Microsoft Word 2007 ichrie conference Proceedings Final-Final 06-06-07. doc

 
2007 Annual International CHRIE Conference & Exposition 
191
 
Table1 
Comparison of Overall Fit Indices for Proposed and Modified Models (N = 198) 
Model 
χ
2
Df 
Χ
2
/
df
NNFI CFI RMSEA 
Δ
χ
2
Model A
 a
1209.1 681 1.78 .827 .841 .063 
- 
Model B
 b
391.2 278 1.41 .931 .941 .045 
817.9
*
Note: NNFI = non-normed fit index; CFI = comparative fit index;
RMSEA = root mean square error of approximation; 
Δ
χ
2 
= 
χ
2
difference 
between two models (i.e., 
χ
2 
of Model 1 minus
χ
2 
of Model 2 = 562.3)
a 
Model A: Initial model containing 39 items derived from the exploratory 
factor analysis.
b 
Model B : Model A minus (V1, V2, V5, V6, V12, V13, V14, V20, V24, V25, V27, V34).
In the first step, a proposed measurement model (Model A) obtained from EFA was tested. According to 
overall fit indices, the proposed model did not produce a good fit with the data, 
χ
2 
(681) =1209.1, p<.001
 
(
χ
2
/df=1.78, CFI = .82, NNFI = .80, RMSEA = .06) in this initial step. The initial model labeled as model A does not 
show acceptable fit indices (CFI = .827 and NNFI = .841), although its chi-square ratio (< 3) and RMSEA (< .10) 
are in acceptable ranges. Thus, the proposed model needed further modifications. When the proposed model was 
required for the improvement in fit –in other words, the fit was found to be inadequate - modification procedures 
were conducted to identify candidate observed variables for deletion from the measurement model. These 
procedures are customary to maximize the fit (Bollen, 1989). Thus, the model modification procedures proceeded to 
identify observed variables that had low factor loadings, significant cross-loadings, and large residuals using 
standardized factor loadings. The MI test objective was to determine whether, in subsequent runs, models would 
better represent the data, with certain parameters specified as free rather than fixed. As the minimum cut-off, it was 
suggested that a standardized factor loading should be greater than .50, and each MI should not exceed 100 (Kline, 
1998). AMOS provided three measures. After removing observed variables based on three criteria, overall fit 
measures (CFI, NNFI, RMSEA, and 
χ
2 
difference test) were used iteratively to determine whether the CFA model 
fitted data well.
In the second step, thirteen observed variables were identified with low factor loadings (below suggested 
level of .50 for the expected constructs.), shared factor loadings, and shared large residuals with other observed 
variable loadings (above MI suggested level of 100). Thus, those variables were removed for modified model 
(Model B). Model B was re-estimated. Results from running this model showed that all fit indices suggested a good 
fit of data, 
Δ
χ
2 
(278) = 391.2, p<.01, 
χ
2
/
df 
= 1.4, CFI = .941, NNFI =.931, RMSEA = .045. The modified model 
surpasses the proposed model on all fit criteria, which confirmed that the modifications were meaningful. Through 
this process to assess a model’s fit, modified model of 
experience 
(Model B) resulted in seven factors (26 items) 
distributed as follows: Sales environment (7 items), Driving Benefit (4 items), Shopping Convenience (3 items), 
Accessibility (4 items), Utility (3 items), Incentive (3 items), and Brand Trust (2 items) as shown in Table 2. The 
modified final CFA model has the
same factor structure with seven factors as does the EFA factor model. It is 
purified from EFA model (39 items to 26 items).