Title: A Comprehensive Framework for Service Quality: An Investigation of Critical Conceptual and Measureme
1A Comprehensive Framework for Service Quality
An Investigation of Critical Conceptual and
Measurement Issues Through a Longitudinal Study
Pratibha A. DabholkarC. David ShepherdDayle I.
Thorpe2000Journal of Retailing, Vol 76(2),
pp139-173
Yoo Jae Heung ITMGT, ICU
2Make it Clean
CS
BI
Personal Attention
Personal Attention
Expectation
Perception
Feature
SQ
Feature
Comfort
Feature
Comfort
Comfort
Gap (Disconfirmation)
Reliability
Reliability
Reliability
3Objective
- Propose and test a comprehensive framework of
service quality - Conceptual
- Critical measure issue
4Summary of Conceptual Issues
- Are relevant factors related to service quality
better conceived as its components or its
antecedents? - Which format increase our understanding of SQ and
which offers greater predictive power? - What role does CS play in the framework of SQ
prediction behavioral consequences? - Does it add greater predictive power over that of
SQ? If so, does CS act as independent determinant
or does it mediate the effect of SQ on behavior?
5Measurement Issues in SQ
- Debate on whether SQ should be measured as
perceptions or as disconfirmation (difference
between perceptions and expectation) - SQ measurement is also tied to disconfirmation.
The computed difference scores for
disconfirmation have problems of reliability,
discriminant validity, and variance restriction - Two-part measures (measures of expectations and
perceptions (Parasuraman, Berry, Zeithaml) - Direct disconfirmation measures (Brown,
Churchill, Peter) - The validity of a cross-sectional vs. a
longitudinal research design (one where
expectations are measured before the service is
delivered and perceptions are captured after the
services) for measuring SQ. - For researchers using the disconfirmation
approach, it is not clear whether both
expectations and perceptions should be measured
in a cross-sectional or whether a longitudinal
design is more appropriate.
6Summary Measurement Issues in SQ
- Are perception measures superior to
disconfirmation measures? - Is measured disconfirmation superior to computed
disconfirmation? - Is a cross-sectional design adequate or does a
longitudinal design offer significant advantage?
7Why antecedents model?
- No study to date has conducted a rigorous
examination of the factors as antecedents and
none has explored which model is better for
understanding SQ - Why antecedents model?
- Component view failed to capture the effect of
the relevant factors and failed to capture
customers overall evaluations of SQ as a
separate, multi-item construct. - P1 Factors relevant to SQ act as antecedents to
overall evaluation of SQ rather than acting as
its components
8Role of CS within the SQ Framework
- Does CS have an incremental effect over that of
SQ on behavioral intentions, and is this effect
an independent or a mediating one? - (SQ - CS or CS-SQ, CS and SQ are independent)
- It is expected that CS will have a mediating role
on behavioral intentions rather than an effect
independent of SQ - P2a CS mediates the influence of SQ on
behavioral intension (BI) rather than both
variables having an independent effect on BI. - P2b CS mediates the influence of SQ on BI rather
SQ mediating the influence of CS on BI.
9Disconfirmation vs. Perceptions
- Traditionally, SQ has been conceptualized as
disconfirmation process but disconfirmation
paradigm suffers due to problems with measuring
expectations. - Increasingly, researchers are simply measuring
perceptions as indicators of SQ - Some researchers have compared computed
difference scores w/ perceptions to conclude that
perceptions are a better predictor of SQ than
disconfirmation. - P3a Perception measures are superior to measured
disconfirmation (direct score ask the gap
directly) - P3b Perception measures are superior to computed
disconfirmation (difference score measure
expectation and perception respectively then
compute the gap)
10Difference Scores vs. Measured Disconfirmation
- Debate in the SQ on the use of difference(computed
) score vs. the measured disconfirmation (Brown,
Churchill, Peter, 1993 Parasuraman, Berry,
Zeithaml, 1993) - Difference Score Perception Expectation
- Measured Score direct mental estimation of
perceptions compared to expectation - If both types of disconfirmation are similar in
their predictive ability, explicit expectations
data are not needed. Also if measured
disconfirmation has the same diagnostic power as
computed disconfirmation, then psychometric
problems associated with the latter can be
avoided. It is clear that there is greater
support for measured disconfirmation - P4 Measured disconfirmation is superior to
computed disconfirmation
11Cross-Sectional vs. Longitudinal Design
- The majority of empirical conducted to measure SQ
have been cross-sectional where both expectation
and perceptions were measured after the service
had been delivered. - This approach assumes that expectations before
the service are identical to expectations after
the service and does not account for the fact
that expectations may change over time or after
the delivery of the service - Longitudinal studies are cumbersome, costly, and
by definition, time consuming - P5 A cross-sectional research design is superior
to a longitudinal design for measuring service
quality.
12Testing Models
Testing Measures
13Methodology
How to make the ugly Frog to a gorgeous Prince?
How?
How??
How???
How????
There is no Magic !
14Institutional customers of the pictorial
directory division churches
Context
Sample
Longitudinal study 397 usable pairs
Qualitative Research
3 FGI ascertaining 4 Factors Reliability,
personal attention, Comfort, Features
Scale Development for Exogenous Variables
Expectation, Perception, Measured Disconfirmation
Measures of Endogenous Variables
Overall SQ, CS, Intention
Data Collection Procedures
Telephone interviews, contacted twice First call
expectation before service delivery Second call
Perceptions and measured disconfirmation after
service
Confirmatory Factor Analysis
For 21 items, CFI( Comparative Fit Index), Drop
items, re-examination Chronbachs alpha for
reliability
Other Tests to Validate Constructs and Measures
Discriminant validity of CS and SQ constructs Was
ascertained using a chi-square difference test
Variance Restriction and Other Patterns
S.D were similar ranged form 0.78 to 1.59 The
gaps for all the items followed a similar
pattern Irrespective of how they were measured.
15Result of CFA and Reliability Test
Chi-Square difference significant at P
0.9 Cronbachs Alpha 0.7
Distinct and Reliable Construct
16Correlations
Additional Chi-square difference test for pairs
of constructs with higher correlations (The
result was significant at p
discriminant validity among the constructs
Collinearity test for the exogenous variables and
VIF (variance inflation factor) test showed Lack
of Collinearity among These variables for all
three mesures (VIF Correlations SQ and CS with CS SQ and CS with
BI CS with BI
17Analytic Results
- Model Fit Regression in SEM with LISREL
- Shows relative importance of the predictor
variables, as long as the fit is acceptable. - Compared to traditional SEM, this technique has
the added advantage of parsimony and is more
robust against violations of assumptions about
multivariate normality - F-value from Regression analysis , RMR, NNFI, CFI
used for check model fit - RMSEA is not an appropriate fit index given
extremely low degrees of freedom - Check the Significant Gammas and absence of
negative Gamma for check model specification
Effect sizes and R2 value to assess the strength
of models - Basic simple correlations and sophisticated
omitted paths analysis
18Test P1 (1/P1)
19Result of Testing (2/P1)
- Model Fit F-value, SD RMR, NNFI, CFI
- Significant Gamma and Absence of negative Gamma
- R2 values for BI
- Correlation of factor with SQ than BI
R2 are higher when CS acts mediating role (see
2/P2)
20Testing P2
21Result of Testing P2 (2/P2)
Acceptable
Fewer Significant Gamma
Higher effect size than Independent Model
Unacceptable
R2 are not much different but this is higher than
the case of SQ as a determininant (see 2/P1)
22Testing Overall Framework
23Result of Testing Overall Framework
- Good Fits
- Factors act as antecedents
- CS plays a mediating role
Computed disconfirmation model is not tested
because part of that dataset is identical to the
perceptions data, and would give misleading
results
24Testing Measures
- Computed Disconfirmation is evaluated against the
other two measures for test of P3b, P4, P5 - Measured Disconfirmation is evaluated against
Perception for a test of P3a - Using same criteria that were used for comparing
models Parsimony in Data collection
See the Proposition Map if you want to remind
25Computed vs. Perception and measured
Disconfirmation (P3b, P4, P5)
Low effect size
- Computed Disconfirmation shows Lowest F-value,
Unacceptable NNFI - Computed and Measured - Negative or
Non-significant Gamma whereas Perception has No
in Antecedents Model - Computed is far lower R2 value for SQ as well as
value for BI despite the overlap of the data - Computed - lower effect size
- Computed - not easier to collect data because
it requires twice (Not parsimonious)
26Measured Disconfirmation vs. Perceptions(P3a)
Higher F-value
Higher R2 but the difference Perception has Higher Effect Size
Perception has Better fit (although both are
unacceptable)
So, It is partially Supported
27Conclusion/Discussion
It looks more clear, now! Thank You!