3.5 Measurement Instruments
3.5.1 Data preparation
3.5.4.1 Descriptive statistics and item analysis
Item analyses of the BSSS revealed a reliability coefficient of .764. This fell comfortably above the generally accepted, yet arbitrary, cut-off point (.70) stipulated by Nunnaly (1978). After inspection of the item analysis statistics, item BSS07 was flagged as a possible poor item.
The inter-item correlation matrix indicated that the inter-item correlations of BSS07 (ranging from .064 to .601) were generally lower in comparison to those of the other items. The deletion of this item would have incurred an increase, albeit very small, in the Cronbach’s alpha (Δ = .001) resulting in .765. Based on this marginal increase, it was decided not to delete the item. The scale is already very short and therefore, it was argued that it would not be sensible to delete item BSS07 from the item pool.
Table 3.8
The mean, standard deviation and reliability statistics for the BSSS
BSSS Number of items M SD ΑΑΑΑ
BSSS 8 24.64 5.912 .764
Note: BSSS = Brief Sensation Seeking Scale
3.5.4.2 Confirmatory factor analysis
3.5.4.2.1 Measurement model specification and data normality
SEM was used to perform CFA on the BSSS measurement model which was specified to consist of eight observed variables (X’s) and one unmeasured latent factor (ξ, i.e. Sensation Seeking) with single-headed arrows from the ξ to the X’s representing the proposed regression of the observed variable onto the latent factor (λs).
The univariate and multivariate normality of the indicator variables were investigated via PRELIS (Jöreskog & Sörbom, 1996a). The null hypothesis of multivariate normality was rejected (skewness and kurtosis: χ2 = 35.374, p = .000). RML estimation was employed to derive the model parameter estimates.
Table 3.9
Skewness Kurtosis Skewness and Kurtosis Value Z-score P-value Value Z-score P-value Chi-square P-value
6.322 5.072 0.000 85.692 3.106 0.002 35.374 0.000
3.5.4.2.2 Evaluation of the measurement model
The measurement model, in this instance, represented the relationship between the
Sensation Seeking variable and its manifest indicators. The aim of the CFA was to
determine whether the operationalisation of the latent variable, i.e. Sensation
Seeking was successful. A graphical representation of the BSSS measurement
model is presented in figure 3.2.
The results of the CFA conducted with LISREL 8.80 (Jöreskog & Sörbom, 2002) are reported in table 3.10. The exact fit of the BSSS measurement model was tested by evaluating the S-Bχ2 statistic. A Satorra-Bentler Scaled chi-square value of 90.792 with 20 degrees of freedom and p = .00 was achieved. Thus, implying that the null hypothesis of exact fit (RMSEA = 0) should be rejected (p < .05).
The null hypothesis of close fit was tested by LISREL and is shown in table 3.10 as the P-Value for Test of Close Fit (RMSEA < .05) = .000. Consequently, the close fit null hypothesis was rejected and it was concluded that the measurement model did not obtain close fit. When a sample has a magnitude of less than 250 with items equal to or less than 12 observed variables, as is the case in the current research study, the CFI and NNFI should be greater than .97 to ensure that a misspecified model is not accepted. In addition, Hair et al. (2006) suggested that the SRMR and RMSEA should be smaller than .08. The CFI value of .880 and NNFI of .832 in this instance were well below the suggested cut-off values. The SRMR (.0845) marginally missed the cut-off value. The RMSEA (.132) and the lower boundary of the 90 percent confidence interval were above the suggested cut-off value. Of the eight items, all but one, item BSS02 (.315) obtained significant factor loadings above .40 (t ≥ 1.64). The completely standardised loadings ranged from .474 to .691.
Table 3.10
Goodness of fit statistics for the BSSS measurement model
Goodness of Fit Statistics
Normal Theory Weighted Least Squares Chi-Square 112.581 (P = 0.00) Satorra-Bentler Scaled Chi-Square (S-Bχ2) 90.792 (P = 0.00)
Degrees of Freedom 20
S-Bχ2/ df 4.5396
Non-Normed Fit Index (NNFI) 0.832
Comparative Fit Index (CFI) 0.880
Root Mean Square Residual (RMR) 0.133
Standardised RMR 0.0845
Root Mean Square Error of Approximation 0.132
90 Percent Confidence Interval for RMSEA (0.105; 0.160) P-Value for Test of Close Fit (RMSEA < 0.05) 0.000
Based on this basket of evidence presented above, mediocre fit was concluded and the uni-dimensionality of the items was questioned. After considering the fact that the measure was developed with the intention of retaining the same basic content structure of Zuckerman’s Sensation Seeking Scale Form-V, it was decided to conduct a multi-dimensional CFA. In this instance the model was specified to consist of four unmeasured latent factors (ξ’s). Each of the four dimensions, i.e. thrill and adventure seeking, experience seeking, disinhibition and boredom susceptibility in the SSS-V are represented by two items in the BSSS. A graphical representation of the BSSS measurement model is presented in figure 3.3.
Figure 3.3. Measurement model of the BSSS (standardised solution; multi-dimensional)
Table 3.11 summarises the fit indices obtained for the measurement model. This model achieved a Satorra-Bentler Scaled chi-square value of 19.689 with 14 degrees of freedom and p = .140 (table 3.11). Thus, implying that the null hypothesis of exact fit should not be rejected (p > .05). It can be concluded that the measurement model provides a perfect account of the manner in which the latent
variables manifest themselves in the indicator variables. All other fit indices reported in table 3.11 corroborated the inference of a well-fitting measurement model. The completely standardised factor loadings revealed that all of the items obtained significant factor loadings above .40 (t ≥ 1.64) and ranged from .618 to .973, with exception of one factor loading being below .40. This was item BSS02, with a factor loading of .330.
Table 3.11
Goodness of fit statistics for the BSSS measurement model (multi-dimensional)
Goodness of Fit Statistics
Normal Theory Weighted Least Squares Chi-Square 23.020 (P = 0.0599) Satorra-Bentler Scaled Chi-Square (S-Bχ2) 19.689 (P = 0.140)
Degrees of Freedom 14
S-Bχ2/ df 1.4064
Non-Normed Fit Index (NNFI) 0.981
Comparative Fit Index (CFI) 0.990
Root Mean Square Residual (RMR) 0.0554
Standardised RMR 0.0381
Root Mean Square Error of Approximation 0.0446
90 Percent Confidence Interval for RMSEA (0.0; 0.0869) P-Value for Test of Close Fit (RMSEA < 0.05) 0.534
3.5.5 Emotional regulation (Emotional Self-Control and Emotional Self-