Heat or Eat ? : An empirical analysis of U.K. cold weather income support

Heat or Eat?:
An empirical analysis of U.K. cold weather income support
PRELIMINARY AND INCOMPLETE - DO NOT CITE
Timothy Beatty ?
Laura Blow
Thomas Crossley
University of York and IFS
IFS
University of Cambridge and IFS
July 5, 2009
Abstract
We investigate whether households trade o? spending on food and spending on heating.
We use a large sample of households from the United Kingdom and Norther Ireland over the
period 1974-2007. We ?nd evidence that low-income households reduce food expenditure during
periods of colder than average weather. In contrast, wealthier households increase spending on
food during colder than average weather. Further we investigate the e?cacy of the Winter Fuel
Payment, a social program designed to mitigate the e?ects of energy costs.
Key Words: Deprivation, Food Expenditure, Heat or Eat.
JEL Classi?cation: D12, Q18.
Selected Paper prepared for presentation at the Agricultural
& Applied Economics Associations 2009 AAEA & ACCI
Joint Annual Meeting, Milwaukee, WI, July 26-28, 2009.
Copyright 2009 by Timothy Beatty. All rights reserved. Readers may
make verbatim copies of this document for non-commercial purposes by
any means, provided this copyright notice appears on all such copies.
?Department of Economics and Related Studies, University of York, York, YO10 5DD, United Kingdom and Insti-
tute for Fiscal studies. Email: tb526@york.ac.U.K., Phone: (+44) (0)1904 434672, Web: www.timothybeatty.name
1

---

Heat or Eat?
An empirical analysis of U.K. cold weather income support
programs.
PRELIMINARY AND INCOMPLETE - DO NOT CITE
1
Introduction
During the winter of 2007-2008, the BBC cited reports that rising energy costs had forced some
households to choose between “heating and eating”. Low income households were said to be forgoing
food in order to pay for the costs of heating their homes1. The issue has received considerable
coverage in the U.K. media with some reports suggesting that energy bills may increase by as much
as 40% in the near future2. Given recent periods of high-energy costs, research into the impacts of
cold weather and heating costs on the quantity and quality of a household’s diet is timely.
The main goal of this paper is to understand the e?ects of cold weather and high energy prices
on the well-being of households, in particular the elderly and low income households. We will
explicitly address the heat vs. eat question and see whether households are forgoing nutritious food
to stay warm. We will also assess the design of the current policy instruments. How e?ective are
these payments at keeping households warm? In brief, we ?nd statistically signi?cant evidence that
low-income households in the U.K. are spending more on energy and less on food during unusually
cold weather.
There is recent evidence outside the United Kindom that low-income households do substitute
between nutritious foods and staying warm. Bhattacharya et al. (2003) and Cullen et al. (2005)
study this question in an American context. Both papers ?nd that low-income households decrease
food consumption in response to increased energy costs via colder weather and higher energy prices.
In addition, Bhattacharya et al. consider changes in the nutritional quality of the diet and ?nd
that low-income households reduce caloric intake by roughly 10% during periods of cold weather.
1http://news.bbc.co.U.K./2/hi/U.K._news/england/7234223.stm
2http://news.bbc.co.U.K./1/hi/business/7461635.stm
2

---

Our approach is novel relative to existing studies for several reasons, not least of which is that
to the best of our knowledge, the “heat or eat” question has not been studied outside of the United
States. This contrast is informative in that it allows us to see whether households bene?tting from
a more generous social safety net face the same “heat or eat” tradeo?. Moreover, the proposed
research focuses not only on the question of “heat or eat” but also on evaluating the e?ectiveness
of policies designed to address the trade-o?. This policy extension is a key innovation relative to
previous studies. Finally, data availability in the United Kingdom is richer, permitting a more
nuanced answer to the questions of interest.
In the U.K. there are two programs designed to mitigate the impact of cold weather on the
well-being of vulnerable households: the Winter Fuel Payment and the Cold Weather Payment.
Winter Fuel Payments are an annual payment intended to help with the costs of “keeping warm this
winter.” At present, households are eligible for a Winter Fuel Payment if they were normally living
in the Great Britain and Northern Ireland and aged 60 or above during a qualifying week (typically
in September). For the winter of 2008/2009, the value of the payment is 250 GBP per household,
which increases in value to 400 GBP at age 80. These payments are stable and predictable. In
addition to pensioners, households receiving welfare are eligible for a Cold Weather Payment in the
event of very cold weather. Households receive a payment of 25 GBP when the average temperature
in their area is recorded below zero over seven consecutive days (from November 1st to March 31st).
Because they depend on the weather, these payments are not predictable.
Measuring the causal impact of policy is di?cult because of the host of potential confounding
factors. Conceptually, one would like to observe the outcome variable of interest for the unit of
analysis in treated and untreated states. Here this would mean observing household food expen-
diture for the same household with and without a Winter Fuel Payment. Di?erences in outcomes
could then be directly attributed to the policy. The trouble of course is that a given household
cannot be in both states simultaneously. But we can use the sharp di?erences in eligibility criteria
to approximate this hypothetical experiment and attempt to assess the e?cacy of heating related
income supplements. For example, this year a person turning 60 on the 21st of September would
3

---

be eligible for a payment, whereas a person who turned 60 on the 22nd would not. More generally,
by comparing the nature of expenditures for otherwise similar households who fall on either side
of an arbitrary eligibility rule, we are able to tease out the causal e?ect of the Winter Fuel Pay-
ment. More formally this approach is known as a regression discontinuity design. Note also that
the Winter Fuel Payment is a relatively recent program. As a result, we can leverage data from
several winters before the program was in place to compare di?erences in household food expen-
diture. This allows further insight into program e?cacy. We approach the evaluation of the Cold
Weather Payment scheme in a similar manner. As before sharp di?erences in program eligibility
are exploited to tease out the causal e?ect of the policy. These di?erences in eligibility are over age
as before but also over weather: seven days of unusually cold weather triggers a payment but six
days does not. Comparing the di?erent e?ects of the two programs is also informative. The Winter
Fuel Payment acts as an income supplement; eligible households receive it no matter the cost. The
Cold Weather Payment acts more like an insurance policy; eligible households only receive it in
the event of unusually cold weather. The relative e?cacy of these di?erent approaches in largely
unstudied.
2
Data
Our data consists of the Family Expenditure Survey (FES) for the years 1974 to 2000 and the
follow-on Expenditure and Food Survey (EFS) for the years 2001 to 2007. The FES/EFS are
large nationally representative surveys of household expenditure, food consumption and income.
The EFS has a number of advantages over other comparable expenditure surveys - notably the
Consumer Expenditure Survey (CEX) - for the purposes of this research. Unlike the CEX, quan-
tity information is observed, which allows one to construct the nutritional pro?le of a household.
Unlike the Canadian Food Expenditure Survey (FOODEX), the EFS collects detailed non-food
expenditure and income information, which will be useful in constructing instruments. Unlike the
Continuing Survey of Food Intakes by Individuals (CSFII) and the The National Health and Nu-
trition Examination Survey (NHANES), the EFS collects information on items other than food.
4

---

This allows a complete picture of household expenditure to be constructed. A ?nal advantage of the
EFS is that it collects information on relatively disaggregated food commodities, the EFS contains
information on expenditure on 231 food-at-home items. A weakness relative to nutritional surveys
is that it is not possible to disentangle consumption among members of a household.
The EFS consists of a series of questionnaires that collect information on recurrent household
expenditures, infrequent expenditures, household demographics and detailed income information.
This provides a detailed picture of non-food consumption items. Of particular importance for the
current study, the survey collects expenditure on various heating fuels (gas, oil, electricity, etc.).
Table 1 summarizes average weekly expenditures on Food-In, Food-Out, Energy and Clothing for
the the households in our sample. In addition we provide information on average after tax household
income and average household size. Where applicable, amounts are in British Pounds, de?ated to
1987 levels.
Table 1: Summary statistics
Variable
Mean
(Std. Dev.)
Food-In Expenditure
28.325
(16.939)
Energy Expenditure
9.891
(7.390)
Food-Out Expenditure
8.096
(10.999)
Clothing Expenditure
14.609
(26.695)
Household Income
203.215
(206.603)
Household Size
2.539
(1.374)
N
236934
We obtain average monthly ground temperature for each of the eight di?erent climatic regions
that make up the United Kingdom and Northern Ireland from the Met O?ce 3. Figure 1 displays
box plots of ground temperatures by region. To these data we add weather data from the British
Atmospheric Data Centre, which allows us to construct measures of anticipated and unanticipated
weather events.
We also merge in price data from the Retail Prices Index for LPG, Oil and
Electricity to control for changes in the price of home heating, and for food.
3http://www.metoffice.gov.U.K./climate/U.K./
5

---

Figure 1: Average Temperature by Region
3
Results
3.1
Expenditure
Initially we consider how poorer households adjust to changes in home heating costs; speci?cally,
unanticipated changes such as cold weather shocks.
Are they less able to bu?er against such
shocks than richer households? And must they therefore reduce expenditures on other essentials
such as food? Our empirical strategy estimates a model of spending on food and on fuel using
the Expenditure and Food Survey (EFS) and the earlier Family Expenditure Survey (FES). Our
identi?cation strategy is as follows: since we have a wealth of information such as year, month,
region and prices over a long time span we can use periods of normal seasonal weather and stable
energy prices to establish “normal” baseline food and fuel expenditure patterns for poorer household
groups and richer ones. A comparison of expenditure behavior during periods of unseasonable
weather and/or rapid change in energy prices with this baseline behavior gives us information
6

---

on how households adjust consumption in response to unseasonable weather or unusual energy
prices. We can then compare the responses of the richer - presumably not resource constrained -
households to that of the poorer households to look at whether poorer households are required to
choose between food and heating in the face of these cost shocks.
As a preliminary step, we follow Bhattacharya et al. (2003) and estimate the following equation
for log household expenditure on food-in, energy, food-out and clothing.
(1) ln (yi)
= ?0 + ?1 Mean Temperature + ?2I<.25 + ?3I<.25 ? Mean Temperature
+?4I>.75 + ?5I>.75 ? Mean Temperature + ?6 ln (Household Size)
8
12
2007
8
2007
+
?jAreaj +
?jMonthj +
?lYearl +
?j,lAreaj ? Yearl + ui
j=2
k=2
l=1975
j=2 l=1975
For low-income households, our main variables of interest are mean Temperature, I<.25 a dummy
variable which takes on a value of one if the households is below the 25th percentile of after-tax
household income, and an interaction between the low-income dummy and mean temperature. We
also consider high-income households, de?ned to be those above the 75th percentile of after-tax
household income. A priori we would not expect these households to trade o? heating or eating.
Finally, we control for the log of household size.
We include dummy variables for year and area and year/area interactions to de-trend expendi-
tures within each climatic area. We include monthly dummy variables to capture normal seasonal
variation. The e?ect of including these dummy variables is that the e?ect of temperature on expen-
diture is identi?ed o? within area temperature variation. In other words, identi?cation is driven
o? the comparison of a cold December to an average December in a given climatic area rather than
an average cold month to an average warm month.
Evidence of a “Heat or Eat” tradeo? will be found in the coe?cient on “Mean Temperature”
and on the interaction between “Mean Temperature” and the income category dummies. The
cumulative e?ect of these dummies will tell us how low and high income households respond di?er-
entially to unanticipated changes in temperature. A positive coe?cient indicates lower expenditure
7

---

when temperatures are low, which would provide evidence of a tradeo?.
Table 2 contains our main regression results. Column 1 reports results for food. First, we note
that the e?ect of mean temperature is small and not statistically signi?cant from zero. However, the
coe?cient on the interaction term between low-income and temperature is positive and signi?cant
at all conventional levels. That is to say we ?nd that higher (lower) than average temperatures
are associated with higher (lower) expenditure on food at home. In contrast, the coe?cient on
the interaction term between high income and temperature is negative and signi?cant, high in-
come households have lower (higher) food expenditures when temperatures are higher (lower) than
average. We note that, lower-income households spend less and high income households spend
more on food relative to the omitted category, those whose income falls between the 25th and 75th
percentile.
Column 2 reports results for energy expenditure. Here the e?ect of mean temperature are sta-
tistically signi?cant for all households and have the expected sign; increases in mean temperature
lead to decreases in energy expenditure. Also, low-income households spend less on energy relative
to the omitted category and high income households spend more relative to the omitted category.
The interaction term on low-income and mean temperature is negative and signi?cant suggesting
that low-income households increase expenditure in response to unusually cold weather. The cor-
responding interaction term for high-income households is positive and signi?cant, suggesting that
high-income households respond less to unusually hot or cold temperatures. Perhaps this higher
level of base energy expenditure provides a bu?er against unusually cold temperatures for these
households.
Column 3 and 4 report results for expenditure on food outside the home and expenditure on
clothing. In contrast to expenditure on food in, expenditure on food out and expenditure on clothing
is positively and signi?cantly related to mean temperature. However, the interaction terms on low-
income and mean temperature are not statistically signi?cant, nor are the interactions between the
high income dummies and mean temperature.
8

---

Table 2: Regression Results: Expenditures
(1)
(2)
(3)
(4)
VARIABLES
Log(Food-In)
Log(Energy)
Log(Food-Out)
Log(Clothing)
Mean Temperature
2.78e-05
-0.00635***
0.00436**
0.00462*
(0.000916)
(0.00103)
(0.00219)
(0.00280)
Income below the 25th Percentile
-0.208***
-0.0662***
-0.834***
-0.608***
(0.00574)
(0.00708)
(0.0158)
(0.0196)
Income LT 25th pct * Mean Temp
0.00281***
-0.00602***
0.00222
-0.00111
(0.000535)
(0.000676)
(0.00150)
(0.00184)
Income above the 75th Percentile
0.217***
0.127***
0.796***
0.555***
(0.00672)
(0.00661)
(0.0135)
(0.0177)
Income GT 75th pct * Mean Temp
-0.00154**
0.00263***
-0.00101
0.00204
(0.000622)
(0.000597)
(0.00122)
(0.00160)
Log Household Size
0.707***
0.354***
0.488***
0.450***
(0.00238)
(0.00253)
(0.00521)
(0.00681)
Constant
2.838***
1.981***
1.590***
1.566***
(0.0181)
(0.0217)
(0.0451)
(0.0565)
Area, Month, Year and Area/Year Interactions are included in all models.
Observations
236371
232147
203990
172975
R2
0.437
0.190
0.236
0.196
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
9

---

Table 3: Regression Results: Calories
(1)
VARIABLES
Log(Calories)
Mean Temperature
-0.00208
(0.00126)
Income below the 25th Percentile
-0.0175*
(0.00988)
Income LT 25th pct * Mean Temp
0.00247***
(0.000950)
Income above the 75th Percentile
-0.0225***
(0.00701)
Income GT 75th pct * Mean Temp
0.000262
(0.000699)
Log Household Size
0.875***
(0.00324)
Constant
9.745***
(0.0295)
Area, Month, Year and Area/Year Interactions are included.
Observations
198739
R2
0.385
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
3.2
Calories
We extend our analysis by considering whether the quantity of calories purchased varies in response
to unanticipated cold weather events. To this end, we re-estimate equation 1, where the dependent
variable is the natural logarithm of calories from all foods purchased during the diary period. Note
that in this preliminary analysis, we use data from the National Food Survey (NFS) only, for the
years 1974-2000. 4
Table 3 summarizes our ?ndings from the calorie model. First we note that the coe?cient
4Future work will combine the NFS and the EFS to present results from 1974-2007
10

---