CHEMICAL ENGINEERING TRANSACTIONS
VOL. 76, 2019
A publication of
The Italian Association
of Chemical Engineering
Online at www.aidic.it/cet
Guest Editors: Petar S. Varbanov, Timothy G. Walmsley, Jiří J. Klemeš, Panos Seferlis
Copyright © 2019, AIDIC Servizi S.r.l.
ISBN 978-88-95608-73-0; ISSN 2283-9216
On the Relationship between Private Transportation
Expenditure and Socio-Demographic Variables
Giorgio Besagni*, Corine Nsangwe Businge, Marco Borgarello
Ricerca sul Sistema Energetico - RSE S.p.A., Power System Development Department, via Rubattino 54, 20134 Milano,
Italy
giorgio.besagni@rse-web.it
A precise understanding of the relationships between household characteristics and transportation expenditures
is of paramount importance to support bottom-up policies. Despite the considerable amount of research
activities carried out in the last decades, an agreement regarding the determinants of the transportation
expenditure is far from being achieved. This paper contributes to the existing discussion, focusing on the Italian
case study, by analysing the relationships between the private transportation expenditure and the socio-
demographic variables. In particular, the impact that the different household characteristics have on the private
transportation expenditures have been examined. The analysis is performed by coupling three statistical
methods: the ordinary least squares method, the variance inflation factor, the least absolute shrinkage and
selection operator. It is found that the geographic area (in terms of the macro-scale and the micro-scale
geographic locations) as well as income-related variables are likely to be determinants for the private
transportation expenditure.
1. Introduction
The so-called “sustainable mobility paradigm”, as defined by Banister (2008), is a cutting-edge research topic
and a priority in the international agenda. Owing to the large share of primary energy consumption and emissions
related to the transportation sector, a considerable amount of research activities is conducted to support
pathways towards emission reduction and “decarbonisation” (viz., reduction in greenhouse gas emissions). Out
of the many studies, Van Fan et al. (2018) discussed sustainable alternatives for freight transportation and
Promjittiphong et al. (2018) analyses the energy consumption of the tourist-sector. To reach these targets,
effective policies are needed: their success relies on the precise understanding of the relationships between the
household characteristics and the transportation expenditure/fuel consumption. The study of these relationships
belongs to the field of the so-called “human dimension” of the energy-intensity in transportation along with the
subsequent “energy metabolism” (Stephenson et al, 2015). This concept was addressed by Tian et al. (2016),
who pointed out that the energy consumption at the “household-scale” determines the carbon emission at the
“country-scale”. In this perspective, Longhi (2015) mentioned that the precise understanding of the determinants
of the energy consumption/expenditure serves as basis for policymakers when planning investments aiming to
reduce the energy consumption at the bottom-level (Longhi, 2015). This paper contributes to the present-day
discussion, by investigating the transportation sector and the existing relationships between the “private
transport expenditure” and both the “socio-demographic dimension” (i.e., household composition, income of the
household, qualification of the occupants…) and the “geographic dimension” (i.e., the macro-scale and the
micro-scale geographic locations). This context is interesting if considering the demographic shift experienced
by European countries. As the population is progressively ageing, this may reflect in the energy metabolism at
the different levels: for example, both Liddle (2014) and Brand et al. (2013) mentioned that the ageing of
population is likely to increase the residential energy consumption and to reduce the transport-related energy
use. This topic was further investigated by Bardazzi and Pazienza (2018), who studied how the changes in the
“socio-demographic dimension” and in the economic drivers would affect the private transport-related fuel
demand in Italy. Unfortunately, despite the considerable amount of research activities carried out in the last
decades, an agreement is far from being achieved on both the methodological point of view and the relationships
1075
DOI: 10.3303/CET1976180
Paper Received: 22/01/2019; Revised: 28/03/2019; Accepted: 28/03/2019
Please cite this article as: Besagni G., Businge C.N., Borgarello M., 2019, On the Relationship between Private Transportation Expenditure and
Socio-Demographic Variables, Chemical Engineering Transactions, 76, 1075-1080 DOI:10.3303/CET1976180
between the transportation expenditure and the many variables describing households. This paper contributes
to the existing discussion, focusing on the Italian case study, by analysing the relationships between the private
transportation expenditure and the household variables. In particular, the impact that the different household
characteristics have on the transportation expenditures have been examined, based on the microdata taken
from the Italian Household Budget Survey published by the Italian Statistical Office. The analysis is performed
by coupling three methods: the ordinary least squares method, the variance inflation factor, the least absolute
shrinkage and selection operator, to select relevant variable. It is found that the geographic area as well as
income-related variables are likely to be determinants for the private and public transportation expenditures.
The paper is organized as follows. Section 2 discusses the dataset and the statistical methods; Section 3
describes the results of the statistical procedure and, finally, Section 4 contains our conclusions.
2. Research design and methods
In this section, the dataset (Section 2.1), the dependant variable (Section 2.2), the predictors (Section 2.3) and
the statistical methods (Section 2.4) are discussed and commented.
2.1 The dataset
The dataset used in this study is the “Household Budget Survey: microdata for research purposes” (reference
year: 2015, ISTAT (2017)), which is representative of the whole Italian population and was obtained by Italian
National Institute of Statistics. The micro-data were collected from 15,015 households, in 502 different
municipalities; for each household more than 1264 variables are available, regarding socio-demographic
information, dwelling characteristics, appliances and monthly expenditures.
2.2 The dependent variables
The dependent variable “private transport expenditure” (coded as ϑ, in the followings) was obtained by summing
different monthly expenditures, to describe the different patterns of the “private transportation”: (a) gasoline
expenditure, (b) diesel expenditure, (c) “other fuel” expenditure, (d) timetable parking expenditure, (e) motorway
tolls, (f) tire expenditure, (g) vehicle spare part expenditure, (h) car accessory expenditure, (i) lubricant
expenditure, (j) maintenance/repair service expenditure, (k) car/motor expenditure. As mentioned above, the
data are available in the dataset in terms of monthly expenditures and refer to a precise month during the year.
Before going ahead with the analysis, it should be verified whether an annual calibration is needed or not
(Besagni and Borgarello, 2018). To this end, Figure 1 displays the relationship between the time variable (month)
and the “private transport expenditure”: the dependent variable does not exhibit high dependency with respect
to the time variable. Hence, an annual calibration is not needed. As the dependent variable has left-skewed
distribution, it is implemented as log-transformed. Another advantage of using logs is that the regression
coefficients refer to the relative changes rather than the absolute changes in per-capita energy expenditures.
This concept was described by Longhi (2015) and by Besagni and Borgarello (2018). Households having
“private transport expenditure” equal to zero were excluded from the analysis.
Figure 1: Relationship between time of the year and “private transport expenditure
150
200
250
300
350
400
1 2 3 4 5 6 7 8 9 10 11 12
P
ri
v
a
te
t
ra
n
sp
o
rt
e
x
p
e
n
d
it
u
re
[
€
]
Month [-]
North-west Norst-east Centre South
Sicily Sardinia Italy
1076
2.3 The predictors
Table 1 lists the variables used as predictors in the regression analysis, with their frequencies and summary
statistics (viz., mean and standard deviation for the continuous variables - the category printed in bold indicates
the reference category). In Table 1, HRP represents the Household Representative Person, which is the
individual that is taken to represent the household. In this study, HRP describes the highest income earner in
the household Summary statistics evaluated on the whole data-set.
2.4 The statistical methods
The statistical approach couples three methods: (a) the ordinary least squares method (OLS), (b) the variance
inflator factor (VIF), (c) the Least absolute shrinkage and selection operator (LASSO). A similar procedure has
been implemented by Besagni and Borgarello (2018). The procedure con consists of four phases, as discussed
in the following:
• Phase#a. OLS relates the dependent variable with the predictors listed in Table 1 as in Eq(1):
y
i
= ln (ϑ) = β
0
+β
1
xi1+β2xi2+… + βkxik+ εi = β0+ ∑ βjxij
w
j=1
+ εi (1)
In Eq. (1), yi is the logarithm of the dependent variable for the i-household; xij is the i-predictor for the j-
household out of w-predictors; β0 is the constant term (viz. the intercept); βj is the j-coefficients for the xij
variable; εi is the error having null mean and constant variance. The model fit is evaluated on the well-
known adjusted coefficient of determination (Radj
2
). After the OLS analysis, owing to possibility of
multicollinearity issues, the variance-inflation factors (VIFj) have been inspected for every βj:
VIFj=
1
1-Rj
2
(2)
If the predictors are uncorrelated, VIF = 1; conversely, if VIF > 1, a possible correlation exist. At present,
there is no agreement on the cut-off point for critical values of VIF (VIFmax). In this study, VIFmax = 3, has
been selected, accordingly with Besagni and Borgarello (2018).
• Phase#b. If multicollinearity has been detected in phase#a, LASSO regression, a variable shrinkage based
on a penalty, is applied to select relevant predictors by using a constrained optimization problem Once the
significant variables have been selected, the OLS is repeated and another Radj
2
is obtained; subsequently,
VIF are inspected again.
• Phase#c. If multicollinearity is detected after LASSO procedure, a progressive selection of predictors is
conducted, by excluding, one by one, the least significant predictors and, at each step, the changes in Radj
2
are inspected and VIF is monitored (a reduction in Radj
2
is considered acceptance only if within 0.5 % and,
in this phase VIFmax = 10).
• Phase#d. If no multicollinearity has been detected after LASSO, a selection of the predictors is obtained
thorough the recursive procedure described in phase#c (a reduction of Radj
2
is considered acceptable if
within 0.2 %).
The interested reader may refer to our previous study for additional details regarding the employed methods
and modelling approaches (Besagni and Borgarello, 2018).
3. Results and discussions
As stated in the introduction, the main goal of this section is to assess the “socio-demographic and geographical
dimensions” of “private transport expenditure” (ϑ); in particular this section tries to answer to the following
question: is the “private transport expenditure” mainly related to the socio-demographic aspects or to the
geographic ones? In order to answer this question, Table 2 displays the final regression model (F(22, 12530 =
222.8, p < 2.2e-16), Radj
2
=28.99 %). In Table 2, for every variable, the values of the coefficients in Eq(1), the
standard error, the t-test value, the p-value (indicated by Pr(>|t|)) and the level of significance are presented;
the first row displays the value of the intercept, β̂0, whereas in the subsequent rows, the other coefficients, β̂j,
are listed (Eq. (1)). When interpreting these results, it should be noted that, as log-transformed dependent
variable are used, interpreting the value of the coefficients is quite straightforward: if we change a certain
coefficient (i.e., β1) by unit, we would expect ϑ to change by 100⋅ β1 percent. Table 2 also presents the results
of the VIF analysis: all the predictors are characterized by VIF < 3, thus suggesting that the OLS-VIF-LASSO
procedure eliminated multicollinearity issues.
1077
Table 1: Socio-demographic variables with their frequencies. (* = reference category).
Variable Summary statistics
Sex of the HRP (a) Male [10,193]*, (b) Female [4,820]
Current economic resources (a) Optimal [279], (b) Adequate [7,912]*, (c) Scarce [5,651], (d) Insufficient
[1,171]
Changes in economic
resources compared to the
previous year
(a) Much improved [30], (b) A little bit improved [512], (c) More or less the same
[8,488]*, (d) A little worsened [4,626], (e) Much worsened [1,357]
Absolute poverty (a) Yes [834], (b) No [14,179]*
Birth place of the household
components
(a) Only born in Italy [13,456]*, (b) At least one born abroad [973], (c) Only born
abroad [584]
Citizenship of the household
components
(a) Only Italian citizens [14,176]*, (b) At least one foreign citizen [257], (c) Only
foreign citizens [580]
Marital status of the HRP (a) Unmarried [2,551], (b) Married or cohabitant [8,252]*, (c) Married but not
cohabitant [355], (d) Legally separated [625], (e) ) Divorced [698], (f) Widow or
widower [2,532]
Qualification of the occupants (a) No member has a qualification [377], (b) At least one member with
elementary school [1,978], (c) At least one member with junior high school
[3108], (d) At least one member with high school [6,483]*, (e) At least one
member with a degree [3,067]
Work contract of the
occupants
(a) There is neither temporary job nor permanent job [7,536]*, (b) At least one
temporary job [1,125], (c) At least one permanent job [6,352]
Source of income of the
occupants
(a) There is no income [83], (b) At least one maintained [413], (c) At least one
pension [4,911], (d) At least one income [9,606]*
Enrolment in study courses (a) No members enrolled in a course [10,930]*, (b) At least one in no title school
[419], (c At least one in elementary school [747], (d) At least one in junior high
school [584], (e) At least one in high school [1,244], (f) At least one in a degree
or post-degree course [1,089]
Expenditure for elderly or
disabled people
(a) Yes [100], (b) No [14,913]*
Household structure (a) Single person 18-34 years [391], (b) Single person 35-64 years [1,817], (c)
Single person 65 years and more [2,240], (d) Couple without children with HRP
18-34 years [178], (e) Couple without children with HRP 35-64 years [1,350], (f)
Couple without children with HRP 65 years and more [2,164], (g) Couple with 1
child [2,276]*, (h) Couple with 2 children [2,184], (i) Couple with 3 children or
more [495], (l) Mono parent family [1033], (m) Others [885]
Number of workers in the
primary sector
(a) No one [13,622]*, (b) One [1,100], (c) More than one [291]
Number of workers in the
secondary sector
(a) No one [9,766]*, (b) One [4,098], (c) More than one [1,149]
Number of workers in the
tertiary sector
(a) No one [4,577], (b) One [6,195]*, (c) More than one [4,241]
Number of managers and
employees
(a) No one [8,227]*, (b) One [4,739], (c) More than one [2,047]
Workers and similar (a) No one [8,166]*, (b) One [4,741], (c) More than one [2,106]
Entrepreneurs and freelancer
workers
(a) No one [13,696]*, (b) One [1,172], (c) More than one [145]
Self-employed workers (a) No one [11,876]*, (b) One [2,583], (c) More than one [554]
Age of the HRP (a) Up to 34 years [995], (b) From 25 to 44 years [2,343], (c) From 45 to 54
years [3,059]*, (d) From 55 to 64 years [2,934], (e) From 65 to 74 years [2841],
(f) From 75 years [2,841]
Type of municipalities (a) Centre of metropolitan area [1,889], (b) Periphery of metropolitan area and
municipalities with 50.001 inhabitants and more [4,032], (c) Other municipalities
until 50.000 inhabitants [9,092]*
Geographic location (a) North-west [3,284], (b) North-east [3,382], (c) Centre [2,791]*, (d) South
[4,385], (e) Sicily [753], (f) Sardinia [418]
Number of cars (a) no one [2,761]*, (b) one [7,324], (c) two [4,226], (d) three or more [702]
Free time expenditures Continuous variable [Mean = 22.68 / Variance = 3027]
1078
Table 2: Details of the final regression model (Significance codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1) –
Code names of the variables in Table 1.
Estimate Std. Error t value Pr(>|t|) Significance VIF
(Intercept) - 𝛽0 4.7258 0.0395 119.7480 < 2e-16 ***
1.07
Sex of the HRP (b) -0.1577 0.0151 -10.4640 < 2e-16 ***
Absolute poverty (a) -0.7241 0.0342 -21.1710 < 2e-16 *** 1.04
Qualification of the occupants (a) -0.3296 0.0862 -3.8230 0.0001 ***
1.65
Qualification of the occupants (b) -0.2251 0.0275 -8.1720 0.0000 ***
Qualification of the occupants (c) -0.1099 0.0177 -6.2000 0.0000 ***
Qualification of the occupants (e) 0.0196 0.0168 1.1660 0.2435
Source of income of the occupants (a) -0.3752 0.0918 -4.0870 0.0000 ***
1.39 Source of income of the occupants (b) -0.2403 0.0478 -5.0310 0.0000 ***
Source of income of the occupants (c) -0.2437 0.0167 -14.5960 < 2e-16 ***
Workers and similar (b) -0.0182 0.0152 -1.1960 0.2317
1.22
Workers and similar (c) 0.0684 0.0196 3.4940 0.0005 ***
Type of municipalities (a) -0.0761 0.0210 -3.6280 0.0003 ***
1.11
Type of municipalities (b) -0.0359 0.0150 -2.4010 0.0164 *
Geographic location (a) 0.0835 0.0201 4.1490 0.0000 ***
1.12
Geographic location (b) -0.0913 0.0199 -4.5860 0.0000 ***
Geographic location (d) 0.0425 0.0192 2.2070 0.0273 *
Geographic location (e) 0.0429 0.0331 1.2970 0.1946
Geographic location (f) -0.0556 0.0405 -1.3730 0.1699
Number of cars (b) 0.5428 0.0344 15.7610 < 2e-16 ***
1.30 Number of cars (c) 0.9838 0.0363 27.1330 < 2e-16 ***
Number of cars (d) 1.2762 0.0445 28.6680 < 2e-16 ***
Free time expenditures 0.0017 0.0001 15.2000 < 2e-16 *** 1.11
The proposed model exhibits Radj
2
=28.99 %; this result suggests that the proposed model is able to explain quite
a small portion of ϑ variance; nevertheless, the reader should consider that Radj
2
represents the proportion of the
variance (of the dependent variable) explained by the selected predictors under the linear modeling approach
expressed in Eq(1). For this reason, a low value of Radj
2
, along with a high significance of the statistical model,
can imply that there is an high relationship between the dependent variable and the predictor, but the
dependence is non-linear and/or additional variables should be include. Looking at the results in Table 2, it is
observed that ϑ is related to the free time expenditures: an increase in the “free time expenditure” equal to 1 €
determines an increase ϑ equal to 0.17 %; this result is somehow expected, as higher “free time expenditures”
is likely to determine higher travel intensity. More importantly, it is observed that the geographic location is
significant both in terms of the macro-geographic location and in terms of the type of municipalities, which
support the geographical dimension of “private transport expenditure”. Concerning the influence of
municipalities, ϑ decreases when passing from small municipalities toward the center of metropolitan cities (-
7.61 %), possibly owing to the higher availability of public transport system in metropolitan cities; conversely,
when passing from small municipalities toward municipalities with more than 50.001 inhabitants (-3.59 %), but
less significant. Concerning the socio-demographic variables, it is found that the household structure is not a
significant variable, which is in disagreement with previous research activities concerning the relationship
between transportation fuel consumption and household variables (see the discussion proposed in the
introduction). The qualification of the occupants is a significant variable: the more the household is “qualified” in
terms of degree and instruction level, the higher is the “private transport expenditure”. Considering the previous
literature, Pachauri and Jiang (2008) found a relationship between the educational level and the energy
consumption. Similarly, households with poor incomes of with low number of sources of income, are more likely
to have lower “private transport expenditures”. It is also observed that, in the case the HRP person is female, ϑ
decreases by 15.77 %. As expected the absolute poverty condition results in a significant decreases of ϑ (-72.4
%), as this is a representation of the household income conditions. In summary, the proposed results, support
that both a “socio-demographic dimension” and a “geographical dimensions” of the “private transport
expenditure” exist, even if the “socio-demographic dimension” is determined by income-related variables rather
than the household-composition variables.
1079
4. Conclusions
Using a nationally representative sample of Italian households, this paper contributes to the existing discussion
regarding “human dimension” of the energy-intensity in transportation, to provide a rational basis to evaluate
the subsequent energy metabolism. In particular, this paper focuses on the Italian case studies and it evaluates
the “socio-demographic and geographical dimensions” of “private transport expenditure”. It is found that the
geographic location is significant both in terms of the macro-geographic location and in terms of the type of
municipalities, thus supporting the geographical dimension of “private transport expenditure”. On one hand, it is
found that the “socio-demographic dimension” is determined by income-related variables (which are also related
to the main occupation of the household components) rather than the household-composition variables; the
household structure is not a significant predictor, in disagreement with the previous literature. These results
might support, even if following a different path, the discussion of Bardazzi and Pazienza (2018): aging of the
population will result in a decrease of household fuel use in Italy in the long-run. This consideration may be of
key interest to policymakers when planning investments aiming to reduce the primary energy consumption at
the household level (in the view of pathways towards emission reduction and “decarbonisation”), by forecasting
the transportation energy consumption. The outcomes of this study are also of practical interest in terms of
environment policies, as they will provide a statistical basis to estimate the impact of “country-scale”
transportation policies at the household-scale. In particular, these results will help forecasting the practical
response of the household towards macro transportation policies.
Acknowledgments
This work has been financed by the Research Fund for the Italian Electrical System in compliance with the
Decree of Minister of Economic Development April 16, 2018.
References
Banister D., 2008, The sustainable mobility paradigm, Transport Policy, 15(2), 73-80.
Bardazzi R., Pazienza, M.G., 2018, Ageing and private transport fuel expenditure: Do generations matter?
Energy Policy, 117, 396-405.
Besagni G., Borgarello M., 2018, The determinants of residential energy expenditure in Italy, Energy, 165, 369-
386.
Brand C., Goodman A., Rutter H., Song Y., Ogilvie D., 2013, Associations of individual, household and
environmental characteristics with carbon dioxide emissions from motorised passenger travel, Applied.
Energy, 104, 158-169.
Fan Y.V., Klemes J.J., Perry S., Lee C.T., 2018, An Emissions Analysis for Environmentally Sustainable Freight
Transportation Modes: Distance and Capacity, Chemical Engineering Transactions, 70, 505-510.
ISTAT, 2017, Household Budget Survey: Microdata for Research Purposes (Reference Year: 2015); Rome,
Italy.
Liddle B., 2014, Impact of population, age structure, and urbanization on carbon emissions/energy consumption:
evidence from macro-level, cross-country analyses, Population and Environment, 35(3), 286-304.
Longhi S., 2015, Residential energy expenditures and the relevance of changes in household circumstances,
Energy Economics, 49, 440-450.
Pachauri S., Jiang L., 2008, The household energy transition in India and China, Energy Policy, 36, 4022-4035,
2008.
Promjittiphong C., Junead J., Hanpattanakit P., 2018, Greenhouse Gas Emission and Mitigation from Sports
Tourism in Benja Burapha Cycling Rally, Sa Kaeo, Thailand. Chemical Engineering Transactions, 63, 397-
402.
Stephenson J., Barton B., Carrington G., Doering A., Ford R., Hopkins D., Lawson R., McCarthy A., Rees D.,
Scott M., Thorsnes P., Walton S., Williams J., Wooliscroft B., 2015, The energy cultures framework:
exploring the role of norms, practices and material culture in shaping energy behaviour in New Zealand,
Energy Research and Social Science, 7 117-123.
Tian X., Geng Y., Dai H., Fujita T., Wu R., Liu Z., Masui T., Yang X., 2016, The effects of household consumption
pattern on regional development: a case study of Shanghai, Energy, 103, 49-60.
1080