Diets, Nutrition, and Poverty - The Oxford Topic of Food, Politics, and Society - page 355

range for this index, a high value implies a monopolistic industry (or, in our case, a more
concentrated diet) and a low value implies a nearly perfectly competitive industry (or, a
more diversified diet).
16. An instrumental variable (IV) robust regression is used.
17. As the middle class is an instrument in the first stage IV regression, it is not used in the
second stage.
18. For further confirmation of dietary diversification reducing calorie intake over the period
1993-2009, see Gaiha, Kaicker, Imai, and Thapa (2013).
19. For details, see Gaiha, Kaicker, Imai, Kulkarni, et al. (2013).
20. We use Rs 300 monthly per capita expenditure as the cut-off point.
21. We use a Heckman model in which two steps are involved: first, the probability of eating
out is determined and then, conditional on it, the amounts spent on eating out. For details,
see Gaiha, Jha and Kulkarni, et al. (2013).
22. The analysis is based on unit record data collected for the 50th and 61st rounds of the NSS
(corresponding to the years 1993-1994 and 2004-2005, respectively). Price effects cap-
ture both own and cross-price effects through substitutions between food commodities.
Briefly, as prices change, demands for commodities change and consequently calorie (and
other nutrients') intakes. Underlying this is a presumption that food choices are informed
by their nutritional content. As Deaton and Dreze (2009) emphasize, people do not buy
calories and other nutrients but food commodities. However, if food choices are informed
by their nutritional values, it is meaningful to talk about demands for calories and other
nutrients
23. For details, see Gaiha et al. (2012).
24. Comparisons of effects of different variables are based on (absolute) elasticities.
25. Note that a significant effect is one that is statistically different from 0.
26. Note that this is a residual time effect. What the graphical representations reflect is the
combined effect of all factors that varied over time whereas the regressions results relate to
the residual time effect.
27. This is not inconsistent with the segments of the protein Engel curve for 2004 lying above
that for 1993 in figure: 13.3.
28. This is not inconsistent with segments of the fat Engel curve lying above those of the 1993 curve.
29. Our explanation is corroborated (with varying food price and expenditure effects) over
the longer period, 1993-2009 (Gaiha, Kaicker, Imai, and Thapa (2013).
30. For example, there is lack of a consensus on what the correct minimum calorie threshold
is, how it should be computed or even whether such a threshold exists (Dasgupta, 1993,
Srinivasan, 1992, 1994, and Svedberg, 2000). On the related issue of a taste for variety, see
Behrman and Deolalikar (1989).
31. Briefly, Lowess is used for locally weighted scatter plot smoothing. For an intuitive exposi-
tion, see Deaton (1995).
32. For details, see Kaicker and Gaiha (2012).
33. Recall that Lowess estimates allow for calorie share variation due to expenditure variation
alone, while our robust regression estimates include the additional effects of food prices.
34. See Dasgupta and Ray (1986, 1987). Srinivasan (1994) offers a cogent critique.
35. For details of the survey and methodology for estimating PNT, see Jha, Gaiha, and Sharma
(2009).
36. This is confirmed by a t-test of mean differences.
37. Noncommunicable diseases include cardio-vascular diseases (CVD), cancers, diabetes,
chronic obstructive pulmonary disease (COPD), asthma, neuro-psychiatric conditions