becoming increasingly confused by
the abundance of material” and “I am also getting very nervous about the final exam.” A
similar pattern emerged for student 2, (medium typical value, high dispersion) who made few
comments during the semester but who, the last class night before the final exam stated
“Scared senseless about the final…” This might reflect the student’s high degree of emotionality
in general. Student 3 had high anxiety and medium dispersion and gave some early warning
of problems, noting in Week 2 that “Very stressed out with this class and course load. This is
the last class I want to worry about.” And in Week 11 “The course material has increasingly
gotten harder and I feel totally lost…” Student 3 also provided evidence of a problem referenced
by Tobias (1994), that of students becoming increasingly frustrated when they cannot
understand. “You think you understand something and then all of a sudden you realize you
don’t. This is a very frustrating experience” and later “…When you think you know something
you end up not really knowing it.”
In their write-in comments, students 4 and 5 reported experiencing little or no anxiety,
although student 5 had a dip in confidence the week midterm results were returned and stated “I
thought I had known the info pretty well, but I realize I am somewhat confused about a few
things and not very confident”. This student seemed to have regained his or her understanding
for the balance of the semester.
Finally, results for student 6 (high affect, high variability) are not depicted in these
graphs. That student stated in Week 8 “I’m feeling that I should drop the class, but I need it for
my program of study.” The student apparently did drop the class. This is a prime example of
the kind of scenarios we hope to help prevent. These findings all suggest that real-time data
collection and evaluation might allow instructors to identify at the start of the semester students
who would benefit from intervention efforts. Additionally, during the semester, instructors could
identify points at which students are developing problems that might be amenable to
intervention.
Another puzzling pattern in the data that merits further consideration was that some
students scoring high on anxiety and negative attitude made no comments at all. This absence
of open-ended comments from a subset of highly anxious students might suggest students’
reticence to reveal details about their affect or a lack of awareness about their affect.
Graduate Statistics Education
23

Conclusions

The present study expands the statistics education knowledge base with respect to the
role changing attitudes and feelings play in the formation of student anxiety and negative
attitudes. Multidimensional constructs, like statistics anxiety and attitudes toward statistics,
require multiple measures. The present study contributes to the “…emerging field of the
assessment of attitudes in statistics education” (Gal & Ginsburg, 1994, p. 10) by assessing the
utility of instrumentation for the measurement of changes in attitudes and anxiety. The data
clearly indicate both the utility of a simple measure such as that provided by the Affect Check
and the need for additional research to provide support for graduate students with statistics
anxiety.
Graduate Statistics Education
24

References

Blalock, H. M. (1987). Some general goals in teaching statistics. Teaching Sociology, 15, 164-
172.
Caine, R. D., Centa, D., Doroff, C., Horowitz, J. H., & Wisenbaker, V. (1978). Statistics from
who? Teaching Sociology, 6, pp. 37-46.
Cruise, R. J., Cash, R. W. & Bolton, D. L. (1985). Development and validation of an instrument
to measure statistical anxiety. Reprinted from the 1985 Statistical Education Section,
Proceedings of the American Statistical Association.
Dauphinee, T. L., Schau, C., & Stevens, J. J. (1997). Survey of Attitudes Toward Statistics:
Factor structure and factorial invariance for women and men. Structural Equation
Modeling, 4, 129-141.
Elmore, P. B., Lewis, E. L., & Bay, M. L. G. (1993, April). Statistics achievement: A function of
attitudes and related experience. Paper presented at the annual meeting of the
American Educational Research Association, Atlanta, GA.
Fitzgerald, S. M., Jurs, S., & Hudson, L. M. (1996). A model predicting statistics achievement
among graduate students. College Student Journal, 30, 361-366.
Gal, I, & Ginsburg, L. (1994). The role of beliefs and attitudes in learning statistics: Towards an
assessment framework. Journal of Statistics Education [On-line serial], 2(2). Available E-
mail: archive@jse.stat.ncsu.edu. Message: send jse/v2n2/gal.
Gal, I, Ginsburg, L., & Schau, C. (1997). Monitoring attitudes and beliefs in statistics education.
In I. Gal & J. B. Garfield (Eds.), The assessment challenge in statistics education (pp.
37-51). Netherlands: IOS Press.
Gaydosh, L. R. (1990). Syllabi and instructional materials for social statistics. Washington, DC,
American Sociological Association.
Lalonde, R. N., & Gardner, R. C. (1993). Statistics as a second language? A model for
predicting performance in psychology students. Canadian Journal of Behavioral Science,
25, 108-125.
Lundgren, T. D., & Fawcett, R. (1980). Statistics from statisticians. Teaching Sociology, 7,
191-201.
McLeod, D. B. (1992). Research on affect in mathematics education: A reconceptualization. in
D. A. Grouws (Ed.), Handbook of research on mathematics teaching and learning. NY:
Macmillan, p. 575 – 596.
Graduate Statistics Education
25

Schacht, S., & Stewart, B. J. (1990). What's funny about statistics? A technique for reducing
student anxiety. Teaching Sociology, 18, 52-56.
Schacht, S., & Stewart, B. J. (1991). What's funny about statistics? Interactive/user-friendly
gimmicks for teaching statistics. Teaching Sociology, 20, 329-332.
Schau, C., Schau, D., Del Vecchio, A., & Stevens , J. (1995, Copyright pending) Survey of
Attitudes Toward Statistics.
Onwuegbuzie, A. J. (2003). Modeling statistics achievement among graduate students.
Educational and Psychological Measurement, 63,1020-1038.
Onwuegbuzie, A. J., & Seaman, M. (1995). The effect of time and anxiety on statistics
achievement. Journal of Experimental Psychology, 63, 115-124.
Onwuegbuzie, A. J., & Wilson, V. A. (2003). Statistics anxiety: Nature, etiology, antecedents,
effects, and treatments: A comprehensive review of the literature. Teaching in Higher
Education, 8, 195-209.
Schau, C. (2003, August). Students’ attitudes: The “Other” important outcome in statistics
education. Paper presented at the Joint Statistics Meetings, San Francisco, CA.
Schau, C., Stevens, J., Dauphinee, T. L., & Del Vecchio, A. (1995). The development and
validation of the Survey of Attitudes Toward Statistics. Educational and Psychological
Measurement, 55, 868-875.
Sgoutas-Emch, S.A. & Johnson, C.J. (1998) Is journal writing an effective method of reducing
anxiety towards statistics? Journal of Instructional Psychology,25, 49-57.
Tobias, S. (1994). Overcoming math anxiety. New York: W. W. Norton and Company.
Watson, F. S., Kromrey, J. D., Lang, T., Hess, M. R., Hogarty, K. Y. & Dedrick, R. F. (2003,
April). Multifaceted Foci: The Antecedents of Statistics Anxiety and Negative Attitudes
Toward Statistics. Paper presented at the annual meeting of the American Educational
Research Association, Chicago.
Watson, F. S., Lang, T. R., Kromrey, J. D., Ferron, J. M., Hess, M. R. & Hogarty, K. Y. (2003,
April). An Assessment Blueprint for EncStat: A Statistics Anxiety Intervention Program.
Paper presented at the annual meeting of the American Educational Research
Association, Chicago.
Graduate Statistics Education
26

Watson, F. S., Kromrey, J. D., Ferron, J. M., Dedrick, R. F., Hogarty, K. Y., Lang, T. R. & Hess,
M. R. (2003, February). Development and Initial Validation of the Encouraging Statistics
Professor Scale. Paper presented at the annual meeting of the Eastern Educational
Research Association, Hilton Head.
Watson, F. S., Kromrey, J. D. & Hess, M. R. (2003, February). Toward a Conceptual Model for
Statistics Anxiety Intervention. Paper presented at the annual meeting of the Eastern
Educational Research Association, Hilton Head.
Wilson, V. A. (1999) Student response to a systematic program of anxiety-reducing strategies in
a graduate-level introductory educational research course. ERIC ED 430 996.
Zanakis, S. H., & Valenza, E. R. (1997). Student anxiety and attitudes in business statistics.
Journal of Education for Business, 72(5), 10-16.
Zeidner, M. (1991). Statistics and mathematics anxiety in social science students - some
interesting parallels. British Journal of Educational Psychology, 61, 319-328.
Zimmer, J. C., & Fuller, D. K. (1996, November). Factors affecting undergraduate performance
in statistics: a review of literature. Paper presented at the annual meeting of the Mid-
South Educational Research Association, Tuscaloosa, Alabama.
Graduate Statistics Education
27

Acknowledgements

This work was supported, in part, by the University of South Florida and the award
program Innovative Teaching Grants: Support for Faculty Creativity in Teaching. The opinions
expressed are those of the authors and do not reflect the views of the University of South
Florida.
Graduate Statistics Education
28
Feelings about Statistics W inter 2004 Date: _____________ Research ID#________

This instrument contains pairs of words that are opposite in meaning. M ark the space between each
pair of words that is closest to your feelings related to learning statistics at this time. Remember that
there are no right or wrong answers. If at any point you feel uncomfortable you may discontinue.

EXAM PLES

If you feel that learning statistics is probably useful, you would mark a space near the word useful.

Useful ___ : ___ : ___ : ___ : ___ : ___ : ___ Useless

If you feel that learning statistics is definitely useless, you would mark the space nearest the word useless.

Useful ___ : ___ : ___ : ___ : ___ : ___ : ___ Useless

Ashamed ____ : ____ : ____ : ____ : ____ : ____ : ____ Proud


Dumb ____ : ____ : ____ : ____ : ____ : ____ : ____ Smart
Encouraged ____ : ____ : ____ : ____ : ____ : ____ : ____ Discouraged
Clear ____ : ____ : ____ : ____ : ____ : ____ : ____ Confused
Mad ____ : ____ : ____ : ____ : ____ : ____ : ____ Pleased
Certain ____ : ____ : ____ : ____ : ____ : ____ : ____ Lost
Calm ____ : ____ : ____ : ____ : ____ : ____ : ____ Anxious
Frustrated ____ : ____ : ____ : ____ : ____ : ____ : ____ Satisfied
Peaceful ____ : ____ : ____ : ____ : ____ : ____ : ____ Irritated
Sad ____ : ____ : ____ : ____ : ____ : ____ : ____ Happy
Depressed ____ : ____ : ____ : ____ : ____ : ____ : ____ Excited
Tranquil ____ : ____ : ____ : ____ : ____ : ____ : ____ Worried
Fascinated ____ : ____ : ____ : ____ : ____ : ____ : ____ Indifferent
Attracted ____ : ____ : ____ : ____ : ____ : ____ : ____ Disgusted
Interested ____ : ____ : ____ : ____ : ____ : ____ : ____ Bored
Confident ____ : ____ : ____ : ____ : ____ : ____ : ____ Afraid
Miserable ____ : ____ : ____ : ____ : ____ : ____ : ____ Delighted
Clever ____ : ____ : ____ : ____ : ____ : ____ : ____ Foolish

Please write a couple of sentences explaining why you feel as you have indicated. If you’ve experienced
something especially stressful this week or today, make a note of that as well.
__________________________________________________________________________________________________
__________________________________________________________________________________________________
__________________________________________________________________________________________________

View publication stats

You might also like