Page 38 - 1990(3) Cesus Of Agriculture, provience report-Punjab
P. 38
between patwar circle com- sample of - 100 farms in which the
ponent of variance is zero and enumerator fails to enumerate all
the results from ·each mouza the large farms or delioe·rately
have to be expanded to half of -,- omits · them with the result that the
the stratum and the square of enumerated farms vary in narrow
differences between them taken range of size. The sampling
in order to get the con tr i- variance _of such a sample would be
but ion of variance between and much lower than obtained if sampie
within mouzas of a . patwar of the large farms had also been
r circle which alone comprises a enumerated. But it is evident that
complete stratum. the estimate of farm .area and farm
size from such a sample would be
ii) One non-certainty - and _ one grossly lower than the true value
certainty mouza in a Certainty and · the resultant low samplihg er-
Patwar Circle in which case ror value for this estimate would
.
.
the raising factor of each be . practic_ally · meaningless . A
mouza has to be adjusted so as "gbod" ·estimate is, therefore, one
to · make each one represent in which the sum total of the sam-
halt of the Patwar Circle to pling errors plus a:11 other errors
· approximate the same situation is. vety low. The errors other than ·
as in . (i) above. those· due to _sampling are alled
non~sampling errors, and - include
iii) Certainty patwar circles cdn~-- error.s ;in sampling frame nd selec-
9
taining only certainty mouza~ 'tion of th~ sample, enumer~titin and .
in which case the within mouza resp9ps_e _ due to Biases ' ·and · ig-
co~ponent of variance -norance·,- errors in recor:ding
comprises the whole cohtribu~ ~nswers., processing errors, et6. In
t{on of that patwar circle to a ' Ceri:sus of Agriculture, one · may
the totc::l.J. variance . In such ·-v1suali-ze that the chances of oc-
cases the non-certainty currence of non-samplin_g. errors
ho~seholds e numerated in each begin with·· planning of the census
mouza wouid have to be bifur~ and .end with · the _publication of
cated in two equal parts in data ~ In others ~ords there are al-
order of their selection. Tha most endless chances bf the occur-
results from each part are r~_nce . ot .. :nori-sarripling errors. It is
then expanded to half of th~ du'e to .this,. reason - that _ in a cen-
mouza and the squar~ of dif...: ·_ sti's\ ,.-:it p<;tys to g"i ve more attention
ference between them taken tci' t9'.'ithe~·-cortt:,ro).. _ of non-sampling er-
estimate the within mouza ·- for:s than "to_ the reduction of sam-
variance component. For two pling errors, and . quite often some
certainty mouzas _within cer- inc.rease in the latter is to be
tainty patwar circl~- the sum ·tol~rated £or the sake of reducing
of the within mouza components the former.
comprises of the .. stratum
variance. It is very difficult to measure
non-sampling errors in quantitative
Non-~ampling _error~ terms and specially to obtain a
comprehensive measure of all the
~~he extent of non-sampling er- errors that might have crept into
rors is measured to judge the ac- the census data at various stages.
curacy of census data. It is con- Only a very elaborate checking and
ceivable that the estimate of a e valuation of the census results
giveri value with very low sampling can yield some usable measure of
error might in fact be a bad _ es- non-sampling errors, but most
timate, as - it is amenable to sub- countries find it difficult to
stantial non-sampling errors: For spare any funds and energy for this
instance, one may visualize a t ype of evaluation after having
(xxxvi)
