Professional Documents
Culture Documents
Farm Size and Productivity
Farm Size and Productivity
Farm Size and Productivity it is just that we have done similar analysis
on the same subject.
Dyer’s assertion about our conclusion
Manabendu Chattopadhyay on the nature of the size-productivity re-
Atanu Sengupta lationship seems to be based on the fact
that none of our zonal regressions is sig-
IN a recent polemic directed against our been better if we could include more crops. nificant, and that the number of significant
paper ‘Farm Size Productivity: A New Admitting this limitation, we can stress cases at the village level is also low. There
Look at the Old Debate’ (EPW, Decem- that paddy is the most important and major is no justification, therefore, to carry out
ber 27, 1997), Graham Dyer (EPW, June crop in West Bengal agriculture.1 Hence, the combined x2 test in such a context,
27, 1998) raises some relevant questions. study on paddy should reveal most of the Dyer concludes. While agreeing that on
His comments seem to cast serious doubts important characteristics of the region the basis of statistical evidence we do not
on the legitimacy of our findings. While where it is a major crop. A number of farm have a strong case to conclude that an
admitting that the results appear to be efficiency studies have been based on major inverse relationship between farm size
quite perplexing, given the current state crop only.2 While data analysis based on and productivity exists, what may be
of the debate, Dyer argues that it has very a number of crops would certainly pro- relevant to state here is that we were guided
slender basis. This note is an attempt to duce results which are closer to reality, more by Bhattacharya and Saini’s asser-
answer some of the questions raised by what we wish to point out is that sufficient tion that most of the village level regres-
Dyer so that most of the probable mis- insights may also be gathered regarding sions are insignificant due to smallness of
understandings which resulted from Dyer’s farmer’s behaviour while concentrating the sample size, and yet one may be
presentations are clarified. on a major crop. interested in carrying out a combined x2
It is clear that Dyer argued on the fol- The zonal classification has been made test for finding the nature of overall re-
lowing important aspects. First, he ques- by Comprehensive Scheme for Studying lation. Results of x2 test for our study
tions the validity of the data set which he Cost of Cultivation (CSSCC) on the basis indeed suggest the existence of overall
considers to be “somewhat compromised” of various agroclimatic factors. A single negative relation between farm size and
[Dyer 1998: A-113]. He feels that data on zone may contain blocks belonging to productivity. Added to this is the fact that
only one crop, viz, paddy, are quite in- different districts.3 Our distinction between there was a preponderance of negative
adequate to draw conclusions regarding developed and less developed zones was coefficients (although often not signi-
size-productivity relationship. These based primarily on rainfall since Indian ficant), particularly in the developed
anomalies in the data set were further agriculture is highly dependent on rain- zones. All these seem to give an indica-
aggravated by the amorphous distinction fall. Systematic rainfall reduces the risk tion that the size-productivity relation may
amongst zones and their classifications to which may provide an impetus towards often be negative, although for various
“developed” and “less developed” ones. adoption of modern technology. We con- constraints and limitations with the data
Furthermore, he is not clear whether we sidered regions receiving normal annual it is not possible to conclude strongly
have used net cultivated area or gross rainfall (1350-2000mm) as developed whether the relation is often negative at
cropped area in our regression exercises. while those receiving above normal (ex- the village level in the sense of statistical
Notwithstanding such data discrepancies, ceeding 2000mm) as well as below nor- significance.
the methodology used by us at times is mal (less than 1350mm) rainfall as less Dyer’s further comment that ANCOVA
not quite correct, as Dyer seems to sug- developed. We agree with Dyer when be analysis is valid only for log-linear model
gest. In fact, he asserts that there has been points out that some other relevant vari- is rather misleading. In a cross-section
a blind following of Bhattacharya and ables may also help to classify the zones study observations across villages/zones
Saini’s (1972) techniques “without fully as developed and less developed. But non- (in our case) are likely to be heteroscedastic.
understanding the implications noted by availability of such data is a major con- Since ANCOVA presumes that variances
the latter” [Dyer 1998: A-114]. He also straint for such studies in India. across such regressions are the same, log-
comments that ANCOVA technique ap- Dyer feels that we have “blindly” fol- linear model would be closer to this as-
plied by us is not valid and this may be lowed the methodology used by sumption than the linear model. This must
applicable only for the log-linear model. Bhattacharya and Saini. In this context not, however, lead one to conclude that
Thus by all counts, our findings, though it may be imperative to note that Chatto- ANCOVA cannot be applied in linear
startling, are laden with serious flaws. padhyay and Rudra (1976) noted that models. Linear models are often used for
In order to clear some serious apprehen- the work by Bhattacharya and Saini was various modelling considerations includ-
sion of Dyer regarding the definitions of statistically most extensive among avail- ing specificational aspect. Appropriately
the regress and regressor used for our able works. However, lack of any dis- speaking, whether linear or log-linear
regression analysis, we state that we in- aggregated data prevented Chattopadhyay models, application of ANCOVA would
deed regressed value of output per unit of and Rudra from undertaking such studies. first require that the hypothesis of equality
net cultivated area (V/A) on net cultivated Availability of disaggregated data for the of variances be tested. One need not,
area (A). Our findings are, therefore, not year 1989-90, although inadequate, however, go into such rigorous applica-
weakened by the introduction of a “sig- prompted Chattopadhyay and Sengupta to tion of statistical tools in a study like ours
nificant bias” [Dyer 1998: A-115]. pursue rigorous statistical analysis similar since the purpose of using ANCOVA was
It is however true that our data set was to that carried out by Bhattacharya and to find the possible effect of aggregation
somewhat limited since we could consider Saini. There is obviously no question of bias at village/zonal level. Moreover,
only one crop, viz, paddy. It would have blindly following Bhattacharya and Saini; consideration of both the linear and log-