For a very long time, I believed that mineral wool is holding the air without any transport of heat due to convection like it is written by Eurima,  European Mineral Wool Manufacturers Association.

Mineral wool thermal insulation prevents convection by holding air still in the matrix of the wool. Still air is a good insulator. Mineral wool also stops radiation and limits the conduction of heat through the body of the insulation. 

But by studying VIP (Vacuum Insulated Panel), I did not understand a graph, found in a paper about VIP’s.

Indeed, it is clearly measured that creating vacuum in mineral wool (removing the air) improves the thermal conductivity with 0.034 W/mK while the thermal conductivity of still air is only 0.026 W/mK. This means that in mineral wool under atmospheric circumstances, natural convection is present. Indeed, on the PAROC-website, we find the following graph with convection mentioned.

According the graph, the conduction should only be 0.008 W/mK while the convection 0.026 W/mK. I guess that conduction and convection are changed. Indeed, with 0.008 W/mK for convection, the VIP graph can be understood: 0.008 + 0.026 = 0.034 W/mK.

But if natural convection is present in normal circumstances, we have to think about the measurement methods for the thermal conductivity. In Europe, we have to live with EN12667, Determination of thermal resistance by means of guarded hot plate and heat flow meter methods. I am not allowed to give a link to this standard because citizen have to pay for digital standards, written free of charge by the industry. This standard includes already the possibility of natural convection in mineral wool:

Wenn die Möglichkeit des Beginns einer Konvektion innerhalb des Probekörpers gegeben
ist (z.B. in Mineralwolle geringer Dichte), können die Lage des Prüfgerätes, die Dicke und die Temperaturdifferenz sowohl den Übertragungsfaktor als auch den Wärmedurchlasswiderstand beeinflussen.

Indeed, in case the cold plate is under the hot plate, there is no natural convection present in the mineral wool because the warm air can not rise. This is the case for the following one sample guarded hot plate system:

Maybe I did not understand something, in that case please correct me. But in the other case, in the above measurement system, natural convection is excluded and this approved system may give a thermal conductivity which is 0.008 W/mK lower than in reality on a dry perfect roof or about 20%. In case of a two sample system, the error could be of the order of 0.004 or only 10%. For higher densities, the deviation should be smaller. I heard that mineral wool samples are used for round-robin test of thermal conductivity measurements. I will not be surprised by large deviations between different EN-approved systems.

It seems that to get the insulation value on the roof (most important part of the building) we want and trust, we have to go to higher densities above 100 kg/m³ or more. But in that case, GLAPOR cellular glass, directly foamed from recycled glass has a comparable price with high density mineral wool and guarantees to be convection free, non-combustible, load bearing, vapour tight and water tight with a stable thermal conductivity.