You are on page 1of 13

An Objective Method fm Determining the Collective

Caste Hierarchy of an Indian Villagel


STANLEY A. FREED
American Museum of Natural Hislory

I. INTRODUCTION
N INDIAN village usually contains representatives of from five to 25
A endogamous social groups called castes (Marriott 1960: 23). Villagers re-
gard the castes as higher or lower than one another in precedence and esteem
thereby forming a hierarchy of castes. Agreement among villagers about the
ranking of castes is close enough so that one can determine the approximate
rank in the hierarchy of most castes, a t least to the extent of placing them as
high, medium, or low. However, agreement is not perfect: some villagers may
assign each caste to a separate rank; some may say that two or more castes
have equal rank; and some may assign relatively high rank to a caste which
others rank relatively low. Disagreement is possible because of the many cri-
teria by means of which castes may be ranked and the fact that villagers may
use different criteria in constructing their rankings; or, even if they use the
same criteria, they may weigh them differently. Some of the criteria entering
into caste ranking are dietary customs, occupation, details of ritual behavior,
wealth, and the interactions of the castes with regard to smoking the same pipe
(hookah) and the giving and taking of food and water. That villagers do not
completely agree in selecting and weighing criteria is apparent from the reasons
villagers give for their caste rankings. Gupta (1956) reports many instances of
disagreements based on conflicting criteria in a village in northwestern Uttar
Pradesh.
The lack of agreement among villagers about caste ranking raises the ques-
tion: what single hierarchy would best express the collective opinion of the
villagers and how many discrete, significant ranks does it contain? Such a
hierarchy of significant ranks is here called a collective caste hierarchy. Signifi-
cant ranks are those whose derivation from the raw data involves a statistical
test of significance. A collective caste hierarchy gives equal weight to the
opinions of all villagers polled. It may have one rank for each caste; or, as is
probably more common, several castes may be effectively tied for particular
ranks, thus making the number of significant ranks in the collective caste
hierarchy less than the number of castes. Castes occupying the same significant
rank in a collective caste hierarchy form a caste bloc. The foregoing describes
the general nature of the collective caste hierarchy; in practice, it is defined
operationally. This paper describes a statistical method for determining the
collective caste hierarchy of a village and presents the results of its use in a
village near Delhi. It also discusses the problems of the effects of membership
in particular castes and of urban experience upon statements of caste ranking
by respondents.
879
880 American Anthropologist [65, 1963
The difficulty of determining the rankings of castes has been frequently
noted. Srinivas (1955: 19-20, 22) comments, “Yet it is difficult, if not impossi-
ble, to determine the exact, or even the approximate, place of each caste in the
hierarchical system,” and any attempt to do so is “fraught with risk” and
“necessarily tentative and arguable.” Mayer (1956: 118) echoes Srinivas about
the perils of constructing caste hierarchies, pointing to the contradictions in
the evidence given to the observer; and in the central Indian village of Ram-
kheri he shows how caste hierarchies constructed according to different criteria
do not completely agree.
Despite the difficulties about determining caste rankings, several authors
have made the attempt for particular villages. The problem of conflicting
opinions in the testimony of informants is usually avoided in various ways.
Sometimes the problem is minimized by disregarding some of the data. Dube
(1955a: 34-42; 1955b) substantially reduces conflicts in dealing with caste
ranking in three villages in Telangana by basing his ranking primarily upon one
criterion: which castes can theoretically take food from which other castes.
Mayer (1960) bases his hierarchy on the criterion of commensality which in-
volves principally the giving and taking of food and water among castes and
the sharing of the same pipe. Another way of dealing with conflicting data is
for the investigator to weigh them and then construct his own picture of the
hierarchy. This is the procedure used by Srinivas (1955) and apparently also by
Majumdar (1958). Srinivas (1955: 25) comments that such evaluations are
somewhat subjective but less so than those of the members of any one of the
castes. Disregarding relevant data in order to minimize conflicts is an approach
that is best avoided when possible; and a subjective weighing of conflicting
opinions is also relatively undesirable because we cannot be sure that different
investigators, studying the same conflicting opinions, would construct the
same hierarchy (Cf. Marriott MS.).
The problem of conflicting opinions in the testimony of informants does not
inevitably force the investigator to disregard some of the data or to make
judgments which are somewhat subjective. Opinions may be systematically
collected so that their variability can be described and measured; and if appro-
priate statistical techniques are used in constructing a collective caste hier-
archy from these opinions, the subjective factor can be minimized and the
amount of uncertainty in the decisions the investigator makes in constructing
the hierarchy can be measured. This kind of approach would result in substan-
tially similar interpretations of the same data by two investigators. Further-
more, if the same techniques of collecting and analyzing data were used in two
or more villages, a comparison of the results would be more meaningful than if
different techniques were used.
There have been two attempts a t formulating an objective approach to the
problem of constructing a collective caste hierarchy for a village. One, by
Marriott (MS. and personal communication), uses caste rankings provided by
a sample of respondents as its basic data. The collective caste hierarchy is
found by determining the median rank for each caste. Marriott recognizes that
FREED] Method for Determining Caste Hierarchy 881
some castes may be effectively tied in rank and groups such castes into “minor
blocs” (1960:16). The component castes of minor blocs are determined by
studying the gap between the medians of castes adjacent to one another in the
hierarchy. Gaps of less than a selected magnitude are interpreted to mean that
two castes are in the same minor bloc. This method is statistically incomplete
because techniques for testing the significance of the differences in median
ranks are not presented.
Mahar (1959) has proposed a method for determining caste rankings quite
different from Marriott’s. Her method is based on the relative placement of
castes with regard to their ritual purity and pollution. Her basic data are
responses to a 13-item questionnaire about kinds of interaction between castes
which involve to a considerable extent ritual purity and pollution. An overall
village hierarchy is derived from the responses to the questionnaire by using a
scaling technique. The major weakness of the method, in its present stage of
development, is that it does not deal effectively with the question of whether or
not castes may be tied in the overall hierarchy.
The method of determining the collective caste hierarchy of a village pro-
posed in this paper follows Marriott’s approach rather than Mahar’s in that
its basic data are opinions about caste ranking given by respondents rather
than statements (or observations) about interaction. Its principal advantage is
its ability to measure the amount of uncertainty involved in the various de-
cisions the investigator makes in constructing the collective caste hierarchy.
Because the collective caste hierarchy is determined independently of an
analysis of interaction, it is then possible to regard the hierarchy and interac-
tion as two variables and study their relationship. This is not possible if the
hierarchy is deduced from interaction, for then their relationship is determined
by the method of analysis. The relationship of the interaction among castes and
the collective caste hierarchy is an important problem and will be discussed in
another paper.
11. THE VILLAGE
The village of Shanti Nagar (a pseudonym) is located about 11 miles north-
west by road from Shakti Nagar, the most northern of the small, named com-
munities which make up greater Delhi. The road between Shanti Nagar and
Delhi is metalled except for about one mile. A bus makes four round trips a
day. The bus usually takes an hour and a quarter for the trip between Shanti
Nagar and the Old Delhi bus station, the place in Delhi where the people of
Shanti Nagar most often board the bus. A railway station is located about two
or three miles from Shanti Nagar; and trains provide cheaper transportation
between Delhi and Shanti Nagar than does the bus. By cycle, Delhi can be
reached from Shanti Nagar in an hour. Hinduism is the only religion in Shanti
Nagar. Agriculture is the principal economic activity although many men have
jobs in Delhi and other cities. Shanti Nagar has 13 castes and a total popula-
tion of 799. One of the 13 castes consists of a family of the Chhipi caste which
moved to Shanti Nagar shortly after we had started our field work. Since the
882 A merican A nthropologist [65, 1963
Chhipi caste is new in the village, we have not included it in this analysis of
caste ranking. Table 1 gives the population of Shanti Nagar by caste and sex.
111. FIELD METHOD
Caste rankings were elicited from 26 randomly selected male respondents. A
set of movable cards upon each of which was written the name of a caste was
presented to each respondent with the request that he arrange the cards in
their order of rank. Marriott (MS.) was the first to use movable cards for
eliciting caste rankings. Such a procedure is clearly superior to questioning
without cards; for in the latter case, the respondent must mentally deal with a
rather large number of items and can easily become confused and tired.
In Shanti Nagar, we varied our technique according to whether the respond-
ents were literate or illiterate. If the respondent were literate, the entire set of
cards was placed before him or handed to him with the request to arrange the
cards in a column with the highest ranking card on top, the next highest under
it, and so on until all the cards were arranged in descending order of rank. The
cards were shuffled before giving them to the respondent. The instructions
emphasized that the arrangement of cards should reflect the respondent’s
opinion. When the respondent had finished arranging the cards, their order was
read to him slowly and clearly to see if they were ranked as he wished. This was
a guard against poor eyesight or marginal literacy which might have led to
bluffing or mistakes.
If the respondent were illiterate or had poor eyesight, the cards were read to
him and their purpose explained. The interviewer asked the respondent which

TABLE1. POPULATION
oF SHANTI
NAGAR
IN 1958 BY CASTE
AND SEX^
-. -
Caste (and traditional occupation) Male Female Total

Bairagi (Beggar) 11 16 27
Baniya (Merchant) 3 3 6
Brahman (Priest) 100 87 187
Chamarb (Leather-worker) 56 42 98
Chhipi (Dyer and Printer of Cotton) 2 3 5
Chuhra (Sweeper) 41 55 96
Gala Kumhar (Potter) 35 23 58
Jat (Farmer) 128 132 260
Jhinvar (Water-carrier) 6 7 13
Lohar (Blacksmith) 6 5 11
Mahar Kumhar (Potter) 1 2 3
Mali (Gardener) 3 7 10
Nai (Barber) 15 10 25

Total 40 7 392 799

The census was taken by interviewing one or more adult members from every family of
Shanti Nagar. The castes are arranged alphabetically.
These are Jatiya Chamars.
FREED] Method for Determining Caste Hierarchy 883
TABLE2. THESAMPLE OF RATERS BY CASTEAND THE POPULATION
OF MEN
25 YEARSOF AGE AND OLDERFROM WHICH THE SAMPLEWAS CHOSEN
_.__ __
Men 25 Years of
Caste Raters
Age and Older"

Bairagi 4 1
Baniya 2 1
Brahman 35 4
Chamar 15 4
Chuhra 12 4
Gola Kumhar 10 4
Jat 39 4
Jh'invar 3 1
Lohar 2 lb
Mahar Kumhar - -
Mali 2 1
Nai 4 1

Total 128 26

a These figures include all Shanti Nagar males 25 years of age or older except one chronically

sick man and five very old men who were variously senile, deaf, and blind. Also excluded were
eight servants from outside Shanti Nagar and one other temporary resident. If a man gave his age
as 24 or 25, he was included; if he gave his age as approximately 20 to 25, he was excluded. The
castes are arranged alphabetically.
Discarded.

caste he regarded as highest and selected the proper card which was placed a t
the top of what would become a column. The interviewer then asked which
caste came next and placed that card below the previous card. After two or
three cards had been placed, the interviewer read the arranged cards; then each
time a new card was placed (if it appeared that the respondent's memory
needed refreshing) the interviewer read those not yet ranked. When all the
cards were arranged in rank order, the interviewer read the final order slowly
and clearly to check for mistakes or last minute changes.
Our procedure differs only slightly from Marriott's. He presented the cards
one by one, whereas we presented them all together. The author, with the
assistance of an interpreter, conducted about half the interviews, and the in-
terpreter, the remainder. The interpreter had nine months experience with us
before he was permitted to interview alone, and we had complete confidence in
his reliability.
IV. T H E SAMPLE
Only men 25 years of age and older were chosen as raters. The population
was stratified by caste and the sample of 26 raters was randomly selected
within the castes, using the table of random digits in Wallis and Roberts
(1956:631-635). Table 2 lists the sample of raters by caste and the population
by caste from which the sample was chosen.
884 American Anthropologist [65, 1963
The sampling procedure was designed to balance several considerations.
First, we decided that 26 random interviews were all that we could afford for
this particular problem and that the expense of additional interviews would not
be justified in terms of added information. Second, we wanted opinions from all
castes to be reflected in the collective caste hierarchy. We, therefore, decided to
select a t least one rater from each caste. These decisions meant that a sample
which reflected relatively faithfully the proportions of the village population in
each caste could not be achieved. However, we did not entirely ignore the
considerable differences in caste populations. We classified the castes as large
(Brahman, Chamar, Chuhra, Gola Kumhar, and Jat) or small (Bairagi,
Baniya, Jhinvar, Lohar, Mahar Kumhar, Mali, and Nai). One rater was chosen
from each of the small castes. No Mahar Kumhar was selected because none
was of the proper age. This meant that 20 raters were left for selection from the
large castes. We assigned four raters to each of the large castes rather than
dividing the 20 among these castes in proportion to their populations. We chose
this procedure because we were interested in the problem of whether or not the
ratings of respondents were influenced by the positions of their castes in the
hierarchy. I n order to investigate this, we had to assess as reliably as possible
the opinion of each caste; it was just as important to know the opinions of the
Gola Kumhar men as the Jat men.
The Lohar’s responses about caste ranking were discarded. He was an old
man and extremely upset a t the time of the interview due to a serious dispute
with a powerful Jat landlord of the village; and he either did not understand
the questions or could not concentrate on them. One Chamar ranked only the
Brahman, Baniya, Jat, Bairagi, Chamar, and Chuhra castes. His rankings were
retained.
Whether or not a completely proportional sampling procedure would have
resulted in a different hierarchy than the one constructed on the basis of the
sampling here used depends upon whether opinion regarding caste ranking
varies from caste to caste. This question has been investigated as described
below; and the results indicate that there is little difference about caste ranking
among the respondents of different castes.
Women raters might give a somewhat different caste ranking than do the
men. This seems likely because of the residence rules after marriage. Men gen-
erally live and die in their village of birth, whereas women shift to their hus-
bands’ villages upon marriage. The wives of Shanti Nagar have come from
some 200 surrounding villages. Since caste rankings vary from village to village,
the wives of Shanti Nagar might differ from the men as regards the placing of
particular castes.

V. STATISTICAL METHOD FOR DETERMINING THE


COLLECTIVE CASTE HIERARCHY
The responses of the 25 raters about the rankings of castes are given in
Table 3. The first step in analyzing these responses is to consider each possible
pair of castes and count to see which caste of the pair is ranked higher b y a
majority of respondents. Those pairs of castes whose members are rated equal
FREED] Method j o r Determining Caste Hierarchy 885
TABLE
3. RESPONSES OF CASTESIN SHANTI
ABOUT THE RANKINGS
OF 25 RATERS NAGAR

Rater CasteR

1 Brahman 27-28 Vil. 1 2 3 4 6 9 5 7 8 1 0 11 12


2 Brahman 25 Urb. 1 2 4 3 8 7 5 1 0 6 9 11 12
3 Brahman 70 Vil. 1 2 3 4 6 5 7 9.10 9.10 8 11 12
4 Brahman 40 Vil. 1 2 8 3 5,6 5.6 4 7 9 10 11 12
5 Baniya 45-46 Vil. 1 2 4 3 6 5 9 7 8 1 0 12 11
6 Jat 32 Vil. 1 2 3 , 4 3 , 4 5.6,75,6.75.6,78.9,108.9.108,9.10 11 12
7 Jat 48 Vil. 1 4 3 2 8 7 6 1 0 9 5 11 12
8 Jat 30 Urb. 1 3 2 4 5 6 7 9 1 0 8 11 12
9 Jat 41 Urb. 1 23,43.4 6 5 7 8 9 1 0 11 12
10 Bairagi 30-32 Urb. 1 4 3 2 5 6 7 1 0 9 8 11 12
11 Mali 35 Vil. 1 2 3 6 5 4 7 9 8 1 0 11 12
12 Jhinvar 60-70 Vil. 1 2 4 3 7 5 6 9 1 0 8 11 12
13 Gola Kumhar 25 Vjl. 1 8 2 3 9 7 4 6 5 1 0 11 12
14 Gola Kumhar 32 Vil. 1 3 2 8 5 4 9 7 6 1 0 11 12
15 Gola Kumhar 30 Vil. 2 4 1 5 3 8 1 9 6 1 0 11 12
16 Gola Kumhar 35 Vil. 3 5 6 4 8 1 0 7 2 1 9 11 12
17 Nai 55 Urb. 1 3 4 2 7 8 6 9 1 0 5 11 12
18 Chamar 25 Urb. 1 2 3 1 0 6 7 8 4 5 9 11 12
19 Chamar 70 Vil. 1 3 4 2 - _ - - - - 11 12
20 Chamar 60 Vil. 1 2 4 3 7 8 5 6 1 0 9 11 12
21 Chamar 30 Vil. 2 3 1 8 4 5 6 7 9 1 0 11 12
22 Chuhra 45 Urb. 1 3 2 5 4 9 8 6 7 1 0 11 12
23 Chuhra 50 Urb. 1 2 3 4 6 5 8 7 9 1 0 11 12
24 Chuhra 36-40 Urh. 1 2 3 7 1 0 4 6 8 9 5 11 12
25 Chuhra 25-26 Urb. 1 2 3 7 5 6 4 1 0 9 8 11 12
______~ ______

Castes arranged according to the preliminary caste hierarchy.

in rank by a respondent are disregarded. The next step in the analysis is the
construction of a preliminary hierarchy for use in arranging data for presenta-
tion in tables. The preliminary hierarchy is based on the paired comparisons
and is constructed in the following manner. The Brahman caste is rated higher
by the majority in comparison with each of the other castes and is therefore
ranked first in the preliminary hierarchy. Of the remaining 11 castes, the
Baniya caste is rated higher in all paired comparisons and hence ranks second.
Of the remaining 10 castes, the Jat caste is rated higher in all paired compari-
sons and thus ranks third. The rest of the preliminary hierarchy is determined
in a similar fashion. The only difficulty concerns the Lohar and Mali castes. Of
22 paired comparisons, half rated the Lohar higher and half rated the Mali
higher. I n order to rank these castes relative to each other, the paired compari-
sons of the next higher caste, the Bairagi, are examined with both the Lohar
and Mali. In comparing the Bairagi and Lohar, 7 of 25 respondents rank the
Lohar higher; in comparing the Bairagi and Mali, 6 of 25 respondents rank the
Mali higher. Therefore, the Lohar caste seems to rate just a bit higher than the
Mali in village opinion. The preliminary hierarchy developed in this way is,
from high to low: Brahman, Baniya, Jat, Bairagi, Lohar, Mali, Jhinvar,
Mahar Kumhar, Gola Kumhar, Nai, Chamar, and Chuhra. The hierarchy is
preliminary because no statistical test of significance is applied in the process
886 American Anthropologist [65, 19631
of deriving the ranks. The collective caste hierarchy as defined here is a hier-
archy of significant ranks which means that a statistical test of significance is
used in establishing the ranks.
The final step in establishing the collective caste hierarchy is to apply a test
of significance to the paired comparisons. We assume (the null hypothesis) that
if there is no difference in the ranks of two castes, A and B, then half of the
respondents will rank A higher and half B higher. The alternate hypothesis is
that if one caste does rank higher than the other, then a significantly greater
number of respondents will so rank it. A five per cent risk of rejecting the null
hypothesis when it is true is taken as the level of significance. A figure of less
than five per cent is interpreted as showing a significant difference in rank be-
tween two castes. The null hypothesis is tested with the binomial probability
distribution.The results of this comparison of the rankings of castes by respond-
ents are given in Table 4.
The application of the foregoing statistical method to the paired compari-
sons results in the collective caste hierarchy of significant ranks. Analysis of
Table 4 reveals seven significant ranks. Four contain only a single caste. The
other three contain blocs of castes. These seven ranks in the order given below
constitute the collective caste hierarchy of Shanti Nagar:
A. Brahman
B. Raniya
C. Jat and Bairagi
D. Lohar, Mali, and Jhinvar
E. Mahar Kumhar, Gola Kumhar, and Nai
F. Chamar
G. Chuhra
The Baniya and Jat castes are considered significantly distinct in rank even
though the five per cent level of significance is exceeded slightly (probability
=.054). This is because the Jats and Bairagis are clearly of the same rank
(probability= .339) and the Baniyas and Bairagis are clearly of separate rank
(probability= .007). The most reasonable interpretation of this situation is that
the Baniyas are of one rank and the Jats and Bairagis of another.
The foregoing method of analysis involves two difficulties. The first is the
fact that respondents sometimes regard two or more castes as equal in rank.
There is no statistically justifiable way of including such ties in a binomial
population of opinions about the superiority-inferiority of a pair of castes.
Therefore, they are dropped from consideration. The respondents of Shanti
Nagar make little use of ties, and thus to disregard them has little effect upon
the analysis. However, there are villages where raters make considerable use of
ties. Marriott's village of Kishan Garhi is one (personal communication). For
such villages, ties cannot be disregarded, for they represent a substantial
amount of collective village opinion.
The second difficulty involves instances which require interpretation such
as the case of the Baniya and Jat castes which are placed in different ranks
even though the five per cent level of significance is slightly exceeded. These are
TABLE
4.FREQUENCY OF A HIGHER RATING
OF A CASTEAS COMPARED TO EACH
O F THE O T H E R CASTES BY 25 RATERSWITH T H E PROBABILITY
OF THE PARTICULAR
OUT COMES^

Casteb

Casteh

25 25 25 25 25 25 25 25 25 25 25
Brahman 25 23 25 25 25 25 24 24 25 25 25
,000 ,000 ,000 .a00 ,000 ,030 ,000 .a00 ,030 .a00 .OOo

25 25 25 25 25 25 25 25 25 25
Baniya 17 19 24 24 24 23 23 25 25 25
,054 ,007 ,000 ,000 ,000 ,000 .OOO .OOO ,000 .OOO

23 25 25 25 25 25 25 25 25
13 24 24 24 23 24 25 25 25
.339* ,000 ,000 .a00 .a00 .a00 .a00 ,000 .ooo
25 25 25 25 25 25 25 25
Bairagi 18 19 21 21 22 23 25 25
,022 ,007 ,000 ,000 ,000 .OOO .OOO ,000

22 23 24 24 24 25 25
Lohar 11 12 19 19 21 25 25
.584* .500' .003 .003 .OOO .OOO ,000

23 24 24 24 25 25
Mali 12 18 17 21 25 25
,500' .011 .032 .OOO ,000 .OOO

24 24 24 25 25
Jhinvar 18 18 21 25 25
.a11 ,011 .a00 ,000 .a00

22 23 25 25
Mahar Kumhar 13 14 25 25
.262* .202* ,000 ,000
_-
23 25 25
Gola Kumhar 14 25 25
,202' .a00 .a00

25 25
Nai 25 25
.a00 ,000

25
Chamar 24
.000

Chuhra

* I n each cell, the top figure (n) is the total number of comparisons; the middle figure (r) is the number of times the
caste ranking higher in the preliminary hierarchy was so ranked; and the bottom figure is the probability of this many
or more occurrences of r by chance in a sample of n from a binomial population where the frequency of the occurrence of
r is .SO, as given in United States Department of Commerce, National Bureau of Standards (1952), Table 2.
An asterisk marks the cells in the Table far castes whose ranks are not significantly distinct. These castes, there-
fore, belong in the same rank and form a caste bloc. Consider, for example, the cell for the Mali and Jhinvar castes. The
top figure, 23, is the number of comparisons provided by the respondents; the middle figure, 12. is the number of com-
parisons which ranked the Mali caste higher than the Jhinvar; and the bottom figure, .500. means that the probability
of this result (12 or more out of 2.1 ranking the Mali higher) by chance is .SO0 if we assume no difference in the ranks
of the Mali and Jhinvar castes. Because the probability is greater than .05, the Mali and Jhinvar castes are by definition
equal and are placed in the same rank.
Castes arranged according to the preliminary caste hierarchy.
888 A merican A nthropologist [65, 1963
of the following general form: caste a is equal in rank to b and distinct from c
and d; castes c and d a r e equal; and caste b is distinct from d but not from c. For
data such as those from Shanti Nagar, the most reasonable way of interpreting
this situation is to regard a and b as forming one rank and c and d another. The
characteristic of the Shanti Nagar data which makes this interpretation rea-
sonable is that pairs of castes, with the exception of the Jat-Baniya pair, are
either sharply distinct or solidly equated. The highest significant probability is
.032 and the lowest nonsignificant one is .202, over six times as great. Had one
more informant rated the Baniya higher than the Jat, this pair would also con-
form to the general pattern. However, there is no reason to believe that all col-
lective caste hierarchies will have ranks, all of whose castes are sharply dis-
tinct, or that their construction will involve only a very few of the difficulties of
the type here described. Under such conditions, modifications in the method of
determining significant ranks might have to be considered.

VI. THE EFFECT OF CASTE MEMBERSHIP ON


OPINIONS OF CASTE RANKING
Several authors have commented that caste membership may influence a
person’s view of the caste hierarchy, or a t least his opinion about the place of
his own caste within it. Srinivas (1955:25) says, “Frequently the position
claimed by a caste differs from the position conceded to it by ,others. . . .”
Mayer (1960:44) states, “Whereas castes in the lowest and highest ranks both
think and say that they are higher than their neighbors, the castes in the
middle may think they are higher, but say publicly that they are equal.” I n a
discussion of the resentment felt by people of low caste because of their status,
Berreman (1960: 126) notes that “NOgroup would admit to being lowest in
the caste hierarchy.”
Inspection of Table 3 shows that most respondents rank their own castes
close to the ranks accorded to them by others. If we compare the rankings
given by each of the respondents with the preliminary caste hierarchy, we find
that all respondents from the Brahman, Baniya, Jat, Jhinvar, Chamar, and
Chuhra castes rank their own castes no further than one rank away from their
ranks in the preliminary caste hierarchy. Respondents of some of the castes in
the middle of the hierarchy (Bairagi, Mali, Gola Kumhar, and Nai) rank their
castes two or more ranks higher than do others. The Bairagi ranks his caste
second while its rank in the preliminary hierarchy is fourth; the Mali ranks
his caste fourth while its rank in the preliminary hierarchy is sixth; the Gola
Kumhars rank their caste variously from first to sixth while its rank in the pre-
liminary hierarchy is ninth; and the Nai ranks hiscaste fifth while itranks tenth
in the preliminary hierarchy. Thus the raters from six of ten castes rank their
own castes as do other villagers while the raters from four castes, all from the
middle of the hierarchy, rank their castes higher than do others.
The question of whether a person’s view of the overall hierarchy, rather
than just his opinion of the position of his own caste, is influenced by caste
membership is investigated as follows. Ten hierarchies are constructed, based
on the median ranks assigned to all of the castes by the raters from each sep-
FREED] Method for Determining Caste Hierarchy 889
arate caste. Each of these hierarchies is then compared to a hierarchy based on
the median ranks assigned to all of the castes by all raters from all castes except
those of the caste whose hierarchy is involved in the comparison. For example,
a hierarchy is constructed based on the responses of the four Brahman raters,
and it is compared to one based on the responses of the 21 other raters. This
procedure is followed for each of the ten castes from which there are raters.
Hierarchies based on medians are used rather than hierarchies based on the
majority opinion in paired comparisons, because the small number of raters
from each caste makes it impossible to construct hierarchies on the latter basis.
For the same reason, significant ranks cannot be used. The hierarchies are com-
pared by using the Spearman rank-correlation coefficient. This coefficient
ranges from - 1 to +1 with values close to +1 indicating a high positive cor-
relation of the two sets of ranks. The results of the analysis are given in Table 5 .
Because of the small number of raters for each caste, any conclusions we
may draw from Table 5 are tentative. However, a rater’s caste membership
appears to have little effect upon his overall view of the hierarchy. All ten co-
efficients are very high; only two fall below +.94. The lower coefficients for the
Gola Kumhar and Nai castes reflect the tendency for the raters from these
castes to rank their own castes considerably higher than the rank assigned by
the rest of the respondents. In summary, the data suggest that (1) caste mem-
bership has little effect upon a person’s overall view of the caste hierarchy, and
(2) most respondents can accurately place their own caste in the hierarchy and
will do so in the kind of interview used in this research.
VII. THE EFFECT OF URBAN EXPERIENCE ON
OPINIONS OF CASTE RANKING
The men of Shanti Nagar have been exposed to considerable urban influ-

TABLE5. CORRELATIONS OF THE CASTE RANKING (ARRANGED BY THE MEDIAN


RANKS OF THE CASTES) BY THE RATERSFROM A SINGLE CASTE AND THE
CASTERANKING BY THE RATERSPROM THE OTHER CASTES

Rank-correlation Coefficient of the


Castes Ranked by Raters Ranking by the Raters of the Castes
of Caste in the First Column and the Ranking
by the Raters from the Other Castes

Brahman .97
Baniya .96
Jat .98
Bairagi .94
Mali .97
Jhinvar .95
Gola Kumhar .86
Nai .85
Chamar .95
Chuhra .98
890 Antericar~Anthropologist [65, 1963
ence. The total male population 20 years of age and older is 165; of these, 81 are
working or have worked or lived in urban areas (primarily Delhi). I will use the
term “urban-employed” to refer to men who have lived or worked in urban
areas and the term “village-employed” for those who have not. Village-em-
ployed men are not completely without urban contacts. Most of them oc-
casionally visit Delhi; and a few visit Delhi several times a week. Therefore,
urban-employed men whose periods of urban work or residence have been
relatively short may have had less urban experience than the few village-em-
ployed men who visit Delhi frequently. However, the criterion of urban work
or residence is relatively unambiguous; and, despite the overlap in urban ex-
perience between some urban-employed men and a few village-employed men,
it is the best criterion available for dividing the villagers into two groups differ-
ing as regards urban experience.
The fact that many of the men of Shanti Nagar have had urban experience
raises the problem of the effect of urban contact upon statements about caste
ranking. Both social scientists (Dube 1955a: 20, 224; 1958: 28; Ghurye
1950: 208-209) and the villagers of Shanti Nagar say that behavior between
castes is different in village and city. As a group, the high caste, urban-em-
ployed men of Shanti Nagar are less concerned with ritual pollution arising
from contacts with low caste people than are high caste, village-employed men.
Some of Shanti Nagar’s high caste urban-employed men testify that they have
eaten and smoked with low caste men, activities which involve ritual pollution;
and they express their willingness to do so in the future. This behavior takes
place in the city and not in Shanti Nagar. Nevertheless it could conceivably
have an effect upon statements by these men about the ranking of castes in
Shanti Nagar.
The problem of the effects of urban experience upon statements of caste
ranking is investigated as follows. Two hierarchies are constructed, based on
median ranks: one for the urban-employed respondents and one for the village-
employed. The hierarchies are compared with the Spearman rank-correlation
coefficient, which in this case is +.98. Thus there is almost no difference in
statements about caste ranking as given by urban-employed and village-em-
ployed men. This probably means that urban experience does not necessarily
affect statements about the rankings of castes in villages, despite the greater
freedom of caste interaction in the city.

VIII. CONCLUSION
This paper discusses the problem of constructing collective caste hierarchies
for Indian villages. A statistical technique is proposed for analyzing the caste
rankings of respondents, which is objective and provides a measure of the
amount of uncertainty involved in the various decisions the investigator makes
in constructing a collective hierarchy. This paper presents the results of using
this technique in developing the collective caste hierarchy of Shanti Nagar.
Further analysis of the data from Shanti Nagar to see whether caste member-
ship or urban experience significantly influence respondents’ statements about
caste ranking shows that these two factors apparently have little effect.
FREED] Method for Determining Caste Hierarchy 89 1
NOTE
l This paper is based upon field work in the village of Shanti Nagar from November, 1957, to
July, 1959. M y wife, Dr. Ruth S. Freed, and I collaborated in all aspects of the field research; when
the pronoun “we” is used in this paper, it refers to my wife and me. I wish to thank the Social
Science Research Council and the National Science Foundation for the post doctoral fellowships
which supported the field work, and the University of North Carolina, Chapel Hill, for a faculty
research grant to help with secretarial expenses. I also wish to thank Professor P. C. Biswas, Chair-
man, and Dr. Indera Pal Singh of the Department of Anthropology, University of Delhi, for their
many kindnesses while I was affiliated with that department as a National Science Foundation
fellow. My thanks are due to Dr. Howard Levene, Department of Mathematical Statistics,
Columbia University, who suggested the use of the binomial probability distribution for analyzing
the paired comparisons, and Drs. Robert L. Carneiro and Joan P. Mencher of the American
Museum of Natural History, who read and criticized the manuscript.
REFERENCES CITED
BERREMAN, GERALD D.
1960 Caste in India and the United States. The American Journal of Sociology 66: 120-
127.
DUBE,S. C.
1955a Indian Village. London, Routledge and Kegan Paul Ltd.
195% Ranking of castes in Telangana villages. The Eastern Anthropologist 8: 182-190.
Calcutta, India, Eka Press.
1958 India’s changing villages. Ithaca, N. Y., Cornell University Press.
GHURYE, G. S.
1950 Caste and class in India. Bombay, Popular Book Depot.
GUPTA, RAGHURAJ
1956 Caste ranking and inter-caste relations among the Muslims of a village in north
western U. P. The Eastern Anthropologist 10:30-42. Calcutta, India, Eka Press.
MAHAR,P. M.
1959 A multiple scaling technique for caste ranking. Man in India 39: 127-147. Calcutta,
India, Sakti Press.
MAJUMDAR, D. N.
1958 Caste and communication in an Indian village. Bombay, Asia Publishing House.
MARRIOTT, MCKIM
1960 Caste ranking and community structure in five regions of India and Pakistan.
Deccan College Monograph Series 23. Poona, India.
(MS.) A technique for the study of caste ranking in South Asia. Paper read at the 56th
Annual Meeting of the American Anthropological Association, Chicago, December,
1957.
MAYER,ADRIANC.
1956 Some hierarchical aspects of caste. Southwestern Journal of Anthropology 12: 117-
144.
1960 Caste and kinship in central India. Berkeley and Los Angeles, California, University
of California Press.
SRINIVAS, M. N.
1955 The social system of a Mysore village. 1% Village India, studies in the little commu-
nity, McKim Marriott, ed. Comparative studies of cultures and civilizations, No. 6.
American Anthropological Association, Memoir 83. Chicago, Illinois, University of
Chicago Press.
UNITED STATES DEPARTMENT OF COMMERCE, NATIONAL BUREAU OF
STANDARDS.
1952 Tables of the binomial probability distribution. Applied Mathematics Series 6.
Washington, D. C., U. S. Government Printing Office.
WALLIS,W. ALLENAND HARRYV. ROBERTS
1956 Statistics, a new approach. Glencoe, Illinois, The Free Press.

You might also like