Edited parts of what Hans-Joachim Brandt wrote:

For LF I can use either the house ground, which is also connected to the water
pipe net, or a radial ground made of 28 insulated litz wires with lengths
between 12 meters and 30 meters depending on the available space in that
direction.

When working on this radial system and continuously measuring the feed point
resistance under resonance conditions, however, I could not confirm Bobs remark
that metal stakes at the far ends would give an improvement.

At my QTH, I too did not notice any difference in "before" and "after"
connection of outer end earth stakes (but I left them connected anyway
:)

At least on my
insulated radial system the feed point resistance increased. Especially with
rather few radial wires the radial ground had been very sensitive against any
ground connection, including a simple earthstake connected  to the centre of the
system. After the number of radials had approached 18 the earth stake
(just a few inches deep) connection to the centre could be tolerated, but never
the connection to the house ground. Once I have tried to elevate all my radials
about one foot from the ground: The feedpoint resistance remained the same, I
needed more inductivity in the variometer, and the sensitivity to any connection
to ground even increased! This sensitivity obviously depends of the capacity of
the insulated radial system to ground.

I use buried bare copper radials.  However, regarding above-the-gound
insulated radials, like those mentioned, the explanation could be that
the "return flux" finds it easier and lower loss to return as current to
the antenna feedpoint via the copper, which being insulated from ground,
has no possibility of contacting lossy soil.

Further, when selected (tuned) inductors are used between the feedpoint
and insulated radials, then it becomes even more attractive for the
"return flux" to terminate on the wire rather than ground, as with
"tuning" the insulated radials can be "-" potential when the driven
antenna is "+" potential, and the soil somewhat neutral potential.  I
have seen some diagrams of commercial stations where the connection to
actual earth is tapped from the central part of the loading coil, and

efficiency with most current collected by the low resistance insulated
radials.  As already mentioned, being "antiphase", they can attract more
flux than if they were neutral.

This favours above-the-ground radials, and more complicated tuning,
however in most domestic situations these type of radials become a
severe nuisance for tripping over, cursing the designer, etc :(

Cheers,

Bob ZL2CA