Copyright 1997 by the American Psychological Association. Inc.
0021-90HV97/S3.00
Journal of Applied Psychology
1997, Vol. 82, No. 2, 247-252
RESEARCH REPORTS
Overall Job Satisfaction: How Good Are Single-Item Measures?
John P. Wanous and Arnon E. Reichers
Michael J. Hudy
Ohio State University
Fisher College of Business, Ohio State University
A meta-analysis of single-item measures of overall job satisfaction (28 correlations from
17 studies with 7,682 people) found an average unconnected correlation of .63 (SO —
.09) with scale measures of overall job satisfaction. The overall mean correlation (corrected only for reliability) is .67 (SD = .08), and it is moderated by the type of measurement scale used. The mean corrected correlation for the best group of scale measures (8
correlations, 1,735 people) is .72 (5D = .05). The correction for attenuation formula
was used to estimate the minimum level of reliability for a single-item measure. These
estimates range from .45 to .69, depending on the assumptions made.
In general, single-item measures can be divided into
single item may also be acceptable. Overall job satisfac-
two categories: (a) those measuring self-reported facts,
tion appears to be an example, particularly because it has
such as years of education, age, number of previous jobs,
a long history of single-item measures, beginning with
and so on, and (b) those measuring psychological con-
the Faces Scale over 40 years ago (Kunin, 1955).
structs, such as job satisfaction. Measuring the former
Scarpello and Campbell (1983) concluded that a sin-
with a single item is commonly accepted practice. How-
gle-item measure of overall job satisfaction was prefera-
ever, the use of single-item measures for psychological
ble to a scale that is based on a sum of specific job facet
constructs is typically discouraged, primarily because
satisfactions. In the years since that article was published,
they are presumed to have unacceptably low reliability.
however, the legitimacy of single-item measures has not
In fact, the use of single-item measures in academic re-
increased. There are probably several reasons for this.
search is often considered a "fatal error" hi the review
First, it is frequently said that one cannot estimate the
process.
internal consistency reliability of single-item measures,
There are exceptions to the norm of using only scales
and this alone is sometimes believed to be a sufficient
to measure psychological constructs, however. If the con-
reason to limit or avoid their use. A second reason may
struct being measured is sufficiently narrow or is unam-
be the increased use of structural equation models that
biguous to the respondent, a single-item measure may
require estimation of the reliable variance.
suffice, as pointed out by Sackett and Larson (1990). for
A third reason is that readers of Scarpello and Campbell
example, most expectancy theory researchers use a single
(1983) may not have been convinced that the moderate
item to measure the perceived probability that effort leads
correlations they reported between the single-item and
to performance (e.g., Ilgen, Nebeker, & Pritchard, 1981).
a scale of job facet satisfactions were due to construct
For more complex psychological constructs (e.g., per-
definitions, as they argued. Rather, readers may have con-
sonality), it is usually recommended that scales with mul-
cluded that low reliability for the single-item measures
tiple items be used. However, there is a middle ground
was a better explanation.
between the extremes of simple (expectancy) and complex (personality) constructs where measurement with a
The Present Study
The present study is a meta-analysis (Hunter &
John P. Wanous and Arnon E. Reichers, Department of Management and Human Resources, Fisher College of Business,
Ohio State University; Michael J. Hudy, Department of Psychology, Ohio State University.
Correspondence concerning this article should be addressed
to John P. Wanous, Fisher College of Business, Ohio State University, Columbus, Ohio 43210. Electronic mail may be sent via
Internet to [email protected].
Schmidt, 1990) of research in which single-item measures
of overall job satisfaction are correlated with scales measuring overall job satisfaction. This was done for two
reasons. First, the average correlation is more representative of the population correlation than a correlation that
is based on a single sample. Second, two pieces of information from the meta-analysis are used to estimate the
247
248
RESEARCH REPORTS
reliability of single-item measures of overall job satisfaction by using a well-known formula (i.e., the correction
for attenuation due to unreliability) in a new way. Specifically, the average observed correlation between the single-item and the scale measures of job satisfaction can
be put into this formula, along with the estimated average
reliability of the scales used to measure overall job satisfaction. By adding a value for the assumed underlying
construct correlation between these two measures of over-
two published studies (Ferratt, 1981; Kalleberg, 1974) because
their multiple-item measures were judged to be poor measures
of overall job satisfaction. This is because an indirect measure
of job satisfaction was used ("should be-is now"), which
involves calculating a raw discrepancy score. This measure of
job satisfaction has been severely criticized (Wall & Payne,
1973). Although judgment calls about the inclusion or exclusion
of data can affect a meta-analysis (Wanous, Sullivan, & Malinak, 1989), omitting these two studies had no effect here, on
the basis of a preliminary analysis.
all job satisfaction, it is possible to solve the equation for
its missing value, that is, the minimum reliability of the
single-item measure. This procedure is described below.
(See also Wanous & Reichers, 1996b.)
The well-known formula for the correction for attenuation can be found in most textbooks on psychometrics
(e.g., Nunnally & Bernstein, 1994, p. 257):
where rxy = correlation between variables x and y, r^ =
reliability of variable x, rw = reliability of variable y, and
f^ = the assumed true underlying correlation between x
Recorded
Information
The sample size, type of job satisfaction measure, reliability
of the scale measure, and the correlation between the single
item and the scale were recorded for each study. The type of
single-item measure was subdivided into (a) 15 correlations
that used the Faces Scale, and (b) 13 correlations that used
some other type of single-item measure. The scales used to
measure overall job satisfaction were subdivided into (a) 8 correlations that used a sum of items with a focus on overall job
satisfaction, (b) 9 correlations that used a sum of facets concerning only the work itself, and (c) 11 correlations that used a sum
of job facet satisfactions.
and y if both were measured perfectly.
This formula is most commonly applied to situations
where variables x and y are different constructs, for example, job satisfaction and job performance. However, the
formula can also be applied when both measured variables
represent the same construct (i.e., job satisfaction, as done
here). When this occurs, Nunnally (1978) reported the
following:
The correlation between two such tests would be expected
to equal the product of the terms in the denominator and
consequently rw would equal 1.00. . . . If rw were 1.00,
rxy would be limited only by the reliabilities of the two
tests: rv = Vr^ • VJv (p. 220)
Method
Meta-Analysis Calculations
A software package developed by Switzer (1991) was used
to make the meta-analytic calculations. This program uses the
Hunter and Schmidt (1990) interactive procedure to produce
both observed and corrected means and standard deviations for
a set of correlations. The program also uses a bootstrap resampling method to re-estimate the parameter estimates of the mean
and variance for comparison to the Hunter and Schmidt (1990)
estimates. Furthermore, the bootstrap method yields 95% confidence intervals for each of these estimates (see Switzer, Paese, &
Drasgow, 1992, for a discussion of the bootstrap method).
The specific corrections made here include (a) sampling error,
and (b) criterion reliability, in this case the reliability of the
scales, which averaged .88 for the six studies reporting this
information.
Literature Search
Results
The following search procedures were used. First, a computerized search of the American Psychological Association's
PsycLit database was conducted. Second, a manual search of
studies cited in various published empirical articles and an unpublished master's thesis (Thomas, 1993) on this topic was
conducted. Third, a second computerized search was done with
the Business Abstracts database.
The Appendix shows the results of the literature search.
Studies are listed alphabetically. In all, there were 17
different samples from 16 sources that reported a total of
28 correlations. There are two ways to report the sample
size: (a) the number of distinct persons in the 17 samples
(N = 4,568; double counting of people is eliminated because some studies report multiple correlations), and ( b )
Criteria for Inclusion
Studies selected for this meta-analysis reported correlations
between two measures of overall job satisfaction, one of which
was a single-item measure. The second measure could be either
a sum of job facet satisfactions or the sum of several items
focused on overall job satisfaction. This led to the exclusion of
the number of persons for all 28 correlations (N = 7,682,
which includes double counting).
Table 1 shows the results of the meta-analysis. The
mean observed correlation is .63 (SD = .09). When this
is corrected for unreliability, the correlation is .67 (SD =
.08). Total artifactual between-studies variance was 24%
249
RESEARCH REPORTS
Table 1
Analysis Results for Correlations Between Two Measures of Overall Job Satisfaction: Single-Item Measure and Scales
Bootstrap estimates
and confidence intervals (Cls)
Corrected mean r
Corrected SD
Distribution
N
No. r,
Observed
mean r
All correlations
All studies
Type of single item
Measure
Faces
Others
Type of scale measure
Sum of facets
Work facet
Overall
7,682
4,568
28
17
.63
.62
.09
.08
.67
.66
.08
.08
.67
.65
.64-.70
.61-.69
.08
.07
.04-. 11
.03-. 11
3,327
4,355
15
13
.62
.64
.08
.09
.66
.68
.06
.09
.67
.68
.63 -.71
.62-.72
.06
.08
.03-.09
.00-. 14
3,690
2,257
1,735
11
9
8
.64
.59
.67
.10
.06
.06
.68
.63
.72
.10
.05
.05
.68
.63
.72
.60- .72
.59- .67
.67-.77
.09
.04
.04
.00-.15
.00-.07
.00-.07
SD
mean r
SD
Estimate
95% CI
Estimate
95% Q
of the variance, of which 18% was attributed to sampling
error.
As a check on the possible effect of double-counting
people when multiple correlations are reported, the calculations were recalculated on the 17 distinct samples. For
those studies reporting more than one correlation, the
results were averaged and weighted by the average sample
size. The results are quite similar to the results using all
28 correlations. The corrected mean correlation is .66 (SD
= .08) for the 17 different samples.
Because 76% of the between-studies variance could not
be accounted for by measurable artifacts, a search for
moderators was undertaken. Two moderators were considered: (a) how the single item was measured: Faces vs.
another type of single-item scale and (b) how the scales
were measured: a sum of job facet satisfactions, a sum
of items focused on overall satisfaction, and a sum of
facets concerning only the work itself, primarily the Work
scale of the JDI.
The type of single-item measure does not moderate the
overall correlation. However, the type of scale used to
measure overall job satisfaction does. Specifically, the difference between using a scale focused only on the work
itself versus a scale that is based on items focused on
overall satisfaction has the classical characteristics of a
moderator. That is, there is a mean difference between
these two groups (r - .63 vs. .72) and the standard deviations for each group are smaller than when they are combined. The third scale measure, sums of job facet satisfactions, had properties similar to the overall results.
Table 1 also shows the bootstrap re-estimates of the
mean and standard deviation from the meta-analysis and
the 95% confidence intervals for each of these re-estimates. This procedure was developed (Switzer; Pease, &
Drasgow, 1992) to deal with the uncertainty associated
with such estimates, particularly when the number of cor-
relations is relatively small. However, in the present study,
the two methods yielded almost identical results.
Finally, it is possible to estimate the minimum reliability of a single-item measure using the correction for attenuation formula, as described above. Several different estimates were made, but in all cases the average observed
correlations were used, not the corrected ones, because
that would have resulted in double-correction and inflated
estimates.
A total of eight estimates of the minimum reliability
of a single-item measure are shown in Table 2. These are
based on (a) which observed mean correlation was used,
(b) which assumption about the underlying construct correlation was used, and (c) which scale reliability was
used. First, the mean observed correlation based on all
28 that were found in our review (.63) was used, as is
the mean correlation for the subgroup using the best scale
measure of overall job satisfaction, that is, those scales
using items that have a focus on overall satisfaction (. 67 ).
Second, one might assume that the true underlying correlation between these two measures of overall job satisfaction is 1.00. However, Scarpello and Campbell (1983)
Table 2
Estimates of Minimum Reliability
Average observed correlation
Assumptions about
the construct correlation
and about scale reliability
Assume true
Reliability
Reliability
Assume true
Reliability
Reliability
M
r = 1.00
= .88 (observed)
= .80
r = .90
= .88 (observed)
= .80
All studies
(r = .63)
Best scale
(r = .67)
.45
.50
.51
.56
.56
.61
.63
.63
.69
.67
250
RESEARCH REPORTS
challenged this assumption. As a result, estimates based
on a construct correlation of .90 are also shown. Third,
only 6 of 17 studies reported scale reliability information,
and it averaged .88, which is fairly high. Because this
average may not be representative, an alternative average
reliability of .80 was also used.
The estimates shown in Table 2 range from a low of
.45 to a high of .69, with an average of .57. Remember,
these estimates are of the minimum reliability for a single
item. The actual reliability could be higher, but it cannot
be lower nor can it be estimated with any greater precision
than a minimum value. The effect of each of the three
assumptions on the estimated minimum values approximately equal, as shown in Table 2.
Discussion
The observed correlations between single items and
scales averaged .63, and the corrected mean correlation
of .67 further indicated convergent validity. This average
is much higher than the results reported by Scarpello and
Campbell (1983) and seems to bolster their argument
that a single-item measure of overall job satisfaction is
acceptable.
The average correlation is highest for the scales composed of items with an overall job satisfaction focus, as
compared with a sum of facets or the Work scale from
the JDI. This is what one would expect to see because
the focus of the items is directly on overall job satisfaction, and thus most similar to the wording of the singleitem measures.
The minimum level of reliability for single-item job
satisfaction measures was also estimated by using the
standard correction for attenuation formula in a somewhat
novel way. Reasonable estimates of this minimum value
are those based on assuming that the true underlying construct correlation is .90 rather than 1.00, combined with
using the best measures available. These estimated minimum reliabilities are .63 and .69, depending on the assumption about the reliability of the scales. In fact, assuming the reliability is closer to .80 than .88 is also realistic
because an average reliability of .82 has been reported
for some job satisfaction scales (Fried & Ferris, 1987).
Thus, a minimum estimated reliability for the single-item
measure close to .70 is reasonable.
It is also interesting to note that differences among
single-item measures had no effect on the meta-analysis
results. However, differences in the ways that scales were
measured did make a difference in the results. Because
of this difference, it seems reasonable to conclude that
the single-item measures are more robust than the scale
measures of overall job satisfaction.
Despite the above data and arguments, there are still
good reasons for preferring scales to single items. Nothing
reported thus far should be interpreted as questioning the
use of well-constructed scales in comparison to a singleitem measure. It should be interpreted, however, as a case
for the acceptability of single-item measures when either
the research question implies their use or when situational
constraints limit or prevent the use of scales.
One example of a research question suggesting the use
of a single-item measure is the measurement of change
in overall job satisfaction. Suppose that a single question
is asked of respondents: ' 'How much did your overall job
satisfaction change in the last year?" Such a single-item
measure might be preferable to a scale, as the following
example illustrates. Imagine a situation where Person A
has an average scale score of 3 by answering 3 to each
of 5 questions. Person B also has an average scale score
of 3 but answered two questions with 1, two more with
5, and the last with 3. These differences are ignored if a
scale score is used.
There may also be practical limitations favoring the
use of a single-item measure. For example, if space on a
questionnaire is limited, researchers have to make choices
about how many items will be included. Specifically, the
Minnesota Satisfaction Questionnaire (Short Form) has
20 job-facet items. If this popular scale is used only to
measure overall job satisfaction, 19 extra questions must
be answered.
There may be shorter alternatives to using a long list
of job facets, such as scales that are focused on overall job
satisfaction (Brayfield & Rothe, 1951; Hoppock, 1935;
Taylor & Bowers, 1972). However, these are all quite
wordy and consume precious space on a survey. The shortest of these is probably the five-item General Satisfaction
Scale from the Job Diagnostic Survey (JDS; Hackman &
Oldham, 1975).
There might also be relevant cost considerations that
would lead researchers to prefer a single-item measure.
For example, our university telephone poll charges $60
per item per sample (faculty, staff, students). Therefore,
using the shortest of the scales (five items from the JDS)
would add $240 to the cost of measuring overall job satisfaction for one sample and $720 for all three.
Finally, there may be issues of face validity. In particular, respondents may resent being asked questions that
appear to be repetitious. This problem would be worse in
an organization with poor employee relations. From a
management perspective, a single item is usually easier
to understand than a scale score, which might appear as
academic nit-picking.
In summary, if neither the research question nor the
research situation suggest the use of a single-item job
satisfaction measure, then choosing a well constructed
scale makes sense. However, if the use of a single item is
indicated, researchers may do so in the knowledge that
they can be acceptable. The use of single-item measures
RESEARCH REPORTS
should not be considered fatal flaws in the review process.
Rather, their appropriateness for a particular piece of research should be evaluated. Future research should be
directed toward examining the validity and reliability of
other single-item measures.
References
References marked with an asterisk indicate studies included in
the meta-analysis.
*Adler, S., & Golan, J. (1981). Lateness as a withdrawal behavior. Journal of Applied Psychology, 66, 544-554.
Brayfield, A. H., & Rothe, H. F. (1951). An index of job satisfaction. Journal of Applied Psychology, 35, 307-311.
*Brief, A. P., & Robereon, L. (1989). Job attitude organization:
An exploratory study. Journal of Applied Social Psychology,
19, 717-727.
*Dunham, R. B., & Herman, J. B. (1975). Development of a
female faces scale for measuring job satisfaction. Journal of
Applied Psychology, 60, 629-631.
Ferratt, T. W. (1981). Overall job satisfaction: Is it a linear
function of facet satisfaction? Human Relations, 34, 463473.
Fried, Y., & Ferris, G. R. (1987). The validity of the job characteristics model: A review and meta-analysis. Personnel Psychology, 40, 287-322.
Hackman, J. R., & Oldham, G. R. (1975). Development of the
Job Diagnostic Survey. Journal of Applied Psychology, 60,
159-170.
*Hom, P. W., Katerberg, R., & Hulin, C. L. (1979). Comparative examination of three approaches to the prediction of turnover. Journal of Applied Psychology, 64, 280-290.
Hoppock, R. (1935). Job satisfaction. New \brk: Harper.
Hunter, J. E., & Schmidt, F. L. (1990). Methods of meta-analysis. Newbury Park, CA: Sage.
Ilgen, D. R., Nebeker, D. M., & Pritchard, R. D. (1981). Expectancy theory measures: An empirical comparison in an experi. mental simulation. Organizational Behavior and Human Performance, 28, 189-223.
*Ironson, G. H., Smith, P. C., Brannick, M. T., Gibson,
W. M., & Paul, K. B. (1989). Construction of a job in general
scale: A comparison of global, composite, and specific measures. Journal of Applied Psychology, 74, 193-200.
Kalleberg, A. L. (1974). A causal approach to the measurement
of job satisfaction. Social Science Research, 3, 299-322.
Kunin, T. (1955). The construction of a new type of attitude
measure. Personnel Psychology, 8, 65-78.
*LaRocco, J. M. (1983). Job attitudes, intentions, and turnover:
An analysis of effects using latent variables. Human Relations, 36, 813-825.
*Miller, H. E., Katerberg, R., & Hulin, C. L. (1979). Evaluation
of the Mobley, Homer, and Hollingsworth model of employee
turnover. Journal of Applied Psychology, 64, 509-517.
*Mossholder, K. W., Bedian, A. G., Morris, D. R., Giles,
251
W. R., & Feild, H. S. (1988). Job performance and turnover
decisions: Two field studies. Journal of Management, 14,
403-414.
Nunnally, J. C. (1978). Psychometric theory (2nd ed.). New
%rk: McGraw-Hill.
Nunnally, J. C., & Bernstein, I. H. (1994). Psychometric theory
(3rd ed.). New %rk: McGraw-Hill.
*0'Reilly, C. A., & Roberts, K. H. (1973). Job satisfaction
among Whites and non-Whites: A cross-cultural approach.
Journal of Applied Psychology, 57, 295-299.
*O'Reilly, C. A., & Roberts, K. H. (1978). Supervisor influence
and subordinate mobility aspirations as moderators of consideration and initiating structure. Journal of Applied Psychology, 63, 96-102.
*Rosse, J. G., & Hulin, C. L. (1985). Adaptation to work: An
analysis of employee health, withdrawal, and change. Organizational Behavior and Human Decision Processes, 36, 324347.
*Russell, J. C., & Farrar, W. E. (1978). Job satisfaction in university employees: A field test of attitude theory. Psychological Reports, 42, 1247-1254.
Sackett, P. R., & Larson, J. R., Jr. (1990). Research strategies
and tactics in industrial and organizational psychology. In
M. D. Dunnette & L. M. Hough (Eds.), Handbook of industrial and organizational psychology (2nd ed., Vol. 1, pp. 419489). Palo Alto, CA: Consulting Psychologists Press.
*Scarpello, V., & Campbell, J. P. (1983). Job satisfaction: Are
all the parts there? Personnel Psychology, 36, 577-600.
*Smith, P. C., Kendall, L. M., & Hulin, C. L. (1969). The measurement of satisfaction in work and retirement. Chicago:
Rand McNally.
Switzer, F. (1991). VGBOOT2.0. Department of Psychology,
Clemson University, Clemson, South Carolina.
Switzer, F. S., DI, Pease, P. W., & Drasgow, F. (1992). Bootstrap
estimates of standard errors in validity generalization. Journal
of Applied Psychology, 77, 123-129.
Taylor, J. C., & Bowers, D. G. (1972). Survey of organizations.
Ann Arbor: Institute for Social Research, University of
Michigan.
Thomas, A. L. (1993). Job satisfaction and its consequences:
A meta-analytic approach. Unpublished master's thesis, Ohio
State University, Department of Psychology, Columbus.
Wall, T. D., & Payne, R. (1973). Are deficiency scores deficient? Journal of Applied Psychology, 58, 322-326.
*Wanous, J. P., & Lawler, E. E. III. (1972). Measurement and
meaning of job satisfaction. Journal of Applied Psychology,
56, 95-105.
*Wanous, J. P., & Reichers, A. E. (1996a). [Correlations between a single-item measure of overall job satisfaction and
the sum of job facet satisfactions.] Unpublished raw data.
Wanous, J. P., & Reichers, A. E. (1996b). Estimating the reliability of a single-item measure. Psychological Reports, 78,
631-634.
Wanous, J. P., Sullivan, S. E., & Malinak, J. (1989). The role
of judgment calls in meta-analysis. Journal of Applied Psychology, 74, 259-264.
{Appendix follows on next page)
252
RESEARCH REPORTS
Appendix
Studies Used in the Mela-Analysis
Study
Adler & Golan (1981)
Brief & Roberson (1989)
Dunham & Herman (1975)
Horn et al. (1979)
Ironson et al. (1989)
LaRocco (1983)
Miller et al. (1979)
Mossholder et al. (1988)
O'Reilly & Roberts (1973)
O'Reilly & Roberts (1978)
Rosse & Hulin (1985)
Russell & Farrar (1978)
Scarpello & Campbell (1983)
Smith et al. (1969)
Wanous & Lawler (1972)
Wanous & Reichers (1996a)
Single item"
82
144
144
113
373
227
227
227
260
235
225
220
139
562
42
507
185
185
80
208
969
966
.98
.92
.89
.92
.88
.89
.81
.91
.86
.77
Scale'
Faces
Faces
Sum of facets
Sum of facets
Faces
Faces
Faces
Item (A)
Item
Item
Faces
Faces
Item
Faces
Faces
Faces
Faces
Item
Item (Yes-No)
Faces
Overall
Work facet
Overall
Overall
Work facet
Overall
Work facet
Work facet
Sum of facets
Work facet
Work facet
Work facet
Item
Sum of facets
Sum of facets
Sum of facets
.74
.71
.71
.75
.53
.75
.76
.67
.58
.63
.59
.71
.45
.59
.44
.63
.32
.42
.54
(JDI)
(JDI)
(MSQ-SF)
(Hoppock Job Satisfaction Blank)
(JDI)
(JIG) .65 (B-R) .68 (JDI)
(JIG) .67 (B-R) .68 (JDI)
(JIG) .60 (B-R) .59 (JDI)
(JDS)
(JDI)
(JDI)
(MSQ-SF)
(JDI)
(JDI)
(JDI)
(JDI)
(MSQ)
(MSQ)
(JDI)
.60
.68
.68
a
Single item measures include (a) the Faces scale—Faces, (b) a numerical item—Item, (c) an adjective item—Item (A), and (d) an item with a
yes-no response—Item (Yes-No). b Scale measures include (a) a sum of facet satisfaction items—Sum of facets, (b) a sum of overall-focused
items—Overall, and (c) a sum of facet satisfaction items concerning only the work itself—Work facet. c JDS = Job Diagnostic Survey; JDI =
Job Descriptive Index; MSQ = Minnesota Satisfaction Questionnaire; MSQ-SF = Short Form of the MSQ; B-R = Brayfield-Rothe Scale.
Received May 23, 1996
Revision received September 13, 1996
Accepted September 17, 1996
•