Here’s just one snippet from the Clark and Gordon document linked in Arthur’s post:
With colour-coded systems, the principal effect of lens scattering is the reduction of the colour differences between the signals. For instance, in a red-white coding scheme, the white signal becomes reddened and the red becomes desaturated, i.e. pink. Because the scattering is angle-dependent, the greatest effect is adjacent to the direction of the transition between the colours. The magnitude and angular variation of the effect depends on the cause of the scattering, viz. lens surface reflections, or dust-water deposits, and in the case of particulate scattering, on the size distribution and optical properties of the particles. Hald (1980) reported water condensation on both sides of the objective lenses in trial PAPI installations. The resultant scattering '. . . totally destroyed the signal, so that the light received over a certain distance happened to be only white with no red sector. In an operational system this will mean that the PAPI system would only give a full "fly down" signal at all angles.' Electrical heating elements have been installed near the lenses in some PAPI boxes in an attempt to overcome the condensation problem. The heaters will not overcome the dust problem: if anything they will exacerbate it because of the extra airflow over the lenses as a result of convection currents.
(My underlining.)