ArticlePDF Available

Psychometric Evaluation of the PSSUQ Using Data from Five Years of Usability Studies

  • MeasuringU

Abstract and Figures

Factor analysis of Post Study System Usability Questionnaire (PSSUQ) data from 5 years of usability studies (with a heavy emphasis on speech dictation systems) indicated a 3-factor structure consistent with that initially described 10 years ago: factors for System Usefulness, Information Quality, and Interface Quality. Estimated reliabilities (ranging from .83-.96) were also consistent with earlier estimates. Analyses of variance indicated that variables such as the study, developer, stage of development, type of product, and type of evaluation significantly affected PSSUQ scores. Other variables, such as gender and completeness of responses to the questionnaire, did not. Norms derived from this data correlated strongly with norms derived from the original PSSUQ data. The similarity of psychometric properties between the original and this PSSUQ data, despite the passage of time and differences in the types of systems studied, provide evidence of significant generalizability for the questionnaire, supporting its use by practitioners for measuring participant satisfaction with the usability of tested systems.
Content may be subject to copyright.
Psychometric Evaluation of the PSSUQ Using Data
from Five Years of Usability Studies
James R. Lewis
IBM Corporation
Factor analysis of Post Study System Usability Questionnaire (PSSUQ) data from 5
years of usability studies (with a heavy emphasis on speech dictation systems) indi-
cated a 3-factor structure consistent with that initially described 10 years ago: factors for
System Usefulness, Information Quality, and Interface Quality. Estimated reliabilities
(ranging from .83–.96) were also consistent with earlier estimates. Analyses of variance
indicated that variables such as the study, developer, stage of development, type of
product, and type of evaluation significantly affected PSSUQ scores. Other variables,
such as gender and completeness of responses to the questionnaire, did not. Norms de-
rived from this data correlated strongly with norms derived from the original PSSUQ
data. The similarity of psychometric properties between the original and this PSSUQ
data,despitethepassageoftimeanddifferencesin the types of systems studied, provide
evidence of significant generalizability for the questionnaire, supporting its use by
1.1. Purpose of This Evaluation
The purpose of this evaluation was to investigate the psychometric characteristics
of the Post Study System Usability Questionnaire (PSSUQ) using data from 5 years
of lab-based usability evaluation. The research emphasis at the time of develop-
ment of the PSSUQ was on enterprise-wide and networked office application suites
(Lewis, Henry, & Mack, 1990). Over the last 5 years the majority of our use of the in-
strument has been in the evaluation of speech recognition systems (with a focus on
speech dictation). The primary research question was whether the instrument,
used for research in an area very different from that for the data used in the previ-
ous psychometric evaluations, would exhibit a factor structure, reliability, sensitiv-
ity, and norms consistent with the previous research. Replication of the previous
findings with this new set of data would provide evidence of significant
(3&4), 463–488
Copyright © 2002, Lawrence Erlbaum Associates, Inc.
Requests for reprints should be sent to James R. Lewis, IBM Corporation, 8051 Congress Ave, Suite
2227, Boca Raton, FL 33487. E-mail:
generalizability for the questionnaire, supporting its use by practitioners for mea-
suring participant satisfaction with the usability of tested systems.
1.2. History of the PSSUQ
The PSSUQ is a 19-item instrument designed for the purpose of assessing users’ per-
ceived satisfaction with their computer systems. It has its origin in an internal IBM pro-
ject called SUMS (System Usability MetricS), headed by Suzanne Henry in the late
1980s. The mission of SUMS was to document and validate procedures for measuring
system usability, including performance, usability problems, and user satisfaction.
At that time, there were a few efforts worldwide to develop instruments for the
measurement of user satisfaction with system usability. In particular, the Question-
naire for User Interface Satisfaction (QUIS) at the University of Maryland (Chin,
Diehl, & Norman, 1988), the Computer User Satisfaction Inventory (CUSI;
Kirakowski & Dillon, 1988), and the Software Usability Measurement Inventory
(Kirakowski & Corbett, 1993) at the University College of Cork in Ireland have had
a significant influence on usability engineering practices. The System Usability
Scale (Brooke, 1996) was also developed during the same time period, but because
there has been no peer-reviewed research published on its psychometric proper-
ties, it has been less influential. (See LaLomia and Sidowski, 1990, for a review of
usability questionnaires published before the PSSUQ; and see Lewis, 1995, for a
comparison of the PSSUQ with the QUIS and CUSI.)
At the time we were working on SUMS, however, we did not know about these
projects, so we developed our own standardized usability questionnaire. Ateam of
IBM human factors and usability specialists working on SUMS created a pool of
items hypothesized to relate to usability, and from those items we selected 18 to use
systematically in usability evaluations as an end-of-study questionnaire named the
PSSUQ (Lewis, 1991, 1992b).
In a separate unpublished study of customer perception of usability, a series of
investigations using decision support systems revealed a common set of five sys-
tem characteristics associated with usability by several different user groups (Doug
Antonelli, personal communication, January 5, 1991). The original 18-item PSSUQ
addressed four of these five system characteristics (quick completion of work, ease
of learning, high-quality documentation and online information, and functional
adequacy), but did not address rapid acquisition of productivity. We subsequently
added an item (Item 8) to the PSSUQ to address this system characteristic and rear-
ranged the order of items to correspond with the questionnaire’s factors, producing
the current version with 19 items (see the Appendix).
The development of the Computer System Usability Questionnaire (CSUQ) fol-
lowed the development of the PSSUQ. Its items are identical to those of the PSSUQ ex-
cept that their wording is appropriate for use in field settings or surveys rather than in
a scenario-based usability evaluation, making it, essentially, an alternate form of the
PSSUQ. The primary reason for the initial development of the CSUQ was to use it in a
mailed questionnaire to obtain sufficient data on the PSSUQ–CSUQ items for a legiti-
mate factor analysis. (At the time, the amount of laboratory data collected with the
PSSUQ did not meet the conventional sample size standards for a factor analysis—
464 Lewis
data from at least five participants for each item in the questionnaire.) For a discussion
of this research and comparison of the PSSUQ and CSUQ items, see Lewis (1995).
1.3. Brief Review of Psychometric Theory and Trade-offs Considered in
the Development of the PSSUQ
The primary purpose of this section is to provide a quick review of the basic ele-
ments of standard psychometric practice. The section also includes some discus-
sion of trade-offs considered in the development of the PSSUQ (Lewis, 1999b). In
this review, the term scale typically refers to a composite measurement based on re-
sponses to a number of items (a Likert scale). An item is a statement for which a par-
ticipant selects a level of response. A scale step is an integer number indicating the
participant’s level of response to that item. See the Appendix for an example of an
item with seven scale steps.
Goals of psychometrics.
The goal of psychometrics is to establish the qual-
ity of psychological measures (Nunnally, 1978). Is a measure reliable (consistent)?
Given a reliable measure, is it valid (measures the intended attribute)? Finally, is
the measure appropriately sensitive to experimental manipulations?
Reliability goals.
In psychometrics, reliability is quantified consistency, typi-
cally estimated using coefficient alpha (Nunnally, 1978). Coefficient alpha can
range from 0 (no reliability)to1(perfect reliability). Measures of individual aptitude
(such as IQ tests or college entrance exams) should have a minimum reliability of
.90 (preferably a reliability of .95). For other research or evaluation, measurement
reliability should be at least .70 (Landauer, 1988).
The initial assessments of the PSSUQ and CSUQ scales (Lewis, 1995) produced
reliabilities exceeding .85, indicating their suitability for use in research and evaluation.
Validity goals.
Validity is the measurement of the extent to which a question-
naire measures what it claims to measure. Researchers commonly use the Pearson
correlation coefficient to assess criterion-related validity (the relation between the
measure of interest and a different concurrent or predictive measure). Moderate
correlations (with absolute values as small as .30–.40) are often large enough to jus-
tify the use of psychometric instruments (Nunnally, 1978).
Previous validity assessment of the PSSUQ indicated a significant correlation (r
= .80) with other measures of user satisfaction obtained at the completion of each
scenario and a significant correlation (r= –.40) with the measure of successful sce-
nario completion (Lewis, 1995).
Sensitivity goals.
A questionnaire that is reliable and valid should also be
sensitive—capable of detecting appropriate differences. Statistically significant dif-
ferences in the magnitudes of questionnaire scores for different systems or other
usability-related manipulations provide evidence for sensitivity.
Psychometric Evaluation of the PSSUQ 465
Analyses of variance conducted on the data used to assess the original PSSUQ
(Lewis, 1995; Lewis et al., 1990) indicated that the PSSUQ was sensitive to user group
and system differences. The CSUQ data (Lewis, 1995) indicated significant sensitiv-
itytousers’ years of experience and breadth of experience withcomputersystems.
Goals of factor analysis.
Factor analysis is a statistical procedure that exam-
ines the correlations among variables to discover groups of related variables
(Nunnally, 1978). These groups of related variables (typically questionnaire items) be-
come the basis for the development of Likert scales designed to reflect the underlying
multidimensional nature of the construct under examination. Because summated
(Likert) scales are more reliable than single-item scales (Nunnally, 1978), and it is easier
to present and interpret a smaller number of scores, it is common to conduct a factor
analysis to determine if there is a statistical basis for the formation of summative scales.
A weakness of factor analysis is that there are no strong methods for assessing the sta-
tistical significance of an estimated principal factor structure (Cliff, 1993), making rep-
lication of results with a different sample (as in this study) especially valuable.
Prior work with the PSSUQ and CSUQ indicated that the similarity of their item
content resulted in very similar factor structures (Lewis, 1995). That work indicated
that the questionnaires tapped into three aspects of a multidimensional construct
(presumably usability). One of the most difficult tasks following this type of ex-
ploratory factor analysis is naming the factors. After considering a number of alter-
natives, a group of human factors engineers named the factors (and their corre-
sponding PSSUQ–CSUQ scales) System Usefulness (SysUse), Information Quality
(InfoQual), and Interface Quality (IntQual).
Number of scale steps.
The more scale steps in a questionnaire item the
better, but with rapidly diminishing returns (Nunnally, 1978). As the number of
scale steps increases from 2 to 20, there is an initial rapid increase in reliability, but it
tends to level off at about 7 steps. After 11 steps there is little gain in reliability from
increasing the number of steps. The number of steps is important for single-item as-
sessments, but is usually less important when summing scores over a number of
items. Attitude scales tend to be highly reliable because the items typically corre-
late rather highly with one another.
As expected, PSSUQ and CSUQ reliabilities were high (Lewis, 1995), with coeffi-
cient alphas exceeding .89 for all scales. A related analysis (Lewis, 1993) showed
that the mean difference of 7-point scales correlated more strongly than the mean
difference of 5-point scales with the observed significance levels of ttests. Because
there might be times when practitioners would be interested in item-level compari-
sons in addition to scale-level comparisons, the current versions of the PSSUQ and
CSUQ use 7-point rather than 5-point scales.
Calculating scale scores.
From classical psychometric theory (Nunnally,
1978), scale reliability is a function of the interrelatedness of scale items, the num-
ber of scale steps per item, and the number of items in a scale. If a participant
chooses not to answer an item, the effect should be to slightly reduce the reliability
466 Lewis
of the scale in that instance. In most cases, the remaining items should offer a rea-
sonable estimate of the appropriate scale score. From a practical standpoint, aver-
aging the answered items to obtain the scale score enhances the flexibility of use of
the questionnaire, because if an item is not appropriate in a specific context and us-
ers choose not to answer it, the questionnaire is still useful. Also, users who do not
answer every item can stay in the sample. Finally, averaging items to obtain scale
scores does not affect important statistical properties of the scores but does stan-
dardize the range of scale scores, making them easier to interpret and compare. For
example, with items based on 7-point scales, all the summative scales would have
scores that range from 1 to 7. For these reasons, it is the practice in our lab to aver-
age the responses given by a participant across the items for each scale.
Based on the factor analyses from Lewis (1995), the rules developed for calculat-
ing scale scores for the PSSUQ and CSUQ (see the Appendix for item format and
content) were
Overall: Average the responses to Items 1 through 19.
SysUse: Average the responses to Items 1 through 8.
InfoQual: Average the responses to Items 9 through 15.
IntQual: Average the responses to Items 16 through 18.
Note that this method for calculating scale scores gives equal weight to each
item in the scale. Although it is a standard practice to weight items equally, one
consequence of this is that the resulting scales will correlate to some extent rather
than being statistically independent.
Although the factors themselves are uncorrelated, this does not mean that estimated
factor scores are uncorrelated. Usually they areonly estimated, not obtained directly.
In these cases the estimated factor scores are likely to correlate substantially even if
the factors themselves are orthogonal. (Nunnally, 1978, p. 434)
This is not usually a problem as long as (a) the correlations are not too close to 1.0
(avoiding multicollinearity) and (b) the scales are useful in interpreting measure-
ment outcomes.
Control of potential response style or consistency in item alignment.
is a common practice in questionnaire development to vary the tone of items so,
typically, one half of the items elicit agreement and the other half elicit disagree-
ment. The purpose of this is to control potential measurement bias due to a respon-
dent’s response style. An alternative approach is to align the items consistently.
A potential criticism of the IBM questionnaires is that they do not use the stan-
dard control for potential measurement bias due to response style. Our rationale in
consistently aligning the items was to make it as easy as possible for participants to
complete the questionnaire. With consistent item alignment, the proper way to
mark responses on the scales is clearer and requires less interpretive effort on the
part of the participant (potentially reducing response errors due to participant con-
fusion). Furthermore, the use of negatively worded items can produce a number of
Psychometric Evaluation of the PSSUQ 467
undesirable effects (Ibrahim, 2001), including “problems with internal consistency,
factor structures, and other statistics when negatively worded items are used either
alone or together with directly worded stems” (Barnette, 2000, p. 363). The setting
in which balancing the tone of the items is likely to be of greatest value is when par-
ticipants do not have a high degree of motivation for providing reasonable and
honest responses (e.g., in many clinical and educational settings).
Thus, first and foremost, the survey or questionnaire designer must determine if using
negatively worded items or other alternatives are needed in the context of the research
or evaluation setting. Unless there is some pervasive and unambiguous reason for not
doing so, it is probably best that all items be positively or directly wordedand not mixed
with negatively worded items. (Barnette, 2000, p. 363)
Obtaining reasonable and honest responses is rarely a problem in most usability
evaluation settings.
Even if consistent item alignment were to result in some measurement bias due
to response style, typical use of the IBM questionnaires is to compare systems or ex-
perimental conditions (a relative rather than absolute measurement). In this con-
text of use, any systematic effect of response style (just like the effect of any other in-
dividual difference) will cancel out across comparisons.
For example, consider a within-subjects design in which participants provide ratings
for two different systems, and the researcher plans to compute a difference score ttest on
the ratings. Assume that each participant has some degree of acquiescence (tendency to
agree) that systematically affects his or her responses to the items. Therefore, each partic-
ipant will produce two scores, each with two components: the true rating (x) and the ef-
fect of the response style (b). To get the difference score for the ttest, the second score is
subtracted from the first for each participant, which has the consequence of removing
the influence of the individual’s response style ((x1+b)–(x2+b)=x1+bx2b=x1x2).
As a second example, consider a between-subjects design in which a researcher
has randomly selected participants from the same population and has randomly
assigned them to work with one of two systems. Suppose that one of the measure-
ments is overall system satisfaction obtained with the PSSUQ. As in the previous
example, each score from each participant has two components: the true rating (x)
and the effect of the response style (b). Averaging across participants, the observed
mean for the first system will be Mean(x1) + Mean(b); and the observed mean for
the second will be Mean(x2) + Mean(b). Note that the expected value of Mean(b)is
the same for both groups because the basis for group membership was random as-
signment from the same population. The value for the numerator of an independ-
ent groups ttest is the difference between the observed means. In this case, that dif-
ference would be (Mean(x1) + Mean(b)) (Mean(x2) + Mean(b)), which equals
(Mean(x1) + Mean(b) (Mean(x2) Mean(b)), which equals (Mean(x1) Mean(x2)).
In this example as in the first, the computations associated with group comparison
have removed the effect of response style. When using the PSSUQ in this way, the
presence or absence of an effect of response style on PSSUQ scores is moot.
Nunnally (1978, pp. 658–672) provided a review of the various types of response
styles. The major types of styles that have been hypothesized to exist are social de-
sirability (self-desirability), the tendency to guess when in doubt, the tendency to
468 Lewis
guess ’true,’ the agreement tendency (acquiescence), the extreme response ten-
dency, and the deviant response tendency.
Social desirability is the tendency for some people to rate themselves as exces-
sively good in self-report inventories, so it does not apply to usability question-
naires. Guessing tendencies can influence the scores on performance tests, but do
not apply to Likert scales (such as those used in the PSSUQ and other usability
questionnaires). Nunnally (1978) discounted the deviant response tendency as ac-
tually being a legitimate response style:
It should be apparent that although the deviant-responding tendency has been dis-
cussedfrequentlyas a response style, it is not a response style accordingtothedefinition
given earlier. It is not an artifact of measurement; rather it comes from a special way of
analyzing the valid variance. (p. 671)
The remaining response styles—the agreement tendency and the extreme re-
sponse tendency—could affect the scores obtained on usability questionnaires. Ac-
cording to Nunnally (1978), however, “The overwhelming weight of the evidence
now points to the fact that the agreement tendency is of very little importance ei-
ther as a measure of personality or as a source of systematic invalidity in measures
of personality and sentiments” (p. 669).
The extreme response tendency is the tendency to mark the extremes of rating
scales rather than points near the middle of the scale. There is some evidence support-
ing the existence of this response style (from small correlations found between the de-
grees of extremeness of ratings made by respondents on different rating tasks such as
picture preferences and attitudes toward minorities; Nunnally, 1978). Nunnally pro-
vided a method for empirically determining if a set of responses from a questionnaire
exhibits evidence for the extreme response tendency. The basis of this method is the
computation of the common shared variance between scores based on the full range of
each item’s scale and scores based on a dichotomous scoring of each item.
An emerging area of research in which extreme response and acquiescence re-
sponse styles have become issues is the area of cross-cultural research (van de
Vijver & Leung, 2001). There is some evidence that responses from members of dif-
ferent cultures exhibit different levels of these response styles, although these dif-
ferences do not always appear (Grimm & Church, 1999). Matters are even more
complicated when the members of the different cultures speak different languages,
necessitating translation of items (and consequent uncertainty about the equiva-
lence of items after translation). After collecting data from different cultures, it is
possible to test for the effects of differential response sets using structural equations
modeling (Cheung & Rensvold, 2000). The problems associated with cross-cultural
testing, however, go beyond the issues of response style and include issues such as
form invariance (whether the responses from the different cultures lead to the same
item-to-factor relation), and the practice of balancing item tone does not guarantee
form invariance between cultures.
Use of norms.
When a questionnaire has norms, data exist that allow re-
searchers to interpret individual and average scores as greater or smaller than the
Psychometric Evaluation of the PSSUQ 469
expected norm scores (Anastasi, 1976). In some contexts (field studies, standard
single-system usability studies), this can be a tremendous advantage. In other con-
texts (multiple-system comparative usability studies, other types of experiments),
it might provide no particular advantage.
Even when norms exist, researchers should be cautious in their use. To apply norms
in the usual way requires a substantial correspondence between the conditions under
which the normative data were generated and those in the measurement situation. A
valid set of norms would require correspondence between the normative and test situa-
tions with regard to participant, system, task, and environmental characteristics. Norms
are of clear value in many situations, but it is important not to overgeneralize their appli-
cability to usability evaluation.
Rather than using normative data for the purpose of specifying the location of a
product on a usability scale (which will generally not be valid for the reasons stated
earlier), it is possible to identify patterns in normative data that can be useful in in-
terpreting PSSUQ results obtained in a usability study. For example, in the data
used to originally evaluate the PSSUQ and CSUQ, the item that consistently re-
ceived the poorest ratings was Item 9 (“The system gave error messages that clearly
told me how to fix problems”). Another normative pattern was that InfoQual
tended to receive poorer ratings than IntQual.
It is important to be careful when interpreting these normative patterns. Al-
though InfoQual scores tend to be poorer than IntQual scores, this is not compel-
ling evidence that the InfoQual of systems is poorer than their IntQual. The under-
lying distribution of scores might differ simply because the scales contain different
items, with the items worded in their specific ways. On the other hand, it is reason-
able to interpret patterns that differ markedly from the observed norms. For exam-
ple, suppose a practitioner has conducted a usability evaluation with a reasonable
sample size and finds that the mean InfoQual scores are about equal to the IntQual
scores. Depending on the circumstances and the accompanying analysis of usabil-
ity problems, this could mean that the IntQual is for some reason worse than nor-
mal or that the InfoQual is better than normal (e.g., if the developers had made a
special effort to provide high-quality documentation).
1.4. Why Apply Classical Test Theory (CTT) Rather than Item Response
Theory (IRT)?
For most of the previous century, the basis for psychometric theory was a set of tech-
niques collectively known as CTT. Most of the psychometric training that psycholo-
gists(including myself) have receivedis in the techniquesof CTT (Zickar,1998). For a
comprehensive treatment of basic CTT, see Nunnally (1978).
Starting in the last quarter of the 20th century (and accelerating in the last de-
cade) was an alternative approach to psychometrics known as IRT. Although not
yet generally accepted in psychology, IRT has had a major impact on educational
testing, affecting the development and administration of the Scholastic Aptitude
Test, Graduate Record Exam, and Armed Services Vocational Aptitude Battery.
Some researchers have speculated that the application of IRT might improve the
measurement of usability (Hollemans, 1999).
470 Lewis
It is well beyond the scope of this article to explicate all the differences be-
tween CTT and IRT (for details, refer to a source such as Embretson & Reise,
2000). One of the key differences is that CTT focuses on scale-level measurement,
but IRT focuses on modeling item characteristics. This property of IRT makes it
ideal for adaptive computerized testing (Zickar, 1998), which is one of the rea-
sons it has become so popular in large-scale educational testing. On the other
hand, obtaining reliable estimates of the parameters of item response models re-
quires data collection from a very large sample of respondents (Embretson &
Reise, 2000), which can make IRT unattractive to researchers with limited re-
sources. Furthermore, current IRT modeling procedures do not handle multidi-
mensional measures very well (Embretson & Reise, 2000; Zickar, 1998). In addi-
tion to these limitations, the typical conception of the construct of usability is that
it is an emergent property that depends on user, system, task, and environmental
variables (the same variables that make it so difficult to develop usability norms).
There are no existing IRT models that can account for all of these variables, and
IRT is better suited for the measurement of latent rather than emergent variables
(Embretson & Reise, 2000).
When the development of the PSSUQ began, IRT was virtually unknown in
standard psychological psychometrics. IRT has made impressive gains in the last
decade, but still does not appear to be adequate to model a construct as intricate as
usability. Even if it were adequate, it is not clear that the additional effort involved
would be worthwhile. Embretson and Reise (2000) observed that raw (CTT) scores
and trait level (IRT) scores based on the same data correlate highly, and “no one has
shown that in real data a single psychological finding would be different if IRT
scores were used rather than raw scale scores” (p. 324). For these reasons, the analy-
ses in this article will continue to apply CTT techniques to the evaluation of the
PSSUQ usability questionnaire.
1.5. Advantages of Using Psychometrically Qualified Instruments
Despite any controversies regarding decisions made in the development of such
questionnaires, standardized satisfaction measurements offer many advantages to
the usability practitioner. Specifically, standardized measurements provide objec-
tivity, replicability, quantification, economy, communication, and scientific gener-
alization. Standardization also permits practitioners to use powerful methods of
mathematics and statistics to better understand their results (Nunnally, 1978).
Although this is also an area of continuing controversy, many researchers
hold that the level of measurement of an instrument (ratio, interval, ordinal)
does not limit permissible arithmetic operations or related statistical opera-
tions, but instead limits the permissible interpretations of the results of these
operations (Harris, 1985). Unless the scales meet certain criteria (Cliff, 1996;
Embretson & Reise, 2000), measurements using Likert scales developed using
CTT are ordinal. Despite this limitation, psychometrically qualified, standard-
ized questionnaires can be valuable additions to a practitioner’s repertoire of
usability evaluation techniques.
Psychometric Evaluation of the PSSUQ 471
1.6. PSSUQ and CSUQ: Summary of Prior Psychometric Evaluations
As discussed earlier, previous psychometric evaluations (Lewis, 1991, 1992a, 1992b,
1995) indicated that both the PSSUQ and CSUQ produced reliable overall composite
scores and had three reliable factors that included the same items.
Investigations into scale validity found that the overall score of the PSSUQ cor-
related highly with other measures of user satisfaction taken after each scenario.
The overall PSSUQ score, SysUse, and IntQual all correlated significantly with the
percentage of successful scenario completion. Sensitivity analyses indicated that
the PSSUQ and CSUQ scales responded appropriately to manipulations of system
and user groups (novices and experts). The normative patterns of relatively poor
ratings for Item 9 and InfoQual were consistent for the PSSUQ and the CSUQ.
The similarities between the outcomes for previous psychometric evaluations of the
PSSUQ (lab data) and CSUQ (survey data) provided support to the generalizability of
the instruments. Replication of the previous evaluations using a completely independ-
ent data set (the primary purpose for this investigation) would provide additional
support regarding their generalizability of use by usability practitioners.
When applicable, the Results section of this article will include detailed data from
the previous evaluations (Lewis, 1995) for comparison with data from this evaluation.
The data analyzed in this report came from 21 unpublished usability studies con-
ducted in our lab, during which participants completed the PSSUQ (paper-and-pen-
cil administration) at the end of the study. All studies used the same version of the
PSSUQ (see the Appendix for details regarding item format and content). Most of the
studies (90%) were investigations of speech recognition systems (IBM and non-IBM
systems), with an emphasis on speech dictation. The other studies were investiga-
tions of a personal communicator (Lewis, 1996) and a pen computing device. The
PSSUQ database created from the questionnaires completed for this study had 210
entries from participants of widely varying backgrounds, computer experience, and
age. With this database, it was possible to investigate the effectof the following inde-
pendent variables on the profile of the PSSUQ scales: Study, Developer, Stage of De-
velopment, Type of Product, Type of Evaluation, Gender, and Completeness of Re-
sponse. See section 3.3. Sensitivity for more detailed descriptions of these
independent variables and the outcomes of their evaluation.
3.1. Factor Analysis
Figure 1 shows the scree plot of the eigenvalues from the analysis. Adiscontinuity
analysis (Coovert & McNelis, 1988) indicated a three-factor solution (note the in-
crease in the difference between the third and fourth eigenvalues relative to the dif-
ference between the second and third, which is indicative of a three-factor solu-
tion). Table 1 shows the varimax-rotated, three-factor solution for this data as well
472 Lewis
FIGURE 1 Scree plot from factor analysis.
Table 1: Varimax-Rotated, Three-Factor Solutions From Factor Analyses
PSSUQ (Current) PSSUQ (Original) CSUQ (Original)
Item Factor 1 Factor 2 Factor 3 Factor 1 Factor 2 Factor 3 Factor 1 Factor 2 Factor 3
Q1 0.83 0.38 0.23 0.77 0.26 0.43 0.74 0.36 0.26
Q2 0.62 0.46 0.20 0.63 0.35 0.46 0.69 0.41 0.16
Q3 0.79 0.35 0.17 0.75 0.38 0.25 0.72 0.21 0.36
Q4 0.82 0.25 0.22 0.81 0.45 0.07 0.74 0.31 0.33
Q5 0.82 0.26 0.32 0.80 0.16 0.36 0.77 0.30 0.32
Q6 0.73 0.40 0.20 0.68 0.38 0.48 0.72 0.22 0.27
Q7 0.47 0.45 0.38 0.69 0.46 0.40 0.63 0.49 0.13
Q8 0.73 0.19 0.29 na na na 0.66 0.39 0.26
Q9 0.32 0.60 0.13 0.05 0.61 0.24 0.23 0.72 0.21
Q10 0.59 0.56 0.14 0.36 0.71 0.24 0.34 0.67 0.28
Q11 0.24 0.89 0.21 0.45 0.63 0.25 0.23 0.81 0.20
Q12 0.28 0.83 0.15 0.44 0.75 0.22 0.24 0.77 0.27
Q13 0.32 0.81 0.13 0.43 0.70 0.32 0.38 0.76 0.17
Q14 0.36 0.79 0.21 0.43 0.74 0.40 0.40 0.73 0.18
Q15 0.15 0.51 0.47 0.30 0.59 0.56 0.34 0.57 0.40
Q16 0.20 0.19 0.86 0.30 0.36 0.75 0.33 0.27 0.81
Q17 0.36 0.10 0.86 0.37 0.36 0.76 0.38 0.26 0.81
Q18 0.38 0.27 0.54 0.22 0.28 0.80 0.34 0.35 0.56
Q19 0.76 0.27 0.37 0.58 0.22 0.64 0.66 0.37 0.50
Note. PSSUQ = Post Study System Usability Questionnaire; CSUA = Computer System Us-
ability Questionnaire. Bold type indicates a large factor loading (0.5 or greater, except for Current
Q7, with the criterion adjusted to 0.45). Bold italics indicates large factor loadings for an item on
more than one factor.
as for the original PSSUQ and CSUQ studies. This three-factor solution explained
72.5% of the variance in the data.
This factor structure was very similar to the structure previously reported for the
PSSUQ and CSUQ (Lewis, 1995), with a few minor differences. In this analysis, the
19th item loaded strongly on the first factor (SysUse), whereas in the past it loaded
about equally on the first and third factors (SysUse and IntQual). Items 7 and 10
loaded about equally on the first and second factors (SysUse and InfoQual). In the
previous evaluations, Item 7 loaded most strongly on SysUse, and Item 10 loaded
most strongly on InfoQual. For purposes of the following evaluations, SysUse in-
cludes all of its former items (1–8), plus the 19th item. For continuity, I resolved the
ambiguities in the factor analysis in favor of the existing PSSUQ scale definitions
(Items 9–15 for InfoQual, Items 16–18 for IntQual).
As expected (Nunnally, 1978), an analysis of the correlations among the esti-
mated factor scores showed substantial correlation: SysUse–InfoQual, r(203) = .72;
SysUse–IntQual, r(207) = .67; InfoQual–IntQual, r(203) = .56; all ps < .000002). In the
initial PSSUQ study (Lewis, 1995), the same pairs of estimated factor scores had
correlations of .71, .68, and .64, respectively; and in the initial CSUQ study, the cor-
relations were .67, .71, and .61. Therefore, across the studies, the intercorrelations
appear to be about .7, .7, and .6, respectively; so the estimated factor scores share
about 36% to 50% of their variance. Although these correlations are significantly
different from zero, they are not so close to one that they would cause
multicollinearity problems during statistical analyses.
3.2. Reliability
Estimates of reliability using coefficient alpha indicated levels of reliability for the
overall PSSUQ and its factors that were consistent with previous estimates (shown
in Table 2—Values for original PSSUQ and CSUQ are from Lewis, 1995). All the
reliabilities exceeded .80, indicating that they have sufficient reliability to be valu-
able as usability measurements (Anastasi, 1976; Landauer, 1988).
Because it is possible to obtain high reliabilities for a scale by including multiple
items that mean the exactly the same (either worded in the same way or in linguisti-
cally similar ways), some critics of the PSSUQ have suggested that this could be the
basis for its highly reliable scales. Specifically, they have noted the similarity
among Items 3 (effective task completion), 4 (quick task completion), and 5 (effi-
cient task completion) in SysUse; and between Items 11 (clear information) and 13
474 Lewis
Table 2: Current and Previous Estimates of Reliability for PSSUQ Scales
Study Overall SysUse InfoQual IntQual
PSSUQ (Current) 0.96 0.96 0.92 0.83
PSSUQ (Original) 0.97 0.96 0.91 0.91
CSUQ 0.95 0.93 0.91 0.89
Note. PSSUQ = Post Study System Usability Questionnaire; SysUse = system usefulness; InfoQual -
informational quality; IntQual = interface quality; CSUQ = computer system usability questionnaire.
(information easy to understand) in InfoQual (see the Appendix for the complete
wording of the items).
To investigate the possibility that the high reliability for the PSSUQ scales was
due to these highly similar items, I recalculated the reliabilities for SysUse without
Items 3 and 5, InfoQual without Item 13, and the revised Overall scale without
these items. Without Items 3 and 5, the reliability of SysUse fell from .96 to .90—still
very high. InfoQual declined slightly from .93 to .91. The effect on the overall mea-
surement of removing the three items was to reduce coefficient alpha from .96 to
.94—a negligible reduction.
The correlation between the original scores and the revised scores was 0.99 for
SysUse, 1.00 for InfoQual, and 1.00 for Overall. The differences between the mean
scores for the original and revised versions of these scales (with 99% confidence in-
tervals) were .05 ± .02, –.06 ± .02, and .01 ± .01, respectively. The SysUse mean
shifted up slightly (somewhere between .03 and .07—less than one tenth of a scale
step), the InfoQual mean shifted down slightly (somewhere between –.04 and
–.08—again less than one tenth of a scale step), and the net effect was that the Over-
all score essentially did not change (somewhere between 0 and .02—less than one
fiftieth of a scale step).
3.3. Sensitivity
The mean values of the PSSUQ factors were 2.8 for SysUse, 3.0 for InfoQual, and 2.5
for IntQual, with 2.8 for both (a) the composite score collapsed across all 19 items
and (b) the mean of the scale scores. Note that the equality of these two ways of
computing the composite scores (averaging across all items or averaging across
scale scores) indicates that there is no need to develop weighting schemes to com-
pensate for the difference in the number of items per scale. In all analyses, lower
scores indicate better ratings, and alpha was .05.
Analyses of variance conducted to investigate the sensitivity of PSSUQ mea-
sures indicated that the following variables significantly affected PSSUQ scores (as
indicated by a main effect, an interaction with PSSUQ factors, or both):
Study(21levels—the study during which participants completed thePSSUQ).
Developer (4 levels—the company that developed the product under evaluation).
Stage of development (2 levels—product under development or available
for purchase).
Type of product (5 levels—discrete dictation, continuous dictation, game, per-
sonal communicator, or pen product).
Type of evaluation(2levels—dictation study or standard usability evaluation).
The following variables did not significantly affect PSSUQ scores:
Gender (2 levels—male or female).
Completeness of responses to questionnaire (2 levels—complete or incomplete).
The details for each of these analyses follows.
Psychometric Evaluation of the PSSUQ 475
Both the main effect, F(20, 184) = 2.2, p= .004, and the interaction, F(40,
368) = 3.2, p= .000000003 (see Figure 2) were significant; overall means across the 21
studies (labeled using the letters A through U for reasons of confidentiality) ranged
from 1.9 to 4.2.
Note that (a) none of the lines were horizontal and (b) the magnitude of differ-
ences among SysUse, InfoQual, and IntQual varied across the studies (in other
words, the lines were not parallel)—patterns that indicate scale sensitivity.
This variable refers to the company that developed the product
under study (with companies coded as CIC, CDC, CKC, and CMC for reasons of
confidentiality). Both the main effect, F(3, 201) = 3.4, p= .02, and the interaction, F(6,
402) = 3.6, p= .002 (see Figure 3) were significant; overall means across developers
ranged from 2.5 to 3.3.
The pattern of results were similar to those for Study in that (a) none of the lines
were horizontal and (b) the magnitude of differences among SysUse, InfoQual, and
IntQual varied across developers—overall patterns indicative of scale sensitivity.
Stage of development.
This variable refers to whether the investigated prod-
uctwas in development or availablefor purchase. Both the main effect,F(1,203) = 4.2,
p= .04, and the interaction, F(2, 206) = 3.1, p= .05 (see Figure 4) were significant. Prod-
uctsunder development received better ratings than products available forpurchase
(overall means of 2.6 and 3.0, respectively). The pattern of the interaction (assessed
usingBonferronittests with α= .017)was that mean ratings of IntQual differed by 0.2
(development, 2.4; product, 2.6—t(207) = 1.1, p= .28), ratings of InfoQual differed by
0.3 (development, 2.9; product, 3.2—t(207) = 1.7, p= .08), and ratings of SysUse dif-
fered by 0.5 (development, 2.6; product, 3.1—t(207) = 3.1, p= .002).
476 Lewis
FIGURE 2 Study × Factor interaction.
This outcome was a somewhat surprising result that might be due to a number
of factors. For example, when evaluating a product under development, the range
of tasks that the product can perform is more limited than will be the case once the
product is complete. This limited functionality affects the number (and possibly
the complexity) of tasks that an evaluator can ask participants to perform with the
product (which is consistent with the significant difference for SysUse).
Type of product.
This variable refers to the type of product under investiga-
tion. The types of products included
Continuous dictation products: Products that allow users to speak continu-
ously when dictating text.
Discrete dictation products: Products that require users to briefly pause be-
tween words when dictating.
Psychometric Evaluation of the PSSUQ 477
FIGURE 3 Developer × Factor interaction.
FIGURE 4 Stage of Development × Factor interaction.
Speech control of computer games: Product that allowed users to control com-
puter games by issuing voice commands.
Personal communicator: A combination cellular phone and personal digital
assistant device.
Pen computing device: A device for capturing and managing handwritten notes.
The main effect, F(4, 200) = 1.9, p= .11, was not significant; but the interaction,
F(8, 400) = 2.3, p= .02, was (see Figure 5).
As was the case for the variables of Study and Developer, (a) the lines were not
horizontal and (b) differences among the scales were not identical across develop-
ers—patterns that provide evidence of sensitivity to the product type.
Type of evaluation.
This variable refers to the type of evaluation conducted
in the study: dictation and standard. Dictation refers to the use of a specific protocol
for the measurement of dictation speed and accuracy (Lewis, 1997, 1999a). The task
in the dictation studies was for a user to dictate from written source text and, in
some studies, to also use the system to compose documents. In most dictation stud-
ies, participants received training in how to dictate and correct, and rarely con-
sulted any system documentation.
Standard refers to the use of a standard scenario-based usability problem dis-
covery protocol (e.g., see Lewis et al., 1990). In this protocol, the typical procedure
was for participants to receive descriptions of tasks to complete with the system
under evaluation. In most cases, the tasks were organized within scenarios de-
signed to provide broad functional coverage. In most standard evaluations, partici-
pants had access to system documentation and used it as required.
The main effect, F(1, 203) = .004, p= .99, was not significant; but the interac-
tion, F(2, 406) = 7.6, p= .001, was (see Figure 6). Post hoc examination of the inter-
action using Bonferroni ttests (with α= .008) indicated that for dictation studies,
478 Lewis
FIGURE 5 Product Type × Factor interaction.
SysUse and InfoQual were not significantly different from one another, t(123) =
.6, p= .54; but both were significantly different from IntQual, t(127) = 4.2, p=
.0001, and t(123) = 4.8, p= .000005, respectively. For dictation studies, SysUse and
IntQual were not significantly different from one another, t(80) = 1.95, p= .05; but
both were significantly different from InfoQual, t(80) = 5.9, p= .0000001, and t(80)
= 5.28, p= .000001, respectively.
Keeping in mind that these data are not from a designed experiment, it seems
reasonable that the difference in the use of system documentation between the
evaluation methods (not used in dictation studies, used in standard studies) could
account for the difference in the PSSUQ scale patterns. Therefore, these results do
not only indicate scale sensitivity by virtue of a significant interaction, but also by
virtue of the different behavior of InfoQual as a function of the type of study.
Neither the main effect, F(1, 194) = .12, p= .74, nor the interaction, F(2,
388) = 1.8, p= .17, were significant. The difference between the female and male
questionnaire means for each of the PSSUQ scales was only 0.1. Although evidence
of gender differences would not affect the usefulness of the PSSUQ, it is of potential
interest to practitioners that the instrument does not appear to have an inherent
gender bias.
Completeness of responses to questionnaire.
Neither the main effect,
F(1, 203) = .26, p= .61, nor the interaction, F(2, 406) = 1.3, p= .28, were significant
(see Figure 7). The difference between the complete and incomplete questionnaire
means for each of the PSSUQ scales was only 0.1.
This finding is important because it supports the practice of including partially
completed questionnaires when averaging items to compute scale scores (rather
than discarding the data from partially completed questionnaires).
Analysis of the distribution of incomplete questionnaires in the analyzed data-
base showed that of 210 total questionnaires, 124 (59%) were complete and 86 (41%)
Psychometric Evaluation of the PSSUQ 479
FIGURE 6 Evaluation Type × Factor interaction.
were incomplete. For the incomplete questionnaires, the mean number of items
(with 95% confidence interval bounds) for Overall (19 items), SysUse (9 items),
InfoQual (7 items), and IntQual (3 items) were, respectively, 15.8 ± 0.5, 8.8 ± 0.1, 4.2 ±
0.5, and 2.9 ± 0.1. Across the incomplete questionnaires, the completion rate for all
SysUse and IntQual items exceeded 85% (averaging 95% and 97%, respectively); but
the average completion rate for InfoQual items was only 60%. These data indicate
that the primary cause of an incomplete questionnaire was the failure to complete
one or more InfoQual items. In most cases (78%), these incomplete questionnaires
came from dictation studies (which did not typically include documentation) or
standard usability studies conducted on prototypes without documentation.
3.4. Norms
Table 3 shows the means and 99% confidence intervals for each item from this
PSSUQ data and from the original PSSUQ and CSUQ data sets (Lewis, 1995). Fig-
ure 8 illustrates the patterns of the means for these three sets of data.
As discussed previously in Section 1.3. Use of norms, there are probably very
few cases in which practitioners can use these norms for the direct assessment of a
product under evaluation. The data for this evaluation come from a variety of
sources that included different types of products at different stages of development
and the performance of different types of tasks. The original PSSUQ data came
from a more consistent source, which included the assessment of three different
systems using a set of benchmark tasks developed for the study of office systems
performed by participants with different levels of computer experience (for details,
see Lewis et al., 1990). The original CSUQ data came from a survey conducted over
a broad range of users at IBM. The original PSSUQ and CSUQ data are, however,
over 10 years old, which casts some doubt on their usefulness as norms for current
480 Lewis
FIGURE 7 Completeness × Factor interaction.
Table 3: Means and 99% Confidence Intervals for PSSUQ and CSUQ Norms
PSSUQ (Current) PSSUQ (Original) CSUQ (Original)
Limit Mean
Limit Mean
Limit Mean
Q1 2.60 2.85 3.09 3.36 4.00 4.64 3.12 3.30 3.48
Q2 2.45 2.69 2.93 3.40 4.02 4.64 3.36 3.54 3.72
Q3 2.58 2.85 3.11 3.07 3.73 4.40 2.73 2.91 3.09
Q4 2.86 3.16 3.45 3.53 4.15 4.76 3.09 3.27 3.45
Q5 2.79 3.06 3.34 3.37 3.98 4.59 3.05 3.23 3.41
Q6 2.40 2.66 2.91 2.75 3.41 4.07 2.77 2.95 3.13
Q7 2.07 2.27 2.48 2.92 3.57 4.22 3.61 3.82 4.03
Q8 2.54 2.86 3.17 na na na 3.40 3.61 3.82
Q9 3.36 3.70 4.05 4.38 4.93 5.48 4.58 4.79 5.00
Q10 2.93 3.21 3.49 3.64 4.18 4.73 3.82 4.03 4.24
Q11 2.65 2.96 3.27 3.87 4.48 5.09 3.94 4.15 4.36
Q12 2.79 3.09 3.38 3.42 4.02 4.63 4.11 4.32 4.53
Q13 2.37 2.61 2.86 3.15 3.79 4.43 3.95 4.13 4.31
Q14 2.46 2.74 3.01 2.81 3.43 4.04 3.70 3.88 4.06
Q15 2.41 2.66 2.92 3.02 3.55 4.08 3.43 3.61 3.79
Q16 2.06 2.28 2.49 2.32 2.91 3.51 3.01 3.19 3.37
Q17 2.18 2.42 2.66 2.37 2.92 3.47 3.02 3.20 3.38
Q18 2.51 2.79 3.07 2.44 3.00 3.56 3.47 3.68 3.89
Q19 2.55 2.82 3.09 3.10 3.69 4.29 3.13 3.31 3.49
SysUse 2.57 2.80 3.02 3.26 3.81 4.36 3.19 3.34 3.49
InfoQual 2.79 3.02 3.24 3.58 4.06 4.54 3.95 4.13 4.31
IntQual 2.28 2.49 2.71 2.42 2.93 3.43 3.17 3.35 3.53
Overall 2.62 2.82 3.02 3.30 3.76 4.22 3.43 3.61 3.79
Note. PSSUQ = Post Study System Usability Questionnaire; CSUQ = Computer System Usability
Questionnaire; SysUse = system usefulness; InfoQual = information quality; IntQual = interface quality.
Means apear in bold face.
FIGURE 8 Means for three sets of Post Study System Usability Questionnaire–Com-
puter System Usability Questionnaire (PSSUQ–CSUQ) norms.
systems, even if the conditions of evaluation were similar to those for the original
The consistently better mean ratings in this PSSUQ compared to the original
PSSUQ data do not necessarily indicate a wholesale improvement in system usabil-
ity over the last 10 years (although this is one possible explanation and might con-
tribute to the differences). It is also possible that the differences are due to differ-
ences in participant populations, system characteristics, or tasks.
Despite this, there are some interesting and potentially useful patterns in the
means from the three sets of data. The item data show substantial correlation, with
the greatest correlation between the PSSUQ means (CSUQ–PSSUQ current: r(19) =
.49, p= .03; CSUQ–PSSUQ original: r(18) = .57, p= .01; PSSUQ current–PSSUQ orig-
inal: r(18) = .81, p= .0005). For all three sets of data, the item that received the poor-
est rating—averaging from .45 to .49 above the next poorest item in the respective
set—was Item 9 (“The system gave error messages that clearly told me how to fix
problems”). Finally, mean ratings of InfoQual tend to be higher (poorer) then mean
ratings of IntQual, with differences for the three data sets ranging from 0.5 to 1.1.
There are several ways in which these findings can be of use to practitioners.
The consistently poor ratings for Item 9 indicate
1. This should not surprise practitioners if they find this in their own data.
2. It really is difficult to provide usable error messages throughout a product.
3. It may well be worth the effort to make the effort to focus on providing us-
able error messages.
4. If practitioners find the mean for this item to be equal to or less than the
mean of the other items in InfoQual, they have been successful in address-
ing the problem.
The consistent pattern of poor ratings for InfoQual relative to IntQual suggest
that practitioners who find this in their data should not necessarily conclude that
they have poor documentation or a great interface. On the other hand, if this pat-
tern appeared in the first iteration of a usability evaluation and the developers de-
cided to emphasize improvement to the quality of their information, then any sig-
nificant decline in the difference between InfoQual and IntQual would be evidence
of a successful intervention.
3.5. Extreme Response Tendency
The extreme response tendency is the tendency to mark the extremes of rating
scales rather than points near the middle of the scale. The procedure for determin-
ing if a set of responses from a questionnaire exhibits evidence for the extreme re-
sponse tendency (Nunnally, 1978) is to
1. Score the responses in two ways—first as the sum of deviations from the
center point of the scale using the number of scale steps in the instrument’s items,
then as the sum of dichotomized scores. Because the PSSUQ uses items with seven
482 Lewis
scale steps, the effect of dichotomization is that ratings from 1 to 3 become 0, a rat-
ing of 4 becomes 0.5; and ratings from 5 to 7 become 1.
2. Divide the squared correlation between the two sets of scores by the product
of their internal reliability coefficients (coefficient alphas) to get an estimate of their
shared common variances. If that ratio is considerably lower than 1.0 (Nunnally
suggests 0.8 as a criterion), it is reasonable to assume that the extremeness tendency
is present to some degree.
The obtained ratios for Overall, SysUse, InfoQual, and IntQual were, respec-
tively, .94, 1.02, 1.18, and .95. Because all ratios were greater than .80, there was no
evidence for an extremeness tendency for any of the scales. (I also performed the
same procedure with truly dichotomous scores, once scoring the central scale point
of 4 as 0 and once scoring it as 1. In both cases, the results were essentially the same
as with the procedure discussed earlier—no evidence for an extremeness tendency
for any of the scales.)
The primary purpose of this research was to investigate the similarity between the
initially published psychometric properties of the PSSUQ (Lewis, 1995) and esti-
mates of the same properties using data from 5 years of lab-based usability evalua-
tion. The key research questions were whether the PSSUQ, used for research in an
area very different from that for the previous psychometric evaluations, would ex-
hibit a factor structure, reliability, sensitivity, and norms consistent with the previ-
ous research. Replication of the previous findings with this new set of data would
provide evidence of significant generalizability for the questionnaire, supporting
its use by practitioners for measuring participant satisfaction with the usability of
tested systems. Failure to replicate would provide information on appropriate lim-
its of generalization for the psychometric properties of the PSSUQ.
4.1. Results Were Acceptable and Consistent With Previous Research
Although the analyzed data came from studies that differed in both content and
protocol from the studies that generated the data for previous analyses, the factor
structure, scale reliabilities, and sensitivity analyses were all consistent with prior
results (Lewis, 1995), and all reached acceptable levels according to standard
psychometric criteria. The reliability of IntQual has been the most variable across
studies, possibly because it has the fewest (only three) items. The profiles of item
means across the evaluations also showed strong similarity.
The investigation into the effect of removing Items 3 and 5 from SysUse and Item
13 from InfoQual indicated that the high PSSUQ scale reliabilities were not de-
pendent on the inclusion of these items and that the removal of the items had no
substantive effect on the scale means. Practitioners can now treat these items as op-
tional when using the PSSUQ, either using the traditional version or a version that
has only 16 items (16% shorter). (Personally, I plan to use the shorter version in fu-
Psychometric Evaluation of the PSSUQ 483
ture studies because the optional items do not provide much additional informa-
tion but do consume study time.)
In summary, the analyses support the continued use by usability practitioners of
the PSSUQ and its historical factors as a measure of user perception of and satisfac-
tion with product usability in scenario-based usability evaluations.
4.2. PSSUQ Ratings Were Insensitive to Gender
Although it would have been acceptable to have detected PSSUQ response differ-
ences as a function of gender, the insensitivity of the PSSUQ to gender makes it eas-
ier to interpret and report PSSUQ scores because there will typically be no reason to
present scores broken down by gender. Despite this, practitioners should still plan
to include gender as a variable in their analyses for cases in which different genders
might react differently to a product.
4.3. PSSUQ Ratings Tended to be Robust Even When Questionnaires
Were Incomplete
Based on psychometric theory (CTT), I had hypothesized in earlier articles that the
failure to complete all items in the questionnaire should not invalidate the re-
sponses or the gathering of the available responses into scale scores by averaging
across the available items (Lewis, 1995). The basis for this hypothesis was that, ac-
cording to CTT, scale reliability is a function of the interrelatedness of scale items,
the number of scale steps per item, and the number of items in a scale (Nunnally,
1978). If a participant chooses not to answer an item, the effect should be to reduce
slightly the reliability of the scale in that instance; and, in most cases, the remaining
items should offer a reasonable estimate of the appropriate scale score. The
nonsignificant main effect and interaction for the completeness variable supported
this hypothesis and, by extension, our current practice of computing mean scale
scores from PSSUQs that participants have not fully completed.
This is an important finding because one of the criticisms made by IRT practitio-
ners regarding CTT scales is that “by not including item properties in the model,
true score can apply only to a particular set of items or their equivalent”
(Embretson & Reise, 2000, p. 53). Taken to its extreme, this would mean that adding
or deleting even one item from a scale developed using CTT could render its scores
invalid (Hollemans, 1999). The essential equivalence of mean scores for complete
and incomplete PSSUQs in this study is consistent with the expectation of equiva-
lence implied by CTT. The questionnaires that result from the decisions of partici-
pants to leave some responses blank are apparently equivalent to the standard
PSSUQ (at least, when averaged over a number of participants). Practitioners using
the PSSUQ need not fear that the failure of participants to complete all items makes
the obtained PSSUQ items worthless for the purpose of computing its scale scores.
These data do not provide information concerning how many items a participant
canignoreandstill producereliablescale scores.The data do suggest that, inpractice,
participants typically complete enough items to produce reliable scale scores.
484 Lewis
4.4. Absolute Normative Data is of Limited Value, but Normative Patterns
Can Be Useful
Practitioners should be cautious when attempting to interpret their own absolute
PSSUQ or CSUQ data against the norms presented in this article. On the other
hand, relative patterns of data that appear consistently in the norms can be of inter-
pretative value. One example is the consistently poor rating for Item 9 (“The sys-
tem gave error messages that clearly told me how to fix problems”). Another is the
consistent difference between the InfoQual and IntQual scores. Because these pat-
terns appeared consistently in the original and these evaluative studies of the
PSSUQ and CSUQ, practitioners should expect to find these patterns in their data.
Deviations from these patterns are potentially meaningful, especially if the practi-
tioner has focused on the development of clear error messages and high-quality in-
formation and finds that the means for these scores are consistent with the means of
the other items and scales and are consistent with observed usability problems.
4.5. Response Styles and PSSUQ Scores
Of the hypothesized response styles (Nunnally, 1978), the ones that might reason-
ably affect PSSUQ scores are the agreement tendency and the extreme response
tendency. The computation of the shared common variance for deviation and di-
chotomous scores indicated no presence of an extreme response tendency in the
current PSSUQ rating data. Nunnally reviewed the evidence for the agreement ten-
dency and concluded that it was of little importance as a source of scale invalidity.
Some recent research (Baumgartner & Steenkamp, 2001; Clarke, 2001; van de
Vijver & Leung, 2001) indicates that there could be significant differences among
different cultures with regard to the agreement tendency and the extreme response
tendency. Practitioners should avoid using the PSSUQ for the purpose of compar-
ing different cultural groups unless there is evidence that the groups do not differ
in these tendencies or the practitioners are prepared to test after the fact for model
equivalence (Cheung & Rensvold, 2000). Note that this is not a limitation that ap-
plies only to the PSSUQ, but is a limitation of any similar questionnaire if used to
compare groups from different cultures.
The strongly nonsignificant outcomes for PSSUQ sensitivity to gender do sug-
gest that, at least in the culture of the United States, there is no difference between
men and women with regard to these potential tendencies when completing the
PSSUQ. There also appears to be no such difference between participants who
complete all of the PSSUQ items at the end of a study and those who do not. It is in-
teresting that of the seven sensitivity assessments, the PSSUQ exhibited evidence
of sensitivity when the method for parsing the data was to divide the systems into
different groups (study, developer, stage of development, type of product, type of
evaluation). In contrast, the PSSUQ was insensitive when the basis for parsing the
data was to divide the respondents into different groups (gender, completeness of
responses). This difference is not strongly compelling evidence of a lack of influ-
ence of response style on PSSUQ scores (as suggested by Nunnally, 1978, for tests of
sentiments in general), but it is consistent with such a hypothesis.
Psychometric Evaluation of the PSSUQ 485
Finally, it is important to keep in mind that when used as a dependent measure
in a standard within- or between-subject experimental design in which cultural dif-
ferences are not an independent variable, any effect of response style will cancel
out across experimental conditions. When used in this way, the presence or absence
of effects of response styles on PSSUQ scores is moot.
The similarity of psychometric properties between the original and current PSSUQ
data, despite the passage of time and differences in the types of systems studied,
provide evidence of significant generalizability for the questionnaire, supporting
its use by practitioners for measuring participant satisfaction with the usability of
tested systems.
Due to its generalizability, practitioners can confidently use the PSSUQ when
evaluating different types of products and at different times during the develop-
ment process. Practitioners should be cautious about using the PSSUQ to compare
the attitudes of different cultural groups. The PSSUQ can be especially useful in
competitive evaluations (see Lewis, 1996) or when tracking changes in usability as
a function of design changes made during development (either within a version or
across versions).
Anastasi, A. (1976). Psychological testing. New York: Macmillan.
Barnette, J. J. (2000). Effects of stem and Likert response option reversals on survey internal
consistency: If you feel the need, there is a better alternative to using those negatively
worded stems. Educational and Psychological Measurement, 60, 361–370.
Baumgartner, H., & Steenkamp, J. B. E. M. (2001). Response styles in marketing research: A
cross-national investigation. Journal of Marketing Research, 38, 143–156.
Brooke, J. (1996). SUS: A quick and dirty usability scale. In P. W. Jordan, B. Thomas, B. A.
Weerdmeester, & I. L. McClelland (Eds.), Usability evaluation in industry (pp. 189–194).
London: Taylor & Francis.
Cheung, G. W., & Rensvold, R. B. (2000). Assessing extreme and acquiescence response sets
in cross-cultural research using structural equations modeling. Journal of Cross-Cultural
Psychology, 31, 187–212.
Chin, J. P, Diehl, V. A., & Norman, K. (1988). Development of an instrument measuring user
satisfaction of the human–computer interface. In Conference on Human Factors in Com-
puting Systems (pp. 213–218). New York: Association for Computing Machinery.
Clarke, I. (2001). Extreme response style in cross-cultural research. International Marketing
Review, 18, 301–324.
Cliff, N. (1993). Analyzing multivariate data. San Diego, CA: Harcourt Brace.
Cliff, N. (1996). Ordinal methods for behavioral data analysis. Mahwah, NJ: Lawrence Erlbaum
Associates, Inc.
Coovert, M. D., & McNelis, K. (1988). Determining the number of common factors in factor
analysis: A review and program. Educational and Psychological Measurement, 48, 687–693.
Embretson, S. E., & Reise, S. P. (2000). Item response theory for psychologists. Mahwah, NJ: Law-
rence Erlbaum Associates, Inc.
486 Lewis
Grimm, S. D., & Church, A. T. (1999). A cross-cultural study of response biases in personality
measures. Journal of Research in Personality, 33, 415–441.
Harris, R. J. (1985). A primer of multivariate statistics. Orlando, FL: Academic.
Hollemans, G. (1999). User satisfaction measurement methodologies: Extending the user
satisfaction questionnaire. In Proceedings of HCI International ’99 8th International Confer-
ence on Human–Computer Interaction (pp. 1008–1012). Mahwah, NJ: Lawrence Erlbaum
Associates, Inc.
Ibrahim, A. M. (2001). Differential responding to positive and negative items: The case of a
negative item in a questionnaire for course and faculty evaluation. Psychological Reports,
88, 497–500.
Kirakowski, J., & Corbett, M. (1993). SUMI: The Software Usability Measurement Inventory.
British Journal of Educational Technology, 24, 210–212.
Kirakowski, J., & Dillon, A. (1988). The computer user satisfaction inventory (CUSI): Manual and
scoring key. Cork, Ireland: Human Factors Research Group, University College of Cork.
LaLomia, M. J., & Sidowski, J. B. (1990). Measurements of computer satisfaction, literacy, and
aptitudes: A review. International Journal of Human–Computer Interaction, 2, 231–253.
Landauer, T. K. (1988). Research methods in human–computer interaction. In M. Helander
(Ed.), Handbook of human–computer interaction (pp. 905–928). New York: Elsevier.
Lewis, J. R. (1991). User satisfaction questionnaires for usability studies: 1991 manual of directions
for the ASQ and PSSUQ (Tech. Rep. No. 54.609). Boca Raton, FL: International Business
Machines Corporation.
Lewis, J. R. (1992a). Psychometric evaluation of the computer system usability questionnaire: The CSUQ
(Tech. Rep. No. 54.723). Boca Raton, FL: International Business Machines Corporation.
Lewis, J. R. (1992b). Psychometric evaluation of the post-study system usability question-
naire: The PSSUQ. In Proceedings of the Human Factors Society 36th Annual Meeting (pp.
1259–1263). Santa Monica, CA: Human Factors Society.
Lewis, J. R. (1993). Multipoint scales: Mean and median differences and observed signifi-
cance levels. International Journal of Human–Computer Interaction, 5, 383–392.
Lewis, J. R. (1995). IBM computer usability satisfaction questionnaires: Psychometric evalu-
ation and instructions for use. International Journal of Human–Computer Interaction, 7,
Lewis, J. R. (1996). Reaping the benefits of modern usability evaluation: The Simon story. In
G. Salvendy & A. Ozok (Eds.), Advances in applied ergonomics: Proceedings of the 1st Interna-
tional Conference on Applied Ergonomics—ICAE ’96 (pp. 752–757). Istanbul, Turkey: USA
Lewis, J. R. (1997). A general plan for conducting human factors studies of competitive speech dicta-
tion accuracy and throughput (Tech. Rep. No. 29.2246). Raleigh, NC: IBM Corp.
Lewis, J. R. (1999a). Streamlining a general test plan for competitive evaluation of dictation accuracy
and throughput (Tech. Rep. No. 29.3158). Raleigh, NC: IBM Corp.
Lewis, J. R. (1999b). Tradeoffs in the design of the IBM computer usability satisfaction ques-
tionnaires. In Proceedings of HCI International ’99 of the 8th International Conference on Hu-
man–Computer Interaction (pp. 1023–1027). Mahwah, NJ: Lawrence Erlbaum Associates,
Lewis, J. R., Henry, S. C., & Mack, R. L. (1990). Integrated office software benchmarks: A case
study. In Human–Computer Interaction—INTERACT ’90 (pp. 337–343). Cambridge, Eng-
land: Elsevier.
Nunnally, J. C. (1978). Psychometric theory. New York: McGraw-Hill.
van de Vijver, F. J. R., & Leung, K. (2001). Personality in cultural context: Methodological is-
sues. Journal of Personality, 69, 1007–1031.
Zickar, M. J. (1998). Modeling item-level data with item response theory. Current Directions in
Psychological Science, 7, 104–109.
Psychometric Evaluation of the PSSUQ 487
The Post-Study System Usability Questionnaire Items
The first item illustrates the item format. The remaining items show only the item
text to conserve space. Each item also has an area for comments (not shown).
1. Overall, I am satisfied with how easy it is to use this system.
2. It was simple to use this system.
3. I could effectively complete the tasks and scenarios using this system.
4. I was able to complete the tasks and scenarios quickly using this system.
5. I was able to efficiently complete the tasks and scenarios using this system.
6. I felt comfortable using this system.
7. It was easy to learn to use this system.
8. I believe I could become productive quickly using this system.
9. The system gave error messages that clearly told me how to fix problems.
10. Whenever I made a mistake using the system, I could recover easily and quickly.
11. The information (such as on-line help, on-screen messages and other docu-
mentation) provided with this system was clear.
12. It was easy to find the information I needed.
13. The information provided for the system was easy to understand.
14. Theinformationwas effectivein helping me complete the tasks andscenarios.
15. The organization of information on the system screens was clear.
Note: The “interface” includes those items that you use to interact with
the system. For example, some components of the interface are the key-
board, the mouse, the microphone, and the screens (including their use of
graphics and language).
16. The interface of this system was pleasant.
17. I liked using the interface of this system.
18. This system has all the functions and capabilities I expect it to have.
19. Overall, I am satisfied with this system.
488 Lewis
... The usability questionnaire is the most frequently used methods for evaluation of mobile app usability, as its simplicity in terms of data collection and analysis. In practice, some studies use the well-validated usability questionnaires designed for general software systems such as the System Usability Scale (SUS) [13] and Post-Study System Usability Questionnaire (PSSUQ) [14]. As they were frequently and extensively used, the validity can be guarantee and certain usability of several aspects can be reliably assessed [15]. ...
Full-text available
Abstract Background Malignant hyperthermia (MH) is a rare anesthetic emergency with a high mortality rate in China. We developed a WeChat applet–based National Remote Emergency System for Malignant Hyperthermia (MH-NRES) to provide a real-time emergency system to help Chinese anesthesiologists deal with MH crises. However, it is imperative that close attention should be paid to the usability of the applet. Purpose The objectives of this study were to (1) evaluate the usability of the applet-based MH-NRES for anesthesiologists; and (2) to test the validity and reliability of a modified mHealth app usability questionnaire. Methods A modified User Version of the Mobile Application Rating Scale (uMARS) was designed. Together with System Usability Scale (SUS) and Post-Study System Usability Questionnaire (PSSUQ), another two well-validated questionnaires, uMARS were then used to evaluate the usability of MH-NRES. The Cronbach alpha of the total score and the subscales of uMARS was calculated to evaluate the internal consistency. The correlation coefficients among three questionnaires were calculated. Results In this study, 118 anesthesiologists provided responses to the questionnaire. The overall mean uMARS score was 4.43 ± 0.61, which ranged from 3 to 5. The mean PSSUQ score were in good to excellent range with mean of 6.02 ± 0.97, which ranged from 3.19 to 7. The overall SUS score was 76.0 ± 17.6, which ranged from 45 to 100. The total uMARS score had excellent internal consistency (Cronbach alpha = 0.984). uMARS and its subscales were strongly correlated with PSSUQ (coefficient 0.758–0.819, P
... Thirdly, usability evaluation was completed through a standard seven-Likert-based Post-Study System Usability Questionnaire (PSSUQ) [32]. The PSSUQ was introduced by International Business Machines Corporation (IBM) to collect users' opinions regarding usability with three main sections, namely system usefulness, information quality, and interface quality with questions ranging from 0 (strongly disagree) to 7 (strongly agree). ...
Full-text available
Background Accurate and timely decision-making in lung transplantation (LTx) programs is critical. The main objective of this study was to develop a mobile-based evidence-based clinical decision support system (CDSS) to enhance the management of lung transplant candidates. Method An iterative participatory software development process was employed to develop the ImamLTx CDSS. This study was accomplished in three phases. First, required data and standard clinical workflow were identified according to the literature review and expert consensus. Second, a rule-based knowledge-based CDSS application was developed. In the third phase, this CDSS was evaluated. The evaluation was done using the standard Post-Study System Usability Questionnaire (PSSUQ 18.3) and ten usability heuristics factors for user interface design. Results According to expert consensus, fifty-five data items were identified as essential data sets using the Content Validity Ratio (CVR) formula. By integrating information flow in clinical practices with clinical protocols, more than 450 rules and 500 knowledge statements were extracted. This CDSS provides clinical decision support on an Android platform regarding inclusion and exclusion referral criteria, optimum transplant time based on the type of lung disease, findings of initial assessment, and the overall evaluation of lung transplant candidates. Evaluation results showed high usability ratings due to the fact provided accuracy and sensitivity of this lung transplant CDSS with the information quality domain receiving the highest score (6.305 from 7). Conclusion Through a stepwise approach, the ImamLTx CDSS was developed to provide LTx programs with timely patient data access via a mobile platform. Our results suggest integration with existing workflow to support clinical decision-making and provide patient-specific recommendations.
... The questionnaire consists of 7 single-answer Likert scale questions (Figure 9), 1 multiple-answer question (Figure 10), and 1 open-ended question. Our usability-related questions were informed by the Post-Study System Usability Questionnaire (PSSUQ), a widely used tool for evaluating the usability of systems [10]. ...
Full-text available
Food waste presents a substantial challenge with significant environmental and economic ramifications, and its severity on campus environments is of particular concern. In response to this, we introduce FoodWise, a dual-component system tailored to inspire and incentivize campus communities to reduce food waste. The system consists of a data storytelling dashboard that graphically displays food waste information from university canteens, coupled with a mobile web application that encourages users to log their food waste reduction actions and rewards active participants for their efforts. Deployed during a two-week food-saving campaign at The Hong Kong University of Science and Technology (HKUST) in March 2023, FoodWise engaged over 200 participants from the university community, resulting in the logging of over 800 daily food-saving actions. Feedback collected post-campaign underscores the system's efficacy in elevating user consciousness about food waste and prompting behavioral shifts towards a more sustainable campus. This paper also provides insights for enhancing our system, contributing to a broader discourse on sustainable campus initiatives.
... The PSSUQ is a 16-item standardised questionnaire devised to measure users' perceived satisfaction of a software system (Tables 4, 5). The PSSUQ has concurrent validity [45], very high scale and subscale reliability and construct validity [46]. Participants were required to complete a 7-point Likert scale with responses ranging from strongly agree (1) to strongly disagree (7) ( Table 5). ...
Full-text available
Background Digital advancement of power assisted exercise equipment will advance exercise prescription for people with stroke (PwS). This article reports on the remote usability evaluation of a co-designed graphical user interface (GUI) and denotes an example of how video-conference software can increase reach to participants in the testing of rehabilitation technologies. The aim of this study was to evaluate the usability of two sequential versions of the GUI. Methods We adopted a mixed methods approach. Ten professional user (PU) (2M/8F) and 10 expert user (EU) participants (2M/8F) were recruited. Data collection included a usability observation, a ‘think aloud’ walk through, task completion, task duration and user satisfaction as indicated by the Post Study System Usability Questionnaire (PSSUQ). Identification of usability issues informed the design of version 2 which included an additional submenu. Descriptive analysis was conducted upon usability issues and number of occurrences detected on both versions of the GUI. Inferential analysis enabled comparison of task duration and PSSUQ data between the PU and EU groups. Results Analysis of the ‘think aloud’ walkthrough data enabled identification of 22 usability issues on version 1 from a total of 100 usability occurrences. Task completion for all tasks was 100%. Eight usability issues were directly addressed in the development of version 2. Two recurrent and 24 new usability issues were detected in version 2 with a total of 86 usability occurrences. Paired two tailed T-tests on task duration data indicated a significant decrease amongst the EU group for task 1.1 on version 2 (P = 0.03). The mean PSSUQ scores for version 1 was 1.44 (EU group) and 1.63 (PU group) compared with 1.40 (EU group) and 1.41 (PU group) for version 2. Conclusions The usability evaluation enabled identification of usability issues on version 1 of the GUI which were effectively addressed on the iteration of version 2. Testing of version 2 identified usability issues within the new submenu. Application of multiple usability evaluation methods was effective in identifying and addressing usability issues in the GUI to improve the experience of PAE for PwS. The use of video-conference software to conduct synchronous, remote usability testing is an effective alternative to face to face testing methods.
... Sensitivity is defined as the capability of an instrument to detect appropriate differences between different systems or between usability manipulations (Lewis, 2002;Sauro and Lewis, 2016), hence representing a vital quality of a usability questionnaire. Sensitivity was determined by comparing usability scores of low and high-usability conditions. ...
... Future work includes the application of the control loop to another use case, such as an intent-based resource usage metrics. Furthermore, an evaluation of the PRG framework could involve potential users interacting with the framework and assessing the intelligibility and usability using standardized methods, such as the Post Study System Usability Questionnaire (PSSUQ) [22]. Moreover, improvements could be made based on the evaluation results. ...
... Regarding the satisfaction domain, various questionnaires may be used (e.g. software usability measurement inventory; [Kirakowski, 1995], post-study system usability questionnaire; [Lewis, 2002], system usability scale [SUS] [Brooke, 1996] (Brooke, 1996), which is a well-established tool and has the advantage of being relatively short (i.e. ten questions) with proven reliability (Bangor et al., 2009). ...
Purpose – Despite the technological paradigm shift presented to the architecture, engineering, construction and operations sector (AECO), the full-fledged acceptance of the building information modelling (BIM) methodology has been slower than initially anticipated. Indeed, this study aims to acknowledge the need for increasing supportive technologies enabling the use of BIM, attending to available human resources, their requirements and their tasks. Design/methodology/approach – A complete case study is described, including the development process centred on design science research methodology followed by the usability assessment procedure validated by construction projects facility management operational staff. Findings – Results show that participants could interact with BIM using openBIM processes and file formats naturally, as most participants reached an efficiency level close to that expected for users already familiar with the interface (i.e. high-efficiency values). These results are consistent with the reported perceived satisfaction and analysis of participants’ discourses through 62 semi-structured interviews.
... The PSSUQ has 4 assessment components: system usefulness, information quality, interface quality, and overall. Lewis [47] asserted that design assessments are the best measurement, with smaller values indicating better quality. Overall, the third iteration results were "good." ...
Full-text available
Background: COVID-19 mitigation protocols, enacted to control the pandemic, have also been shown to have a negative impact on mental health, including the mental health of adolescents. The threat of being infected by SARS-CoV-2 and substantial changes in lifestyle, including limited social interaction due to stay-at-home orders, led to loneliness as well as depressive symptoms. However, offline psychological assistance is restricted, as psychologists are bounded by mitigation protocols. Further, not all adolescents' guardians are open to their children attending or have the means to pay for psychological service; thus, adolescents remain untreated. Having a mobile health (mHealth) app for mental health that uses monitoring, provides social networks, and delivers psychoeducation may provide a solution, especially in countries that have limited health facilities and mental health workers. Objective: This study aimed to design an mHealth app to help prevent and monitor depression in adolescents. The design of this mHealth app was carried out as a high-fidelity prototype. Methods: We used a design science research (DSR) methodology with 3 iterations and 8 golden rule guidelines. The first iteration used interviews, and the second and third iterations used mixed method approaches. The DSR stages include the following: (1) identify the problem; (2) define the solution; (3) define the solution objective; (4) develop, demonstrate, and evaluate the solution; and (5) communicate the solution. This study involved students and medical experts. Results: The first iteration resulted in a wireframe and prototype for the next iteration. The second iteration resulted in a System Usability Scale score of 67.27, indicating a good fit. In the third iteration, the system usefulness, information quality, interface quality, and overall values were 2.416, 2.341, 2.597, and 2.261, respectively, indicating a good design. Key features of this mHealth app include a mood tracker, community, activity target, and meditation, and supporting features that complement the design include education articles and early detection features. Conclusions: Our findings provide guidance for health facilities and to design and implement future mHealth apps to help treat adolescent depression.
Objetivo: verificar a usabilidade do Guia de Bolso Automático (GBA) como método de revisão anatômica para alunos do curso de medicina de uma instituição privada da cidade de Salvador. Metodologia: o estudo foi realizado com alunos da Escola Bahiana de Medicina e Saúde Pública (EBMSP) do 1º e 2º ano através da aplicação de um formulário virtual disponibilizado por e-mail e aplicativo multiplataforma de mensagens instantâneas. A escala Self Directed Learning Readiness Scale (SDLRS) adaptada foi utilizada para traçar o perfil de estudo dos estudantes, a System Usability Scale (SUS) e o Post Study System Usability Questionnaire (PSSUQ) foram mensuraram a usabilidade do GBA. A pesquisa ocorreu no período de outubro a dezembro de 2021. Resultados: 103 participantes responderam o formulário virtual, sendo 68,9% do sexo feminino e com idade média entre 18 e 20 anos (72,9%), que possuíam um perfil de estudo ativo e familiaridade com recursos digitais. A SUS evidenciou que o GBA foi considerado um aplicativo OK - Score SUS de 68,37 - e, na PSSUQ, a satisfação geral com o sistema obteve média de 2,61 (±0,09), representando uma usabilidade positiva do GBA. Ainda na PSSUQ, a qualidade da informação (2,51 ±0,13) e a qualidade da interface (2,52 ±0,15) obtiveram resultados satisfatórios. Conclusão: o Guia de Bolso Automático é uma ferramenta com boa usabilidade geral. A partir disso, mais estudos são necessários para avaliar seu impacto no desempenho e retenção de conteúdo dos estudantes de medicina.
Conference Paper
Full-text available
Simon (TM-Bellsouth Corp.) is a commercially available personal communicator (PC) combining features of a PDA (personal digital assistant) with a full suite of communications features. This paper describes the involvement of human factors engineering in the development of Simon, and summarizes the various approaches to usability evaluation employed during its development. Simon has received a considerable amount of praise from the industry and won several industry awards, with recognition both for its innovative engineering and its usability.
Full-text available
Response styles are a source of contamination in questionnaire ratings, and therefore they threaten the validity of conclusions drawn from marketing research data. In this article, the authors examine five forms of stylistic responding (acquiescence and disacquiescence response styles, extreme response style/response range, midpoint responding, and noncontingent responding) and discuss their biasing effects on scale scores and correlations between scales. Using data from large, representative samples of consumers from 11 countries of the European Union, the authors find systematic effects of response styles on scale scores as a function of two scale characteristics (the proportion of reverse-scored items and the extent of deviation of the scale mean from the midpoint of the response scale) and show that correlations between scales can be biased upward or downward depending on the correlation between the response style components. In combination with the apparent lack of concern with response styles evidenced in a secondary analysis of commonly used marketing scales, these findings suggest that marketing researchers should pay greater attention to the phenomenon of stylistic responding when constructing and using measurement instruments.
Full-text available
This study is a part of a research effort to develop the Questionnaire for User Interface Satisfaction (QUIS). Participants, 150 PC user group members, rated familiar software products. Two pairs of software categories were compared: 1) software that was liked and disliked, and 2) a standard command line system (CLS) and a menu driven application (MDA). The reliability of the questionnaire was high, Cronbach’s alpha=.94. The overall reaction ratings yielded significantly higher ratings for liked software and MDA over disliked software and a CLS, respectively. Frequent and sophisticated PC users rated MDA more satisfying, powerful and flexible than CLS. Future applications of the QUIS on computers are discussed.
Technical Report
Full-text available
This report describes a general plan for conducting human,factors studies of accuracy and throughput for competitive speech dictation systems. The report covers the topics of (a) distinguishing between studies that are appropriate for measuring accuracy/throughput and those that are appropriate for discovering usability problems, (b) defining different measures of accuracy and throughput, (c) reviewing briefly previous tasks used at IBM to assess speech dictation systems and selecting those to continue to use in future studies, (d) developing efficient experimental designs that are appropriate for the study of up to four dictation conditions in a single study and are also appropriate for between-studies comparisons of measurements,and (e) describing a protocol for collecting accuracy and throughput measures (both performance,and satisfaction). Use of these designs should allow developers of speech dictation systems to collect competitive dictation accuracy and throughput in an efficient and flexible manner. ITIRC Keywords
This chapter discusses the conduct of research to guide the development of more useful and usable computer systems. Experimental research in human-computer interaction involves varying the design or deployment of systems, observing the consequences, and inferring from observations what to do differently. For such research to be effective, it must be owned—instituted, trusted and heeded—by those who control the development of new systems. Thus, managers, marketers, systems engineers, project leaders, and designers as well as human factors specialists are important participants in behavioral human-computer interaction research. This chapter is intended as much for those with backgrounds in computer science, engineering, or management as for human factors researchers and cognitive systems designers. It is argued in this chapter that the special goals and difficulties of human-computer interaction research make it different from most psychological research as well as from traditional computer engineering research. The main goal, the improvement of complex, interacting human-computer systems, requires behavioral research but is not sufficiently served by the standard tools of experimental psychology such as factorial controlled experiments on pre-planned variables. The chapter contains about equal quantities of criticism of inappropriate general research methods, description of valuable methods, and prescription of specific useful techniques.