This study used data from the HIES, collected between July 2010 and June 2011. This survey includes all areas within four provinces of Pakistan (i.e., Balochistan, Khyber Pakhtunkhwa, Punjab, and Sindh). A two-stage stratified random sampling scheme was adopted in this survey; separate sampling strategies were used for urban and rural areas. Enumeration blocks in urban areas and mouzas/dehs/villages in rural areas were selected at the first stage (defined here as “primary sampling units,” PSUs) from which households were selected at the second stage (defined here as “secondary sampling units,” SSUs). In urban areas, sample PSUs from each stratum were selected using a probability proportional to size (PPS) method, based on households in each enumeration block as a measure of size (MOS). Similarly, in rural areas, the population of each village was taken as the measure of size (MOS) used to select sample villages using a probability PPS method. Overall, 16 households from each rural PSU, and 12 enumeration blocks from each urban PSU, were selected for this survey using a systematic sampling scheme.

The nationally representative sample included in our study thus comprised 16,341 households overall, covering 14 large cities and 81 districts, including both urban and rural areas. The HIES reports information on a variety of social issues, including education, health, employment and income, immunization, use and satisfaction with facilities and services, and details of household consumption. Within the consumption module, the survey collects information on the quantities and values of 69 food items. We used these data (quantities and values of 69 food items) in our analysis.

Household food intake data were recorded every 14 days (fortnightly), with a 30-day recall period for food items. The fortnightly measure of food consumption was converted to monthly consumption. The quantities and nutritional values for more than 69 food items were recorded and stratified according to their sources, for example, from own production, market purchases, and gifts of staple food received in kind. Expenditure on purchased foods was also recorded.

The present analysis aggregated food items into 11 main groups: milk and milk products, meat, fruit, vegetables, spices, sugar, wheat, rice, pulses, oil, and other foods (formation of food groups is given in Supplementary Table A1). To calculate nutrient consumption from the reported food quantities, we applied conversion factors from the Food Composition Table for Pakistan, provided by Pakistani government (32), which contains data on nutrient contents for various foods items. The quantity of a given nutrient consumed was calculated as follows:

where N is the quantity of the nutrient, θi is the average nutrient content of a unit of food i, and Qi is the number of units of food i that were consumed (food items given in HIES food consumption sheet are in multiple units, g/kg/L). The food composition table for Pakistan (Table 1) provides the number of calories and nutrients (macro and micro) in 100 g of edible portion and each nutrient has also different units of measurement. We converted this 100 g into per kg unit to hold the consistency in the units of food items consumed. After the extraction of nutrients from food items, we computed the total daily household energy acquisition per capita. On the basis of this information, per capita daily calorie consumption and per capita consumption of nutrients (macro and micro) were computed (last two columns of Table 2). The current expenditure rather than income was used as a measure of household welfare; this is because expenditure tends to be a more dependable estimate of the permanent income of a household.

Given the possible non-linear relationship between income and calorie intake, we used two different approaches to assess this. First, we used a flexible specification, in which the parameters were linear but which permits elasticity to vary with income. According to Engel's law, as income increases, the proportion of income spent on food items diminishes. Thus, it is expected that as income rises, the calorie–income elasticity also decreases.

Differentiating Equation (2) with respect to ln X_i yields:

And differentiating Equation (4) with respect to ln X_i yields:

We obtained estimates using Equations (3) and (5) for each nutrient. If the coefficients attached to per capita expenditure in Equations (3) and (5), θ₁ and γ₁, respectively, are significant, then nutrient income elasticities vary with the level of income. For example, in model (5), if γ₀ > 0 and γ₁ < 0, then nutrient income elasticity decreases as income rises. If γ₀ = γ₁/X₁, the elasticities will be 0 and if γ₀ = γ₁/X₁, the elasticities will be negative. Moreover, when γ₁ = 0, the elasticities become constant and equal to γ₀ [for a detailed discussion on the advantages of log linear inverse model see (33)].

Finally, we addressed the issue of non-linearity in nutrient–income relationships using non-parametric and semiparametric regression. Non-linearity in the relationship between nutrients and income indicates that relatively less well-nourished individuals are more likely to make larger nutritional changes with budget shifts than those with better nourishment (34). In other words, people with greater economic security gain little nutritional benefits, while those with less economic security gain more from increased nutrient consumption. It is therefore imperative to understand the full range of nutritional responses instead of aggregating them into a single-point estimate. In order to explore the nutrient consumption patterns and deficiencies, it is of greater importance to carry out summary statistics of sampled households. Other issues such as the behavior of nutrient consumption to income changes are discussed in the section next to summary statistics.

Food consumption patterns of an individual or household indicate the composition of the available food, consumed by the population as a whole. In Pakistan, food consumption patterns generally indicate a greater risk of malnutrition. Addressing the first objective of this study, Tables 4–7 show the increase or decrease in average consumption for key nutrients (macro and micro) compared with the average recommended intake, and the contribution of each food group in the per capita daily intake of calories and nutrients. We have also observed the contribution of each food group in average calorie consumption at different income levels.

The consumption patterns in Table 4 indicate the low diversification of the diet of people in Pakistan: cereals were found to constitute a major part of their diet, within which wheat and wheat products alone provide 46% of the total daily per capita calories consumed. On average, 50% of total protein intake was found to come from wheat consumption. Wheat was also a significant source of carbohydrate (contributing 59% of total carbohydrate intake), iron (44% of total iron intake), and zinc (75% of total zinc intake). Pulses contributed 6% of the total protein intake. Overall, the consumption of animal protein in our sample was relatively low, at 6.4 g per person per day, with households obtaining the majority of their protein from cereals. The results of the study are consistent with a previous study (3) in which cereal was found to provide 54% of total protein intake, with low consumption of high-protein foods, such as animal products, when considered as a proportion of total protein. This indicates that the quality of the diet of many Pakistanis is nutrient-poor.

Fruits and vegetables accounted for 9 and 26%, respectively, of total vitamin A consumption; this is low compared with the recommended intake. Low intake of animal products, fruits, and vegetables often results in micronutrient deficiencies (5). Table 5 shows the mean calorie consumption from different food groups in the 25% of households with the lowest and highest incomes, respectively. These data show that households with the highest incomes obtain more calories from expensive food items including milk and meat compared with the lowest income households, who obtain most of their calories from wheat.

Along with calculating patterns of nutrient consumption overall, to explore the association between income and nutrient consumption patterns, we also created two further groups in our analysis based on income: the 25% of households with the lowest expenditure and the 25% of households with the highest expenditure. Although for very poor households differences might be less noticeable in current expenditure and current income (8). Table 6 shows the mean, median, and interquartile range for calories, major macronutrients (protein, fat, and carbohydrates), and micronutrients (calcium, iron, iodine, zinc, and vitamin A). In order to calculate calories, we included only households with a per capita calorie consumption of 600–8,000 kcal inclusive. The value of the mean was found to be greater than the median for total calories and all nutrients (macro and micro). This indicates that the distribution of all nutrients was positively skewed, and consideration of the mean alone would have resulted in an overestimation of nutrient consumption; the median, interquartile range (IQ = Q7525–Q25), and significance in nutrient intake for low- and high-income groups are therefore also reported (Table 6).

The mean per capita daily calorie consumption was 2,227 kcal in the study period (2010–2011), and the households with the highest incomes had a mean calorie consumption approximately 36% higher than those with the lowest incomes. The greatest difference in the consumption of any specific nutrient between these income groups was for calcium, with high-income households consuming 62% more than lower-income households. Calcium is a vital micronutrient, and deficiency may result in bone or tooth weakness and serious conditions such as osteoporosis. The high prevalence of osteoporosis in Pakistan owing to inadequate calcium intake in the older population is a significant public health issue. Low-income households may therefore be at greater risk of this condition owing to reduced calcium intake observed in our study. Iron consumption was also 39% higher and vitamin A consumption 55% higher in households with the highest vs. lowest incomes. However, on average, iron intake was found to be adequate for the majority of the population according to nutritional guidelines, whereas the consumption of vitamin A was found to be lower than recommended daily allowance. Vitamin A deficiency results in corneal deterioration and childhood blindness in many parts of the developing world, including Pakistan; the results of the study, therefore, highlight this as an important consideration in addressing malnutrition in Pakistan. Other differences in the consumption of specific nutrients were found to be relatively small between the highest and lowest income groups. In our sample, the median person suffers from micronutrient deficiencies (based on a comparison with the recommended daily allowance) in calcium, iodine, zinc, and vitamin A, as well as in overall calories, protein, and fat. Deficiencies were found to be worse in the lowest-income group, which included all macronutrients and micronutrients as well as overall calories. Conversely, the highest-income group was found to be deficient in calcium, iodine, zinc, and vitamin A. These results support our hypothesis that income affects nutrient consumption, with lower-income households experiencing a greater number of deficiencies. The per capita consumption of nutrients in the lower- and higher-income households was found to be significantly different at 1% level of significance (p < 0.01). Having information about the prevailing nutrient consumption patterns among these segments of the sample, we further study the nutrient consumption adequacy (Table 7) for the same groups of income distribution.

In addition, we calculated nutrient consumption adequacy as the ratio between the median consumption of nutrients and the recommended daily allowance obtained from nutritional tables. An important implication of these results is that individual households in the highest-income group could have a nutrient consumption adequacy >100%, while those of the lowest-income group could be substantially <100%, so remedial interventions for nutrition may be very different for both groups. The data in Table 7 show that the highest-income households had full nutrient consumption adequacy for macronutrients (all >100%), whereas the corresponding values were all below 100% in the lowest-income households, apart from carbohydrates. Some nutrients, such as protein, carbohydrates, and iron, had a strikingly high adequacy ratio in the highest-income group. On average, calorie consumption was 90% of the reference consumption level, but this value was lower for the lowest-income households (74%). Protein consumption did not appear to be a major nutritional issue in our study sample, probably because wheat is a major component of the food consumed in this population, and wheat is a reasonably good source of protein (consistent with the first EDA result in Table 4). Fat consumption was found to be greater than the reference level (135%) for higher-income households, but lower-income households consumed lower than the reference level.

Overall, these data show that total calorie intake may not be a serious threat for lower-income households; however, major deficiencies in key micronutrients, particularly calcium, iodine, zinc, and vitamin A, affect lower-income households to a substantially greater extent than those with higher incomes.

The ordinary least squares (OLS) estimation results of Equation (2) reported in Table 8 show that estimated income elasticities for all households were significant and positive for calories, all macronutrients, and all micronutrients. These results are in accordance with those from previous studies. For example, calorie income elasticity was found to be 0.38, which is similar to calorie income elasticity estimates for India (0.35) (10), Indonesia (0.43) (35), and rural Mexico (0.44) (8). The nutrient income elasticity for fat, calcium, iodine, and vitamin A were all in the range of 0.54–0.65; these results are consistent with adequacy ratio results, that is, larger elasticity estimates are associated with a lower adequacy ratio (below 100%) of nutrients.

We estimated Equation (2) with an instrumental variable (IV) two-stage least square (2SLS) method. The choice of a valid instrument, includes the relevancy and exogeneity of the instrument, is always a critical job. We used two different instruments; non-food expenditure and square of non-food expenditure. The results of IV approach are given in Table 8. The value of the F-test was >10, indicating that these instruments are valid. Elasticities estimated by IV regression for calories and all macro- and micronutrients were found to be lower in magnitude than the OLS estimates. This pattern indicates upward bias in estimates obtained using the OLS approach. An interesting result is that IV approach also shows a high magnitude (but less than OLS) for those macronutrients and micronutrients for which the magnitude is high in the OLS approach. These results are also consistent with our descriptive result and reveal that improvement in income appears to be most valuable particularly for those key nutrients for which the median person is most deficient.

Ordinary least squares results using specifications 3 and 5, presented in Tables 9, 10, were similar, but the IV results had a different magnitude for these specifications between specifications 3 and 5. Income elasticity was found to be high, sizeable, and significant for fat, calcium, iodine, and vitamin A using both OLS and IV results for specification 3. However, income elasticity for calories was found to be insignificant according to our IV results for specification 5. Intake of carbohydrates, calcium, iron, and zinc did not appear to change with income (IV estimates did not reach significance). In almost all models, per capita expenditure had a significantly positive coefficient.

We used nutritional survey data to estimate the nutrient–income elasticity. This enabled us to obtain detailed nutritional data from a very large sample size from diverse regions across Pakistan, which would not have been feasible otherwise. However, using household economic surveys for nutritional assessments has some limitations affecting accuracy. First, respondents may not remember the exact quantities consumed, especially when the recall period is long, as with the 14- and 30-day recall periods used in the HIES. Second, food consumption recall relates to the total food purchased by the household, not all of which is actually consumed by household members. Some may be fed to pets, wasted, or given to visitors or hired helpers. This could lead to an overestimation of actual food intake. Conversely, the receipt of food gifts and absence of some household members during the survey period may lead to underestimation of food intake. Third, we did not measure consumption for individual household members, as the household consumption survey does not report information about the intra-household food distribution; therefore an equal distribution of food among household members must be assumed. Fourth, the data used in this study were from household surveys conducted in 2010 and 2011, thus further studies are required on the household surveys conducted after this. In addition, this study focused on households whose caloric consumption fell between 600 and 8,000 kcal/day. However, owing to the large sample size and comprehensive measurement of specific nutrients as well as overall calorie intake, this study was able to produce realistic estimates and important insights into the overall nutritional status of the Pakistani population.

The results obtained in this study point toward some important policy actions that could be taken to improve nutrition in Pakistan. Of primary concern is increasing and upholding income to reduce food insecurity and malnutrition, which we have demonstrated to be associated with improved nutrition, which could be achieved by strengthening the three primary economic areas in Pakistan: agriculture, industry, and services, on a sustainable basis. Another concern is reducing nutrient shortfalls between average consumption and the recommended daily allowance in low-income households. This could be achieved by strengthening subsidization of food, for example, through networks of utility stores to provide discounted food, direct nutrient supplementation programs, or in-kind transfer of food items, price interventions, cash transfer programs, and social safety net programs. Finally, increased income may not reduce malnutrition alone; attention should also be given to the other socioeconomic and environmental factors including safe drinking water, better healthcare, and education. These factors may help improve food utilization.