The authors also used a funnel plot to show that smaller studies more prone to error were more likely to show an association between saturated fat intake and CVD than were larger studies less prone to error.
In the picture below, each square or diamond represents a study. If it is to the right of the vertical line down the middle, it reported an increase of CVD risk with increased intake of saturated fat. If it is to the left of the vertical line, it reported a decrease in risk. The studies plotted towards the top were larger and the studies plotted toward the bottom were smaller.
If there is no publication bias, we would expect the studies to be distributed symmetrically around the average result of the total pooled data (in this case, a risk ratio of 1.0, meaning no effect). Publication bias tends to affect smaller studies — everyone wants to know the results of large, expensive, extensively publicized studies, but small studies will often go unpublished or ignored if they have negative results. In the picture, you can see that the smaller studies were greatly skewed towards finding an increase in CVD risk with increased intake of saturated fat, while the larger studies were more likely to find no effect.
This does not prove, but suggests, that many small studies went unpublished or otherwise lost down the memory hole if they found no association or a negative association between intake of saturated fat and risk of CVD.
All in all, however, we must remember that correlation never demonstrates causation. As I will be discussing in the upcoming sequel to my PUFA Report, the controlled intervention trials substituting polyunsaturated fats for saturated fats suggested that replacing foods like butter with foods like vegetable oil would increase the risk of cancer and possibly even hasten the development of atherosclerosis.
It will be interesting to see how extensively the media publicizes this analysis — or will it be ignored?