The item analysis was performed separately for the self-deception enhancement and impression
management subscales of the SAPI SD-scale. The objective of the item analysis was to evaluate
the extent to which the items of the SD-scale successfully fulfil the task they were allocated under the design intention assumed under operational research hypothesis 1. The critical cut-off value for Cronbach alpha was set at .80. According to Field (2013) an alpha value of .70 is also seen as suitable. According to Nunnally and Bernstein (1994) reliability coefficient values are acceptable when they are equal to or above .95 and unacceptable at values below .70. Vorster (2010) claimed that the researcher can accept a reliability coefficient value of .65 if an
83 exploratory research paradigm is used, however a reliability coefficient value lower than .60 cannot be tolerated (Vorster, 2010).
As seen in Table 14 the Cronbach alpha values for both subscales were below the critical cut-off value of .80. Especially the self-deception enhancement subscale returned an extremely low Cronbach alpha value of .444. The impression management subscale returned a Cronbach alpha value of .553. This resulted in the examination of the inter-item correlations and the items statistics in order to determine whether the internal consistency of the subscales could be improved through the refection or deletion of any items.
Table 14:
Reliability Statistics for the SD Subscales
Subscale
Self-deception enhancement Cronbach's Alpha Cronbach's Alpha Based on
Standardized Items
N of Items
.444 .504 5
Impression management Cronbach's Alpha Cronbach's Alpha Based on
Standardized Items
N of Items
.553 .629 7
In Table 15 the inter-item correlations for the two Social Desirability subscales are given. For both subscales the inter-item correlations were generally low but especially so in the case of the
self-deception enhancement subscale. The mean inter-item correlation for the self-deception enhancement subscale was .145 (table not shown) and for the impression management scale,
.220 (table not shown). The item-total statistics are shown in Table 16. In the case of the self-
deception enhancement subscale deletion of none of the items would bring about an increase in
the Cronbach alpha value. In the case of the impression management subscale the deletion of RQ53_SD was suggested.
According to Worthington and Whittaker (2006) it is unusual to retain all the initial research items and researchers should remove items that fail to contribute meaningfully to any potential factor solutions. Not deleting items that need to be removed makes it difficult to arrive at a final decision regarding the numbers of factors to retain. Worthington and Whittaker (2006) recommended that researchers retain only the potentially meaningful items early in the process in order to optimise scale length once the factor solution is clear.
In the current study, however, all the items were retained for further analysis. The reasoning behind this decision is argued below.
Earlier it was argued that operational research hypotheses 2, 3 and 4 assume that all 12 items of the SAPI SD scale are complex items that load on at least two latent dimensions.
Table 15:
Inter-item correlations for the two SD Subscales
Self-deception enhancement RQ27_SD Q035_SD RQ168_SD Q188_SD Q223_SD RQ27_SD 1.000 .098 .123 .251 .157 Q035_SD .098 1.000 .225 .117 .103 RQ168_SD .123 .225 1.000 .127 .104 Q188_SD .251 .117 .127 1.000 .166 Q223_SD .157 .103 .104 .166 1.000 Impression management RQ53_SD Q090_SD Q102_SD Q109_SD RQ129_SD RQ131_SD RQ252_SD RQ53_SD 1.000 .024 .071 .163 .037 .018 .026 Q090_SD .024 1.000 .167 .120 .203 .191 .279 Q102_SD .071 .167 1.000 .198 .073 .171 .163 Q109_SD .163 .120 .198 1.000 .112 .202 .168 RQ129_SD .037 .203 .073 .112 1.000 .203 .235 RQ131_SD .018 .191 .171 .202 .203 1.000 .264 RQ252_SD .026 .279 .163 .168 .235 .264 1.000
Moreover it is hypothesised that the general factor is unrelated to the two narrower social desirability factors in the measurement models hypothesised by operational research hypotheses 2 and 4. Likewise, under operational research hypothesis 3 the two method factors are hypothesised to be unrelated to the two narrower social desirability factors in the hypothesised measurement model.
Table 16:
Item-Total Statistics for the two SD Subscales
Self-deception enhancement Scale Mean if Item Deleted Scale Variance if Item Deleted Corrected Item-Total Correlation Squared Multiple Correlation Cronbach's Alpha if Item Deleted RQ27_SD 14.35925 5.361 .244 .085 .390 Q035_SD 15.05712 4.158 .237 .065 .396 RQ168_SD 15.20403 4.390 .254 .071 .374 Q188_SD 14.16343 5.279 .259 .091 .381 Q223_SD 14.20594 5.060 .205 .051 .409
85 Impression management Scale Mean if Item Deleted Scale Variance if Item Deleted Corrected Item-Total Correlation Squared Multiple Correlation Cronbach's Alpha if Item Deleted RQ53_SD 19.28997 12.176 .091 .029 .578 Q090_SD 20.19398 9.974 .318 .121 .498 Q102_SD 19.01086 10.977 .260 .080 .522 Q109_SD 19.19856 11.185 .296 .103 .511 RQ129_SD 20.40030 10.861 .278 .095 .515 RQ131_SD 18.80709 10.187 .335 .130 .492 RQ252_SD 20.09176 9.548 .370 .159 .474
In terms of operational research hypotheses 2, 3 and 4 relatively lower levels of internal consistency should therefore be expected. Categorising items in smaller groups based on their hypothesised loadings on two factors will therefore also not solve the problem since the two factors involved are assumed to be orthogonal. The noisy, low internal consistency data for the two social desirability subscales serves as evidence against the position hypothesised under operational research hypothesis 1 that the SD-scale of the SAPI comprises a set of simple items that measure one of two correlated social desirability factors. The results depicted in Table 14, Table 15 and Table 16 rather unequivocally lead to the conclusion that the items of the SD-scale do not successfully fulfil the task they were allocated under the design intention assumed under operational research hypothesis 1. Although the low internal consistency reliability findings are consistent with operational hypotheses 2, 3 and 4, these findings cannot be regarded as sufficient to conclude support for any these hypotheses. Neither do they assist in deciding which position provides a more plausible account of the mechanism that produced the observed inter-item variance-covariance matrix.