Chapter 6 Expenditure components

In this chapter, we present a metadata analysis focusing on the household surveys’ ability to measure food, non-food, durables, and housing expenditures through different data collection methods. The characteristics of each household survey and their ability to construct a nominal consumption aggregate were collected by filling a metadata formulary.⁴ This metadata collection tool may be accessed in the folder metadata_collection_tool.

We provide an empty formulary and an example of a filled out formulary that you may use as reference. See “SAR_PAK_2015 (Example).xlsm”.
You must keep the filled out formularies in the same folder where you also keep the file “consumption surveys metadata.xlsx”. Once you have filled out the information for a formulary, click on the Submit button. This will record all of the content of your formulary as a new row in “consumption surveys metadata.xlsx”.
The Stata metaform.ado program transforms “consumption surveys metadata.xlsx” into a shape to be used in Tableau.

6.1 Food

The first component is food expenditures, which should include not only food produced and/or consumed at home, but also food purchases outside of home, and food transfers to and from the household. This component is sometimes thought to be easier to measure than non-food items. When household members eat from a common pot, it is common to find a single well-informed individual who can act as respondent and provide information about how much the household has consumed. Household surveys sometimes include a food expenditure diary, but this is not common in South Asia where most information on food is collected based on recalling what a household has consumed and purchased over the last seven days.

The ideal reference period should be a week, but in this analysis we have found reference periods still vary from yearly, to monthly, to biweekly, to daily across surveys. In recent years, the Indian National Sample Survey Organization has experimented with 7-day and 30-day recall periods for a number of items, and discovered that the 7-day recall cuts Indian poverty rates by half, removing some 200 million people from dollar-a-day poverty add reference. This is an issue that has not been much investigated, but has recently moved into the forefront of research. In other words, methods matter and it is important to standardize data collection as much as possible.

In most cases, countries modify their food baskets and change their data collection methods over time. A wide mixture of data collection methods characterizes the most recent rounds of household surveys. Bangladesh HIES 2016 collects daily household food consumption (both quantity and value) for a period of 14 days. In addition, it collects weekly consumption of a series of spices. Bhutan LSS 2017 collects food consumption (both quantity and value) by asking the respondent to recall their consumption in the last seven days, the last thirty days, and the last year. In some cases, such as Afghanistan LCS 2016, the female respondents are the ones asked about household food consumption. These methods can be very diverse and we provide a brief description for the latest survey available for each country in Table 6.1.

Table 6.1: Mixture of collection methods for food at home
Country, Year	Number of food items	Collection method	Quantity and/or Value
Afghanistan, 2016	92	Recall (weekly)	Quantity
Bangladesh, 2016	151	Diary (daily)	Both
Bhutan, 2017	115	Recall (weekly, monthly, yearly)	Both
India, 2011	134	Recall (multiple)	Both
Maldives, 2016	146	Recall (weekly)	Both
Nepal, 2015	97	Recall (weekly)	Both
Pakistan, 2015	170	Recal (biweekly)	Both
Sri Lanka, 2016	198	Recall (weekly)	Both

Food is composed of all edible goods that are purchased and consumed by the household with the purpose of nourishing. Food baskets are usually organized by categories into:

Cereals and cereal products;
Meat;
Fish and other seafood;
Milk, other dairy products and eggs;
Oils and fats;
Fruit and nuts;
Vegetables, tubers, plantains, cooking bananas and pulses;
Sugar, confectionery and desserts;
Salt, sauces and condiments, spices and culinary herbs and seeds.

Alcohol and tobacco are considered non-food items for measuring purposes, but it is common to find them in the food section of a questionnaire. In most surveys, households report both quantities and expenditures for most of the foods they purchase (e.g. three kilograms of rice for 5 rupees). However, this is not the norm. For example, Afghanistan LCS 2016 asks what was the amount of 92 items used in the last seven days, but the price of each item is collected separately through a market price survey.

Besides food produced and consumed at home, the food component should include food consumed outside the home, both formally such as at restaurants and cafés, and informally, such as small snacks and drinks. This component is usually recorded separately and is more likely to suffer from measurement error. An agreement on how these expenditures should be recorded has not been reached. For example, Afghanistan LCS 2016 asks for the total amount spent on food and drinks outside of home in the last month. In contrast, Bhutan LSS 2017 asks how many times did the household consume food for breakfast, lunch, dinner, or snack, outside of home in the last 7 days and what was the average value of the purchase.

Another component that is prone to measurement error is food transfers, which are currently measured unevenly with different collection methods. We have found that it is more common to find questions regarding transfers to the household than transfers from the household to other households. For example, Bhutan LSS 2017 asks what is the total value of a food item that was received as gift over the past 12 months. However, it does not ask what amount of the food item the household has transferred to others.

We provide a Tableau dashboard where the user may compare surveys according to:

The number of food items in the food basket
Whether they collect the quantity and/or value of food consumed at home and outside of home
Data collection methods (diary vs. recall) as well as their reference periods (daily, weekly, bi-weekly, monthly, and/or yearly)
Information on self-production
Information on transfers to and from the household

Instructions: The user may interact with this dashboard by filtering the observations. Try clicking on Bangladhesh in the bottom panel. This will filter the data in the upper and bottom graphs to show only the characteristics of the surveys in Bangladesh. The same is possible by selecting a single question, a single answer to a question, a single year, etc.

Figure 6.1: Food expenditure

6.2 Non-food

The second component, non-food expenditures, includes frequently purchased goods and services like soap, cleaning supplies, newspapers, and personal care items. It also includes some less frequent but regular purchases like clothing, footwear, kitchen equipment, curtains, bedcovers, etc. In most cases, the non-food expenditures component must be constructed by aggregating expenditures on goods and services from different sections of a survey.

A homogenous definition of what constitutes non-food expenditures is required. The Classification of Individual Consumption According to Purpose (COICOP) is the international reference classification for household expenditures. We define non-food expenditures as categories 2-12 of the COICOP:

Food and non-alcoholic beverages
Alcoholic beverages, tobacco and narcotics
Clothing and footwear
Housing, water, electricity, gas and other fuels
Furnishings, household equipment and routine household maintenance
Health
Transport
Information and communication
Recreation, sport and culture
Education services
Restaurants and hotels
Miscellaneous goods and services

Still, each country may determine its own definition of what constitutes non-food expenditures. In Bangladesh 2016, non-food expenditures included fuel and lightning, cosmetics and hygiene items, transport and travel, ready-made garments, clothing materials, footwear, household-use textiles, health treatment expenses, housing related expenses, education, recreation and leisure. In Bhutan 2017, non-food expenditures included tobacco and doma, clothing and footwear, transport and communications, household operations, recreation, furnishings and household equipment, agricultural input and machinery, miscellaneous expenditure, educational expenses, health expenses, rental expenses, energy for the home, and remittances abroad.

Similarly to what found studying the food component, non-food items are also collected for different reference periods, for example, from consumption in the last 30 days, past 3 months, or the last year. Monthly is the most frequent collection period for non-food expenditures. Both food and non-food expenditures have insufficient coverage on transfers.

We provide a Tableau Dashboard where the user may compare surveys according to:

The number of non-food items in the non-food basket
Whether they collect the quantity and/or value of non-food items
Data collection reference periods (daily, weekly, bi-weekly, monthly, and/or yearly)
Whether they provide information on transfers to and from the household

Figure 6.2: Non-food expenditure

6.3 Durable assets

The third component is consumer flows from durable assets. The consumer flows from durable assets (also referred to as user costs or rental equivalents) represent the opportunity cost of money invested in these goods taking into account their lifetime and depreciation. The most important finding regarding the collection of data on durable assets through household surveys in South Asia is that the data collection methods followed can be very different.

It is expected for data collection methods to vary over time and from one country to another. However, even for items within a country’s particular survey we can find different data collection methods. For example, even though both cows and cars may be considered productive assets, the information available for these two assets within a survey may be very different and would be contained in different sections of the survey. Even for more similar goods, like a washing machine and a sewing machine, the information provided by a survey may differ. For that reason, we found it necessary to characterize each item’s data collection method separately. Data available for household assets (e.g. a car) may be different from data available for agricultural assets (e.g. a tractor). Some assets are grouped together, while others are collected individually, and some may even have their own section in the survey. This is the case for cellphones, which are durable assets, but are usually given special treatment and collected separately in the surveys.

Long-lived goods (automobiles, appliances, furniture) have a positive and significant impact on living standards. These durable assets may deliver useful services to consumers through repeated use over an extended period of time. For that reason, surveys should collect data on all assets available to the household, not just the ones purchased recently. This is unfortunately the case for India 2011 and Pakistan 2015. Pakistan 2015 collects information on durable assets only if they were purchased in the last year. India 2011 collects specific details only for assets acquired in the last 30 days or in the last year.

To measure the flows from using durable goods over extended periods of time, we need to know their quantity, date of purchase, and how much they cost at the time of purchase vs. their current value. Nepal AHS 2016 represents the standard practice a survey should follow. Nepal AHS 2016 asks whether the assets are available, their quantity, age, purchase and current values, and whether the assets were paid for or received as gift. Asking how many years ago was an asset acquired provides more information than asking whether the asset was purchased in the last 12 months. The most incomplete example would have to be Pakistan 2015, which focuses only on assets acquired in the last year and only measures values, not quantities.

Table 6.2 provides a quick comparison between countries by presenting the number of assets collected, and their collection method (i.e. quantity and/or value). In addition, we provide a Tableau dashboard where the user may identify which surveys collect insufficient information regarding durable assets.

Table 6.2: Durables
Country, Year	Number of assets	Quantity	Value
Afghanistan, 2016	18 <br> 7	Yes <br> Yes	Yes <br> Yes
Bangladesh, 2016	27	Yes	Yes
Bhutan, 2017	13 <br> 9 <br> 23	No <br> Yes <br> Yes	No <br> No <br> Yes
India, 2011	53	Yes, but only if recently purchased	Yes, but only if recently purchased
Maldives, 2016	16	Yes	Yes
Nepal, 2016	23	Yes	Yes
Pakistan, 2015	67	No	Yes, but only if recently purchased
Sri Lanka, 2016	25	Yes	Yes

Figure 6.3: Durables assets expenditure

6.4 Housing

The fourth component is rent, which is in some cases observed directly for households who rent their house or apartment. For the rest, rent is obtained either by asking a household how much they would expect to receive each month for this house if they rented it out to someone, or by using a hedonic housing regression model based on dwelling characteristics and actual rent or house values. Therefore, there are three types of rent:

Rent paid by households when they are not owners of their dwelling: i.e. How much do you pay per month?
Self-reported or imputed rent by households who are owners of their dwelling: i.e. How much do you think you would have to pay if you had to rent this dwelling?
Predicted rent by households who are owners of their dwelling and did not self-report their rent: i.e. How much does the hedonic regression model predict a household would have to pay for a dwelling of these characteristics?

Most households in South Asia do not report values on paid rent because most households are owners of housing rather than renters. One area of South Asia where we find low levels of house ownership is Thimphu in Bhutan. According to Bhutan LSS 2017, in Thimphu, 59% of the households pay rent, while 17% of households live in rent-free dwellings. Of rent-paying households, 85% live in dwellings owned by private individuals and 14% live in housing owned by the government and by public corporations.

The methodologies for a hedonic regression model vary by country. The methodology followed by Afghanistan differs from the rest of the countries in that they base their hedonic pricing model on the dwelling’s values and then convert it to a monthly rent, instead of trying to directly estimate rents. In Afghanistan, actual rent for renters is collected by asking “How much money per month does your household pay to live in this dwelling?”. However, there is a low number of renters. Self-reported rent was not included in the survey. Instead, rents are predicted by a hedonic regression model. Half of all owners in the LCS 2016 report the value of their dwelling by answering “If you were to purchase this dwelling today, how much would it cost?”. For these households, a hedonic housing model is estimated and used to predict the value of the dwelling based on the characteristics of the dwelling. A hedonic housing model relates the housing price to factors such as size, location, construction materials, etc. Separate regressions are estimated for urban, rural and tent dwellings. The actual or predicted housing values are converted to a monthly rent by imposing a relationship based on interest and depreciation rates. For ALCS 2016 a depreciation rate of 1.5 percent and an interest rate of 2.5 percent are assumed.

For Bangladesh HIES 2016, the housing expenditure component may include the three types or rent depending on the homeownership status of each of the households: actual rent, imputed rent (i.e., the amount that homeowners report they would like to get if they could rent their house) or predicted rent. For households that did not report rent or self-report their rent, a predicted rent was estimated using a hedonic regression model. This regression model was estimated using the (log of) reported rent on the left-hand side and was regressed against a set of housing characteristics, including number of rooms, wall materials, access to electricity and tap water, kitchen, dining room, telephone connection, dwelling’s land size, and a vector of the 16 original strata dummy variables. The value of dwellings was collected by asking “If you want to buy or construct a dwelling just like this today, how much money would you have to pay?”, but it is not clear whether this value was used in the hedonic regression model.

In the rest of the countries, households are asked for their hypothetical rental values, not for the value of their dwelling. Table 6.3 summarizes how each survey collects rent or dwelling values for households who own their dwelling. For Bhutan LSS 2017, households were asked for the amount of rent they pay for their dwelling in a month. People who own their dwelling or stay in rent-free houses were asked to estimate the monthly house rent for their dwellings. The question was “How much would you pay if you had to rent this dwelling?”. A similar question is included in the surveys Maldives HIES 2016, Nepal AHS 2016 and Sri Lanka HIES 2016. India 2011 and Pakistan 2015 still lack enough data to measure rents.

Table 6.3: Housing
Country, Year	How is rent calculated for owners?
Afghanistan, 2016	If you were to purchase this dwelling, how much would it cost?
Bangladesh, 2016	Imputed rent; also asks If you want to buy or construct a dwelling <br> just like this today, how much money would you have to pay?
Bhutan, 2017	How much do you think you would have to pay if you had to rent <br> this dwelling?
India, 2011	NA
Maldives, 2016	How much would you expect to receive each month for this house <br> if you rented it out to someone?
Nepal, 2016	If any person want to take this house in rent, how much money <br> would have to pay?
Pakistan, 2015	NA
Sri Lanka, 2016	What is the estimated rent of owner-occupied house?

The most relevant dwelling characteristics displayed in the following Tableau dashboard are:

Rent captured by households’ reported rent or estimated by fitting a hedonic pricing model (regressing information available on housing characteristics on housing values)
Type of dwelling (house, part of house, separate apartment, shared apartment)
Tenure status
Area of dwelling, number of bedrooms, bathrooms, and kitchen
Material of the walls, roof, and floor
Sources of drinking water
Sanitation (access to flush toilet, pit latrine)
Access to electricity
Travel time to reach services
Others such as access to an Internet connection

Instructions: The user may interact with this dashboard by filtering the observations. Try selecting Bangladhesh in the bottom panel. This will filter the data in order to show only the characteristics of the surveys in Bangladesh. The same is possible by selecting a single question, a single answer to a question, a single year, etc.

Figure 6.4: Housing expenditure

We would like to thank Gabriel Nicolás Camargo Toledo for his support in the recollection and organization of the input data for this metadata analysis.↩