Also, they note that much of sugar, edible oils, milk & products and fruit & vegetables are consumed not directly by households but are purchased after processing either by hotels and restaurants or by other manufacturers. In such cases, these would appear differently in the NSS and NAS data, with the former including these under "other foods" while the latter would include them directly under the item concerned.
 
This explains also why the NSS has tended to measure higher expenditure under "other food". The relative over-estimation by the NSS of fuel & light has likewise been explained by failure of the NAS to adequately capture fuel wood and twigs collected directly by households. Thus, for most of the above items, the differences are not particularly surprising or unexpected, especially given that the NSS does not capture all consumer expenditure since it leaves out institutional consumption such as in hostels, prisons and ceremonials.
 
However, for certain items such as clothing and "other non-food" the differences are large and have been attributed in past analysis both to a failure of NAS to measure household consumption correctly and to a failure of the NSS to adequately capture the consumption of the relatively richer household who consume relatively more of these.
 
Indeed, it is on the basis of the these observations, that the Expert Group on Poverty Estimates had decided in 1993 that the differences between the NSS and NAS were unlikely to cause any serious bias in poverty measures estimated directly from NSS estimates, and had accordingly decided to end the practice till then of adjusting these estimates to conform to the NAS.
 
However, the issue of alternative reference periods has now again opened up this issue since apparently much of the difference in food consumption between the NAS and the NSS can be resolved if the one week rather than the one month schedule is used in the latter. Indeed, in its Report No. 477, the NSSO reports higher errors for week-based estimates as compared to estimates based on 30 days, but on noting that "the substantial and systematic differences between the week and month based estimates indicate that one or both methods are not depicting the real life situation", goes on to claim some support for the Type 2 schedule since the total of the one week estimates is closer to the NAS than the totals of the 30 day estimates.
 
Although the NSSO itself is careful on the matter, suggesting that further methodological surveys would be advisable, others may not be so careful. They may not only ignore the fact that the relative standard error of each item canvassed on the one week basis is higher than by the 30 day schedule, but also fail to notice that the recent experiments merely reproduce what Mahalanobis and his associates had found almost 50 years ago.
 
As mentioned earlier, it was found in the 1950s that the one month estimate of most food items was lower than the one week estimate but was in greater conformity with the physical (weight) measures of actual household consumption of foodstuffs, than the one week estimate. Thus, the bias between results from the two reference periods continues to remain in the same direction, and with no experiment through physical weighment repeated, there is no further evidence for judging the relative plausibility of the two estimates. Comparison with NAS is a poor substitute given the past judgement of researchers such as Minhas and the assessment of the Expert Group on Poverty Estimates, both of which found strong reason not to accept the NAS as necessarily giving reliable estimates.
 
Table 2 gives the perentage differences between the NSS and NAS (1980-81) using both the schedules in the former. It may be noticed that although the Type 2 NSS estimates are fairly close to the NAS for all food items taken together, while the Type 1 NSS estimates are about 20 per cent lower, this result comes about because the Type 2 estimate gives higher estimates for all food items including for those where the Type 1 estimate is higher than the NAS estimate. As a consequence, the apparent concordance between the Type 2 NSS estimates and the NAS is something of statistical artifact because these week-based results continue to show large shortfalls from the corresponding NAS estimates for sugar, edible oils, milk & products and meat, fish & eggs but almost double the estimate for "other food". If both positive and negative divergence are given equal weight to measure the difference between the NSS and NAS, the results from the Type 1 schedule turn out to be closer to the NAS for food items. Also the large gap between the NAS and NSS for non-food is further widened when the Type 2 rather than the Type 1 results are considered.
Table 2 >>

Table 3 gives the absolute value of consumption estimates for 1995-96 from both the Type 1 and Type 2 NSS schedules and from both the NAS estimates with base 1980-81 and base 1993-94. This not only shows the patterns discussed above but also certain inherent infirmities in the NAS data.
Table 3 >>
 
Thus, for two items, "pan, tobacco and intoxicants" and "clothing", the NAS has substantially revised downwards its consumption estimates, between the 1980-81 and 1993-94 series, bringing these closer to the NSS estimates. But for two others, "fruits and vegetables" and "other non-food" the NAS has revised upwards its estimates and thus increased the gap with the NSS. For "other non-food" there is at least the likelihood that new goods and services were being underestimated earlier and may not be captured in the NSS which does miss out on the rich who consume these more, but the doubling of fruit and vegetable consumption is intriguing and highly suspicious.
 
As discussed in an earlier Macroscan, not only does this not correspond with the known area under horticultural crops, it has the effect of making the NAS estimate three times the NSS Type 1 estimate and more than double even the NSS Type 2 estimate. In view of these large and sometimes inexplicable revisions, the NAS can hardly be said to represent the type of benchmark that Mahalanobis had set himself when he actually carried out physical weighment to check the validity of reference periods.

Finally, as Charts 3A and 3B show, there is the intriguing fact that the Type 2 schedule sample results show richer households consuming relatively more food and less non-food as compared to the Type 1 schedule, thus overturning much of what is now accepted wisdom regarding changing consumption habits along the Engels Curve. This too should suggest a more careful look at the estimates emanating from the Type 2 schedules, and by implication, of the overall consumption results of the 55th Round.
Chart 3a >> Chart 3b >>