Wot?

The study:
A Randomized Study of How Physicians Interpret Research Funding Disclosures 
http://www.nejm.org/doi/full/10.1056/NEJMsa1202397#t=article

Method:Participants [MD internists; n=263] were presented with three abstracts, each of which described a trial of a different hypothetical new drug. Participants were told to assume that the hypothetical drug had recently been approved by the Food and Drug Administration and was covered by insurance and that the abstract was from a “high-impact” biomedical journal and written by academic physicians at established U.S. universities. We randomly varied the level of methodologic rigor and the disclosure [pharma-funded, NIH-funded, or no disclosure] so that the three abstracts that the physicians received described trials at each level of methodologic rigor and with each disclosure variation.

There were problems with the method, the most relevant of which was external validity--do internists normally prescribe new drugs based on abstracts?

The authors reported an effect for pharma versus NIH (Figure 2, below) by doing this:

For each question, we estimated a hierarchical proportional-odds regression model, using the appropriate Likert-scale response as the outcome. This model included a random intercept for each physician to account for within-physician correlation of responses across abstracts, as well as fixed effects for methodologic rigor (low, medium, or high), disclosure statement (industry funding, NIH funding, or no statement), and drug.

In the appendix, they offered the means for the various rigor x disclosure combinations, which don't seem to show a pharma versus NIH effect (Figure 1, below). But, I lack knowledge. Seeking yours.

--Why didn't they compare mean ratings? Is that precluded in a study of this design, in which every subject was exposed to three stimuli showing three rigor levels and three disclosure types?

--Why did they "account for" within-physician correlations? It's over my head, but such correlations could mean that rigor and disclosure didn't much affect them.

--did the random intercept also account for fixed effects for rigor, disclosure, and drug? It;s hard to tell by how it was worded.

Figure 1. The means, available in the supplemental appendix but not analyzed for the results section of the paper, graphed. The colors (purple, red, and pink) denote low, medium, and high rigor, in that order.

Figure 2. Results as depicted in the study (Note relative scalings of Y-axes)