A recent article (below) has challenged the "multidimensionality" of a popular instrument. Actually, people are used to calculate just one overall score, implying just one factor, so it does not change much the current practice. However, what could be the basis to aggregate different factors in one overall score?
Article The Oral Health Impact Profile-14: A unidimensional scale?
The is a lot of research in HRQoL where Rasch measurement has been used to develop / confirm the unidimensionality of instruments. Try
http://scholar.google.com/scholar?hl=en&q=HRQoL+Rasch&btnG=&as_sdt=1%2C5&as_sdtp=
to see articles adopting that position.
You might be interested in: https://iadr.confex.com/iadr/13iags/webprogram/Paper174560.html
Thanks Daniel, that was essentially the same results we found. I wonder if others have also found this and also if OIDP also has 1 dimension.
Dear Daniel,
I noticed that there are two conclusions in the abstract, which one looks better?
a) "OHIP was unidimensional enough to be sufficiently described with one summary score";
b) "a four-factor structure of Psychosocial Impact, Orofacial Pain, Oral Function, and Orofacial Appearance as dimensions appeared as an OHRQoL model..."
As you have worked with OHIP49, I wonder if all 49 items loaded into one factor, and, if you fitted a four-factor model, did the items load nicely into the theoretical dimensions? Sometimes, we get the postulated number of factors, but the items load into factors that are very difficult to interpret.
Dear Roger,
I wouldn't say that one conclusion is better than the other - both conclusions belong together. The finding of the four-dimensional structure for OHIP is an important contribution to the theoretical understanding of perceived oral health. It suggests that the OHIP measures four kinds of patient-perceived oral health – Orofacial Pain, Oral Function, Orofacial Appearance, and Psychosocial Impact. However, from a practical point of view, one summary score is sufficient to characterize oral health-related quality of life and is suggested for research and patient care.
We explored the dimensionality of the OHIP in an exploratory factor analysis and confirmed the models in a confirmatory factor analysis. The four-factor model showed strong item loadings, sufficient model fit, and good interpretability, with high inter-factor correlations. We also found a good fit of the unidimensional model suggesting that OHRQoL can be sufficiently described with one score.
We already prepared two papers for these analyses and will submit the papers within the next couple of days. Hopefully, it will not take so much time to get our results published.
So, you have four distinctly different factors which are meaningful, but it's good enough to summarise these to a single score. Surely this is contradictory.
...maybe at first glance, but I don't think so. The good fit of the unidimensional model and the high inter-factor correlations in the four-factor solution suggest that OHRQoL can also be sufficiently described with one score. This is also in line with the extensive experience in using the OHIP summary score as a valuable measure of patient-reported outcomes. Furthermore, our finding is not too different from other HRQoL instruments such as the SF-36, which has also different dimension, but scores can be combined into sum scores for physical and mental health. However, our findings warrant further research for refinement of the dimensions.
Since the discussion regarding our findings is difficult without access to the actual results, I will post a link to the corresponding papers as soon as possible within this thread.
An instrument cannot be, at the same time,uni and multidimensional. On the other hand, we can test, using Mplus, if there is a supra dimension. However, we have tested in our paper, the 3-factor model with a supra factor, combining the three factors described in literature and it was not better. Furthermore, it is known, that the more factors you have, the better the fit because the saturated model is the best model. Therefore, using OHIP-49, perhaps 15 factors will be provide a even better fit. We have to combine theory and data. Some people are data-driven, other are only theory-drive.
As I already stated, we applied a EFA to derive the OHIP's dimensions. We applied several approaches to determine dimensionality such as the ratio of the first-to-second eigenvalue, Cattell’s scree plot, and Horn’s modified scree plot (parallel analyses). Therefore, the generation of the models was data-driven, and the models were confirmed in CFA using a different set of data. Fortunately, the obtained models fitted very well with the theory. Since we first applied objective criteria to determine dimensionality in the EFA and second tested the gained models in the CFA, we did not create domains based on an optimizing approach in the CFA (e.g., using 15 factors or even more).
Even though our analyses revealed four highly correlated factors suggesting four dimensions of the OHIP, the applied criteria also suggest the presence of a dominant, higher-order general factor. Together with the other findings (e.g., model fit, inter-factor correlations), we interpret this as an indicator that OHRQoL can be sufficiently described with one score.
Just to complete: we used R (r-project.org) for our factor analyses.
- Badges
- Science topic
- Similar topics
- Psychology
- Psychometrics