The problem is that since the 55th Round has canvassed both the Type 1 and Type 2 schedules from all households, there would have been a pressure for consistency between the answers to the "one week" and "one month" reference periods on the part of both respondents and investigators. It is obvious that when the household is questioned using both the one week and one month reference period, the answers are likely to be tested by simple multiplication of the one week reply for the monthly response as well. This means that the two schedules can no longer be seen as independent and may give misleading results depending upon how the conflation of the referenced periods affects the responses.
 
If such pressure for consistency has led to primacy being given to the "one week" response, a 25 per cent rural poverty incidence in the 55th round would correspond to a 45 per cent poverty incidence by the procedure followed in 1993-94 so that the proponents of liberalising reforms could end up claiming massive poverty reduction while in fact poverty might have increased massively.
 
Even if the pressure for consistency has worked evenly across the schedules, so that the 55th Round results are an average of the two, a 25 per cent rural poverty incidence in this Round would correspond to a 36 per cent poverty by the method used in earlier rounds, so that a large poverty reduction could be claimed without there having been any significant reduction from the actual poverty level in 1993-94.
 
In view of this, the 55th Round stands severely compromised in its ability to give estimates of mean consumer expenditure and poverty comparable to that in earlier Rounds. This is particularly the case because, as Charts 2A and 2B show, the errors associated with measuring the population in lower expenditure classes is much higher by the Type 2 schedule. These errors are computed by the NSSO on the basis of the variance obtained across sub-samples in the same survey and for the same schedule type, and show the much greater inherent unreliability of poverty estimates calculated with data from the Type 2 schedule.
Chart 2a >> Chart2b >>
 
The National Sample Survey Organisation needs, therefore, to conduct another large sample survey on the basis of the earlier schedule exclusively, to give comparable estimates. In the meantime, the 55th Round should be treated as an experimental survey whose comparability with past surveys is poor, but which may further elucidate the intriguing differences revealed by the results of the 51st to 54th Round surveys. This is important because much mud has been slung at the NSS consumer expenditure data in the course of the recent debate on poverty, and the air needs to be cleared so that questions about the reliability of data does not continue to cloud assessment of such an important issue.
 
Much of the criticism of NSS data in the recent past has concentrated on the fact that these give lower estimates of total consumer expenditure than the alternative estimates from the National Accounts Statistics (NAS), and some have even claimed that the difference between these two estimates have grown alarmingly during the nineties. In an earlier edition of Macroscan (February 22nd, 2000), it had been pointed out that although the NSS does indeed give a lower estimate of overall consumption than the NAS, and although the difference did increase during the seventies and eighties, the ratio between these two estimates have remained fairly stable during the nineties so that the 1990s trends in poverty obtained from NSS data cannot be challenged on this ground.
 
Nonetheless, since this has been the main thrust of the attack on the NSS data so far, we present in Table 1 the ratios of the NSS to NAS estimates of consumption by broad items and over successive full-year NSS rounds beginning 1977-78. Since only the NAS estimates with base 1980-81 cover this entire period, these ratios have been calculated with these NAS estimates rather than the new estimates with 1993-94 as base. As mentioned earlier, the ratio of total NSS consumption to NAS consumption is seen to decline from 0.81 in 1977-78 to 0.69 in 1990-91 but remains almost constant thereafter. However, the more important differences between these two estimates relate to the item-wise results.
Table 1 >>
 
As may be seen there are certain persistent differences between these two data sets at the level of individual items. Thus, for sugar, edible oils, fruits & vegetables, milk & products, pan, tobacco & intoxicants, and other goods & services, the NSS has consistently measured lower consumption but with no obvious time trend in these ratios. In the case of meat,fish & eggs and clothing, NSS has lower consumption and the ratio has fallen over time. For cereals, the ratio has always been close to unity but with some tendency to decline over time. But, for pulses, other food and fuel & light, the NSS has consistently measured higher consumption than the NAS.
 
These persistent differences between NSS and NAS data have been analysed thoroughly by resarchers, notably B.S. Minhas and his associates, who have noted that it is normal all over the world for items like intoxicants to be under-reported by respondents and that something similar is probably true also for non-vegetarian items in a country such as India.