I have been struggling to ensure conceptual alignment in my research.
I'm working on questionnaire development in behavioural science (management) and have specified the area of interest as a formative construct (i.e. it is formed by it's indicators/items, rather than the items/indicators being a reflection of as in personality).
Index construction with formative measures typically employs stats such as PLS-SEM, but emphasizes that item selection is not resolved exclusively with statistics. The items must 'fully' represent the final construct, therefore, items may not be eliminated from a survey purely on the basis of poor statistical performance.
For me, this (and the role of qualitative research in item generation), lead to a mixed-methods approach that aligns with my ontology.
However, I've been trying to situate my research in a paradigm. I've been told pragmatists (which is where I originally sat fundamentally disagree with the existence of formative constructs). And as far as I can tell critical realism aligns well, right up until the point where I'm using any form of statistics.
Whilst I understand CR has a largely subjectivist epistemology and a belief that we operate in an open system, there is a consensus that the 'real' exists. My instinct is to argue that in gathering the views of lots of people and applying statistical modeling (such as SEM), then I'm attempting to get as close as possible to causal tendencies. In this way, I'm proposing that these mechanisms combine in this way in a great many instances, whilst acknowledging that it is not a 'law' and there will be some contexts in which these general tendencies don't apply...
However, my other 'problem' with CR and formative constructs is that index development in formative constructs requires sampling the entire nomological network (i.e. all aspects of the construct are identified). Whereas, CR appears to suggest it's highly unlikely/impossible to capture the whole of reality. This feels conflicted.
I'm really struggling to find a theoretical stance in which feels like it fits with my world view - that there is a reality; that we all see that reality through our own lens, but if we combine lots of different perspectives together we get pretty damn close to reality.
Any help would be very much appreciated as I'm feeling lost at present.
Here a few thoughts. First, as a pragmatist, I have never encountered any problems with formative constructs. Indeed, one's thinking has to begin in a formative fashion in order to put anything into action.
Second, with regard to CR and sampling the full nomological network, I would treat that as an "aspirational" goal, rather than ever knowing for certain how close we are to reality.
Finally, I am not sure what you mean when you say that, "items may not be eliminated from a survey purely on the basis of poor statistical performance." If your goal is to construct a measure for a survey that captures some underlying concept, then each item has to be correlated adequately with the other items that are measuring that concept. Taking coefficient alpha as a basic example, adding items that have a "poor statistical performance" to a scale will degrade the performance of the scale as a whole. To my way of thinking, if your goal is to generate a measure that will be used in a statistical framework, then its performance will have to be assessed according to statistical standards.
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Energy