Page 25 - 1960 (3) Pakistan Census Of Agriculture Vol 2 West Pakistan

xx .

,corresponding measure the connected items such as. that be anywhere between 96 and 104; it might as welt-
of total area of farms; total area under crops, total be 100.
n umber of livestock and several other totals.
52. The size of the sampling error must vary
47 .. Having discovered the extent and nature of from item to item according to the total value of each
1.1nder-enumeration in the number of farms which item, the frequency with which each item is likely to be
,correspondingly affected other Census items, it was encountered during enumeration and the inherent
felt th'at the quality of the Census Totals could be variability of its values. Similarly the size of the-
i mproved appreciably by using the Ratio Estimates sampling error would vary ·with the number of villages
:method. It may, however, be noted that the use of in the· sample and would, therefore, be the lowest at
-the ratio' estimates did not by any means disturb the the Provincial level. ·
Telationships established in the Census.
53. The precision of the Census Estimates is
48. Part B of the questionnaire relating to farm
f amilies, use of manures, indebtedness, livestock, etc., generally studied throughtheir sampling errors. The·
-which was filled in by · direct interviews with the far- less is the sampling error the more efficient is the sample
mers did not inherit any faults of the primary source design. In order to develop an efficient sample design
~nd, therefore, the incidence of error in these items was the technique of stratification is involved. But in the
-of an insignificant nature. Agricultural Census. wher_e the number of c}laracteristics
to be studied was so large and varied in nature, a proper
stratification with due Weight to all of the variables, was
49. In the use of the Ratio Estimates method, impossible; Therefore, the sampl_e design adopted
-the most important question· was the selection of a reli- for the Census based on . the stratification of villages
.;able benchmark. , Since the Census depended on the by size of their culturable area and by assessment circles
.Revenue records, for information on all questions was considered to be most efficient from the practical
])ertaining to land, it was considered that the totals point of view. The reason for giving weight to the
,obtained from the Revenue records, could serve as a factor of culturable area in the selection. of sample
·valid benchmark for developing ratio estimates . villages was that the majority of it-ems under study
. Among the numerous items included in the Census in the Census were closely related to it. However,
t hat of the cultivated area was found to be the one with since all the questions asked in the census were not
·which other items were closely related. Therefore, the apparently correlated with the culturable area to the
,,Census Organization obtained from the Revenue re- same degree their values as estimated from the sample
,-cords the totals of the cultivated area for all villages were not supposed to show a uniform level of
:and developed that as benchmark figures. . · accuracy.

50. In connection with the use of the ratio esti- 54. If the sampling errors were to be calculated
'.mates another question to be considered was whether from all the Census observations which were over
t he ratios should be developed at Province level, dis- nine lakhs in number for each of the 84 Census items,
1:rict level or at the tehsil level. The Post Enumeration ·that would have made the (ask well-nigh impossible.
-;Survey had indicated that the extent of under- Therefore from the original sample, a sub-sample had
,-enumeration varied from district to district arid tehsil to be sele;ted for the sake of calculating the sampling
·to tehsil. As the benchmark figures at tehsil level were ·errors for some of the important items only. According-
not easily available, the ratios at tehsil level could not ly a sub-sample of 144 villages approximating 2 percent
,be developed. It was, therefore, considered advisable of the total sample villages was used in order to obtain
·to develop 'ratios' separately for each of the district . sampling errors of 21 important items for_the.Provin~ial
-~and that was not difficult to achieve since benchmark Census · estimates: A statement showing sampling·
:figures were available for every district. Our analysis errors in this _ respect is given in Appen<l,ix B of our
,further established the fact that the course followed West Pakistan Report 1. ·
-<lid not disturb the ulti.Ip.ate results. Consequently, ·
"the sample observations of the tehsils/talukas were
:raised with the help of the respective district ratios. 55. Since in the · Census the information about
a number of items has been obtained from the Revenue
records, the users of the Census data may sometimes
51. The third · factor to be considered in con-
mection with the quality of the Census data is that of be tempted to compare the data with previously
:the sampling error, even though it may not be of much published statistics which have also been derived from
..-consequence in a sample as large as 20 percent. Be- the same records. In this connection it may be kept
in view that there are a number of factors on account
Jfore this factor was e'xamined in detail it is necessary of which the Census Totals should be expected to be
rto point out that 'sampling error' does not mean an different from the corresponding published figures.
><error in the usual sense of the word. It is in fact only The main factors are:- ·
;a measure of probable variation of the sample esti-
J'll1ates from the true value. For example, if the value
~of an item as estimated from the sample census is 100 (i) Differences in coverage and reference
z.and the sampling error pertaining to this item is cal- period; and
<;ulated as 2 percent it would be safe to say that in
~ 5 percent chances the true value of that _item would (ii) Differences of concepts.