Advances in Health Sciences Education 7: 63–80, 2002.
© 2002 Kluwer Academic Publishers. Printed in the Netherlands.
Measuring Self-assessment: Current State of the Art
MYLÈNE WARD1, LARRY GRUPPEN2and GLENN REGEHR3∗
1Department of Surgery, University of Toronto, Toronto, Ontario, Canada;2Department of Medical
Education, University of Michigan, Ann Arbor, Michigan, USA;3Department of Psychiatry,
University of Toronto, Toronto, Ontario, Canada (∗author for correspondence; Centre for Research
in Education, University Health Network, 200 Elizabeth St., 1 Eaton South, Toronto, Ontario M5G
2C4, Canada. E-mail: email@example.com)
Abstract. The competent physician pursues lifelong learning through the recognition of deficien-
cies and the formulation of appropriate learning goals. Despite the accepted theoretical value of
self-assessment, studies have consistently shown that the accuracy of self-assessment is poor. This
paper examines themethodological issues that plague themeasurement of self-assessment abilityand
presents several strategies that address these methodological problems within the current paradigm.
In addition, the article proposes an alternative conceptualization of self-assessment and describes its
associated methods. The conclusions of prior research in this domain must be re-examined in light
of the common pitfalls encountered in the design of the studies and the analyses of the data. Future
efforts to elucidate self-assessment phenomena need to consider the implications of this review.
Key words: education, educational measurement, medical, self-assessment, self-evaluation
As with any profession that operates under the principles of self-regulation
and autonomy, the competent physician must be a self-directed, lifelong learner
(Moore and Cordes, 1992). The first step in this process is the diagnosis of one’s
own learning needs, which enables the formulation of appropriate learning goals
(Spencer, 1999). Therefore, the ability to accurately assess one’s strengths and
weaknesses is critical to the enterprise of lifelong learning (Gordon, 1992). Despite
this theoretical argument for the critical importance of self-assessment in the
professions, the conclusions drawn from research regarding professionals’ ability
to self-assess are mixed at best.
Two previous reviews offer a comprehensive summary of findings in the self-
assessment literature. In 1989, Falchikov and Boud published a meta-analysis of
quantitative self-assessment studies in higher education. They reported the results
of forty-four studies in a variety of subject areas, including medicine, guidance
counseling, law, engineering, behavioral science, psychology, and dietetics. Corre-
lations between self-assessed and external measures of performance ranged from
MYL`ENE WARD ET AL.
–0.05 to 0.82, with a mean correlation of 0.39, suggesting that, on average,
self-assessors were poor to moderate judges of their performance. Subsequently,
Gordon (1991) published a literature review of eighteen self-assessment studies in
the health professions. Studies of self-assessment of factual knowledge reported
correlations in the range of 0.02 to 0.65. The accuracy of global self-assessments
based on an extended period of performance was even worse, with the highest
correlation reported among the six studies in this category being 0.32. Both reviews
suggest that the measurement of self-assessment often yields less than promising
Since self-assessment is fundamental to the concept of self-directed learning
and the maintenance of professional competence, educators find it troubling that
researchers who have attempted to establish the accuracy of self-assessment have
often observed incongruities between self-evaluations and external measures of
achievement. But researchers’ conclusions about the accuracy of self-assessment
have been tempered by multiple considerations. In particular, the measurement
of self-assessment encounters several methodological problems. This may limit
the literature’s capacity to support conclusions about self-assessment ability. Our
objective in this paper is to examine the quality of the measurement of self-
assessment. We begin by providing a synopsis of the methodologies employed
in the self-assessment literature. We then examine the problems inherent in the
predominant methodological approach. In the last section, we present several
strategies that address these methodological issues. In addition, we propose
an alternative conceptualization of self-assessment and describe its associated
Summary of the Literature: Methodologies
In order to investigate approaches to the measurement of self-assessment, we
conducted a review of the methods employed in the self-assessment literature.
We included 41 of the studies reviewed by Falchikov and Boud (1989), elim-
inating 3 unpublished studies. Of the 18 studies identified by Gordon (1991),
six appeared in the Falchikov and Boud (1989) meta-analysis, three did not
provide a quantitative measure of self-assessment accuracy, and three studies
were excluded since they could not be located, leaving a total of 6 unique
articles. Additional studies published since 1990 were identified by a computerized
literature search of the MEDLINE, CINAHL, Education Resources Information
Center, and PSYCHINFO databases. These searches used combinations of the
key words: self-assessment, self-evaluation, self-concept, higher education, and
medical education. Bibliographies were searched for relevant articles. We included
all studies if they met the following criteria: a) a study population of individuals in
higher education and, b) a quantitative measurement of self-assessment accuracy.
A total of 67 studies (41 from Falchikov and Boud, 6 from Gordon, and 20 new
studies) were included in this analysis. A review of the methodologies employed
in each indicates that the measurement of self-assessment accuracy is approached
in a similar fashion across a range of disciplines in higher education and the health
professions. Table I presents a classification of studies by method.
The most common methodology (41 studies) used to evaluate self-assessment
involves correlational analyses. In this common design, aself-assessment score and
a score based on some external measure (often an expert evaluation) is generated
for each individual in the group. Across the group the self-ratings are correlated
with the expert ratings to obtain a single numerical value for the group. This
numerical value is interpreted as a measure of the group’s self-assessment ability.
The assumption is that if all students are effectively evaluating their ability relative
to their peers, the collection of student self-assessments should correlate well with
the external measure. Conversely, if the correlation is low, it is assumed that the
group as a whole is poor in determining the quality of their performance relative to
As a separate but related methodology, some studies (N = 16) reported the
proportion of self-ratings that corresponded with expert ratings. Falchikov and
Boud (1989) observed that the definition of ‘agreement’ is not consistent across
this group of studies. Some require identical ratings on a 100 point scale, but
most studies adopt a slightly more liberal definition of ‘agreement’, whereby self-
ratings are deemed equivalent to expert ratings if they assign matching scores on
a Likert-type scale. Both ‘percentage agreement’ and ‘correlation’ are techniques
that are used to describe the relationship between two variables, in this case self
and expert assessment. Both approaches generate conclusions for the group with
respect to self-assessment ability, as defined by close agreement between scores
generated by trainees and experts. Thus, while somewhat different in presentation,
the correlational and agreement designs share many of the methodological issues
that we will be discussing. A total of 55 studies (82% of studies identified in the
review) employed one or both of these methods.
A third methodological approach that appears frequently in the self-assessment
literature (27 studies) involves the direct comparison of the absolute values of
self-ratings and some external standard. This approach provides information about
whether student self-ratings tend to match the external standard, or whether the
self-assessors tend to overrate or underrate themselves. As with the other meth-
odologies, the intent generally involves comparing the self-assessing group as a
whole to the external standard by comparing means.
It is predominantly on the basis of these three quantitative methodologies that
claims about self-assessment ability have been made. With the exception of a small
number of articles that we have published recently, every paper making quantita-
tive evaluations of self-assessment used one or more these techniques exclusively.
Yet there has been relatively little reflection on the nature of these methodolo-
gical paradigms and the assumptions under which they function. The following
section will describe some of the methodological pitfalls that face these models for
evaluating self-assessment ability.
MYL`ENE WARD ET AL.
Table I. Summary of methodologies employed in quantitative studies of self-assessment∗
Health Professions Antonelli 1997 Cochran 1980
Subtotal 37 studies 24 (64.9%)
Pitishkin-Potanich Israelite 1983
Subtotal 30 studies 17 (56.7%)
TOTAL 62 studies 41 studies (66.1%) 16 studies (25.8%) 27 studies (43.5%) –
9 (30.0%) 12 (40.0%)–
*Listed by first author only for ease of presentation.
This discussion begins with a focus on the predominant approach to the measure-
ment of self-assessment, the correlational model. As elaborated above, correla-
tional analyses are used to determine whether a group of individuals are accurate
judges of their own performance, as compared to an external measure of perfor-
mance. A weak or absent relationship between self-ratings and the external
measure suggests that the individuals in this group are poor self-assessors. We have
identified three methodological issues with the correlational design that cast doubt
on this conclusion: problems with the gold standard, problems with differential use
of the scale by participants, and problems with group level analyses.
PROBLEMS WITH THE ‘GOLD STANDARD’
First, it is worth noting that expert (e.g. medical faculty, clinical preceptors) eval-
uations are a common source of the ‘objective’ measure against which student
self-assessments are compared. For evaluation of a specific task, experts are chosen
to observe and judge the performance. For the evaluation of performance during a
clinical rotation, preceptors are asked to provide a global assessment. Thus, all
studies that compare self-assessment to expert assessment and draw a conclusion
about self-assessment ability are making the same basic assumption: expert judge-
ment isthe gold standard by which tomeasure all aspects of clinical competence. In
other words, faculty and supervisors provide the ‘true rating’ (Palmer et al., 1985).
As many authors recognize, the reliability claims of this gold standard are
suspect (Abrams and Kelly, 1974; Arnold et al., 1985; Bergee, 1997; Falchikov
and Boud, 1989; Farnill et al., 1997; Harrington et al., 1997; Hay, 1995; Johnson
and Cujec, 1998; Kolm and Verhulst, 1987; Martin et al., 1998; Regehr et al.,
1996). However, only a handful of studies report the reliability of the gold standard,
and for these few studies, there is evidence of inconsistency among expert raters.
Harrington et al. (1997) applied a relative ranking model to the global assessment
of performance over an orthopedic residency rotation and reported a mean inter-
rater reliability of only 0.27. The authors hypothesized that study designs using
longitudinal evaluations are particularly prone torater unreliability. Overthecourse
of a residency or clerkship rotation, each preceptor observes trainees in different
clinical settings (e.g. at rounds, in the OR, on the ward) performing different clin-
ical responsibilities. Evaluations take place at the end of a lengthy rotation and are
thus limited by recall.
By contrast, expert raters are far more likely to agree given the chance to
evaluate a short, structured, and relatively simple task. The study by Regehr and
colleagues (1996) lends support to this argument. The relative ranking model was
applied to the assessment of a standardized patient interview. The mean inter-rater
reliability was much higher, at 0.70. Martin et al. (1998) also reported high inter-
reliability between two communication experts (Cronbach’s alpha = 0.94) who
viewed thevideotapes ofresidents performing astandardized patient interview. The
MYL`ENE WARD ET AL.
question of expert reliability is not limited to medical education, as Bergee (1997)
reported mixed inter-rater reliabilities (coefficient alpha 0.23 to 0.93) for evalu-
ations of applied music performances. Thus, the idea of an infallible ‘expert rater’,
the traditional gold standard, should be viewed with a degree of skepticism and
this places important limitations on the interpretation of low correlations between
expert and self ratings.
Furthermore, the validity of the ‘gold standard’ in these studies may be simi-
larly questioned. For the purposes of self-assessment evaluation, the gold standard
scores must satisfy two related conditions regarding validity. First, experts must
provide a valid measure of the dimension that they claim to evaluate. This issue
is addressed in an extensive body of literature on global rating scales (Gray, 1996;
Keynan et al., 1987; Turnbull et al., 1998), however as just one example from the
self-assessment literature, Risucci and colleagues (1989) reported that the mean
supervisor ratings of cognitive achievement correlated only moderately with the
total raw score on the in-training examination of the American Board of Surgery
(ABSITE) (r = 0.55, p < 0.01). In this study, cognitive qualities accounted for
30% of the variance in faculty ratings, leaving 70% of the variance unexplained.
With respect to the evaluation of the non-cognitive areas of competence, such as
professionalism and communication skills, there are no real ‘gold standards’ (Gray,
1996; Keynan et al., 1987; Turnbull et al., 1998). Therefore, the validity of expert
assessment remains elusive and uncertain.
However, even if this first validity concern is addressed, the validity of the
expert rater presupposes a second condition: that the expert’s concept of the
dimensions that are relevant to a ‘good’ performance are reasonable and appro-
priate. For example, several studies suggest that clinical supervisor assessments
reflect an emphasis on cognitive achievement while medical students emphasize
non-cognitive abilities (Arnold et al., 1985; Kegel-Flom, 1975). Under these
circumstances, a measure of the relationship between self-ratings and supervisor
ratings would likely generate a low correlation, which, typically, implies inaccurate
self-assessment. Arnold and colleagues (1985) argue that ‘the validity of self-
evaluations can not be established solely against criterion measures of cognitive
achievement or faculty assessments that contain a cognitive emphasis’. In other
words, even if one assumes that clinical supervisors are in fact accurate (valid)
judges of cognitive achievement, any conclusion with respect to the accuracy of
self-assessment presumes that experts are providing a fair measure of clinical
performance by focusing on cognitive achievement.
In summary, it appears that there are significant problems with the assump-
tion that experts’ assessments qualify as the gold standard against which to judge
self-assessment. Few efforts have been made to study the reliability of this gold
standard, when this reliability functions as a theoretical upper limit on the corre-
lation of the students with the gold standard. In addition, the validity claims of
the gold standard must be further examined. There is good reason to believe that
experts are not measuring what they intend to measure, or that what they intend to
measure is necessarily what is really important.
PROBLEMS WITH DIFFERENTIAL USE OF THE SCALE AMONG STUDENTS
The correlational approach to evaluating self-assessment also rests on the notion
that it is appropriate to consider a group of individual self-assessment scores as
a set of coherent scores. This notion makes two assumptions. If either of these
assumptions is incorrect, the paradigm is severely limited in its ability to produce
a true measure of self-assessment ability in the population being assessed.
First, in order to treat the individual self-assessments as a set of coherent scores,
we must assume that the group of trainees are all evaluating themselves by tapping
into the same aspect of performance. As mentioned earlier, differences between
rater types (self, expert) have been explored, but no one has asked whether or not
differences exist within the group of students. It seems improbable that students
or residents constitute a monolithic group, that self-assessments would reflect a
common understanding of the dimensions of performance. Thus, the finding that
students tend to focus on non-cognitive aspects of performance (Arnold et al.,
1985; Risucci et al., 1989) should not be interpreted as a claim that all students do
so. Such an interpretation likely over-generalizes the consistency among students.
If so, each individual student might be accurately evaluating the dimensions that
he or she chooses to evaluate, but there is no reason to believe that the numbers
generated from such accurate self-assessments would have any coherent collective
However, it would not be sufficient to show that all individuals are measuring
themselves based on the same criteria. This research design assumes that all indi-
viduals measure these dimensions of competence in a consistent manner, and yet,
even the best scale is subject to interpretation. As an extreme case, consider the
following theoretical example (Table II). Four students (A, B, C, D) evaluated both
themselves and each other member of the group on a 7 point scale. An expert used
an identical instrument to evaluate the students. This table shows that each student
gave his or her own performance a 5 out of 7. Thus, if we look only at the students’
self-assessment scores (ignoring, as the paradigm usually does, the scores that the
students would give to their colleagues), only student B is a good judge of her
own performance, relative to the expert’s assessment. Student A underestimated
her performance and Students C and D overestimated their performances. On the
surface, the group as a whole is self-assessing very poorly. Since all four students
gave themselves 5 out of 7, the correlation of the students’ self-assessments with
the expert’s assessments is zero. However, a closer look at the self-evaluations
in the context of the students’ peer evaluations shows that, in fact, self and peer
assessments were all perfectly accurate (assuming the experts’ assessments qualify
as the gold standard). Student A, rated the highest by the experts, did recognize
her performance was superior to her peers’; however she failed to use the upper-
MYL`ENE WARD ET AL.
Table II. Self, Peer, and Expert Assessments: an example to illustrate the impact
of differential use of the scale.
*SAdenotes scores generated by Student A, SBdenotes scores by Student B, etc.
most values of the scale. Student D, using a very narrow range of the scale, also
demonstrated an understanding that his performance was poor relative to the group.
This example illustrates four different interpretations of a score of 5 out of 7. It
could be considered superior, average, or poor, depending on the individual and
the context. This example also shows that inconsistent use of the scale among
students attenuates the correlation of expert and self-ratings, regardless of the
group’s self-assessment ability.
By themselves, numbers are meaningless. The way in which each rater (self
and expert) interprets and applies the scale must be defined before the numbers can
be meaningfully combined and conclusions can be drawn about the accuracy of
PROBLEMS WITH GROUP-LEVEL ANALYSES
Evenassuming that‘experts’ arereliable and valid assessors of trainee performance
and that students are measuring the same dimensions of performance using the
scale in the same way, a conclusion about the accuracy of self-assessment based on
the group correlation remains potentially problematic. This prominent methodolo-
gical approach rests on yet another assumption: that every individual in the group
is equal in terms of self-assessment ability.
Group-level correlation between self and expert ratings can make claims only
at the level of the group. Either the correlation is low, suggesting that the group
as a whole cannot self-assess effectively, or the correlation is high, suggesting that
the group, as a whole, can self-assess accurately. Of course, the question of what
is an acceptable correlation as a measure of self-assessment ability is likely to be a
matter of further controversy. However, whether self-assessment is determined to
be poor, moderate or good, the determination of this fact with a single correlation
is of limited value. Further analyses and interpretation of a potentially complex
phenomenon are difficult at best when we cannot understand individual variation
within the group.
Figure 1. Self vs. Expert Evaluation: an example to illustrate the impact of a few ‘poor
self-assessors’ on the overall correlation of self and expert evaluation scores.
Not only is individual variation within the group masked in this methodological
paradigm, variation may also have a significant negative impact on overall group
correlation. A few outliers could easily spoil the measurement of self-assessment
ability for the whole group. The following hypothetical example illustrates this
problem (Figure 1). The scatterplot shows that most individuals in this group (100
of 120) made accurate self-assessments of their performance. The correlation of
these 100 scores with the gold standard is perfect, r = 1.0. However, ten individuals
were above average but grossly underestimated their performances, and conversely,
ten individuals were below average yet overestimated their performances. The
correlation between students and experts for the whole group was only 0.236.
These few outliers exercised a hugely negative impact on the overall correlation,
even though the vast majority of individuals in the group were ‘perfect’ self-
assessors. Although this example has been made extreme to demonstrate the point,
it is nonetheless the case that group-level analyses fail to reflect an appreciation for
group heterogeneity and therefore may skew the conclusions inappropriately.
PROBLEMS WITH THE DIRECT COMPARISON OF THE SCORES
Almost 50% of the studies reviewed make a direct comparison of student self-
ratings to scores from some external criterion instead of or in addition to using
the correlational method. This form of comparison is subject to many of the same
methodological criticisms that are inherent in the correlational paradigm. Certainly,
the assumption that the gold standard is a reliable and valid measure of the indi-
vidual remains unsubstantiated. This model also assumes that all individuals are
evaluating the same dimensions and using the scale in the same manner as the
expert (as demonstrated in Table II). Further, the comparison of group means as
generated by the participants and some external standard clearly hides individual
MYL`ENE WARD ET AL.
differences (every member of the group may be wildly inaccurate but the group
mean could be identical to the mean of the external standard).
It is worth noting that two studies that directly compare scores draw conclusions
about subgroups, often discovering that the high achievers tend to underestimate
their performance relative to the gold standard, and underachievers tend to over-
estimate their performance relative to the gold standard (Arnold et al., 1985;
Wooliscroft et al., 1993). These findings, however, may simply be a restatement of
the fact that there is a weak correlation between self-assessment scores and scores
generated by the gold standard. In statistical terms, a correlation of less than 1
necessarily implies regression to the mean, such that higher scores tend downward
and lower scores tend upward. There sometimes appears to be a suggestion that
overestimation by underachievers and underestimation by overachievers has motiv-
ational or ego defensive roots (e.g. Gordon, 1992). It is important to remember,
however, that the same phenomenon would occur if a random number generator
produced the self-assessment scores.
Dealing with the Pitfalls
We have described several major methodological issues associated with the tradi-
tional approach to the study of self-assessment. We believe that these problems
are sufficiently severe to raise questions regarding the validity of the paradigm
as it is usually instantiated, and to raise questions regarding the legitimacy of the
general conclusions that might be drawn from the literature. It may be that some
of these problems are insurmountable. However, several can be ameliorated, if
not eliminated. In this section, we present several strategies for dealing with at
least some of the ‘pitfalls’ that we have identified above. These strategies fall into
two general categories. The first involves maintaining the existing paradigm and
making alterations that minimize some of its obvious methodological weaknesses.
The second involves the presentation of an alternative paradigm that we have been
exploring for several years.
SOLUTIONS WITHIN THE PARADIGM
‘Solutions within the paradigm’ refers to strategies that tackle the issues outlined
in the previous section: Is our gold standard infallible and how would we know?
Are students uniform in their evaluation of performance and their use of the scale?
Is group-level analysis an oversimplification of self-assessment phenomena?
Correcting for the unreliability of expert ratings
It was suggested earlier that ‘expert’ evaluations of students might be neither valid
nor reliable measures against which to compare student self-evaluations. Although
we have no claims to more valid alternatives to evaluations provided by experts
(there are no true gold standards in educational evaluation), it is relatively easy to
address issues regarding the reliability of these expert evaluations. The best way to
improve the gold standard is to optimize reliability through the use of multiple
expert raters. Agreement between raters will never be perfect, however expert
unreliability can be taken into account when subsequently calculating the corre-
lation between self and expert ratings. Regehr and colleagues (1996), for example,
describe the use of the ‘correction for attenuation’ formula, whereby the student-
expert correlation is divided by the square root of the expert inter-rater reliability.
In this particular study, correlation between students’ and experts’ scores was 0.43.
But when the student/expert correlations were corrected for the attenuation due to
experts’ unreliability, the mean self-assessment score increased to 0.58.
Stabilizing students’ use of the scale
As with experts, we questioned the consistency with which the target group of
students was evaluating the same criteria and using the scale consistently. Perhaps
the best solution to the possibility that students are evaluating different dimensions
of performance was enacted by Henbest and Fehrsen (1985), who asked medical
students to create their own criteria for evaluation at the beginning of a family
medicine rotation. When these same criteria were used to evaluate the students
at the end of the rotation, a high positive correlation was found between self and
faculty ratings (0.74; p < 0.01).
Problems with differential use of the scale among students can be addressed
through the provision of explicit anchors for the evaluation criteria. Martin et al.
(1998), for example, attempted to resolve this methodological issue by providing
residents with exposure to ‘benchmarks’ of performance. Family practice residents
performed an interview with a standardized patient and rated their performance.
They were then shown videotapes of four performances of the same scenario that
varied in quality and were asked to rate the four performances. To account for
inconsistent use of the scale, the resident’s self-assessment scores were rescaled
based on their evaluations of the videotaped performances. The mean and standard
deviation of each resident’s scores for the videotapes were calculated, and the
resident’s self-evaluation was expressed as a z score relative to this mean and
standard deviation. Thus the scale was standardized across residents through the
calculation of z scores, an expression of where each resident felt they stood relative
to the four performances. In this study, rescaling the residents’ self-assessments
into z scores did not improve the resident-expert assessment correlations (although,
see Hodges et al., 2001, for a reanalysis of these data). However, this should not
deter future efforts to apply these strategies or develop new innovative ways to
ensure consistent use of the scale.
It may not be necessary, however, to use such an elaborate method for rescaling
students’ scores to ensure consistency in their use of the scale. Another possible
strategy is to provide examples of videotaped performances with predetermined
scores. These explicit anchors would likely help to standardize the scale.
MYL`ENE WARD ET AL.
Comparing self-assessment to peer assessment
It is worth acknowledging that the study of peer assessment has traditionally
employed the same methodological paradigm as we have been describing and,
thus, falls prey to the same potential pitfalls. Yet, several studies show that peer
assessment is more accurate than self-assessment (Bergee, 1997; Falchikov and
Goldfinch, 2000; Linn et al., 1976; Martin et al., 1998; Morton and MacBeth, 1977;
Risucci et al., 1989). These studies lend support to the argument that individuals
can identify good and bad performances, but are unable or unwilling to apply
the same standards to their own performance. This use of peer assessment as a
‘control’ condition to evaluate self assessment may be an interesting alternative
method to deal with the pitfalls that we have been describing. Rather than elim-
inating or ameliorating the pitfalls, we might control for them instead. Of course,
correlations are notoriously unstable and demonstrating difference between two
correlations (self assessment vs. peer assessment) may be difficult and may require
more power than most studies can feasibly produce with the limited number of
subjects available. In principle, however, this is certainly a useful option when the
power is available.
AN INTRAINDIVIDUAL APPROACH
Rather than attempt to deal with the problems inherent to the ‘traditional’
paradigm, it is also possible to step outside the paradigm. We can search for new
ways to conceptualize self-assessment and develop methods associated with this
The development of a new framework for the study of self-assessment has been
the recent focus of research in this domain. Regehr et al. (1996), for example,
offered a reconceptualization of self-assessment that focused not on the indi-
vidual’s ability to rate herself relative to her peers, but on her ability to identify her
ownstrengths and weaknesses relative to each other. Theysuggested that the ability
to identify areas of performance that require the greatest degree of improvement
would lend greater efficiency to self-directed learning efforts. Consistent with this
framework, Gruppen and colleagues (1997, 2000; Fitzgerald et al., 2000) make a
distinction between the conception of self-assessment as an interindividual process
or as an intraindividual process. The study of self-assessment according to the
common methodological paradigm is an interindividual process. The subject asks,
‘How good am I?’ and generates a self-rating based on his perception of his ability
relative to others or to some ideal standard. The intraindividual process is more
consistent with the model suggested by Regehr et al. (1996) where self-assessment
is defined as the ability of each individual to identify his or her own relative
strengths and weaknesses. In other words, the self-assessor asks the question ‘What
aspects of my performance need the most work? Which aspects need the least
Methodologically, this new intraindividual perspective requires multiple self-
assessments from each subject in order to calculate individualized estimates of
self-assessment accuracy. This is accomplished through the evaluation of multiple
tasks or multiple aspects of a single performance.
An example of evaluating intraindividual self-assessment across a set of related
tasks is provided in a study by Fitzgerald et al. (2000). Medical students completed
an OSCE-type examination with ten stations (e.g. breast examination, EKG inter-
pretation). The students estimated their scores on each station. Self-assessment
accuracy was determined for each subject through a correlation of each subject’s
self-ratings with the expert ratings’ provided for that individual over the ten
stations. Thus, each student’s correlation with the expert over the ten stations
reflects the extent to which that student was able to identify the stations at which
he performed well and the stations at which he performed poorly compared to his
own performance at the other stations.
The application of the intraindividual approach to the measurement of self-
assessment across a set of skills within a single task was illustrated in a study
by Gruppen et al. (1997). Students were evaluated on their performance of a
standardized patient (SP) interview. The evaluation form contained seven items.
The correlation coefficient between each student’s self-assessments and the SP’s
assessments was calculated over the seven items. Again, each individual’s correla-
tion with the SP reflects her ability to identify the dimensions of performance on
which she was relatively effective or ineffective relative to her performance on the
The major limitation of this new paradigm as described in the studies above
is that it depends heavily on variation of scores. For example, suppose a student
gives all aspects of his performance a 3 out of 5. Even if this represents an accurate
self-assessment of performance (i.e. the student has correctly determined that all
aspects of performance are moderate), the correlation between student and expert
will be zero (because there is no variation in the multiple scores generated by
the student). This may be a significant issue. One of the most common errors in
global assessment is the halo effect, which describes the tendency of a rater to
give similar evaluations to separate aspects of an individual’s performance (Gray,
1996). Therefore, if this approach is employed in the study of self-assessment, it is
important to ensure that subjects and experts are using the full range of the scale
In an effort to circumvent this potential problem, Regehr and colleagues (1996)
proposed a modified Q-sort method, which they termed the ‘relative ranking
model’. The Q-sort method has been used for the study of psychological character-
istics (Bem and Funder, 1974; Block, 1978; Mowrer, 1953). Regehr and colleagues
(1996) applied this technique to the evaluation of communication skills in a psychi-
atric interview. The evaluation form listed ten skills and a horizontal row of ten
boxes. Above box 1 appeared the descriptor ‘needs most work’, above box 10
‘needs least work’, and boxes 5 and 6 were paired with the descriptor ‘mid-level
MYL`ENE WARD ET AL.
skills’. Following a standardized patient encounter, students were asked to identify
skills that belonged at the extremes and center of the scale, and then place the
remaining skills using one item per box. The expert rated the performance using
the same form. A correlation was calculated for each individual student’s rankings
of skills with each of the experts’ rankings.
The authors concluded that the relative ranking model was not only useful as an
alternative measure of self-assessment, but also ideal for the provision of feedback
and could be used for educational purposes. However, as an assessment tool, this
method has limitations. It is still important to study self-assessment as it pertains
to the recognition of overall competence or incompetence, as this is key to safe
practice. The Q-sort technology, however, is entirely intraindividual and therefore
makes no effort to assess overall ability against some relevant criterion. In addi-
tion, this model may generate a somewhat ‘artificial’ hierarchy of strengths and
The importance of self-assessment in education stands undisputed. The health
professions and higher education literature testify to this fact, as reflected in a
sustained interest in this domain for over 30 years. Despite the theoretical value
of self-assessment, the traditional measures employed in this literature could lead
to the conclusion that self-assessment ability is poor. However, problems inherent
in the traditional approaches for measuring self-assessment call into question this
verdict on self-assessment.
In this paper, we have proposed several ways to address the methodological
issues that arise within the traditional paradigms for studying self-assessment. For
instance, the use of multiple expert raters can assess the reliability (or lack thereof)
of the gold standard, the problem of differential use of the scale among students
may be addressed by providing methods to ‘calibrate’ their self-evaluations, and
the pitfalls as a cluster can be controlled by comparing self-assessment to peer
assessment in the same individuals. In addition, we have offered one alternative
framework for research in this domain that conceptualizes self-assessment as an
intraindividual (as opposed to interindividual) process. This new framework faces
its own limitations, but it exemplifies the potential that exists to take the study of
self-assessment in new directions. We do not claim to have solved all the problems
inherent in the self-assessment literature. However, we do strongly suggest that the
current status quo is insufficient. Studies that make use of the traditional designs
to study self-assessment without accounting for the potential methodological flaws
inherent in these approaches will not be able to contribute meaningfully to the
self-assessment literature in the future.
MEASURING SELF-ASSESSMENT Download full-text
Abrams, R.G. & Kelley, M.L. (1974). Student self-evaluation in a pediatric-operative technique
course. Journal of Dental Education 38: 385–391.
Antonelli, M.A.S. (1997). Accuracy of second-year medical students’ self-assessment of clinical
skills. Academic Medicine 72(10 Supplement 1): S63–S65.
Arnold, L., Willoughby, T.L. & Calkins, E.V. (1985). Self-evaluation in undergraduate medical
education: A longitudinal perspective. Journal of Medical Education 60: 21–28.
Bem, D.J. & Funder, D.C. (1974). Predicting more of the people more of the time: assessing the
personality of situations. Psychological Review 8: 506–520.
Bergee, M.J. (1997). Relationships among faculty, peer, and self-evaluations of applied perform-
ances. Journal of Research in Music Education 45: 601–612.
Bishop, J.B. (1971). Another look at counselor, client, and supervisor ratings of counselor effective-
ness. Counselor Education and Supervision 10: 319–323.
Block, J. (1978). The Q Sort Method in Personality Assessment and Psychiatric Research. Palo Alto,
CA: Consulting Psychologists Press.
Boud, D.J., Churches, A.E. & Smith, E.M. (1986). Student self assessment in an engineering design
course: An evaluation. International Journal of Applied Engineering Education 2(2): 83–90.
Boud, D.J. & Tyree, A.L. (1979). Self and peer assessment in professional education: A preliminary
study in law. Journal of the Society of Public Teachers of Law 15(1): 65–74.
Burke, R.J. (1969). Somepreliminarydata on theuseof self-evaluations and peer-ratingsinassigning
university course grades. Journal of Educational Research 62(10): 444–4448.
Calhoun, J.G., Ten Haken, J.D. & Wooliscroft, J.O. (1990). Medical students’ development of self
and peer assessment skills: a longitudinal study. Teaching and Learning in Medicine 2: 25–29.
Calhoun, J.G., Wooliscroft, J.O., Ten Haken J.D., Wolf F.M. & Davis, W.K. (1988). Evaluating
medical student clinical skill performance: Relationships among self, peer, and expert ratings.
Evaluation & the Health Professions 11: 201–212.
Chiu, L.H. (1975). Influence of student teaching on perceived teaching competence. Perceptual and
Motor Skills 40: 872–874.
Cochran, S.B. & Spears, M.C. (1980). Student self-assessment and instructors’ ratings: A compar-
ison. Journal of the American Dietetic Association 76: 253–257.
Coutts, L. & Rogers, J. (1999). Predictors of student self-assessment accuracy during a clinical
performance exam: comparisons between over-estimators and under-estimators of SP-evaluated
performance. Academic Medicine 74(10 Supplement): S128–S130.
Daniel, S.J., Scruggs, R.R. & Grady, J.J. (1990). Accuracy of student self-evaluations of dental
sealants. Journal of Dental Hygiene 64: 339–342.
Das, M., Mpofu, D., Dunn, E. & Lanphear, J.H. (1998). Self and tutor evaluations in problem-based
learning tutorials: Is there a relationship? Medical Education 32: 411–418.
D’Augelli, A.R. (1973). The assessment of interpersonal skills: A comparison of observer, peer and
self ratings. Journal of Community Psychology 1: 177–179.
Davis, J.K. & Rand, D.C. (1980). Self-grading versus instructor grading. Journal of Educational
Research 73(4): 207–211.
Doleys, E.J. & Renzaglia, G.A. (1963). Accuracy of student prediction of college grades. Personnel
and Guidance Journal 41(6): 528–530.
Everett, M.S. (1983). Influence of trait anxiety on self-grading. Educational Directions 8(1): 4–9.
Falchikov, N. (1986). Product comparisons and process benefits of collaborative peer group and self
assessments. Assessment and Evaluation in Higher Education 11(2): 146–166.
Falchikov, N. & Boud, D. (1989). Student self-assessment in higher education: A meta-analysis.
Review of Educational Research 59: 395–430.
Falchikov, N. & Goldfinch, J. (2000). Student peer assessment in higher education: A meta-analysis
comparing peer and teacher marks. Review of Educational Research 70: 287–322.