In this study, short forms of the Child Perceptions Questionnaire for 11–14-year-olds (CPQ11–14) have been developed, tested for cross-sectional validity and reliability, and compared with the original instrument in terms of measurement sensitivity and discriminative properties. Each of the shortening techniques that were used, the item impact method and the stepwise regression, produced a 16-item and an 8-item measure. Measures of different lengths were developed to facilitate the administration of the questionnaire in clinical settings (16-item short-form) and in epidemiological surveys involving general populations (8-item short-form). To preserve the multidimensionality of the instrument so that it continues to conform to the WHO definition of health and the contemporary conceptualization of child health, the questions were selected from all domains in the CPQ11–14. Each domain contributed four questions for the 16-item short-forms and two questions for the 8-item short-forms. Previous research has indicated that versions of short-form questionnaires generated by the two approaches we used often differ in their content and measurement properties. The 16-item short forms generated in this study, i.e. CPQ11–14-ISF-16 and the CPQ11–14-RSF-16, had 14 questions in common (Table 1). The questions specific to these two questionnaires concern functional limitations and social well-being. On the contrary, the 8-item versions shared only 2 questions (Table 2). However, this difference in content had little effect on the performance of the two versions, reflecting the fact that Cronbach's alphas in each domain in the long form of the CPQ11–14 were high.
The questionnaires demonstrated considerable measurement sensitivity as the range of the scores showed that the short forms are detecting substantial variability in children's perceptions of their OHRQoL. The 16-item measures did not show floor-effects, while they were minimal for the 8-item questionnaires: 0.8% (CPQ11–14-ISF-8) and 4.1% (CPQ11–14-RSF-8). On average, all short forms detected higher levels of impact on the quality of life than the CPQ11–14. This can be explained by the fact that the questions selected for the short forms concern problems that children reported as the most frequent and the most bothersome. The lower scoring questions that were deleted when generating the short forms contribute to the CPQ11–14 scores and, consequently, lower the values of its standardized score.
The high correlations between the CPQ11–14 and the short-forms suggest that they are measuring the same construct. The association was somewhat stronger for the regression short-forms in comparison to impact short-forms, which can be explained by the fact the questions selected for the regression short-forms are those that explain the most variation in the overall scores of the CPQ11–14.
Reducing the number of questions in a questionnaire inevitably affects its content validity. Although content relevance remains intact, content coverage (i.e. the extent to which the questionnaire represents the construct of interest) is diminished. This, in turn, has the potential to compromise a measure's construct validity. Furthermore, since the reliability of a measure is a function of its length, the reduced number of questions may further attenuate construct validity by increasing the measurement error. However, the findings presented in this paper indicated that all short-forms have good construct validity since they were positively correlated with both global ratings. The correlation coefficients, as predicted, were lower for the rating of oral health than the rating of well-being. They were also either identical or very similar to the correlation coefficients found for the long form of the CPQ11–14 (0.23 and 0.40 for these two global ratings, respectively).
The construct validity of the short forms is further supported by the results of testing their ability to detect the hypothesized gradient in the impact of paedodontic, orthodontic and oro-facial conditions on children's quality of life. Although the score differences found on the CPQ11–14-RSF-16 were not statistically significant, they were in the expected direction and similar to the differences found on the CPQ11–14-ISF-16. The RV coefficients indicated that the statistical precision of the short forms in this study was similar to the statistical precision of the CPQ11–14, since all had values close to one. Gradients were also observed within the three clinical groups according to the severity of the condition. However, because clinical data were not available for some children, sample sizes were small and the differences mostly non-significant.
Although the reliability coefficients for the short forms were lower than those estimated for the CPQ11–14 (Cronbach's α = 0.91; ICC = 0.90), they all exceed standards for group-level comparisons [6, 21]. However, they suggest possible limitations of the short forms for smaller-scale cross-sectional studies, especially when the samples involved show low variations in their OHRQoL. The same holds for individual-level assessments since they require that reliability coefficients are at least 0.90 [6, 21].
A weakness of this study is that none of the short forms was administered on its own. Instead, the data collected in the validation study for the original questionnaire were used to evaluate their measurement properties. The possibility is that children may have responded differently had the short forms been the data collection instruments. However, it seems reasonable to assume that this is not very likely as Schofield et al.  found no significant differences in the mean summary scores when the SF-12 was embedded in the SF-36 as opposed to when it was administered by itself to an equivalent independent sample.
The study provides evidence about measurement sensitivity and discriminative properties (i.e. construct validity and reliability) of the 16-item and 8-item short forms of the Child Perceptions Questionnaire for 11–14-year-old children developed using the item impact method and stepwise regression. However, these are preliminary findings based on convenience sampling of a clinical population and further testing in replicated studies involving clinical and general samples of children in various settings is necessary. If the cross-sectional properties of the short forms are confirmed then, since they perform equally well but vary in their content, the one that is selected for a study would depend on the purpose of the investigation, the population studied and research context. This is of a particular importance with respect to the 8-item versions as they share only two questions. Moreover, if an 8-item version is used analysis of overall scale scores is possible but not analysis at the level of the individual domains. The number of items per domain is insufficient for this purpose.
A final consideration is whether the item impact or regression approach is better when developing a short form measure. From a statistical point of view the latter may be contraindicated because the distribution of the data derived from a quality of life questionnaire will, more likely than not, violate the assumptions of linear regression analysis. Moreover, the use of forward stepwise regression in this context may be compromised by the part-whole correlation effect (10) since it often results in the wrong variables being selected. Because of these problems Coste et al (10) suggest that an expert-based approach if preferable. While these statistical considerations are important, the study reported here suggests that, in practice, the regression approach performs reasonably well. The advantage of the item impact approach is that it selects those items of most importance to the people who will be completing the questionnaire who may be considered to be the ultimate experts concerning the impact of a given condition on the quality of life (11). Juniper et al (11) suggests that the choice of approach is largely a philosophical matter in which an investigator must decide whether patients' views or statistical considerations are of most importance. Locker and Allen (15) take the view that the method of developing a short form questionnaire is less important than its content and properties, a view that is supported by the results of this study. However, since different approaches can result in different short form instruments which may vary in their items and their properties, investigators shortening a measure should consider using more than one approach to determine the effect of method on outcome.