Predicting behaviour: comparing the performance of factual versus attitudinal approaches

Abstract

Despite years of refinement and improvement to questionnaire design, the need remains to identify effective, but ‘user-friendly’ questioning procedures to predict behaviour, without compromising predictive performance. Questionnaires developed to predict behaviour typically use an attitudinal approach. However, those types of questionnaires are associated with increased respondent fatigue, decreased question comprehension, opposition to question tone and demand effects. An alternative approach involves use of factual questions. This study compares ease of application in the field as well as predictive performance of the two approaches to questionnaire design in the context of predicting willingness to donate blood. Results indicate that factual questions are quicker and easier to apply in the field than attitudinal questions, while predictive performance of the approaches is equivalent.

Keywords:

• Survey design
• question type
• attitudinal questions
• factual questions
• predicting behaviour
• blood donation

Notes

1. Three variables were highly correlated, with values ranging from r = .74 to r = .86 (p =  < .01): donation experience (categorisation of a respondent’s level of blood donation experience from ‘never donated’ to ‘regular donor’), donation frequency (how many times a respondent had donated) and last donation (the recency of a respondent’s last donation).

2. Additional variables are not typically included in TPB methodology as it assumed that the effect of other variables on behaviour is mediated through the main variables attitude, subjective norm and perceived behavioural control (Ajzen, 1991). However, in an attempt to improve the predictive validity of the model over and above the standard variables and moral obligation, we entered age and gender into the regression equation. We found that inclusion of the additional variables had little effect on the model’s predictive performance as estimated by Nagelkerke R² .21 with age and gender included, compared with Nagelkerke R² .20 without the additional variables.