Understanding energy consumption in Mexico: an …iussp.org/sites/default/files/event_call_for_papers/L...1 Understanding energy consumption in Mexico: an age-period-cohort analysis1

1

Understanding energy consumption in Mexico: an age-period-

cohort analysis1

Landy Sánchez Peña

El Colegio de México

There is increasing interest on the environmental implications of households’ energy consumption,

given the impact that global fuel demand has on green gas emissions. Residential energy demand

has been growing and is expected to keep doing so over the next decades, particularly in

developing countries (IEA 2013). This draws attention to the environmental implications of

consumption practices, as well as to the need for better understanding their drivers and

heterogeneity across the globe. Different studies establish the impact of income and other socio-

demographic characteristics on the level and fuels used by households. Studies also show the

relevance of cultural norms and values for explaining consumption (Ropke and Reish 2005). A less

explored component is the time dimension of consumption: how it changes as people ages, across

cohorts and historical periods. Asking questions about changes in energy behaviors over the life

course is necessary in order to understand better how environmental practices are shaped; the

analysis could also provide insights about the possibilities of transforming them.

Over the last two decades, Mexico’s population became more urban and educated, with

lower fertility rates and smaller household size, and experienced notable changes in its labor force

and age structure. At the same time, educational and public discourses around the environment

1 Paper presented at the XXVII IUSSP International Population Conference, Busan Korea, August 26-31, 2013

2

have also changed importantly while the country experienced deep variations on income

distribution and economic performance. Altogether, these transformations could have altered

household energy demand, as a result of combined effects of changes on population socio-

demographic profiles, historical shifts on economic wealth, and cohort lifestyles and consumption

practices. Therefore, it is relevant to examine the dynamics of household energy consumption

simultaneously investigating the existence of effects of age, period, and cohort (APC).

To differentiate these effects, I pooled data from eight National Income-Expenditure

Household Surveys (1992-2008) to assemble synthetic cohorts of urban Mexican residents.

Building on previous work (Sanchez and Jasso 2012), I first examine whether there is a significant

cohort effect and to what extent of energy consumption differ across cohorts; I analyze the

argument that younger cohorts have higher energy consumption. Second, I examine period effects

by looking at whether Mexico’s economic context translates into adjustments of household energy

consumption. Finally, I assess if cohorts respond differently to period effects. Following the

methodology developed by Yang and Land (2006), Yang (2008 a, b), I implemented an age-period-

cohort hierarchical model for cross-classified records, where households are nested

simultaneously in cohorts and periods (year of the survey). This work is part of a broader

discussion on whether economic development transformed lifestyles and population’s

consumption habits, and it hopes to contribute to the discussion on the new challenges to balance

development, equality and environment.

Determinants of household energy consumption

Households are a better suited unit of analysis to understand consumption that aggregated

national statistics since consumption varies significantly across population groups, making it

imperative to consider households’ characteristics to explain such variation (Currand and de

3

Sherbinin 2004). Moreover, households are the domains where budgetary; saving and

expenditures decisions are made; albeit not except of inner conflict and inequalities.

Household income is a clear determinant of how much energy households use. It has direct and

indirect effects. On one hand, income impacts the fuels households acquire: higher income units

buy modern fuels (electricity and gas) and less traditional ones such as wood and coal (Jiang and

O’Neill 2004). Income also affects household capacity to acquire larger homes and more

appliances that tend to increase their energy demand, although it could also imply buying more

efficient ones ((O'Neill Brian & Chen; Gram-Hanssen, K. 2005). Across countries, and particularly

developing ones, higher incomes are associated with larger energy consumption Guertin, et al.

2003, Lenzen et al. 2006, Sari and Soytas 2007)

However, at the same income level there are significant differences in household energy

demand since they differ on their saving rates, demographics and practices (Ropke and Reish

2005). While income represents households’ spending power, they could have different saving and

debt patterns, indicating also their needs and preferences (Pachauri 2006). Other household

characteristics also has been shown to play a role energy demand, such as family structures and

living arrangements, household size, occupation and working hours and, broadly speaking, their

lifestyles (Lenzen et al., 2006, Jensen 2008, van den Berg 2008, Przkawetz et al., 2001, Pucher et al.

1998; Greening and Jeng 1994). However, the weight of these factors varies across countries and

historical moments. Studies for Mexico, (see Sánchez 2012) show that household size is a key

determinant of household energy use, as in other countries. This relationship is not linear since

larger households can generate "economies of scale": there is a basic energy required to sustain a

home, which is apportioned between a greater number of individuals in bigger families; such

economies of scales compensate the increases on total energy consumed by larger households

(Ironmonger et al. 1995). Additionally, family arrangements and household life cycle can also

4

influence energy consumption both represent alternative ways of organizing daily routines and

needs, which involve different energy loads (Shipper 1996 of Sherbinin 2007). Such findings are

also present in the Mexican case (Sanchez 2012)

On the other hand, research also suggests that the characteristics of the household head,

such as sex, education, occupation and age are associated with energy use that home does.

However, the direction and magnitude of these effects is not as clear. On one hand, it is possible

to expect that higher educational and occupational levels will be correlated with greater levels of

energy consumption, because a higher social status is often associated with intensive energy

lifestyle, such as more use of electronic technologies, intensive private transport use and larger

homes (Ropke and Reisch 2005). Other studies, however, have suggested that education could

also imply a greater environmental awareness or sensitivity to fuel prices, which could result in

lower energy consumption (Bhattacharjee and Reichard 2011). Meanwhile, studies of energy

consumption in transportation have found that female-headed households tend to have lower

levels of energy demand (Przkawetz et al. 2004). It is not clear that such behavior occurs in the

case of Mexico, but studies in the country have found that households headed by women if they

have a distinct pattern of spending, reducing wasteful spending (Tuirán 1993).

Age of the household head is often used as an indicator of household life cycle, and results

for some countries found that age is linearly associated with higher energy consumption;

suggesting that households in more advanced stages increase their consumption net of other

effects (Pachauri 2007). Those increments are theorizing to reflect changes in lifestyles and needs.

Changes in energy consumption with aging points to a larger issue about the need to understand

better consumption over the life course. To the best of my knowledge, there is not longitudinal

analysis about household energy demand, although time series analysis provides a look to

5

differences in aggregated levels of energy consumption. On one hand, several studies show an

association between changes in country’s income levels and energy demand (Erbas and

Irongmonger 1995, Ramazan and Ugur 2007), suggesting the need to examine how changes on

development levels affects energy consumption. Moreover, studies also suggest that increments

on living standards impact households’ demographics and lifestyles (Lenzen et al. 2006), which in

turn shape how peoples consume. In this sense, it is necessary to differentiate the temporal

components of these changes, elucidating between effects of age, period and cohort.

Age, period, and cohort interplays in energy consumption

In population studies, the age, period, cohort distinction is well-known. APC is a model of societal

transformation since changes are experienced by individuals due to aging and period effects, and

through the succession of cohorts (see Mason & Fienberg, 1985). People changes as they get

older as a result of combined effects of biological, social and psychological mechanisms (Alwin and

McCammons 2003). At the same time, individuals respond to historical events and processes, that

is, period impacts that cross population groups. On the other hand, the idea of cohort effects

refers to how formative experiences contribute to shape individuals values and behaviors

(Mannenheim 1952). This effect arises because of shared experiences, similar timing and

conditions of life transitions, and socio-demographic composition of the cohort itself. As Alwin and

McCammons (2003, p.23) claim “Members of a birth cohort share a social history, that is,

historical events and the opportunities and constraints posed by society at a given time. Further,

members of a birth cohort share the experience of the life cycle at the same time […]. And finally,

members of a birth cohort share the experience of the cohort itself, that is, the distinctive aspects

of the cohort, for example, its size or its level of education, are something unique to the cohort.”

6

Cohort replacement is a mechanism of social change since it could implicate not only a change in

population composition but also in attitudes and practices.

Changes in energy consumption could be explained by a combination of individual, period,

and cohort effects. First, as household move to the life course energy needs and practices

transform; it could reflect, for example, differences in domestic workloads or time allocation. On

the other hand, birth cohorts could have distinct energy practices as a result of their formative

experience on energy availability, prices, and public policy, as well as the economic and cultural

context they grew up. On this regard, it is common to suggest that cohort exposed to higher living

standards would have more intense energy practices; in the Mexican case, this implies expecting

that younger cohorts will have higher energy consumption since the country’s wealth has

increased over time. Nevertheless, studies conducted in Europe suggest that cohorts exposed to

post-materialist values and high levels of pollution during childhood are more inclined to support

pro-environmental positions (Menz and Kühling 2011) and, therefore, one would expect a lower

energy consumption. Menz and Kühling’s argument is potentially extensible to other contexts

where exposure to environmental problems increased and there is public discussion on the topic,

as it has happened in Mexico in the recent decades (Molina 2005, Lezama 2011). Analytically it is

necessary to distinguish between cohort composition and a “net” cohort effect. For example,

Mexican younger cohorts have smaller household size than their predecessors and size is

associated with larger per capita consumption. Still, it remains to be seen if there is a cohort

membership effect beyond differences in their demographic profile, which could be a sign of

distinct energy practices shaped by formative experiences shared by the cohort such as energy

shortages, environmental discourses or prices. Thus, we could expect cohort differences on

energy consumption in Mexico due to differences on their socio-demographic profiles, as well as

differences on exposure to environmental problems and discourses. However, it is unclear

7

whether younger cohorts will have higher energy demands given the conflicting nature of the

forces driving energy consumption.

On the other hand, period effects could also shape energy consumption. Time-series

analysis suggests a link between economic growth and household energy demand at the country

level: higher income countries have higher energy demand and increases in living standards also

translates in greater fuels consumption (Sari and Soytas 2007). Still, it is less clear how economic

growth relates to individual household consumption, that is, how households respond to their

economic context. The evidence about household response to prices is mixed while several studies

show that households respond to price changes, but not as quick and largely as economic theory

will predict (for a review see Gillingham et al 2009). We have limited knowledge on whether

macro-economic variables impact household energy consumption; in the Mexican case this issue is

relevant given recurrent economic crisis that have impacted overall consumption levels in the

country, as well as a highly unequal income distribution that has hindered economic development

of the poor (Scott 2009). Moreover, historical times could matter differently for households; for

example, if energy consumption decreases for all households in times of economic downturn, it

may do even more so for middle-income households.

This paper examines the following questions:

a) Are there age, period and cohort effects on household energy consumption in Mexico?

b) Do cohort variations hold once that period and life course patterns are taking into

account? Do younger cohorts have higher energy consumptions than older ones?

c) Does Mexico economic development matter to explain variations on energy consumptions

across households?

d) Do cohorts respond differently to historical times?

8

Data and Methods

To analyze these questions, I pooled data from eight National Household Income and Expenditure

Surveys (ENIGH by its Spanish acronym), years 1992, 1996, 1998, 2000, 2002, 2004, 2006 and

2008. The Survey is collected by the National Institute of Statistics and Geography and it is a

national sample, with representativeness for urban and rural areas. In this exercise, I only use the

subsample of urban households (residents in areas of 2,500 or more inhabitants). Because the

Survey is used to estimate poverty levels in the country, it received large attention by the Statistics

office and its sample grew considerably over time: the 1992 ENIGH includes 10,530 households

(6,554 urban), while the 2008 ENIGH included 29,468 households (22,734 urban).

The ENIGH collects household demographic information and characteristics of their homes

and collects detailed information about frequency, amounts and details of income and

expenditure items. This article analyzes the direct energy demand of households, i.e. the one used

in the home for cooking, heating or cooling and maintenance. To estimate energy consumption I

employed a common standardized approach (see Jiang and O'Neill 2004); I first divide fuels

expenditures by year prices, in order to obtain the quantities purchased by the household. Then, I

multiply the quantities by its corresponding calorific power, which is a measure of the energy

contained in each fuel unit and is measured in megajoules. The calorific energy intends to measure

how much energy is used when a fuel is used (SENER 2009a) 2; it provides a proxy to the energy

efficiency and environmental implications of each fuel to the extent that firewood has a higher

caloric content than electricity, reflecting its less efficient combustion. This procedure allows for a

2 Calorific values are published annually by the Ministry of Energy and adjusted over time to reflect

efficiency gains in the production, supply or distribution of fuels.

9

comparable measure over time and across households, which cannot be obtained by looking at

household spending. The caloric energy indicator accounts for the fact that households may be

spending the same amount but in different types of fuels and with different environmental

impacts. For example, two household may have spent the same amount, but one bought firewood

and the other electricity; thus the environmental impact of the first household would be higher

than the second one. Similarly, over time households can change the fuels they consume, so that

the same level of spending could represent different energy impacts. Also, spending variable could

hide changes in fuel prices. In this sense, a caloric content measure addresses these issues and

better summarizes the environmental impact of energy consumption.

The benefits of this methodological strategy, however, are limited by the availability of

data, particularly fuel prices. In Mexico fuel prices vary by regions, electricity tariffs are fixed

depending on area of residence and level of consumption. ENIGH data does not include

information on the actual fuel price paid by the household. Therefore, I have to use the average

price paid at the regional level or national, if the former was not available; most data published by

Ministry of Energy. Additionally, there are no officially published time series for firewood and coal;

instead I used data reported on regional and local studies. It is necessary to mention that firewood

and coal consumption only accounts for 8% of household spending in urban areas. To account for

regional variations in local markets fuels, I introduce a series of dummies that indicate the

geographical area of residence in the country; those regions are associated fuel availability and

prices and weather differences. These regions were defined based on the energy balances

published by the Secretary of Energy in Mexico.

I pooled the ENIGHs and construct synthetic cohorts. Following the approach Yang and

Land (2006) Yang (2008a and 2008b) I used a crossed-classified hierarchical model where

10

households are simultaneously nested within cohorts and periods: cohort members can be

interviewed in multiple replications of the survey and the members of a cohort can be interviewed

on several occasions (Land 2008). Using this cross-classification multilevel structure is possible to

use different time variables for age, period and cohort and consequently find a solution to the

classic identification problem in APC models(Yang and Land 2006, 2008). This method has the

advantage of using microdata rather than aggregating information, as in other APC methods do.

Also, by considering the hierarchical nature of the data we can ask questions about how group

membership (cohort) and historical context (periods) matter for household behavior.

Furthermore, these models allow estimating fixed and random effects and, therefore, it is possible

to average effects and variations between levels. As I shall discuss in the results section, I

estimated a mixed effect model to account for the cohort and period effects and to test the

hypothesis of differential period effects. The full model is estimated as follow:

Level 1

Level 2

Combined or mixed model

Where i= households, j=birth cohorts, k=periods

11

As I dependent variable I used the log of household per capita energy consumption. I use

the age of the household as a continuous variable indicating the life course of the household, and I

construct cohorts in five years intervals based also on the age of the household head. Periods

correspond to the year of the survey. At the household level variables previously used and tested

to explain household energy consumption were introduced, namely, total per capita income of the

household, per capita expenditure (as proxy of budgetary constraints); sex and education of the

household head, household size, number of rooms (as proxy of dwelling size); family structure and

geography of residence. At the period level, I will introduce GDP growth and income distribution

indicators. The precise way these variables were operationalized will be discussed in the results

section.

Results

An initial exploration of household direct energy consumption shows that it grew between 1992

and 2008, from 3222.48 megajoules quarterly in 1992 to 3939.5 in 2008. There were some

fluctuations over these years, but an incremental trend prevail (See table 1). The same table

shows per capita energy consumption by cohort. Differences between categories are notorious,

and younger generations have a lower per capita energy demand than older cohorts. In fact, the

average energy consumption of the cohort 1913-1917 is 37% larger than the 1984-1988 cohort,

and 31 % greater than the youngest cohort (1989-1993). In addition, table 1 also shows that as the

head of the household gets older energy consumption increases. This may reflect the life cycle of

the household (Pachauri 2004), and in this table it may also be associated with household income,

that increases with age, even at older stages.

12

Of course, behind these averages there are notable changes in the socio-demographic

composition of the population across cohorts and over time, such as differences in average

household size, income and educational levels. An exploratory look at some characteristics by

cohorts illustrates this point (Figure 1 and 2). Younger cohorts are more educated as the higher

education category gained weight in detriment of those in the lower category (Figure 1). The graph

also gives a sense of speed of the educational transition in the country. Cohorts also differ in their

household size composition; older cohorts have a higher number of single households and one

and two members units. Middle cohorts have a larger proportion of 4- 5 members units, as well as

6-7 ones, while younger cohorts have also a large proportion of 2-3 members and single units

households. These differences reflect both differences in fertility rates as well as the life course

stage at which households are (graph 2). This cohort composition illustrate the relevance of an

By year By cohort By age groups

1992 3222.5 before 1913 4577.3

1996 3285.4 1913-1917 4691.1 less than 20 3022.0

1998 3439.7 1918-1922 4890.2 20-29 2868.2

2000 3747.4 1923-1927 4703.4 30-39 2989.7

2002 3559.4 1928-1932 4571.1 40-49 3546.5

2004 3749.3 1933-1937 4494.5 50-59 4186.8

2006 3826.0 1938-1942 4112.8 60-69 4472.8

2008 3939.5 1943-1947 4244.8 70+ 4772.8

Average 3628.5 1948-1952 3942.2 Average 3628.0

1953-1957 3641.4

1958-1962 3390.8

1963-1968 3171.6

1969-1973 2998.0

1974-1978 2912.5

1979-1983 2846.8

1984-1988 2951.9

1989-1993 3234.5

Average 3628.0

Table 1 Mean Household per capita energy consumption

13

APC analysis once that cohort differences are apparent but also indistinguishable from age and

period effects with aggregated data.

Graph 1

Graph 2

020

40

60

80

100

Befo

re 1

913

1913-1

917

1918-1

922

1923-1

927

1928-1

932

1933-1

937

1938-1

942

1943-1

947

1948-1

952

1953-1

957

1958-1

962

1963-1

968

1969-1

973

1974-1

978

1979-1

983

1984-1

988

1989-1

993

urban

No education Primary education

Some secundary education Higer education

N.E

Perc

ent

Graphs by location

Graph. Education by cohort

020

40

60

80

100

Befo

re 1

913

1913-1

917

1918-1

922

1923-1

927

1928-1

932

1933-1

937

1938-1

942

1943-1

947

1948-1

952

1953-1

957

1958-1

962

1963-1

968

1969-1

973

1974-1

978

1979-1

983

1984-1

988

1989-1

993

urban

1 2-3

4-5 6-7

+7

Perc

ent

Graphs by location

Graph. Household size by Cohort

14

HAPC Model Results

As mentioned previously, a cross-classified hierarchical model uses microdata to identify

the effects of age, period and cohorts independently and it takes advantage of the nested

structure of the data to explore questions about group membership and context impacts. In the

appendix a there is a table showing the nested structure of the pooled data, which adds to 97120

households. Model 1 introduces Age as a level 1 predictor and assesses the significance of Cohort

and Period effects through testing the significance of the random components. Results show that

across all households –years, cohorts of typical age3- the average per capita energy consumption

is 2125.68 megajouls quarterly.4 The variable age was introduced as log variable, and results

indicate that it has a positive impact on the dependent variable, to the extent that an 10%

increase on age yields an increment of 4% on per capita consumption (see models results in table

B at the appendix). Model 1 also support significant cohort and period effects and shows sizable

variations in both of them. Figure 3 graphically represents cohort effects, that is, the “added”

effects of cohort membership . It is apparent that middle rank cohorts have a

stronger effect than older and younger cohorts once that age and period effects are taking into

account. Also, belonging to younger cohorts (1963-1968 and up) has a “negative” impact over the

overall mean, that is, tend to reduce the consumption mean. On the other hand, the period effect

also shows that 1992, 1996 and 2002 decrease average per capita consumption,

once that age and cohort periods are considered. Those years are, in fact, times of economic

downturn in the Mexican economy, as we can see below. In contrast, the other years increase

average household per capita energy consumption, particularly the year of 2004.

3 Age is centered at the gran mean, as all other continuous variables are.

4 The depended variable is log of per capita energy consumption, I always present the results

exponenciated. When the exploratory variables were also introduced also as log, the results are discussed on percent change.

15

Graph 3

Graph 4

0

500

1000

1500

2000

2500

3000B

efo

re 1

91

3

19

13

-19

17

19

18

-19

22

19

23

-19

27

19

28

-19

32

19

33

-19

37

19

38

-19

42

19

43

-19

47

19

48

-19

52

19

53

-19

57

19

58

-19

62

19

63

-19

68

19

69

-19

73

19

74

-19

78

19

79

-19

83

19

84

-19

88

19

89

-19

93

me

gajo

uls

, qu

arte

rly

Birth Cohorth Effect

mean + cohort

overall mean

0

500

1000

1500

2000

2500

3000

19

92

19

96

19

98

20

00

20

02

20

04

20

06

20

08

me

gajo

uls

qu

arte

rly

Period Effect

mean + period

overall mean

16

While the previous model presents the “pure” APC effects, Model 2 introduces other

predictors at the household levels. Variables previously used in studies of energy consumption

were added to the model. As in the previous model, age has a positive and significant effect on per

capita energy consumption of the household, after considering the impact of other variables, a

10% increase in age still has 4% increment on per capita energy consumption. This supports the

argument that the life course the household has an impact on the level of energy consumption,

net of other characteristics, and that such effect remain stable across cohorts and periods. Also, as

other studies show, income increases energy consumption. Figure 4 graphs the marginal effect of

this variable, holding constant other variables at their means for the continuous variables and at

the category of reference for categorical ones. This is the average effect across all cohorts and all

periods. The variable was introduced as log, so I estimated its effect at different values. It is

noticeable that income increases consumption and it does so more rapidly in the middle and

upper income ranks. This is similar to previous findings that indicate a positive non-monotonic

association between energy consumption and income, where higher incomes have a larger effect

than lower ones (Brounen, Kok and Quigley 2012)

Graph 5

0

500

1000

1500

2000

2500

3000

3500

10

0

50

00

10

00

0

15

00

0

20

00

0

25

00

0

30

00

0

35

00

0

40

00

0

45

00

0

54

00

0

70

00

0

Income average marginal effect

per capita enregyconsumption

17

In addition, this model adds total per capita spending as an indicator of the budgetary

constraints faced by the household, as suggested by Pachauri (2004). The results show that as

spending increases (fewer restrictions) so does household consumption, net of the effect of

household income: a 10% increment in per capita household spending yields a 5% increase in

energy consumption. Additionally, results show that increments on household size are associated

with lower per capita consumption, which is consistent with findings in previous work (O’neill and

Cheng 2002). This would point to the presence of economies of scale, which are explained by the

existence of a basal energy needed to sustain the operation of a home. The size of the dwelling,

indicated by the number of rooms also has a positive impact on per capita energy consumption, to

the extent that an increment of a 10% will increase household consumption by 0.5%. This variable

indicates energy demand related to the physical and space properties of the homes, since

socioeconomic status of the households are being accounted by other variables in the model.

Moreover, education of the household head also has significant effects on energy

consumption. Compared to those with no education, being headed by a person with basic

education increases 12% per capita energy consumption while having some high school education

do so by 16%, interestingly having higher education increases 10% in contrast to those with no

education. This effect is net of the increases do to income; hence, it is possible to postulate that it

accounts for practices and perceptions about consumption. Something similar happens with family

arrangements that are correlated significantly with energy demand, beyond household average

size. Compare to single households, households without children increase per capita energy

consumption by 33% while couples with children do 20%. Compose households also have a larger

impact (28%) than extended households (20%). These differences could be associated with

everyday practices and organization arising from their composition and structure, to the extent

that households with less traditional cores are more likely to have larger energy loads. Results

18

also show that region of residence also impacts consumption. The Northwest and Northeast have

the largest energy demand compared to the Western region (reference category), those regions

experience more extreme temperatures year around, and it could also reflect differences on

energy prices and subsidies.

Despite the relevance of household-level characteristics to explain energy consumption,

the model still support significant cohorts and periods effects, suggesting that cohort membership

and year of observation mediates the association between sociodemographis and energy

consumption. Though, these effects are small particularly in for cohorts. A better look to these

effects can be taken by assessing them individually. The following figure graphs the cohort effects

(mean and 95% confidence interval). It is apparent that once we consider household

characteristics, the majority of the cohorts no longer have a significant, independent effect over

the household energy consumption (graph 6). These indicates that the cohort effect previously

identified was mainly a compositional effect; that is, group differences in education, household

size, income, etc. However, it is noticeable that there are a few cohorts that still have a significant

effect beyond those compositional differences. Those are the two at the extreme right hand size

of the graph (1943-1947, 1948-1952). Those cohorts were born and rise in times of economic

expansion and oil foster growth, and they increase average energy consumption.

19

Graph 6

On the other hand, model results suggest that period effects remain significant above and beyond

household characteristics and cohort membership; in fact only three time points are no longer

significant (1998, 2000, 2008). It is interesting to notice that the effect of the remaining years

change once that we consider first-level variables. For example, the 1992 and 1996 that in model 2

decrease average energy consumption now have a positive impact on the mean. This suggests that

those previously observed effects were due to differences in the composition of households across

time. Overall, the fact that period effect remains highly significant, point to the need for better

explore how time characteristics may impact energy consumption (graph 7).

19

23

-192

7

19

84

-198

8

Ante

s d

e 1

913

19

74

-197

8

19

89

-199

3

19

79

-198

3

19

13

-191

7

19

28

-193

2

19

18

-192

2

19

38

-194

2

19

63

-196

8

19

69

-197

3

19

58

-196

2

19

53

-195

7

19

33

-193

7

19

43

-194

7

19

48

-195

2

-.05

0

.05

.1

Pre

dic

ted C

ohort

effect

0 5 10 15 20Rank

20

Graph 7

Model 3 starts looking at variations in the way households respond to economic context, by

introducing explanatory variables at the period level. Since the 1980s, Mexico’s economic

development has been characterized by growth with instability and persistently high levels of

inequality; even if Mexico has increase its average per capita income 22% between 1990 and 2010

in real terms, it remains one of the most unequal countries in the world. Across different

specifications, I tested whether GDP per capita or GDP growth rates had an impact on household

energy consumption; none of these variables were significant and the model fit did not improve

[results not showed]. In contrast, Income share held by highest 20 percentile, an indicator income

distribution inequality, did have a sizable impact: the higher the inequality, the larger the energy

consumption. Although it is not fully clear the mechanism that will produce this result, there are a

couple of arguments that could be explored. The 20-top share variable may be an indicator of a

1992

2008

2002

2000

200

4

1996

199

8

200

6

-1-.

50

.51

Pre

dic

ted p

eri

od e

ffect

0 2 4 6 8Rank

21

growth pattern, signaling not only wealth distribution but also indicating a context where high

consumption is an indication of social status given the concentration of resources. Also, it is worth

notice that other studies have found that with high income inequality there is higher and more

concentrated energy consumption (see Jacoboson et al. 2005). Despite the significance of this

period variable, the overall fit of the model did not improve suggesting that it may not contribute

importantly to explain household energy consumption.

Finally, model 5 tests the hypothesis that cohorts experience period effects differently. In

order to do I add a random cohort-by-period interaction classification (see Leckei 2013). In

previous models, cohort and periods have additive effects, but here I consider the possibility that

the period effect depends on cohort membership. For this model, I dropped inequality period

variable, since it did not add significantly to the model fit. The significance of the random

interaction supports this idea; thus, how much historical times impact household energy

consumption vary by cohort.

Final remarks

Given the centrality of greenhouse energy consumption, particularly CO2, we need to better

understand its determinants and temporal dynamics. An age-period- cohort analysis can provide

an insightful look on the way household life course and their social and historical context influence

their energy practices. This paper shows that there are significant differences across cohorts and

periods in household energy consumption. To a great extent those differences respond to changes

in the socio-demographic profile of households over time. Cohorts compositional differences are

particularly relevant to explain patterns of energy consumption, but there is also a detectable

cohort “pure” effect for some cohorts that came to age during times of economic expansion: these

22

cohorts add to the mean energy consumption, above and beyond other characteristics. This

suggests lasting energy practices. In addition, neither the exploratory analysis nor the model runs

indicate that younger cohorts have more intensive energy patterns than older ones. Further work

is necessary to understand better why this may be the case, particularly in relation to energy

efficiency practices.

The results also suggest that historical contexts shape energy use; changes over time do

not solely respond to changes in population composition, but also on shifting social and economic

context. The analysis points to the need of examining carefully this link. At the same time, the

results suggest that cohorts respond differently to historical moments, suggesting the need to

consider both dimensions to understand consumption and develop public policies to counteract it.

Overall, an APC analysis provides valuable insights for public policies. It allows advance the

knowledge of how households form their environmental practices and perceptions. On one hand,

“pure” cohort effects could provide an explanation for energy practices perdurability, despite

multiple public attempts to decrease energy consumption: formative experiences could imprint

energy use practices in lasting ways. On the other hand, based on the Mexican data, cohort

replacement may contribute to reduce household energy consumption in the long run. At the

same time, the life course plays a role that need to be addressed to design better policies while

understanding the socioeconomic contexts households inhabit.

References

Barber, J. 2003, "Production, Consumption and the World Submit on Sustainable Development", Environment, Development and Sustainability, vol. 5, pp. 63-93.

Bernard, G. 1995, "World Population, Economic Growth, and Energy Demand, 1990-2100: A Review of Projections", Population and Development Review, vol. 21, no. 3, pp. 507-539.

23

Brounen, Dirk, Nils Kok, and John J. M. Quigley. 2012. "Residential Energy Use and

Conservation: Economics and Demographics." European Economic Review 56:931-945.

CAIT 2010. Climate Analysis Indicators Tools. Washington, D.C. World Resource Institute. On line:cait.wri.org/cait.php, acceded between January and March 2010.

Cramer, J. 1998, "Population Growth and Air Quality in California", Demography, vol. 35, pp. 45-56.

Curran, S. & de Sherbinin, A. 2004, "Completing the Picture: The Challenges of Bringing "Consumption" into the Population- Environment Equation", Population and Environment, vol. 26, no. 2, pp. 107-131.

de Sherbinin, A. & Curran, S. 2004, Completing the Picture: the Challenges of Bringing

"consumption" into the Population-Environment Equation. Essay Prepared for Consideration

by PERN Cyberseminar 17-31 May 2004, published online,

www.populationenvironmentresearch.org/seminar.jsp.

de Sherbinin, A., Carr, D., Cassels, S. & Jiang, L. 2007, "Population and Environment", Annual Review of Environment and Resources, vol. 32, pp. 345-347.

Esquivel, G. 2009, The Dynamics of Income Inequality in Mexico since NAFTA, United Nations Development Programme, Regional Bureau for Latin America and the Caribbean.

Gram-Hanssen, K. 2005, "Domestic electricity consumption -consumers and appliances" in The Ecological Economics of Consumption, eds. L. Reisch & I. Ropke, First ed. Edward Elgar Publisher, Cheltenham, UK ; Northampton, MA. pp. 132

Greening, L. & Jeng, T. 1994, "Lifecycle analysis of gasoline expenditures patterns.", Energy Economics, vol. 16, no. 3, pp. 217-228.

Guertin, C., Kumbhakar, S. & Duraiappah, A. 2003, Determining Demand for Energy Services: Investigating income driven behaviors, International Institute for Sustainable Development.

Hernández, F., Daniel & Pérez, García, María, 2003, En el año 2000, Gasto de los hogares y pobreza en México, SEDESOL.

Irongmonger, D., Aiteken, C. & Erbas, B. 1995, "Economics of Scale in energy use in adult-only households", Energy Economics, vol. 17, pp. 301-310.

Jensen, J. 2008, "Measuring consumption in households: interpretations and strategies", Ecological Economics, vol. 68, pp. 353-361.

Jiang, L & O'Neill, B. 2004 "The energy transition in rural China", International Journal Global Energy Issues, vol. 21, no. 1/2, pp. 3-26.

Jiang, L. & O'Neill, B. 2004, "The energy transition in rural China", International Journal Global Energy Issues, vol. 21, no. 1/2, pp. 2-26.

Jiang, L. 1969, "Urbanization, energy consumption and climate change", The 1967 Report on the World on the World Social Situation, pp. 131-133.

Joachim, S. 2001, "The Environmental Kuznets Curve: A Methodological Artefact?", Population and Environment, vol. 23, no. 2, pp. 175-191.

24

Leckie, G (2013) Module 12: Cross-Classified Multilevel Models, Centre for Multilevel Modelling, University of Bristol

Lenzen, M., Wier, M., Cohen, C., Hayami, H., Pachauri, S. & Schaeffer, R. 2006, "A comparative multivariate analysis of household energy requirements in Australia, Brazil, Denmark, India and Japan", Energy, vol. 31, pp. 181-207.

Liddle, B. 2004, "Demographic Dynamics and Per Capita Environmental Impact: Using Panel Regressions and Household Decompositions to Examine Population and Transport", Population and Environment, vol. 26, no. 1, pp. 23-39.

Liu, J., Wang, R., Yang, J. 2009, "Environment consumption patterns of Chinese urban households and their policy implications", International Journal of Sustainable Development & World Ecology , vol. 16, no. 1, pp. 9-14.

Loren Lutzenhiser & Bruce Hackett 1993, "Social Stratification and Environmental Degradation: Understanding Household CO2 Production", Social Problems, vol. 40, no. 1, pp. 50-73.

MacKeller, F., Lutz, W., Prinz, C. & Guajon, A. 1995, "Population, households and CO2 emissions", Population Development Review, vol. 21, pp. 849-865.

Menz Tobias & Jan Kühling, (2011) “Population aging and environmental quality in OECD countries: evidence from sulfur dioxide emissions data”, Population & Environment, Vol.33, 1, pp.55-79

Michael, D., Brian, O. & Katarína, Z. 2008, Effects of Household Age and Size on the Elasticity of Energy Consumption, International Institute for Applied Systems Analysis.

O'Neill Brian & Chen Belinda 2002, "Demographic Determinants of Household Energy Use in the

United States" in Population and Environment. Methods of Analysis. A Supplement to Vol. 22 of Population and Development Review, eds. Lutz Wolfgang, Prskawetz Alexia & Sandersin Warren, First ed., Population Council, New York, pp. 53.

O'Neill, B., MacKeller, F. & Lutz, W. 2001, Population and Climate Change, Cambridge University Press, New York.

Pachuri, S. 2004, "An analysis of cross-sectional variation in total household energy requirements in India using micro survey data", Energy Policy, vol. 32, pp. 1732-1735.

Prskawetz, A., Leiwen, J. & O'Neill, B. 2002, Demographic Composition and Projections of Car Use in Austria, Working paper 2002-034, Germany.

Pucher, J., Evans, T. & Wenger, J. 1998, "Socioeconomics of urban travel: evidence from the 1995 NPTS", Transportation Quarterly, vol. 52, no. 3, pp. 15-33.

Ropke, I. 2005, "Consumption in ecological Economics", Internet Encyclopedia of Ecological Economics, Published online, December 2008.

Ropke, I., Reisch, L. 2005, "The place of consumption in ecological economics" in The Ecological Economics of Consumption. eds. L. Reisch & I. Ropke, First ed., Edward Elgar Publishing, Cheltenham, UK ; Northampton, MA., pp. 1.

Sánchez, L and Jasso. 2012. “Un análisis edad-cohorte-periodo del consumo energético de los hogares urbanos en México”. Ponencia presentada en el V Congreso de la Asociación Latinoamericana de Población, Montevideo, Uruguay, del 23 al 26 de octubre de 2012.

25

Sánchez, Landy .2010. "Inequality and household energy consumption un urban Mexico"

Ponencia presentada en la Reunión de Population American Association Dallas, Estados Unidos, 1 de abril de 2010

Sánchez, Landy. 2012. “El consume energético de los hogares en México” en Coyuntura Demográfica No.2, Sociedad Mexicana de Demografía, México

Sari, R.and, Soytas, U. 2007, "The growth of income and energy consumption in six developing countries", Energy Policy, vol. 35, pp. 889-898.

Satterthwaite, D. 2009, "The implications of population growth and urbanization for climate

change" in Population dynamics and climate change, eds. J.M. Guzmán, G. Martine, G. McGranahan, D. Schensul & C. Tacoli, UNFPA, IIED, , pp. 45-63.

Scott, J. 2009, Redistributive Constraints under High Inequality: The Case of Mexico, United Nations Development Programme Regional Bureau for Latin America and the Caribbean ed.

SENER 2009, Balance Nacional de Energía 2008, SENER.

Sheinbaum, C. Martínez, D & Rodríguez, L (1996) “Trends and Prospects in Mexican residential Energy”, Energy, Vol.21, No.6, pp. 493-504

Sheinbaum, P. Claudia 2000, Metodología para la elaboración del inventario de gases de efecto invernadero asociado al consumo de energía en México, Instituto Nacional de Ecología.

van den Bergh, J. 2008, "Environmental regulation of households: An empirical review of economic and psychological factors", Ecological Economics, vol. 66, pp. 559-574.

Waldkirch, A., Ng, S. & Cox, D. 2004, "Intergenerational Linkages in Consumption Behavior", The Journal of Human Resources, vol. 39, no. 2, pp. 355-381.

Wilks, R. 2005, "Questionable Assumptions about Sustainable Consumptions" in The Ecological Economics of Consumption, eds. L. Reisch & I. Ropke, First eedn Edward Elgar Publisher, Cheltenham, UK ; Northampton, MA.

Yan, W. & Minjun, S. 2009, "CO2 Emission Induced by Urban Household Consumption in China", Chinese Journal of Population, Resources and Environment, vol. 7, no. 3, pp. 11-19.

Yang, Yang and Land, K 2006. "A Mixed Models Approach to the Age-Period-Cohort Analysis of Repeated Cross-Section Surveys, with an Application to Data on Trends in Verbal Test

Scores." Sociological Methodology 36:75-97. Yang, Yang and Land, K. 2008. “Age-Period-Cohort Analysis of Repeated Cross-Section Surveys:

Fixed or Random Effects?” Sociological Methods and Research 36 (special issue): 297-326.

Yang, Yang. 2008a. “Social Inequalities in Happiness in the U.S. 1972-2004: An Age-Period-Cohort Analysis.” American Sociological Review 73: 204-226.

Yang, Yang. 2008b. “Trends in U.S. Adult Chronic Disease Mortality: Age, Period, and Cohort Variations.” Demography 45: 387-416.

26

Appendix

Table A. Cross-classified nested structure of pooled data

27

Table B. Models results

Random effects income

Fixed Effects Coefficient

standard

error Coefficient

standard

error Coefficient

standard

error Coefficient

standard

error Coefficient

standard

error

Intercept 7.662 0.08 *** 7.664 0.04 *** 7.691 0.16 *** 7.602 0.15 *** 7.686 0.16 ***

Age 0.370 0.04 *** 0.401 0.01 *** 0.401 0.01 *** 0.394 0.01 ***

Per capita Income 0.079 0.01 *** 0.079 0.01 *** 0.076 0.01 ***

Per capita expenditure 0.566 0.01 *** 0.566 0.01 *** 0.570 0.01 ***

Household size 1 (Ref)

Household size 2 -0.216 0.09 ** -0.216 0.09 ** -0.217 0.09 **

Household size 3 -0.239 0.09 ** -0.239 0.09 ** -0.239 0.09 **

Household size 4 -0.286 0.09 *** -0.286 0.09 *** -0.284 0.09 ***

Household size 5 -0.348 0.09 *** -0.348 0.09 *** -0.344 0.09 ***

Household size 6 -0.419 0.09 *** -0.419 0.09 *** -0.414 0.09 ***

Household size 7+ -0.518 0.09 *** -0.518 0.09 *** -0.512 0.09 ***

Rooms numbers 0.050 0.00 *** 0.050 0.00 *** ***

No education (ref)

Elementary education 0.119 0.01 *** 0.119 0.01 *** 0.118 0.01 ***

at least some high school 0.154 0.01 *** 0.154 0.01 *** 0.153 0.01 ***

Higher education 0.103 0.01 *** 0.103 0.01 *** 0.104 0.01 ***

Female headed Household 0.064 0.01 *** 0.064 0.01 *** 0.064 0.01 ***

Single households (ref)

Couples withouth children 0.282 0.09 *** 0.282 0.09 *** 0.283 0.09 ***

Couples with children 0.194 0.09 * 0.194 0.09 * 0.197 0.09 *

single parents households with children 0.133 0.09 0.133 0.09 0.137 0.09

Extended households 0.186 0.09 * 0.186 0.09 * 0.185 0.09 *

Composed Households 0.250 0.09 ** 0.250 0.09 ** 0.250 0.09 **

West (ref)

Northeast 0.224 0.01 *** 0.224 0.01 *** 0.225 0.01 ***

Northwest 0.390 0.01 *** 0.390 0.01 *** 0.390 0.01 ***

Center -0.070 0.01 *** -0.070 0.01 *** -0.070 0.01 ***

South -0.025 0.01 ** -0.025 0.01 ** -0.024 0.01 **

Share of Income Top 20% 0.158 0.08 *

Random effects (reml estimation)variance

estimate

Standard

error

variance

estimate

Standard

error

variance

estimate

Standard

error

variance

estimate

Standard

error

Cohort 0.074 0.03 0.009 0.01 0.0002 0.000 0.0002 0.00 0.0000 0.00

Periodo 0.022 0.01 0.013 0.01 0.209 0.112 0.1524 0.09 0.2106 0.112759

Individual residual 1.618 0.01 1.619 0.01 0.900 0.004 0.8996 0.00 0.0019 0.00

Cohort * period 0.8984 0.00

Log likelihood -161245.4 -161239.4 -131990.1 -131990.1 -131967.1

df 4 5 27 28 28

BIC 322536.700 322536.2 264290.1 264301.5 264255.5

Random Cohort-

Period interaction

Cross-classified Hierarchical APC Model

M5M4

Unconditional Model Age-period-cohort pure model Household covariates

M1 M2 M3

28

Table C Descriptive statistics

mean Sd

Age 45.835 15.244

Per capita Income 7198.030 22127.9

Per capita expenditure 8707.038 15994.1

Household size 1 (Ref) 0.077 0.267

Household size 2 0.138 0.345

Household size 3 0.181 0.385

Household size 4 0.236 0.425

Household size 5 0.179 0.384

Household size 6 0.089 0.285

Household size 7+ 0.098 0.297

Rooms numbers 3.463 1.726

No education (ref) 0.089 0.285

Elementary education 0.389 0.487

at least some high school 0.239 0.426

Higher education 0.278 0.448

Female headed Household 0.217 0.412

Single households (ref) 0.076 0.265

Couples withouth children 0.075 0.263

Couples with children 0.502 0.500

single parents households with children 0.100 0.300

Extended households 0.226 0.418

Composed Households 0.021 0.142

West (ref) 0.209 0.406

Northeast 0.164 0.370

Northwest 0.088 0.283

29

Center 0.335 0.472

South 0.179 0.384

Income share Top 20% 54.887 1.242