Psychology
2011. Vol.2, No.7, 706-712
Copyright © 2011 SciRes. DOI:10.4236/psych.2011.27108
Factorial Validation of Warr’s (1990) Well-Being Measure: A
Sample Study on Police Officers
Sónia P. Gonçalves1,2, José Neves3
1Centro de Investigação e Intervenção Social, ISCTE-IUL, Lisboa, Portugal;
2Instituto Piaget, Almada, Portugal;
3Departamento de Recursos Humanos e Comportamento Organizacional, ISCTE-IUL, Lisboa, Portugal.
Email: sonia.goncalves@iscte.pt
Received August 2nd, 2011; revised September 4th, 2011; accepted October 6th, 2011.
The purpose of this study was to test the factorial validity of the job-related affective well-being scale—the IWP
Multi-Affect Indicator. The sample was composed of 1466 police officers and collected through self-report
questionnaires. With the objective of validating the factorial structure of the IWP Multi-Affect Indicator, several
models were tested using confirmatory factor analysis. The results supported a four-factor structure: anxiety,
comfort, depression and enthusiasm, as well as a five-factor structure including the same four factors plus a
second-order factor called global affective well-being.
Keywords: Factorial Structure, Job-Related Affective Well-Being, Police
Introduction
The conceptualization and operationalization of measures
related to well-being at work are a real challenge for research-
ers. As Warr, Cook and Wall (1979: p. 129) stated, “The need
to examine a large number of subjective variables has often led
investigators to devise their own items or to select from previ-
ous measures small segments with unknown psychometric pro-
perties. An additional difficulty arises from the complexity and
ill-defined scope of many concepts in the area”. These authors
also emphasize the importance of developing strong, but small-
sized, instruments suitable for different kinds of samples and
with acceptable psychometric properties as well, so as to allow
the comparison of results from different samples.
The general model of affective well-being has been at the
core of mental health (Keyes, 2005) and human experience
(Muchinsky, 2000), and it has been used to describe the subjec-
tive estimation of whether a person is feeling well or unwell
(Warr, 1987). The general structure of well-being experience
has been a topic of research since the 50’s (Mäkikangas et al.,
2007). In most theoretical models, general affective well-being
has been conceptualized as the level of pleasure and activation.
For example, Watson and Tellegen (1985) divide emotions into
two predominant supplementary dimensions: positive affect and
negative affect.
Within occupational health psychology, the structure of af-
fective well-being has been classified in the same way as gen-
eral affective well-being. Moreover, empirical studies have
shown that affective well-being is the most central aspect in oc-
cupational well-being (Van Horn, Taris, Schaufeli, & Schreurs,
2004). The model of general affective well-being was intro-
duced to the workplace by Peter Warr (1987; 1990). This con-
ceptualization classifies work-related emotions into the same
two dimensions: pleasure and activation (Figure 1). A certain
degree of pleasure/satisfaction or displeasure/dissatisfaction
(horizontal dimension) may be accompanied by high or low
levels of activation (vertical dimension), and in turn these levels
of activation may be accompanied by different levels of pleas-
ure.
Four quadrants result from the combination of the axis of
pleasure and the axis of activation level: anxiety (high active-
tion and low pleasure), enthusiasm (high activation and high
pleasure), depression (low activation and low pleasure), and
comfort (low activation and high pleasure). Consequently, this
combination forms two orthogonal axes: (2a) anxiety/(2b) com-
fort and (3a) depression/(3b) enthusiasm.
IWP Multi-Affec t Indicator
Based on his conceptualization of affective well-being Warr
(1990) developed the IWP Multi-Affect Indicator. Its aim is to
operationalize this multidimensional conceptualization of work-
related affective well-being based on axes 2 and 3, since axe 1
is usually assessed by means of general work satisfaction meas-
ures.
This scale comprises 12 items to measure affective well-be-
ing: six positive feelings (comfortable, calm, relaxed, motivated,
enthusiastic and optimistic) and six negative feelings (tense,
anxious, worried, depressed, melancholic and unhappy). Re-
spondents were asked to assess how often their job had made
them experience any of these twelve feelings over the past
weeks (e.g., “In the past few weeks, to what extent has your job
(2a)
anxiety
(3b)
enthusiasm
(3a)
depression
(2b)
comfort
(+) activation
(-) activation
(1b) pleasure
(1a) displeasure
Figure 1.
Conceptualization of job-related affective well-being (adapted from
Warr, 1990).
S. P. GONÇALVES ET AL. 707
made you feel ...?”) on a 6-point Likert scale (1 = Never to 6 =
All the time).
According to Warr and Parker (2010), this scale allows us to
obtain scores for each of the quadrants (anxiety, enthusiasm,
depression and comfort); the authors have defined this as “pri-
mary score for general use” (p. 3). Furthermore, the authors
also refer to a “secondary specialized use” (p. 3), i.e., the scores
of axes/dimensions (anxiety-comfort and depression-enthu-
siasm; the anxiety and depression items are reversed), the score
of negative affects (left quadrant in Figure 1) and positive af-
fects (right quadrant in Figure 1), as well as an overall score of
well-being at work (all quadrants) where higher values reflect
greater well-being (the anxiety and depression items are re-
versed). The scores are created using the average of the corre-
sponding items for each factor. These dimensions are necessar-
ily intercorrelated and, according to the authors, it is desirable
to analyze multiple defaults.
This measure of affective well-being at work has a set of
valuable assets (Warr & Parker, 2010): 1) It is a context-spe-
cific measure (“context-specific”) that addresses the feelings
experienced at work, and not general feelings, and therefore
should be a better predictor of work-related outcomes than
free-context measures (“free-context”); 2) It is based on the
model of affection—a circumplex model of affect (Russel,
1980), which has already been thoroughly studied, hence theo-
retically and strongly supported; 3) It covers all four quadrants,
while similar measures only cover them partially. For instance,
PANAS does not cover the low activation quadrants (i.e., de-
pression and comfort).
Studies on the factorial structure of the job-related affective
well-being scale (Table 1) have resorted primarily to principal
components analysis, which has resulted in two- or three-factor
solutions. Within a large sample, Warr (1990) found a two-
factor solution, in which the six items “tense”, “anxious”, “wor-
ried”, “calm”, “comfortable”, and “relaxed” compose the anxi-
ety-comfort factor, whereas the remaining six items (depressed,
melancholic, unhappy, motivated, enthusiastic, optimistic) con-
stitute the depression-enthusiasm factor. Nevertheless, items
“anxious” and “comfortable” gave great weight to both factors.
This same structure was later replicated in other studies, except
for differences in terms of the contribution exerted by some
items (Cifre & Salanova, 2002; Sevastos, Smith, & Cordery,
1992).
Two studies (Daniels, Brough, Guppy, Peters-Bean, &
Weatherstone, 1997; Mäkikangas, Feldt, & Kinnunen, 2007)
were found on the factorial structure of affective well-being at
work through confirmatory factor analysis, which is “consid-
ered the best way to examine the dimensionality of the prede-
fined structure of a scale” (Mäkikangas et al., 2007: p. 199).
The study by Daniels et al. (1997) confirmed that the three-
factor structure was the one with better adjustment: positive
affect (enthusiasm and optimism), negative affect (tense, wor-
ried, anxious, calm and relaxed), and pleasure-displeasure fac-
tor (comfortable , motivated, depressed, melancholic and un-
happy); this structure is in line with the model of humor by
Watson and Tellegen (1985). The results of the confirmatory
factor analysis study by Mäkikangas et al. (2007) showed that
the four-factor model is the one that best describes the structure
of affective well-being at work, i.e., anxiety, comfort, depress-
sion and enthusiasm.
There are other studies that use the IWP Multi-Affect Indi-
cator together with other measures to explore, on the one hand,
different dimensions of affective well-being in a workplace
context, e.g., anxiety-comfort, depression-pleasure, boredom-
enthusiasm, tiredness-vigor and angriness-placidness (e.g.,
Daniels, 2000) and, on the other, different aspects of occupa-
tional well-being, including, for instance, cognitive, profess-
sional, social and psychosomatic dimensions (e.g., Van Horn et
al., 2004).
Objective
The purpose of this study was to test the factorial structure of
Warr’s (1990) IWP Multi-Affect Indicator scale, which has
operationalized affective well-being at work. To accomplish
this task, various models were tested using confirmatory factor
analysis.
Method
Participants
The study involved 1 466 police officers (91.9% males),
aged 20 to 59 years (M = 35.91, SD = 8.33) and with different
tenure levels, from less than one year to 41 years (M = 13.90;
SD = 8.06). The majority of participants (58.7%) were married,
and 48.7 % had from 10 to 12 years of schooling.
Measures
Affective well-being at work. The IWP Multi-Affect Indi-
cator scale contains 12 items proposed by Warr (1990) to
measure affective well-being at work according to the concept-
tualization already outlined (Table 2). The participants’ task
was to indicate to what extent their job had made them experi-
ence any of those feelings over the past weeks on a Likert scale
ranging from 1 (never) to 6 (all times).
Procedure
Data were collected through a questionnaire. Telephone and
face-to-face contacts were established with the commanding
officers, and they were explained the study as well as the pro-
cedure for collecting data. The questionnaires were distributed
and, after a period of approximately three weeks, we collected
the questionnaires that had already been completed, either in
person or via internal mail. The response rate was approxi-
mately 80%.
Results
Descriptive Statistics and Correlations between Items
Table 3 shows mean and standard deviations and item corre-
lation matrix. All the items are significantly correlated with
each other. The correlations between items of positive and ne-
gative affect are negative, and the correlations between items of
the same pole are positive. The coefficients are high, thus sug-
gesting the existence of factors underlying the relationships
between variables.
Analysis of the Factorial Structure
In order to validate the factorial structure of the IWP Multi-
Affect Indicator various models were tested using confirmatory
factor analyses. In all analyses, the estimated parameters were
based on the variance-covariance matrix of the items in the
scale. The method of maximum likelihood was used to estimate
these parameters. The factors were specified as latent variables
and represented by the components hypothesized by Warr
S. P. GONÇALVES ET AL.
708
Table 1.
Summary of the studies on the IWP Multi-Affect Indicator.
Authors (date) Sample Main results
Warr (1990) 1686 workers of
different professions
Explanato r y analysis:
– Factor analysis with varimax rotation confirmed the two dimensions provided: anxiety-comfort and
depression-enthusiasm, thus corroborating the pleasure/activation model;
– Items “anxious” and “comfortable” gave weight to both dimensions;
– The remaining items carried the expected weight.
Reliability:
– Cronbach’s Alphas for anxiety-comfort: .76
– Cronbach’s Alphas for depression-enthusiasm: .80
Sevastos, Smith,
& Cordery (1992)
3044 civil servants
and white collar
workers
Exploratory a nalysis:
– Factor analysis with varimax rotation replicated Warr’s (1990) results;
– Factor analysis with oblimin rotation: only the item “anxious” added weight to both dimensions.
Confirmatory analysis:
– Single-factor model: χ2(54) = 14689.56, p < .001, GFI = .784, AGFI = .688, RMSR = .076
– Two-orthogonal factor model: χ2(53) = 13614.91, p < .001, GFI = .846, AGFI = .773, RMSR = .067
– The two-factor model showed better adjustment than the single-factor model (χ2(1) = 1074.65, p < .001)
– Items suggested by modification indices are responsible for the mismatch: “comfortable”, “motivated”,
“anxious”
Reliability:
– Cronbach’s Alphas for anxiety-comfort: .82
– Cronbach’s Alphas for depression-enthusiasm: .85
Daniels et al. (1997) 267 civil servants
156 police officers
Confirmatory analysis (workers/police officers):
It tested the model proposed and found by Warr, Sevastos et al. and Watson & Tellegen
– Model with better adjustment: Watson & Tellegen
– Warr’s Model: χ2(41) = 176.02, p < .001, AIC = 94.02; NFI = 0.91; NNFI = 0.89; CFI = 0.93/χ2(41) =
84.64, p < .001, AIC = 2.64; NFI = 0.93; NNFI = 0.94; CFI = 0.96
– Sevastos et al.’s Model.: χ2(40) = 152.54, p < .001, AIC = 72.54; NFI = 0.93; NNFI = 0.91; CFI =
0.94/χ2(40) = 81.47, p < .001, AIC = 1.47; NFI = 093; NNFI = 0.94; CFI = 0.96
– Warson & Tellegen’s Model: χ2(38) = 95.79, p < .001, AIC = 19.79; NFI = 0.95; NNFI = 0.91; CFI =
0.97/χ2(38) = 1.75, p < .001, AIC = –9.37; NFI = 0.94; NNFI = 0.95; CFI = 0.97
Cifre &
Salanova (2002)
299 factory workers
and blue-collar
workers
Exploratory a nalysis:
– Factor analysis supports the existence of two dimensions;
– Items ‘comfortable’ and ‘motivated’ gave weight to both dimensions;
– The solution without rotation refers to a single factor.
Reliability:
– Cronbach’s Alphas for anxiety-comfort: .86
– Cronbach’s Alphas for depression-enthusiasm: .77
Mäkikangas, Feldt,
& Kinnunen (2007)
615 managers
(longitudinal study)
Confirmatory a nalysis (t1/t2)
– Model with better adjustment: four-factor model (anxiety, comfort, depression, enthusiasm): χ2(48) =
270.62, p < .001, CFI = .97; RMSEA = 0.08; NNFI = .95; AIC = 330.62/χ2(48) = 205.66, p < .001, CFI
= .98; RMSEA = 0.07; NNFI = .97; AIC = 265.66
– Two-factor model (anxiety-comfort/depression-enthusiasm): χ2(53) = 1722.63, p < .001, CFI = .84;
RMSEA = 0.23; NNFI = .80; AIC = 1772.63/χ2(53) = 1784.56, p < .001, CFI = .88; RMSEA = 0.23; NNFI
= .85; AIC = 1834.56
– Two-factor model (positive/negative): χ2(53) = 665.01, p < .001, CFI = .92; RMSEA = 0.14; NNFI = .91;
AIC = 715.01/χ2(53) = 594.89, p < .001, CFI = .90; RMSEA = 0.21; NNFI = .87; AIC = 1421.07
– Three-factor model (negative affect/positive affect/pleasure-displeasure): χ2(51) = 1394.57, p < .001, CFI
= .86; RMSEA = 0.21; NNFI = .82; AIC = 1448.57/χ2(51) = 1367.07, p < .001, CFI = .90; RMSEA = 0.21;
NNFI = .87; AIC = 1421.07
– Single-factor model: χ2(54) = 1715.05, p < .001, CFI = .82; RMSEA = 0.22; NNFI = .78; AIC =
1763.05/χ2(54) = 1753.19, p < .001, CFI = .87; RMSEA = 0.23; NNFI = .84; AIC = 1801.19
Reliability (t1/t2):
– Cronbach’s Alphas for anxiety: .74/.78
– Cronbach’s Alphas for comfort: .80/.83
– Cronbach’s Alphas for depression: .83/.85
– Cronbach’s Alphas for enthusiasm: .82/.85
(1990) to conceptualize affective well-being at work. The cor-
relations between the factors were specified to be estimated
freely. To guarantee the statistical identification of the meas-
urement model, the variance of the factors was fixed at 1.00.
Five-factor models were tested based on Warr’s alternative
conceptualization of affective well-being and on the factorial
validation studies of the IWP Multi-Affect Indicator, previously
systematized.
Model 1 (M1) supports a model of four correlated factors
(anxiety, comfort, enthusiasm and depression), which are as-
sumed to represent the four quadrants of emotional-affective
states, based on the level of pleasure and activation as hypothe-
sized by Warr (1990). Furthermore, this factor structure has
demonstrated good adjustment in previous and recent studies
(e.g., Mäkikangas et al., 2007). This model was used as a basis
for comparison with the other models. Given the conceptualize-
tion proposed by Warr, the emergence of these four basic fac-
tors and their scores are considered of primary importance.
Model 2 (M2) specifies a factor structure consisting of two
correlated factors (anxiety-comfort and depression-enthusiasm)
which were not only conceptualized by Warr (1990) but also
found in other exploratory studies (e.g., Cifre & Salanova,
2002).
Model 3 (M3) supports a factor structure composed of five
S. P. GONÇALVES ET AL. 709
Table 2.
Items of the IWP Multi-Affect Indicator of affective well-being.
Items Factor (quadrant) Factor (axis)
Tense
Anxious
Worried
Anxiety
Comfortable
Calm
Relaxed
Comfort
Anxietya-comfort
Depressed
Melancholic
Unhappy
Depression
Motivated
Enthusiastic
Optimistic
Enthusiasm
Depressiona-enthusiasm
Note: aIn this case the negative items should be reversed.
factors. It contains the four first-order factors hypothesized in
Model 1 plus a general second-order factor. The hypothesis
tested was that the four first-order factors are different dimen-
sions of the same latent factor: the global affective well-being
at work.
Model 4 (M4) specifies a factor structure consisting of two
factors that include both positive and negative affects, respect-
tively. It is in line with Watson and Tellegen’s (1985) work and
also with Warr and Parker’s (2010) recent proposals.
Finally, Model 5 (M5) specifies a factor structure consisting
only of a first-order factor. This model assumes that all items in
the scale measure just one latent variable.
One of the important phases when using confirmatory factor
analysis is to assess the model’s general quality of adjustment.
This is not, however, a consensual issue. Various authors refer
to different cutoff values for the quality levels of adjustment
(e.g., Bollen, 1990; Hu & Bentler, 1999), whereas others point
up values more or less restrictive.
Bollen (1990) considers the RMSEA values of <.08; GFI,
NFI, and CFI > .90 to be acceptable. The authors Hu and
Bentler (1999) believe that the PCFI and PGFI values above 0.6,
RMSEA close to .06, and CFI and TLI near .95 reveal a good
adjustment. The AMOS manual (SPSS, 2006) mentions CFI
values above .90 and GFI values above .80. Marôco (2010)
reports the following as reference values: X2 and p-value, the
smaller the better; for the CFI, GFI and TLI values <0.8 reveals
poor adjustment; [0.8; 0.9] values reveal poor adjustment; [0.0;
0.95] values show a very good adjustment; 0.95 shows a very
good adjustment; RMSEA >0.10 means unacceptable adjust-
ment: [0.05; 0.10] equal good adjustment and 0.05 very good
adjustment; finally, the AIC is used for comparing models and
the lower the better.
The reference values presented in the literature refer to mod-
els with excellent adjustment, which does not mean that values
slightly below should be excluded (Marsh, Kit-Tai, & Wen,
2004). One must consider value combination instead of ex-
cluding values that are just a notch bellow excellent. The values
of our models are based on this reading. In addition, as stated
by Howieson (2008: p. 222):
Table 3.
Descriptive statistics and it em-correlation matrix f o r the job-related affe c tive well-being scale.
Descriptive statistics Correlation matrix
M SD 2 3 4 5 6 7 8 9 10 11 12
1. Tense 2.95 1.15 .773** .641** –.358**–.425**–.426**.596**.575**.564** –.342** –.327**–.328**
2. Anxious 3.04 1.15 .667** –.304**–.366**–.383**.577**.562**.528** –.313** –.261**–.281**
3. Worried 3.33 1.21 –.257**–.320**–.334**.496**.456**.453** –.262** –.209**–.223**
4. Comfortable 3.26 1.21 .671**.664**–.337**–.289**–.309** .618** .614**.556**
5. Calm 3.55 1.24 .812**–.374**–.339**–.345** .562** .552**.558**
6. Relaxed 3.41 1.25 –.323**–.284**–.325** .541** .547**.543**
7. Depressed 2.46 1.30 .820**.727** –.361** –.314**–.354**
8. Melancholic 2.39 1.19 .770** –.369** –.323**–.365**
9. Unhappy 2.32 1.33 –.390** –.345**–.377**
10. Motivated 3.51 1.32 .862** .750**
11. Enthusiastic 3.46 1.30 .796**
12. Optimistic 3.63 1.36
N
ote: **p < .05. M = Mean, SD = Standard Deviation
S. P. GONÇALVES ET AL.
710
“It is important to recognise that global fit indexes alone
cannot possibly envelop all that needs to be known about a
model in order to judge the adequacy of its fit to the sample
data. As Sobel and Bohrnstedt (1985: p. 158) so cogently stated
well over a decade ago: ‘Scientific progress could be impeded
if fit coefficients (even appropriate ones) are used as the pri-
mary criterion for judging the adequacy of a model’ (…) Fit
indexes yield information bearing only on the model’s lack of
fit. More importantly, they can in no way reflect the extent to
which the model is plausible; this judgment rests squarely on
the shoulders of the researcher. Thus, assessment of the model
adequacy must be based on multiple criteria that take into ac-
count theoretical, statistical, and practical considerations”.
The results presented in Table 4 demonstrate that Model 1
and Model 3 have reasonable indexes of adjustment. Of greater
importance, these indexes are the best of all the models tested.
In fact, Model 1 adjusts itself considerably better than the other
models.
Figure 2 below presents the estimated standardized parame-
ters for the hypothetical factor models that revealed better ad-
justment. As expected, there are significant correlations be-
tween factors, and they follow the expected trend.
In general, the CFI, GFI and TLI indicators reveal very good
adjustment to M1 and good adjustment to M3 whilst the
RMSEA indicator reveals good adjustment to M1 and M3. The
indicator values that are adjusted to the other models are far
from being acceptable, especially for models M2 and M5. They
demonstrate that a four-factor correlated structure and the fac-
torial structure of this second-order compound translate best the
factorial structure of the IWP Multi-Affect Indicator.
Descriptive Statistics and Reliability
The analysis of descriptive statistics (Table 5) shows that po-
lice officers experienced mostly affective well-being (M =
3.861, SD = .894) since the emotions of enthusiasm (M = 3.531;
SD = 1.238) and comfort (M = 3.406; SD = 1.114) prevailed.
They also experienced emotions of anxiety for some time (M =
3.105; SD = 1.045) but rarely experienced emotions of depress-
sion (M = 2.385; SD = 1.168). The internal consistency of each
dimension was assessed by considering inter-item correlations,
inter-total correlations, average inter-item correlation, and the
value for Cronbach’s Alpha. The average inter-item correlation
ranges between .467 and .803, thus exceeding Briggs &
Cheek’s (1986) ideal reference values that are between 0.2 and
0.4. The minimum value of the item-total correlation is .536
and the maximum is .885, both fitting into the ideal reference
value suggested by Hair et al. (1998), which is greater than 0.30.
Inter-item correlations meet the criteria of Hair et al. (1998), for
they set the optimal value above 0.50, except for the minimum
value .216 regarding general affective well-being. The values
for Cronbach’s Alpha ranged from .871 of anxiety to .924 of
enthusiasm; therefore measures have good internal consistency.
It is generally assumed that the internal consistency analysis of
work-related indicators shows good psychometric properties
(Table 5).
Table 4.
Adjustment indicators of the different factorial structures of work-related affective well-being.
Models X2 gl CFI GFI TLI RMSEA AIC ΔX2gl
M1: Four correlated factors: anxiety, comfort,
depression, enthusiasm 394.943*** 48 .975 .956 .965 .709 454.943 -
M2: Two correlated factors: anxiety-comfort
and depression-enthusiasm 5715.769*** 53 .588 .508 .487 .272 5765.769 5.320.826(5)***
M3: A general second-order factor based on M1 863.548*** 50 .941 .912 .922 .106 919.548 468.605(2)***
M4: Two correlated factors: Positive and
Negative 2496.877*** 53 .821 .739 .777 .179 2546.877 2.101.934(5)***
M5: Single Factor 6092.624*** 54 .561 .493 .463 .279 6140.624 5.697.681(6)***
Note: ***p < .001.
M1: Four correlated factors: anxiety, comfort, depression, enthusiasm M3: A general second-order factor based on M1
Figure 2.
Confirmatory factor analysis of the measure of work-related affective well-being. Notes. All correlations are significant at p < .001.
S. P. GONÇALVES ET AL. 711
Table 5.
Cronbachs Alphas, inter-item and total-item correlations, and descriptive statistics.
Inter-Item
Correlations
Total-Item
Correlations
Dimension1 No.
Items
Cronbach’s
Alpha
Min Max
Inter-Item Average
Correlations
Min Max
M2 SD MinMax
Anxiety 3 .871 .640 .773 .693 .693 .793 3.105 1.045 1.006.00
Comfort 3 .883 .665 .813 .716 .702 .815 3.406 1.114 1.006.00
Depression 3 .912 .730 .824 .715 .785 .857 2.385 1.168 1.006.00
Enthusiasm 3 .924 .750 .861 .803 .801 .885 3.531 1.238 1.006.00
Affective
well-being 12 .913 .216 .866 .467 .536 .696 3.861 .894 1.006.00
Note: M = Mean, SD = Standard Deviation; Min = Minimum; Max = Maximum;1Scale response from 1 to 6; composite variables based on the average of the correspond-
ing items for each factor; Scale response from 1 to 5; 2Higher values translate higher levels of the dimension.
Discussion and Conclusions
The IWP Multi-Affect Indicator was put forward as the fac-
torial validation of the job-related affective well-being scale.
The validation was performed in three steps. First, the psycho-
metric parameters were estimated for different types of items.
Second, the adjustment (goodness of fit) of the basic model was
compared with the adjustment of several alternative models.
Finally, additional tests were conducted with the aim of im-
proving the adjustment of the hypothesized factor models that
had shown better adjustments in the previous step.
The analysis demonstrated that the best factorial solution for
affective well-being at work is the four factors one: depression,
comfort, anxiety and enthusiasm, as stated by Mäkikangas et al.
(2007). The solution of a general second-order factor also
showed acceptable adjustment values. It suggested that the above
four factors are different dimensions of the same latent factor:
general work-related affective well-being, as already proposed
by Warr (1990; Warr & Parker, 2010). These results reinforce
the multi-dimensionality of work-related affective well-being,
as already suggested by several authors who advocate the
abandonment of one-dimensional vision to represent emotional
experiences at work (e.g., Daniels et al., 1997; Warr, 2002; Van
Horn et al., 2004), where “measures predicted upon multidi-
mensional models may help to provide a more precise descrip-
tion of the relationship between work-related affective well-
being and other organizational phenomena” (Daniels et al., 1997:
p. 137).
The factor structure comprises two factors, i.e., negative and
positive affects, thus revealing itself to be the second best
model according to the study by Mäkikangas et al. (2007).
However, our study did not reach acceptable values; the two-
factor structure—anxiety-comfort and depression-enthusiasm—
did not reach acceptable adjustment values. This structure has
shown inconsistent results; the results of Sevastos et al. (1992),
for example, support it, yet the results of Mäkikangas et al.
(2007) do not. It should be examined in future studies. Finally,
as theoretically expected, the single-factor model did not pre-
sent a good adjustment. These last two factorial structures
proved to be the least adjusted structures of data and revealed
the lowest adjustment value indicators.
In summary, the estimated parameters indicate that two of
the factorial structures reported are valid and can be used in
investigations that need to measure work-related affective
well-being. Although further studies are necessary to examine
other types of validity (e.g. predictive validity), this study con-
tributes to reducing the lack of scales with good psychometric
parameters in Portugal, particularly when it comes to assessing
well-being at work. This tool would benefit from the broaden-
ing of the validation process of its sample to other more differ-
entiated samples, so as to generalize its use to the Portuguese
working population. Until now there were no studies on the
factorial analysis of Warr’s scale based on a Portuguese sample,
although the scale and some dimensions have already been used
on Portuguese samples (e.g., Chambel & Curral, 2005).
The present study and its factorial analysis have revealed that
work-related affective well-being, as general affective well-
being, may be understood as a multi-dimensional phenomenon.
Thus, broader conceptualizations of affective well-being may
help occupational health professionals develop a correspond-
ingly extensive repertoire of intervention strategies. In this
sense, as previously mentioned, these conceptualizations enrich
both the theory and practice of occupational psychology. The
IWP Multi-Affect Indicator supports this multi-dimensional
approach, and so it can potentially capture subtleties, complexi-
ties and changes in the experience of work that general, uni-
dimensional measures can/may not.
Acknowledgements
This work was funded by The Foundation for Science and
Technology (FCT—Lisbon, Portugal) through a PhD scholar-
ship granted to the first author (ref. SFRH/BD/38801/2007).
The authors thank Isabel Carvalho for providing the Portu-
guese version of the Multi-Affect Indicator used in this work.
References
Bindl, U. K., Warr, P. B., Parker, S. K., & Inceoglu, I. (submitted for
presentation). Multiple patterns of affect-behavior associations. An-
nual SIOP Conference, Atlanta.
Bollen, K. A. (1990). A comment on model evaluation and modifica-
tion. Multivariate B ehavioral Research, 25, 181-185.
doi:10.1207/s15327906mbr2502_5
Briggs, S. R., & Cheek, J. M. (1986). The role of factor analysis in the
development and evaluation of personality scales. Journal of Per-
sonality, 54, 106-148. doi:10.1111/j.1467-6494.1986.tb00391.x
Chambel, M. J., & Curral, L. (2005). Stress in academic life: Work
characteristics as predictors of student well-being and performance.
Applied Psychology: An International Review, 54, 135-147.
doi:10.1111/j.1464-0597.2005.00200.x
Cifre, E., & Salanova, M. (2002). Multidimensionality and bipolarity of
a Spanish Version of Warr’s (1990) well-being measure. The Journal
S. P. GONÇALVES ET AL.
712
of Psychology, 136, 69-74. doi:10.1080/00223980209604139
Daniels, K. (2000). Measure of five aspects of affective well-being.
Human Relations, 53, 275-294.
Daniels, K., Brough, P., Guppy, A., Peters-Bean, K. M., & Weather-
stone, L. (1997). A note on a modification to Warr’s measures of
work-related affective well-being. Journal of Occupational and Or-
ganizational Psychology, 70, 129-138.
doi:10.1111/j.2044-8325.1997.tb00638.x
Hair, J. F. Jr. et al. (1998). Multivariate data analysis. New Jersey:
Prentice-Hall International, Inc.
Howieson, W. B. (2008). A quantitative evaluation of the reformulated
1996 path-goal theory of work unit leadership via structural equa-
tion modeling. Edinburgh: University of Edinburgh.
Hu, L.-T., & Bentler, P. (1999). Cutoff criteria for fit indexes in co-
variance structure analysis: Coventional criteria versus new alterna-
tives. Structural Equation Modeling, 6, 1-55.
doi:10.1080/10705519909540118
Keyes, C. L. M. (2005). Mental illness and/or mental health? Investi-
gating axioms of the complete state model of health. Journal of Con-
sulting and Clinical Psychology , 73, 539-548.
doi:10.1037/0022-006X.73.3.539
Mäkikangas, A., Feldt, T., & Kinnunen, U. (2007). Warr’s scale of
job-related affective well-being: A longitudinal examination of its
structure and relationships with work characteristics. Work & Stress,
21, 197-219. doi:10.1080/02678370701662151
Marôco, J. (2010). Structural equation analysis: Theoretical founda-
tions, software & application s. Pêro Pinheiro: Report Number, Lda.
Marsh, H. W., Kit-Tai, H., & Wen, Z. L. (2004). In search of golden
rules: Comment on hypothesis testing approaches to setting cutoff
values for fit indexes and dangers in overgeneralizing Hu and
Bentler’s (1999) findings. Structural Equation Modeling, 11, 320-
341. doi:10.1207/s15328007sem1103_2
Muchinsky, P. M. (2000). Emotions in the workplace: The neglect of
organizational behavior. Journal of Organizational Behaviour, 21,
801-805.
doi:10.1002/1099-1379(200011)21:7<801::AID-JOB999>3.0.CO;2-
A
Russell, J. A. (1980). A circumplex model of affect. Journal of Person-
ality and Social Psychology, 39, 1161-1178. doi:10.1037/h0077714
Sevastos, P., Smith, L., & Cordery, J. L. (1992). Evidence on the reli-
ability and construct validity of Warr’s (1990) well-being and mental
health measures. Journal of Occupational and Organizational Psy-
chology, 65, 33-49. doi:10.1111/j.2044-8325.1992.tb00482.x
SPSS (2006). Guide on how to u s e the AMOS. Lisboa: SPPSS.
Van Horn, J. E., Taris, T. W., Schaufeli, W. B., & Schreurs, P. J. G.
(2004). The structure of occupational well-being: A study among
Dutch teachers. Journal of Occupational and Organizational Psy-
chology, 77, 365-375. doi:10.1348/0963179041752718
Warr, P., & Parker, S. (2010). IWP multi-affect Indicator. Sheffield:
Institute of Work Psychology, University of Sheffield. Paper no pub-
lished.
Warr, P. (1987). Work, unemployment, and mental health. Oxford:
Oxford University Press.
Warr, P. (1990). The measurement of well-being and other aspects of
mental health. Journal o f Oc cu pa ti onal Psychology, 63, 193-210.
doi:10.1111/j.2044-8325.1990.tb00521.x
Warr, P., Cook, J., & Wall, T. (1979). Scales for the measurement of
some work attitudes and aspects of psychological well-being. Jour-
nal of Occupational Psychology, 52, 129-148.
doi:10.1111/j.2044-8325.1979.tb00448.x