Professional Documents
Culture Documents
Pubblicazione depositata presso gli Uffici Stampa della Procura della Repubblica e della Prefettura di Milano
A Formative Model for Measuring Customer Satisfaction with a
Degree Course
1. Introduction
1
individual. In SERVQUAL the problem of weighting is concerned only with the
dimensions and is often solved by asking the interviewed to give a weight to the
dimensions themselves (with the constraint that the sum of weights is equal to one).
The index proposed in this paper is within the logic of the formative models, but
nonetheless presents some particulars regarding: a) a second level of aggregation, b) a
different number of respondents. In fact, in this case the means of the dimensions do not
lead to a CS global measure (as in SERVQUAL), but only to a partial measure
regarding the teaching; it is the combination of these latter measures that lead to the
overall CS index for DC. Lastly the different teaching courses are attended by a
different number of students and moreover often item non-response is verified.
While point a) is easily solved, point b) requires careful examination to establish
opportune weights for the different levels.
In paragraph 2 the index G is briefly set out, in paragraph 3 the method for
calculating the overall index is shown and lastly, in paragraph 4, the variability within
and between the intermediate levels (dimensions and teaching) is analysed to obtain the
overall variability measure.
2. Distance Indexes G
1
1 K
= y k M nk
*
G . (1)
n k =1
The index (1) can be normalised taking it to its maximum value (Fabbris, 2000):
1
K
* y k M nk
1
G 1 K
y k M f k ,
k =1
G = = = (2)
Max(G ) ( M m) n
*
M m k =1
where f k = n k n . We observe 0 G 1; in fact, G is equal to zero when all the units
of the population are gathered at the maximum value, it is equal to one when all the
units are gathered at the maximum distance from the optimum and take on increasing
values because of their gap from the optimum. Moreover such an index can be used
with any score scale and it is independent of the type of scale chosen (Fabbris, 2000).
Taking in formula (2) = 1 and = 2 we have the following indexes:
2
K
1
G1 = (M yk f k ) (3)
M m k =1
and
1
1 K 2
G2 =
M m k =1
( yk M ) 2 f k
, (4)
Referring to the CS of a Degree Course (DC), index (3) or (4) is calculated by the
distribution of answers associated to each question of the questionnaire submitted to the
students attending a teaching course. Within the logic of formative models a linear
combination of question indexes has to be considered to define those of dimension, a
linear combination of the latter to define the teaching indexes and lastly combining the
teaching indexes to obtain the overall DC index. To this end, since the base of the
formative model is the question index and since for each question there is a different
number of respondents1, a system of weights must be defined for the combination of the
indexes. The idea is to split up the data gathered into question clusters with a different
number of answers. In this way the population referred to is no longer made up of the
students but of the questions in the questionnaire, which will then be collocated on
different levels, corresponding to a different frequency of answers. For example the
dimension index is calculated by the arithmetical mean of the question indexes
weighted for the different frequencies of answer. This kind of weighting holds good at
any level we want to consider the index.
Before analysing in detail the partial indexes we should introduce the following basic
symbols:
k=1,,K answer score index;
1
The number of respondents to the single questions is variable both because of the different number
of students present at the single courses, as well as for the item non-response.
3
hd n jd total of answers to question j of dimension d for each teaching
h;
1 K
hd G1jd =
M m
M (y k )( hd f jd k ) (5)
k =1
and is the question index, that is the index made considering the distribution of the
answers to the j-th question of the d-th dimension in the questionnaires collected for
teaching course h.
To define a dimension index we have to aggregate the indexes (5) for the Jd
questions of the dimension we are examining. For this a simple mean of the indexes
hd G1 jd could be used if all the dimension questions have the same number of answers;
otherwise it would be necessary to fall back on a weighted mean, using as weighting the
total number of answers hd n jd to the single questions2.
Therefore, under the hypothesis of question clustering with different number of
answers, the dimension index is as follows:
Jd
( hd G 1j d )( hd n jd )
1 Jd K
(y
j d = 1d
hd G 1 = = M k )( hd f jd k )( hd jd ) (6)
Jd
M m
j d = 1d k =1
hd n jd
j d = 1d
2
As above hd n jd is very variable according to the questions jd; if not and every student answered all
*
the questions, hd n jd would equal to h n .
4
Jd
where the values hd j d = hd n j d
j d = 1d
hd n jd are the relative frequencies of answers of the
same dimension.
We note that formula (6) can also be defined considering the overall answers to Jd
questions, as following. In fact, starting from formula (6) and simplifying suitably, we
get:
1 Jd K
hd G1 = M
M m
(y
)( hd f jd k )( hd jd ) =
k
j d =1d k =1
1 Jd K
hd n j d k hd n j d
1 K
n.k
=
M m
M k n y =
M yk hd
=
j d =1d k =1 hd jd hd n M m k =1 hd n
1 K
=
M m
M ( y k )( hd f .k ) (6bis)
k =1
where hdn.k is the answer frequency of score k for all the questions of dimension d and
hdf.k is the relative frequency.
Naturally, using formula (6bis) the information about the question indexes is lost,
that is the information about the satisfaction level of each single attribute is lost.
Furthermore, in the absence of blank answers (see note (2)), formula (6) is like a
simple mean of the indexes hd G1 j d , as follows:
Jd
j d = 1d
hd G 1jd
hd G1= . (7)
Jd
Now we can define an index of a higher level, that is the teaching index hG1
including all dimensions. In the same way as before, we calculate the weighted mean of
Jd
dimension indexes hd G 1 , using as weighting the quantities
jd =1d
hd n jd = hd n . Therefore
we have:
(
d =1
hd G1 )( hd n)
h G1 = D
. (8)
d =1
hd n
In formula (8) for hd G1 we put formula (6) and, simplifying appropriately, we get:
5
D Jd
(
d =1 jd =1d
hd G1jd )( hd n jd )
h G1 = D Jd
; (8bis)
d =1 j d =1d
hd n jd
that is like the weighted mean of the question indexes, with weights equal to the number
of answers. So, we can arrive at the teaching index without considering the dimension
indexes; in this way however the information about indexes hd G1 , useful in comparing
the different teaching, are lost.
Moreover, with no blank answers, we simplify formula (8) in the following way:
d =1
hd G1 J d
h G1 = D
(9)
Jd =1
d
and the teaching index is still a weighted mean of the dimension indexes, but with
weights equal to the number of questions of each dimension.
In the same way, the overall CS index is made (Degree Course index): it is the
weighted mean of the teaching indexes h G1 , with weights equal to the total answers of
all the questions of each teaching course:
(
h =1
h G1 )( hn)
DC G1 = H
. (10)
h =1
h n
In this case too, with no blank answers, formula (10) is thus simplified:
( h G1 )( hn* )
DC G1 = h =1
H
, (11)
h =1
h n *
If the overall CS index is made by formula (4), with the symbols given, we have the
quadratic question index:
6
1
1 K 2
hd G2 j d = k
M m k =1
( y M ) 2
hd f jd k ; (12)
we must then find a suitable method of summary for such indexes, always under the
hypothesis of question clustering. Since formula (12), excluding the factor 1/(M-m), is a
weighted quadratic mean of the gaps ( y k M ) and since for such a mean, like the
arithmetical one, associative properties are valid, we propose a dimension index hd G2
given by the weighted quadratic mean of the indexes (12), with weighting equal to the
total number of the answers to Jd questions of the dimension itself; that is:
1
Jd 2
hd 2 jd hd jd
2
( G )( n )
G = jd =1d . (13)
hd 2
Jd
n
hd j d
jd =1d
1
1 K 2
hd G 2 = k
M m k =1
( y M ) 2
hd f . k . (13bis)
Like formula (6bis), formula (13bis) gives again the dimension index without taking
into account the clustering hypothesis and considering the overall distribution of the
answers to Jd questions of the same dimension; but in this case, as pointed out, the
information about the single attributes is lost.
Likewise the quadratic dimension indexes can be gathered to calculate the quadratic
teaching indexes and, from these, the quadratic degree course index, with the same set
of weighting as indexes G1.
As shown in paragraph 3, the overall CS index, calculated by the natural index (3) or
the quadratic one (4), is a complex index, made under the hypothesis of question
clustering for every level.
In fact it becomes interesting to evaluate the variability on those levels thereafter
suggesting opportune breakdowns.
7
distribution of the answers to the each question of the questionnaire may be substituted
by index hd G1 jd with weighting hd n jd 3. So the overall variability is:
Jd
( )
H D
2
hd G1jd DC G1 hd n jd
Var ( hd )
G1jd = E ( hd G1jd DC G1 ) =
2 h =1 d =1 jd =1d
H D Jd
. (14)
h =1 d =1 j d =1d
hd n jd
Jd
If we consider the partial indexes hd G1 and h G1 , and
jd =1d
hd n jd = hd n , the numerator
Jd Jd
( ) ( )
H D H D
hd n jd =
2 2
hd G1jd DC G1 hd G1jd hd G1 hd n jd +
h =1 d =1 jd =1d h =1 d =1 j d =1d
(15)
H D H D
(
h =1 d =1
hd G1 h G1 )
2
hd n + ( h G1 DC G1 )
h =1
2
d =1
hd n.
H D
If formula (15) is divided by h =1 d =1
hd n , the three addends to the second side are
[
E h E d ( hd G1 h G1 ) = Var ( d G1B )
2
] (17)
E h ( h G1 DC G1 ) = Var ( h G1B ).
2
(18)
Formula (16) is a measure of the variability within the dimensions, formula (17) instead
is the variability between the dimensions of each teaching course, while formula (18) is
the variability between the teaching courses. It follows that formula (16) together with
formula (17) is a measure of the variance within the teaching courses:
For formulas (19) and (18), the overall variability is broken down as follows:
3
The question index is a synthesis of the answer distribution ( y , n ), with variance
hd j d k hd jd k
E ( hd )
y jd k , where = h d j
2
d
k
( hd y jd k )( hd n jd k ) h d j k hd n j k . However this
d d
last represents the variability of answer scores; in this case it is of no interest, as the aim of this paper is to
analyse the variability of the indexes hd G1 jd .
8
Var ( hd )
G1 jd = Var ( h G1W ) + Var ( h G1B ). (20)
Var ( h G1 ) = E d (Var ( hd G1 )) + E d ( hd G1 h G1 ) ,
2
(21)
where the first addend to the second side of formula (21) represents the variability
within the dimensions and the second addend the variability between the dimensions of
the same teaching course.
The above formulas show that, when the CS index is obtained using the measure of
distance (3) it is possible, according to clustering, to breakdown the overall variance
within the partial variances and such breaking down holds good also for the CS indexes
at different levels. In fact, the higher level partial indexes are equal to weighted means
of the lower level partial indexes. Such a coincidence does not exist when the CS index
is given by the distance measure (4) which is a quadratic mean. In this case CS partial
indexes and the total one are quadratic means, while for the variance we have to refer to
arithmetical means. So with indexes G1 the means of the squares of the differences
between the lower level partial indexes and the higher-level partial index are variances,
while for the indexes G2 they are not. It follows that, for the latter the breakdown of the
overall variance under question clustering hypothesis is still valid, however formulas
(16), (17) and (18) must be formally different. In particular, indicated as hd G2 the
weighted arithmetical mean of the question indexes hd G2 jd with weighting hd n jd ; with
h G2 the weighted arithmetical mean of the dimension indexes hd G2 with weighting
hdn and, lastly, with DC G2 the weighted arithmetical mean of the teaching indexes h G2
with weighting h n , the formulas become as follows:
[ (
E h E d E jd ( hd G2 jd hd G2 ) )] = E [E (Var (
2
h d hd G2 ))] = Var ( d G2W ) (22)
E [E ( ]
G2 h G2 ) = Var ( d G2 B )
2
h d hd (23)
E h ( h G2 DC G2 ) = Var ( h G2B )
2
(24)
These are the variances within and between the partial levels.
Instead, if we want to compare the variability between the quadratic partial indexes,
formulas (22), (23) and (24) must be modified as follows:
9
H D
( G2 hd G2 )
2
n
[ ( ( ) )] = Var (
hd hd
G2W ) +
2 h =1 d =1
E h E d E jd hd G2 jd hd G2 d H D
(25)
h =1 d =1
hd n
( G2 h G2 ) h n
[ ]
h
E h E d ( hd G2 h G2 ) = Var ( d G2 B ) +
2 h =1
H
(26)
h =1
h n
E h ( h G2 DC G2 ) = Var ( h G2B ) +
2
( DC G2 DC G2 ) .
2
(27)
5. Conclusions
Two indexes to measure the CS of a university degree course have been proposed in
this paper using the same formative model. Such indexes are normalised and are made,
the first DC G1 , on a linear function of the distances between the actual satisfaction for
the different attributes and the expected satisfaction, objectively measured; the second
DC G 2 with the same distances of satisfaction but using a quadratic function. The first is
therefore like an arithmetical mean of absolute standardised gaps stating dissatisfaction
gaps with maximum quality (Fabbris, 2000); the second instead considers the same
gaps as a quadratic mean. For the properties of the means we observe that DC G1 DC G2 .
With the index DC G1 both the high and low gaps are considered equally, while with
DC G 2 the higher gaps take on greater importance. So the second index is used if we
want to emphasise a situation of great dissatisfaction; if the opposite is the case it seems
more natural to use DC G1 .
In the formative models it seems possible to breakdown the two indexes into three
levels which can be considered in different ways, thus providing different possibilities
for analysis, such as comparisons within the same level or between different levels.
The system of weighting, very important for a formative model, is equal for both
indexes, and has been created hypothesising the data collected from question clustering
to which different weight is given according to the actual number of answers received.
The indexes proposed are descriptive of the satisfaction of the student population
who attend teaching for a degree course. However the reference to the student
population is relevant only when the data is gathered, not when it is elaborated. In fact
for this the basic unit is not the student, but the single question of the questionnaire. The
number of students appears as number of different answers; so this coincides with the
weighting adopted.
10
REFERENCES
Cronin J.J., Taylor S.A. Jr. (1992), Measuring Service Quality: a Re-examination
and Extension, Journal of Marketing, 56, 55-68.
Cronin J.J., Taylor S.A. Jr. (1994), SERVPERF versus SERVQUAL: Reconciling
Performance-Based and Perceptions-Minus-Expectations Measurement of Service
Quality, Journal of Marketing, 58, 125-131.
Fabbris L. (2000), Un indice per misurare il divario tra la qualit attuale e quella
ottimale di un servizio, edited by Civardi M. and Fabbris L., Metodi e tecniche per le
rilevazioni assistite da computer (volume 3), Valutazione della didattica con sistemi
computer-assisted, CLEUP, Padova, 169-178.
Montinaro M., Nicolini G. (2002), La Customer Satisfaction: analisi storica e
validazione campionaria, edited by Frosini B.V., Magagnoli U. and Boari G., Studi in
onore di Angelo Zanella, Vita e Pensiero, Milano, 523-543.
Parasuraman A., Berry L.L., Zeithmal V.A. (1988), SERVQUAL: a Multiple-Item
Scale for Measuring Customer Perceptions of Service Quality, in Journal of Retailing,
1, 11-40.
Zanella A. (2001), Valutazione e modelli interpretativi di customer satisfaction: una
presentazione dinsieme, Universit Cattolica del S. Cuore, Istituto di Statistica, Serie
E.P., N. 105.
11
La serie dei Working Papers del Dipartimento di Economia Politica e Aziendale pu essere richiesta
al seguente indirizzo: Sezione Working Papers - Dipartimento di Economia Politica e Aziendale -
Universit degli Studi di Milano, Via Conservatorio 7 - 20122 Milano - Italy - fax 39-02-50321450
- Email: Dipeco@unimi.it. A partire dal numero 98.01, i working papers sono scaricabili dal sito
Internet del dipartimento, allindirizzo:
http://www.economia.unimi.it
The Working Paper Series of the Dipartimento di Economia Politica e Aziendale can be requested
at the following address: Sezione Working Papers - Dipartimento di Economia Politica e Aziendale
- Universit degli Studi di Milano, Via Conservatorio 7 - 20122 Milano - Italy - fax 39-02-
50321450 - Email: Dipeco@unimi.it. From number 98.01, working papers are downloadable from
the Internet website of the Department at the following location:
http://www.economia.unimi.it