Conclusion

Assessing scale reliability with the coefficient alpha (or some other reliability index) should be one of the first tasks you undertake when conducting questionnaire research. If responses to selected scales are not reliable, there is no point performing additional analyses. You can often improve suboptimal reliability estimates by deleting items with poor item-total correlations in keeping with the procedures discussed in this chapter. When several subscales on a questionnaire display poor reliability, it might be advisable to perform a principal component analysis or an exploratory factor analysis on responses to all questionnaire items to determine which tend to group together empirically. If many items load on each retained factor and if the factor pattern obtained from such an analysis displays a simple structure, chances are good that responses to the resulting scales will demonstrate adequate internal consistency.