Construct Validity
Perhaps the most precious possession in marketing research, and in particular quantitative analysis, is a construct with some form of validity; say predictive validity. For example, the marketing research construct customer value purports to have predictive validity. That is, it purports to be a leading indicator of changes in market share or in other words, actual changes in market share correspond with the predicted change arising from the marketing research.
When Constructs Fail
What happens when a research supplier claims to provide constructs that predict outcomes, but there is no such validity? Who is setting the standards and moreover, who is policing them? Is it the responsibility of the research buyer to catch out such suppliers? One could assume that those who have a caveat emptor philosophy would argue that the research buyer who does not insist on evidence of construct validity deserves all they get. Also, that the market will ultimately discipline such suppliers and indeed, buyers. But does it? The research buyer can find themselves in an awkward situation and become innocently complicit with the supplier’s failure.
The Source of Failure
It is often difficult to anticipate exactly how a respondent will interpret any given question, even though it surely was written with the intention of addressing a specific construct. Take for example loyalty; in some markets the buyer’s prediction of remaining is confounded by their inability to leave due to say, contracts for mobile phones or security for business banking.
In practice, the whole process that involves the identification and subsequent measuring of a construct, should begin with qualitative research. Here, the construct can be discussed and the driving forces teased out. At the end of a successfully run qualitative stage, the researcher should have identified the respondents’ stated behavioural intentions and developed a construct that will capture that intention.
Understood as Intended
It is then a matter of capturing the construct in words and using it to survey the relevant customers. All being well, the dependent variable question, that is, the construct question, will be understood by the respondents as intended by the researcher and the subsequent modelling of the drivers against this construct will provide a theoretical model that reasonably approximates actual behaviour.
However, if the construct question (dependent variable) is poorly defined, the likelihood of a useful model is low. This is because a poorly defined construct is open to misinterpretation by the respondent, which can result in their understanding of the construct being different from the intended meaning.
Importantly, a successful model is determined by two criteria. Firstly, it must be statistically valid. Secondly, it must also have construct validity. At its origin, construct validity arises from wise qualitative research and an experienced quantitative modeller.
This still leaves the matter of accountability when a construct fails.
