What will you do?

The theory and practice of how to ask questions about intended behaviours

Last time we looked at best practice for how we ask people about their behaviour. This will inevitably involve remembering, as we are asking about the past. In this post we are asking about the future - how we ask about the the likelihood of changing behaviour.

This is important, not least as when conducting a Randomised Control Trial we often do not have the luxury of measuring change in actual behaviour, instead we need to look at intent to change behaviour. For example we may be testing communications and want to have a measure of which of the comms options is likely to be more effective. Running a trial and then measuring subsequent behaviour change is often too onerous in terms of logistics, elapsed time and investment required.

This means that we need to measure intention as a means of determining whether participants will enact the behaviour, and how frequently they will perform it over time. Differences between the cells in terms of intentions indicates which comms option is most likely to result in behaviour change.

So how should we ask about behavioural intentions? Some simple guidance on best practice here:

How Likely/How Often?: Yes/no questions and likelihood questions are the most common form of behavioural intention questions, the former often acting as a useful screening question. They force the respondent to come down off the fence, but often result in inaccurate data. Likelihood questions are more respondent-friendly, but the responses don’t always transfer into percentages or projections. One way to overcome the pitfalls of both is to ask them together, in sequence. Another solution is to ask two different types of likelihood question: one could be a more general question, the other more specific, depending on a ‘what if’ scenario.

How Often/When?: Another way of asking about behavioural intent is to ask respondents how often they will engage in the behaviour over a specific time period. The questionnaire can either leave a blank for the respondent to fill in a numerical value or feature a frequency question with fixed intervals (0, 1-2, 3-5, 6-10, 11-20, 21-30). The former offers fewer constraints, but more possibility for error, and therefore used less frequently than the fixed interval question. This, though, requires planning to ensure that the intervals are appropriate, with zero always having its own separate category.

Likelihood Questions vs Frequency Questions: Researchers can ask how likely respondents are to perform a behaviour within a certain time period (from highly unlikely = 1 to highly likely = 9), or how often they might perform that behaviour within a similar period. There are pros and cons to both. Infrequent behaviours can skew frequency estimates towards zero, and therefore less variance and information than an estimate of likelihood. Yet in the context of frequent behaviours the data produced can be much more worthwhile. Wansink and Ray (1992) found that when products were consumed frequently, numerical estimates were more accurate, when consumed infrequently, likelihood estimates were more valuable. If the frequency of behaviour varies, it can be useful to employ both forms of questioning and use them to triangulate on the behaviour.

Conclusions

Behavioural scientists are often not survey methodologists so there is a danger that the all important dependant-variable of behavioural intention is not thought about carefully enough. Again, we can see the way in which market & social research is necessarily a critical element of behavioural science.

For those that are looking for more detail, this book is a great resource.