The second generation of statistical modeling
namely Structural Equations Modeling (SEM) distinguishes two measurement models:
reflective and formative latent measurement constructs (Edwards & Bagozzi,
2000).
In reflective construct, the construct is the
cause of the items designed to measure the construct. In other words, the items
can reflect the concept in the construct. However, all the items meant to
reflect the construct are expected to be correlated and, therefore, some of them
can be deleted without affecting the concept in the construct.
As an example, if job satisfaction construct is defined to be measured as a reflective construct, then one
can use items such as
- I like my job.
- I’m happy in my work,
- I am unlikely to want to leave this position. As illustrated in Figure 1
Figure
1: Reflective Construct
On the other
hand, in the formative construct, the items are the causes of the construct.
Meaningthat, the items meant to measure the construct form the concept in the
construct. These items, however, might not be correlated and, therefore, deleting
any item(s) may cause that some of the construct aspects are ignored.
For example, if job satisfaction construct is conceptualized as a formative construct,
one can use items such as
- I am satisfied with my pay,
- I have a good boss
- My work hours are ideal.
- I have many promotion opportunities.
- I enjoy working with my co-workers,….and so on as illustrated in Figure 2
As a researcher, the first and most important
step is to clearly define what we are planning to measure and whether our
construct is to be defined reflectively or formatively BEFORE designing a
questionnaire or generating items or questions. Specifically, we have first to
establish a clear conceptual definitions of our constructs and plan how we are
going to measure them.
Reference
Edwards, J. R., & Bagozzi, R. P. (2000). On
the nature and direction of relationships between constructs and measures.
Psychological Methods, 5, 155-174.
that's a great explanation sir....now that we have software available (like Smart PLS 2.0) to measure them, we can't afford to go ahead with the traditional research style of measuring almost everything reflectively...old research which usually predicted everything through reflective measures, will be still called as classics....but now if it is ignored by the researchers, it will be called as a blunder...
ReplyDeleteYou are right dear Pranav Kumar, that is why there has been a trend among the researchers to revisit many latent variables that were defined only reflectively to put them in there correct format. Doing that can be a good contribution to the knowledge.
Delete@ Pranav Kumar: is there anything called blunder in research?
Deletehi anonymous friend ...blunder is not used here as a research term, just an English word which seeks to explain here the deliberate mistake a researcher can commit….u see, we have clear cut methodology description for quantitative, qualitative and mixed methods….but if someone does not follow them properly due to lack of reading/exposure, and to make it worse, if he uses wrong research techniques, it would be kind of blunder…. isn’t it…. If not, then what it wud be?....E.g. conducting lived experience related phenomenological interviews and using only that data for coming up with a model by using grounded theory techniques? Is it not blunder? Term it in your way of course….same way in quantitative research, earlier operationalization usually meant using reflective style of dimension ascertainment and item wordings because there were not many software option available for evaluating formative constructs. But now, if someone ignores this issue, it will be a mistake according to me….so marketing mix has to be taken as a formative construct formed by 4Ps….direction of causality will be from dimension to construct….and regressing it is possible now using PLS….The simple question which researcher should ask is: whether these dimensions are forming the construct or the construct is reflected in these dimensions….the answer should guide us to take a final call….but if we finalize the nature of our construct without asking this question to ourselves, then there are chances of leading towards an operationalization mistake….thanks….
DeleteYes Dr. Abdulla sir.... but running models with formative constructs look quite tricky and challenging as well....
DeleteThis comment has been removed by the author.
ReplyDeleteThis comment has been removed by the author.
ReplyDeleteExcellent Sir. Can we design construct where items to First order variable are formative and First order variables (i.e. Many) to Second Order variable is Reflective than Second order variable to Third order variables are Formative and so up to Five Levels Formative?
ReplyDeleteExcellent Sir. Can we design construct where items to First order variable are formative and First order variables (i.e. Many) to Second Order variable is Reflective than Second order variable to Third order variables are Formative and so up to Five Levels Formative?
ReplyDelete