We present an assessment of the CMB large scale anomalies in polarisation
using the two-point correlation function as a test case. We employ the state of
the art of large scale polarisation datasets: the first based on a Planck 2018
HFI 100 and 143 GHz cross-spectrum analysis, based on SRoll2 processing, and
the second from a map-based approach derived through a joint treatment of
Planck 2018 LFI and WMAP-9yr. We consider the well-known $S_{1/2}$ estimator,
which measures the distance of the two-point correlation function from zero at
angular scales larger than $60^{\circ}$, and rely on realistic simulations for
both datasets to assess confidence intervals. By focusing on the pure
polarisation field described by either the $Q$ and $U$ Stokes parameters or by
the local $E-$modes, we show that the first description is heavily influenced
by the quadrupole (which is poorly constrained in both datasets) while the
second one is more suited for an analysis containing higher multipoles up to
$\ell \sim 10$, limit above which both datasets become markedly noise
dominated. We find that both datasets exhibit a lack-of-correlation anomaly in
pure polarisation, similar to the one observed in temperature, which is better
constrained by the less noisy Planck HFI 100$\times$143 data, where its
significance lies at about $99.5\%$. We perform our analysis using realizations
that are either constrained or non-constrained by the observed temperature
field, and find similar results in the two cases.