In addition to the fundamental limit due to photon statistics, the measurement of polarization is affected by several factors some of which are discussed in the following subsections. However, the effects of flat-field errors and their remedy needs particular attention and is treated in detail in Sect. 5.
In IMPOL, two orthogonal polarization components are measured simultaneously and the Stoke's parameter is obtained from the ratio of the fluxes in these components (Eq. 2). Since the atmosphere is not birefringent, this eliminates the effects due to atmospheric scintillation, or for that matter any effect which changes both the polarization components by the same factor - like variations in the effective exposure times of observations, presence of thin clouds etc.
All the optical elements are anti-reflection coated to minimize the polarization efffects due to reflection at their surfaces, and care is taken to minimize the stray light, reflected from the walls etc., reaching the detector. In particular, all the aperture stops in the beam path are chosen to be non-metallic, including the grid in the focal plane, so as to avoid polarization of light scattered from these (Pospergelis 1965). The mounts for field lens, the half-wave plate and the Wollaston prism have been designed to minimize stress-birefringence due to differential thermal expansion.
The retardance introduced by the half-wave plate could deviate from 180
\
either because of the finite angle of incidence or because of the chromatic
effects. If the beam is incident at a small angle i, the
maximum change in retardance is given by (derived from expression in
Serkowski 1975)
where 
and 
are the refractive indices of the material
for ordinary and extraordinary rays. For the aperture used, the maximum
angle of incidence is about 5 so that 
rad.
The chromatic effects give 
rad for the wideband and
rad for the V-band. It can be shown
(Serkowski 1974)
that the depolarization 
due to an uncertainty of 
rad
in the retardance is to the lowest order given by
, p being the fractional
polarization.
Therefore even for the wideband 
. Further, circular
polarization in the incident light is converted to a linear polarization of
magnitude 
, where V is the circularity parameter.
But in typical observations this
does not pose a serious problem because the circular polarization
is usually less than the linear polarization.
If the position angle of the half-wave plate fast axis changes with
wavelength, it renders the measurement of 
erroneous.
The achromatic half-wave plate used in IMPOL does not produce any appreciable
dispersion in its fast-axis position angle over the wavelength range of
interest.
The uncertainty 
involved in the positioning of the half-wave
plate leads to a maximum error in the measurement of linear polarization
given by 
. Thus, 
for 
.
From Eq. (3) it can be shown that the variance due to
photon noise in the measurement of each of the Stoke's parameters 
is given by
where 
and 
are the number of photoelectrons and k is the ratio
of the flux from the background to that from the source. For small values
of p, and are approximately
equal and denoting 
, the standard deviation of p can be
written as
where NB = kN for observations with the grid and 
for observations without the grid (see Sect. 5). Similarly, we can also show
that
It is worth noting here that p, as defined above, is a positive definite
quantity and follows the Rice probability distribution given by
Here, p0 is the true value of fractional polarization being estimated by
p and I0 is the modified Bessel function of order zero. Since p is
a biased estimator, several schemes have been suggested
(Simmons & Stewart 1984 and references therein) for debiasing, but none of
them is fully satisfactory. For values of 
larger than
about 4, almost all the debiasing schemes agree and reduce approximately to
the relation
where 
is the "debiased" estimate of p0. Thus it is desirable to
work with the normalized Stoke's parameters as far as possible
and use the quantities p and only to present the final results.
However, in general, the normalized Stoke's parameters themselves might not be
normally distributed and might have bias
(Clarke et al. 1983). Besides, for
photon-noise dominated measurements, the positive kurtosis of the distribution
will lead to erroneous estimates of the confidence levels unless more than
a few thousand photoelectrons are collected. Thus, in order to arrive at an
optimum procedure, it is essential to carefully study the nature of the
dominant sources of noise in the measurement and their effects on the
analysis (see Sect. 5).