Approaches to measuring overall job satisfaction: The COM premium

1
Approaches to measuring overall job satisfaction
The COM premium MICHAEL ROSE University of
Bath Coordinator WAM Network
SUMMARY
1 Single item measures are imperfect
reliability is passable, yes but only just
2 Validity is also a serious worry SIMs do not
measure in an unbiased way levels of satisfaction
with a JOB
3 They do, however, perform far better as
measures of satisfaction with WORK
4 The conceptual differences between JOB, WORK,
and OCCUPATION should be more often recognised
2
CONFIRMING WANOUS et al.
1 John Wanouss samples were small (averaging
around 200), limited to particular
organisations, and not always random is that a
problem?
2 No, apparently. UK is fortunate in having
several employee surveys not just with large
random samples occupations, industries fully
represented
..but also data on satisfaction with many job
facets (pay, variety, security, etc.) to create
good composite measures of overall job
satisfaction (COMs)
3 And single-item questions about overall job
satisfaction level (SIMs)
4 Data from these surveys confirm the likely
value range of a pseuo-Alpha for SIMs
SCELI Social Change and Economic Life
(1985)0.69
EiB -Employment in Britain (1992)0.63
WiB Working in Britain (2000)0.68
If anything, then, the data available to the
Wanous group led to slight under-estimation of a
correct value for pseudo-Alpha
3
Once you start correlating facet measures of
satisfaction with a SIM for overall job
satisfaction you very soon have the sense that
something, somewhere, is not quite right with the
data
4
(No Transcript)
5
Worrying!
Associations between a SIM and other important
work related variables may be distorted
In particular, the effect of certain types of
variable on overall job satisfaction may be
exaggerated or understated or even minimised
So lets look at some associations
A For this, a COM measure for the EiB data will
be banded into 7 levels to create a mimic of
the SIM in terms of response distribution
the SIM carries a banding penalty
correlation is attenuated by it
we want a fair test, dont we!
B The most appropriate statistic for this
comparison will be ? (eta) value
? measures non-linear associations as well as
linear associations
? is non-monotonic, i.e. when test variable is
ordered, interval properties are not
assumedthough such variables are easily handled
technically, ? measures the ratio of between
category variance (sum of squares) to total
variance in this sense it is THE fundamental
measure of all correlational association (Nunally
Bernstein)
6
(No Transcript)
7
(No Transcript)
8
SUMMARY OF RESULTS
In the EiB data set almost 80 work related
variables have statistically significant
associations with overall job satisfaction, as
measured by the SIM
In 5 cases the association with a SIM is closer
than with the COMSIM MIMIC
but remember, the tables shown use a 7 level
mimic form of the COM
well come back to this in a moment
In 7 cases, the association of the COM is closer
but barely so (i.e. pgt0.10 to 0.05)
In all 65 other cases the association of the
COMSIM MIMIC with second variable is closer and
the variance accounted for greater

often MASSIVELY closer and greater
The mean increase in variance accounted for from
use of the COM form is
57
THAT IS MORE OR LESS EXACTLY WHAT YOUD EXPECT
FROM THE DIFFERENCE OF THE RELIABILITY VALUES
9
Q. How far does banding reduce measured
association?
Quite a bit, guv!
Overall, the difference in variance accounted for
(?2) averages13, but rises to almost 30 for
fair pay
10
CONCLUSION Banding does significantly reduce
measurement accuracy
Implications of the finding that banding reduces
associations with a SIM?
NB Wanous does not specifically deal with such
banding effects, while arguing that sources of
unreliability other than internal consistency
justify an increase of up to 10 in the real
value of pseudo-Alpha.
His value estimates for pseudo-Alpha are of
minimum values
The percentage penalty (average attenuation of
?) for unsquared associations is 6.2. This value
might be used to correct the pseudo-Alpha value
for the SIM in EiB
personally, Ive no objection whatsoever to
that.
The value of the pseudo-Alpha in EiB might thus
be increased, from 0.63 to (0.63/100)106.20.68,
just two points short of target (0.70)
Allowing for further sources of unreliability
like non-random framing effects, interviewer
recording error, data processing mistakes, etc.
the EiB SIM easily clears the 0.70 hurdle.
11
BUT THATS NOT THE POINT HERE!
At least in the form used in EiB (which is not
used by everybody ), a SIM is not just less
reliable in the strict sense (internally
consistent, meaning predictable in application)
It seriously understates the effects on overall
job satisfaction of such factors as
pay
job security
training opportunities
under-use of skills and experience
management competence
influence over decisions
favouritism of supervisor
and even childcare arrangements
Its a pretty long and worrying list
So, not just borderline for reliability, but also
a biased measure of overall job satisfaction and
therefore of questionable validity as a measure
of overall job satisfaction
12
END OF STORY?
Perhaps not though an imperfect, at times even
unsafe - measure of overall satisfaction with a
JOB, the SIM form used in EiB and elsewhere may
be a far better measure of OVERALL SATISFACTION
WITH WORK
Recall the disproportionate influence of the
facet THE WORK ITSELF in the regression model of
relative facet importance
This requires further examination
The word JOB, in both UK and American English,
can denote
A contractually held position in a work
organisation a POST
The generic skills required in doing the job
successfully - WORK
And in a still wider perspective, a community of
people with comparable terms of employment and
similarly skilled work an OCCUPATION
13
Thanks for your attention!